109
Estimating healthselective migration in patients with systemic lupus erythematosus or Sjogren’s from administrative data Jeremy Labrecque Department of Epidemiology, Biostatistics and Occupational Health McGill University October 2012 A thesis submitted to McGill University in partial fulfillment of the requirements of the degree of Master of Science © Jeremy Labrecque 2012

Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  • Upload
    vudien

  • View
    222

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

               

Estimating  health-­selective  migration  in  patients  with  systemic  lupus  

erythematosus  or  Sjogren’s  from  administrative  data  

   

Jeremy  Labrecque  Department  of  Epidemiology,  Biostatistics  and  Occupational  Health  

McGill  University  October  2012  

                 

 A  thesis  submitted  to  McGill  University  in  partial  fulfillment  of  the  requirements  of  the  

degree  of  Master  of  Science        

           

©  Jeremy  Labrecque  2012  

Page 2: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  1  

Table  of  Contents    Table  of  Contents..............................................................................................................................1    List  of  figures  and  tables ................................................................................................................3    Abstract ...............................................................................................................................................5    Résumé ................................................................................................................................................7    Acknowledgements .........................................................................................................................9    1  Introduction................................................................................................................................. 11  1.1  Description  of  systemic  lupus  erythematosus  and  Sjogren’s.............................................14  1.2  Rationale..............................................................................................................................................15  1.3  Specific  Objectives ............................................................................................................................17  1.4  Thesis  structure.................................................................................................................................18  

 2  Literature  review....................................................................................................................... 19  2.1  Search  strategy ..................................................................................................................................20  2.2  Historical  research  (1871-­1970).................................................................................................20  2.3  Modern  research  (1970-­2011).....................................................................................................21  

2.3.1  Migration  as  the  outcome..................................................................................................22     2.3.2  Comparative  health  of  migrants  and  non-­migrants .................................................29     2.3.3  Surveys  on  reasons  for  migration ..................................................................................34  

2.3.4  How  selective  migration  can  affect  estimates ............................................................37  2.4  Summary  of  health  selective  migration ....................................................................................38  2.5  Administrative  databases ..............................................................................................................39  2.6  SLE,  Sjogren’s  and  health-­selective  migration ........................................................................40  

 3  Methodology................................................................................................................................ 43  3.1  Overview ..............................................................................................................................................43  3.2  Data  sources........................................................................................................................................44     3.2.1  Administrative  databases .................................................................................................44     3.2.2  Systemic  autoimmune  rheumatic  disease  cohort .....................................................46     3.2.3  Montreal  CMA  comparison  cohort .................................................................................47     3.2.4  Outcomes ................................................................................................................................49     3.2.5  Potential  confounders ........................................................................................................50  3.3  Statistical  methods ...........................................................................................................................51     3.3.1  Descriptive  and  bivariate  analyses................................................................................52     3.3.2  General  population  comparison .....................................................................................53     3.3.3  Pre-­diagnosis  comparison ................................................................................................55     3.3.4  Regional  migration  comparison .....................................................................................56     3.3.5  A  note  on  comparisons.......................................................................................................56  

Page 3: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 2  

 4  Results ........................................................................................................................................... 59  4.1  Descriptive  and  bivariate  analyses.............................................................................................59  4.2  Health-­selective  migration  comparisons..................................................................................67     4.2.1  General  population  comparison .....................................................................................69  

4.2.2  Pre-­diagnosis  comparison ................................................................................................72     4.2.3  Regional  migration  comparison .....................................................................................75  

 5  Discussion .................................................................................................................................... 79  5.1  How  the  results  compare  to  the  literature...............................................................................80  5.2  Strengths ..............................................................................................................................................82  5.3  Limitations ..........................................................................................................................................83  5.4  Overall  conclusions ..........................................................................................................................86  5.5  Future  research .................................................................................................................................87  

 Bibliography ................................................................................................................................... 89    Appendix  A  –  Forward  sortation  areas  and  ICD-­9  codes  used  in  this  thesis ............. 99    Appendix  B  –  Models  considered  in  analyses ....................................................................101    Appendix  C  –  Sample  WinBUGS  code ....................................................................................103    Appendix  D  –  Coefficients  from  all  models .........................................................................107                                          

Page 4: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  3  

List  of  figures  and  tables      Figures  Figure  1  Regional  migration  rates  among  SLE  and  Sjogren’s  patients  compared  to  provincial  estimates ......................................................................................................................... 63  Figure  2  Predicted  probabilities  of  migration  in  women  diagnosed  at  30  and  50  years  in  the  general  population  comparison  with  95%  CrI  error  bars .................................................. 72  Figure  3  Predicted  probabilities  of  migration  in  people  diagnosed  at  30  and  50  years  in  the  pre-­diagnosis  analysis  with  95%  CrI  error  bars.......................................................................... 74  Figure  4  Predicted  probabilities  of  migration  of  people  diagnosed  at  30  and  50  years  in  the  regional  migration  analysis  with  95%  CrI  error  bars. ............................................................... 77  

     Tables  Table  1  Summary  of  the  three  comparisons  performed ........................................................... 51  Table  2  Sample  sizes  for  each  analysis ........................................................................................ 59  Table  3  Number  of  between-­FSA  moves  and  crude  between-­FSA  migrations  rates  (in  moves  per  year  with  95%  CI  in  parentheses)  in  the  frequency-­matched  general  population  sample,  SLE  and  Sjogren’s  patients  by  disease,  disease  duration  and  age  in  the  general  population  comparison ...............................................................................................................................................................61  Table  4  Number  of  moves  and  crude  migrations  rates  (in  moves  per  year  with  95%  CI  in  parentheses)  by  disease,  disease  duration  and  age  in  A)  the  pre-­diagnosis  comparison  for  between-­FSA  moves  and  in  B)  the  regional  comparison  for  regional  moves .......................... 62  Table  5  Comparison  between  populations  in  the  full  Montreal  CMA  comparison  cohort  to  the  2001  Canadian  Census.  Proportion  of  the  population  found  in  each  age  group  is  found  in  parentheses  to  allow  comparison  of  age  distribution  by  sex ................................................ 64  Table  6  Descriptive  statistics  of  variable  in  the  general  population  and  pre-­diagnosis  data  sets....................................................................................................................................................... 65  Table  7  Coefficients  (and  95%  CrI)  from  bivariate  hierarchical  logistic  regression  analyses  in  the  healthy  control,  pre-­diagnosis  and  regional  migration  analyses .................................. 67  Table  8  Odds  ratios  (95%  CrI)  of  migration  relative  to  controls  by  age  and  disease  duration  strata   ................................................................................................................................. 68  Table  9  The  progression  of  odds  ratios  (95%  CrI)  relative  to  controls  for  people  diagnosed  at  30  and  50  years  old  for  all  three  comparisons ........................................................................ 70          

Page 5: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 4  

         

Page 6: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  5  

 Abstract    

Canadian   public   health   agencies   have   a   mandate   to   monitor   the   prevalence,  

incidence   and   patterns   of   chronic   disease.   These   agencies   are   increasingly   using  

administrative  health  data  for  these  purposes.  However,  valid  use  of  administrative  data  

for  chronic  disease  surveillance  requires  an  understanding  of  some  inherent  limitations.    

Health-­‐selective  migration,   which   occurs  when   people  migrate   differentially   by   health  

status,   is   a   limitation   that   has   not   been   estimated   in   administrative   data   sources.     To  

investigate   this   issue,   we   estimated   health-­‐selective  migration   in   a   cohort   of   systemic  

lupus  erythematosus  (SLE)  and  Sjogren’s  patients,  identified  from  physician  and  hospital  

claims  databases   in  Quebec   and   compared   them   to   rates   in   an   age   and   sex   frequency-­‐

matched   sample   from   Montreal,   Quebec   using   hierarchical   logistic   regression.   The  

association   between   disease   and   migration   was   modified   by   both   age   and   disease  

duration.  Both  SLE  and  Sjogren’s  patients  migrated   less  than  controls  when  young.  For  

example,  30-­‐year-­‐old  SLE    (OR  0.54,  95%  CrI  0.45-­‐0.64)  and  Sjogren’s  (OR  0.41,  95%  CrI  

0.28-­‐0.56)  patients  with   two  years  of  disease  duration  had   lower  odds  of  moving   than  

frequency-­‐matched  controls.  Above  age  50,   the  odds  of  migration   in  SLE  and  Sjogren’s  

patients  was  comparable  or  slightly  higher  than  in  controls.  Patients  at  age  70  with  two  

years  of  disease  duration  had  an  OR  of  moving  of  1.29  (95%  CrI  1.04-­‐1.58)   in  SLE  and  

1.09  (95%  CrI  0.81-­‐1.42)   in  Sjogren’s.  The  associations  between  migration  and  disease  

duration  in  SLE  and  Sjogren’s  were  qualitatively  different.  One  year  of  SLE  duration  was  

associated  with   an  OR  of   0.96   (95%  CrI   0.93-­‐0.98)   and  one   year   of   Sjogren’s   duration  

Page 7: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 6  

was  associated  with  an  OR  of  1.05  (95%  CrI  1.00-­‐1.10).  Results  were  similar  when  using  

SLE  and  Sjogren’s  patients  pre-­‐diagnosis  as  the  control  and  when  looking  at  migration  on  

a   regional   scale.    Overall,   SLE  and     Sjogren’s  have  an   impact  on  migration   rates  which  

varies  by  age,  disease  and  disease  duration.  

Page 8: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  7  

 Résumé    

Les   organismes   canadiens   de   santé   publique   ont   le   mandat   de   surveiller   la  

prévalence,  incidence,  et  les  tendances  des  maladies  chroniques  dans  notre  pays. De  plus  

en  plus,  ces  agences  utilisent  des  bases  de  données  administratives  sur  la  santé  à  ces  fins.  

Cependant,   l’utilisation   valable   de   ces   sources   de   données   pour   la   surveillance   des  

maladies   chroniques   exige   une   compréhension   de   certaines   limites   inhérentes,   en  

particulier   la  migration   sélective  par   l’état  de   santé,   ce  qui   se  produit   lorsque   les   gens  

migrent  de  façon  différente  du  à  leur  état  de  santé.  Nous  avons  effectué  une  évaluation  

des   déménagements   chez   des   patients   ayant   le   lupus   érythémateux   disséminé   ou   le  

syndrome  de  Sjogren’s  tel  qu’identifiés  dans  les  données  des  réclamations  des  médecins  

et   des   hôpitaux     du   Québec.   La   régression   logistique   hiérarchique   a   été   utilisée   pour  

comparer   les  taux  de  migration  chez  des  patients  atteints  de   lupus  ou  de  Sjogren’s  aux  

taux  de  déménagement  dans  un  échantillon  de   concordance  des   fréquences  par   âge   et  

par  sexe  de  la  population  de  Montréal.  

Dans  notre  échantillon,  l’association  entre  les  maladies  et  la  migration  a  démontré  

des  modifications  par  l’âge  du  patient  et  la  durée  de  sa  maladie.    Les  patients  lupiques  ou  

ayant  le  Sjogren’s  déménagent  moins  quand  ils  sont  plus  jeunes.  Par  exemple,  à  30  ans  et  

atteint   de   l’une   de   ces   maladies   depuis   deux   ans     les   patients   lupiques   (RC:   0.54,  

intervalle  crédible  (ICr)  95%  0.45-­‐0.6)  et  ceux  atteints    de  Sjogren’s  (RC:  0.41,  ICr  95%  

0.28-­‐0.56)   ont   des   cotes   de   déménagement  moins   élevés   que   ceux   du   groupe   témoin.    

Par   contre,   à   50   ans   et   plus,   les   cotes   de   migration   des   patients   lupiques   et   ayant   le  

Page 9: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 8  

Sjogren’s  sont  comparables  et  même  légèrement  supérieures  à  celles  du  groupe  témoin.    

Les   patients   âgées   de   70   ans   étant  malades   depuis   2   ans   ont   un   rapport   de   cotes   de  

déplacement   de   1,29   (ICr   95%:   1,04-­‐1,58)   pour   ceux   atteints   du   lupus   et   de   1,09   (ICr  

95%,  0,81-­‐1,42)  pour  ceux  ayant  le  Sjogren’s.  Aussi,  l’association  entre  la  migration  et  la  

durée   de   la   maladie   était   qualitativement   différente   entre   les   patients   lupiques   et   les  

patients  atteints  de  Sjogren’s.    Par  exemple,  les  patients  ayant  le  lupus  depuis  un  an  ont  

montrés   un   rapport   de   cotes   de   0,96   (ICr   95%,   0,93-­‐098)   et   les   patients   atteint   de  

Sjogren’s  ont  montrés  un  rapport  de  cotes  de  1,05  (ICr  95%  1,00-­‐1,10)  par  an  de  durée  

de  maladie.  

Les   résultats   étaient   similaires   lorsque   le   groupe   témoin   était   composé   de  

patients  qui  n’avaient  pas  encore  été  diagnostiqués  ou  lorsqu’on  regarde  les  migrations  à  

l’échelle   régionale.   Ce   mémoire   supporte   l’idée   que   les   maladies   chroniques   peuvent  

affecter   les   taux   de   migration   et   que   ceux-­‐ci   peuvent   varier   en   fonction   de   l’âge   du  

patient,  de  la  maladie  et  de  la  durée  de  cette  maladie.  

Page 10: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  9  

Acknowledgements  

First  and  foremost  I  would  like  to  thank  my  co-­‐supervisors:  Dr.  Sasha  Bernatsky  

and  Dr.  Lawrence  Joseph,  each  of  which  has  made  invaluable  contributions  to  my  

progress  as  a  student  of  epidemiology  and  to  my  research.  Dr.  Bernatsky  allowed  me  the  

independence  to  explore  my  own  ideas  while  always  being  a  source  of  inspiration,  new  

ideas  and  criticism.  Dr.  Joseph  shared  his  immense  wealth  of  knowledge  of  statistics  and  

methodology  as  well  as  invaluable  guidance  through  the  writing  of  my  thesis.  

Thank  you  to  my  committee  members,  Dr.  David  Buckeridge  and  Dr.  Nancy  Ross  

who  were  available  when  I  wanted  to  tap  their  expertise.  A  special  thank  you  to  Dr.  

Buckeridge  and  his  student  Aman  Verma  for  giving  me  access  to  data  and  taking  the  time  

to  teach  me  the  skills  needed  to  use  it.  

Thank  you  to  Jennifer  Lee  for  all  her  help,  administrative  and  otherwise,  from  the  

beginning  to  the  end  of  my  thesis.  

A  special  thanks  to  my  family  and  friends  for  all  of  their  support  and  

encouragement  through  the  ups  and  downs  of  thesis  writing.  Their  support  makes  this  

type  of  work  possible.  

Lastly,  this  work  would  not  have  been  possible  without  the  Graduate  Masters  

Training  Award  I  received  from  the  Canadian  Arthritis  Society.  

Page 11: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 10  

Page 12: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  11  

 

1  Introduction  

     

 

Health-­‐related   mobility   lies   at   the   crossroads   of   epidemiology   and   health  

geography,  aiming  to  describe  how  health  can  affect  movement  between  geographic  or  

socioeconomic  strata.  Health-­‐related  mobility  occurs  when  people  with  different  health  

statuses   move   differentially   between   socioeconomic   or   geographic   strata   (Fox et al.,

1982).  When  the  variable  in  question  is  geography,  this  phenomenon  can  be  referred  to  

more  specifically  as  health-­‐selective  migration.  Health-­‐selective  migration  encompasses  

both   overall   health   measures,   for   example   over-­‐all   quality   of   life,   and   more   specific  

health  states,  such  as  a  given  disability  or  diagnosis.  

Health-­‐selective   migration   is   important   to   epidemiology,   especially   when  

estimating  associations  between  disease  and  spatial  characteristics,  such  as  investigating  

possible  variations   in  disease  prevalence  over  different   regions   (Bentham, 1988; Lewis,

2003; O'Reilly & Stevenson, 2003; Ocaña-Riola et al., 2009; Patrick, 1980; Polissar, 1980;

Tiefelsdorf, 2007; Veugelers & Guernsey, 1999).     Whether   health-­‐selected   or   not,  

migration  can  cause  misclassification  or  measurement  error  of  residential  exposure  over  

time   (Polissar, 1980; Rogerson & Han, 2002; Tousignant et al., 1994),   possibly   creating  

biases  in  estimated  associations.        

Research  on  health-­‐selective  migration  has  included  diverse  populations  such  as  

the  elderly   (Flynn, 1980; Meyer & Speare, 1985),   cancer   survivors   (Haenszel & Dawson,

Page 13: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 12  

1965; Kliewer, 1992; Polissar, 1980),   people   with   long-­‐term   disabilities   (Longino et al.,

1991; Speare et al., 1991),  chronic  illnesses  (Larson et al., 2004; Yiannakoulias et al., 2007)  

and  poor  self-­‐rated  health  (Halliday & Kimmitt, 2008).  Despite  the  quantity  of  research,  

the   relationship  between  health   and  migration  often   remains  unclear   in   terms  of   both  

magnitude  and  direction.  There  are  number  of  factors  that  contribute  to  this.  

The   first   and   likely   most   important   factor   is   that   different   specific   health  

conditions   can   be   associated   with   either   increased   or   decreased  migration   (Barsby &

Cox, 1975; Nelson & Winter, 1975),   and   the  net  effect  across  a  given  population  can  be  

either  positive  or  negative   (Patrick, 1980).  Therefore,   lessons   learned   from  research  on  

one  set  of  health  conditions  may  not  necessarily  be  applicable  to  other  conditions.  Even  

within  one  condition,  estimates  from  one  region  may  not  apply  to  other  regions.  

Second,   the  relationship  between  health  and  migration  may  be  non-­‐linear  when  

health   is   measured   on   an   ordered   scale,   such   as   self-­‐rated   health,   or   when   using   a  

continuous   measure   such   as   number   of   disabilities   (Findley, 1988; Silverstein &

Zablotsky, 1996).  The  health-­‐migration  association  can  also  be  modified  by  factors  such  

as   age   (Kolcić & Polasek, 2009; Norman et al., 2005),   sex   (Bentham, 1988; Halliday &

Kimmitt, 2008)   and   disease   duration.   In   addition,   it   may   be   important   to   consider  

migration   in   terms   of   distance   migrated   (Bentham, 1988; Boyle et al., 2001).   Effect  

modification  by  age  is  particularly  important  given  that  a  large  portion  of  the  literature  

on  health-­‐selective  migration  focuses  on  the  elderly  making  it  difficult  or   impossible  to  

extrapolate   to   younger   cohorts.   If   non-­‐linearity   or   effect   modification   is   present,   any  

estimate  ignoring  these  phenomena  will  be  inaccurate.  

Page 14: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  13  

Third,   the  potential   for  a   large  number  of   confounders  depends   strongly  on   the  

health  measure   in   question.  Many   variables   are   related   to  migration,   for   example   age,  

sex,  education,   income,  marital  status  and  employment   (Patrick, 1980).     In   turn,  each  of  

these  may  or  may  not  also  be  related  to  different  measures  of  health,  and  therefore  may  

or  may  not   be   confounders.   For   instance,   income  might   be   related   to   self-­‐rated  health  

(Kennedy et al., 1998)  and  therefore  is  likely  a  confounder  in  that  context,  but  might  not  

be  related  to  a  genetic  susceptibility  to  a  disease,  and  therefore  not  a  confounder  in  that  

context.    

Fourth,   diseases   with   long   latency   periods   from   initial   symptoms   to   clinical  

manifestation  are  associated  with  increased  misclassification  because  the  disease  may  be  

attributed  as  incident  in  a  location  different  form  where  it  was  in  fact  incident  (Polissar,

1980; Rogerson & Han, 2002).   This   is   particularly   important  when   assessing   regionally  

varying   environmental   exposures   and   their   relationship   with   disease;   that   is,   invalid  

estimates  of  association  could  arise   if  persons  affected  by  that  disease  move  out  of   the  

exposed  region  because  of  subclinical  symptoms.  For  example,  persons  living  in  regions  

where  air  quality   is  poor  may  develop   respiratory   symptoms,   and  move   to  a   region  of  

better  air  quality  where  their  ailment  is  then  clinically  diagnosed.  

Further   complicating   the   association   between   health   and   migration   is   the  

possibility   that   migration   itself   may   cause   a   decline   in   health   (Ferraro, 1983; Findley,

1988).  Some  health-­‐selective  migration  research  that  has  been  done  on  health-­‐selective  

migration   uses   data   that   cannot   establish   the   temporal   order   of   health   and  migration  

events,  and  so  is  only  able  to  estimate  associations.  In  order  to  properly  establish  that  a  

Page 15: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 14  

change   in  health  has   either  precipitated  or  prevented   a  migration   event,   data  must  be  

able  to  resolve  whether  migration  happened  before  or  after  a  decline  in  health.    

 

1.1  Description  of  Systemic  Lupus  Erythematosus  and  Sjogren’s    

This   thesis  will   look   specifically   at   health-­‐selective  migration   in   systemic   lupus  

erythematosus   (SLE;   ICD9   710.0,   ICD10   M32)   and   Sjogren’s   syndrome   (ICD9   710.2,  

ICD10   M35)   patients,   both   of   which   fall   under   the   rubric   of   systemic   autoimmune  

rheumatic  diseases.  Health  selective  migration  within   these  chronic  disease  states   is  of  

considerable  interest,  as  will  be  discussed  below.  

SLE  is  a  multi-­‐systemic  autoimmune  disorder  of  unknown  etiology.  Its  prevalence  

in   the   province   of   Quebec   is   estimated   to   be   about   45   cases   per   100,000   and   is  most  

common   among   women   of   childbearing   age   (Bernatsky et al., 2007).   SLE   is   associated  

with   many   physical   manifestations,   including   arthritis,   as   well   as   constitutional  

symptoms  and  psychosocial  problems  such  as  fatigue  or  depression  (Tench et al., 2002).  

In  a   sample  of  SLE  patients   in   the  Netherlands,   two-­‐thirds  reported   “that   their  disease  

had  either  a  periodic  or  permanent  effect  on  their  ability  to  perform  everyday  activities  

at   present”   (Boomsma et al., 2002).   Of   829   SLE   patients   sampled   from   clinical   and  

community-­‐based   sources   in   California,   28%   reported   that   their   disease   affected   their  

ability  to  perform  basic  self-­‐care,  42.5%  reported  it  affected  their  ability  to  walk  to  get  

around   and   72.9%   said   it   affected   their   ability   to   sleep.   SLE   disease   duration   is   also  

associated  with  work  disability   (Bertoli et al., 2006; Mau et al., 2005; Yelin et al., 2007).  

Page 16: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  15  

Overall,  the  relationship  between  disease  duration  and  physical  and  emotional  health  is  

unclear  (Mcelhone et al., 2006).  

Sjogren’s  syndrome  is  another  autoimmune  rheumatic  disease  but  tends  to  affect  

older   individuals,  again,  predominantly  women,  and  have   less  severe   involvement.  The  

hallmark  characteristic  of  Sjogren’s   is   inflammation  of  the   lacrimal  and  salivary  glands,  

with  resultant  decrease  in  tears  and  saliva,  leading  to  very  marked  dry  eyes  and  mouth  

(Gran, 2002).  In  a  small  proportion  of  Sjogren’s  patients,  more  severe  involvement  (e.g.  of  

kidneys  or  the  central  nervous  system)  can  occur.  Estimates  of  Sjogren’s  prevalence  are  

less   consistent   because   they   depend   on   the   definition   and   criteria   used   as  well   as   the  

target  age  group  and  range  from  4-­‐2,700  per  100,000  (Avina-Zubieta et al., 2011; Hansen,

1991; Kabasakal et al., 2006; Pillemer et al., 2001).   Female   Sjogren’s   patients   have   been  

found  to  have  increased  functional  disability  (Strömbeck et al., 2003)  and  lower  perceived  

health-­‐related  quality  of  life  (Meijer et al., 2009).  

   

1.2  Rationale  

There   are   two   important   reasons   to   study   health-­‐selective   migration   in   SLE   and  

Sjogren’s  patients.  The  first  is  that  geographical  patterns  such  as  clustering  or  differences  

in   urban/rural   prevalence,   incidence   or  mortality   have   been   used   to  make   inferences  

about   possible   etiological   factors   or   patterns   of   health   care   use   in   rheumatic   diseases  

(Alamanos et al., 2003; Andrianakos et al., 2003; Barnabe et al., 2012; Gómez-Rubio &

López-Quílez, 2010; Hart et al., 2009; Kurahara et al., 2007; Labrecque, Joseph, et al., 2010;

Labrecque, Smargiassi, et al., 2010; Walsh & Gilchrist, 2006; Ward, 2010).   If   rheumatic  

Page 17: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 16  

diseases,  such  as  SLE  or  Sjogren’s,  affect  migration  rates  or  destinations,  the  inferences  

made   from  clustering  or  urban/rural  differences  may  be   limited.  For   these  reasons,  an  

assessment  of  whether  and  how  health-­‐selective  migration  may  occur  is  very  important  

in  these  conditions.  Of  interest  as  well,   is  whether  health-­‐selective  migration  could  also  

occur   in   patients   with   rheumatic   diseases   pre-­‐diagnosis   due   to   the   presence   of   pre-­‐

clinical   symptoms.   If,   for   instance,   we   observe   similar   estimates   of   health-­‐selective  

migration   relative   to   the   general   population   and   also   relative   to   SLE   and   Sjogren’s  

patients  before  their  diagnosis,   it  would  provide  evidence  for  weak  or  no  pre-­‐diagnosis  

health-­‐selective   migration.   Secondly,   decisions   about   moving   often   consider   opposing  

reasons   to   stay   or   move   (Hull, 1979; Patrick, 1980).   Therefore,   from   a   public   health  

perspective,  an   increase   in  migration  among  those  with  these  diseases  may  indicate  an  

aspect   of   their   disease   is   increasing   their   necessity   to  move.   This   may   occur   because  

healthcare  resources  are  either  absent,  inaccessible  or  more  distant  (Borders et al., 2000;

Buchanan et al., 2006)   and  may   alert   public   health   planners   of   areas  where   additional  

resources  are  required.  Canadian  studies  have  found  that  the  elderly  in  rural  areas  tend  

to   have   less   access   to   medical   specialists   (Allan & Cloutier-Fisher, 2006; McDonald &

Conde, 2010).   One   study   in   Iowa,   however,   suggested   this  may   not   be   the   case   among  

arthritis  sufferers  (Saag et al., 1998).  From  the  same  perspective,  a  decrease  in  migration  

may  indicate  that  some  aspect  of  SLE  and  Sjogren’s  is  decreasing  their  ability  or  desire  to  

move   rendering   migrations   that   would   have   happened   for   other   reasons   (e.g.   work,  

retirement)  difficult  or  impossible.    

Page 18: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  17  

There   is   also   evidence   that   health-­‐selective   migration   may   depend   on   the  

geographical   distance   migrated.   For   this   reason,   it   is   important   to   estimate   health-­‐

selective  migration  on  more  than  one  scale.  

   

1.3  Specific  objectives

The   overall   objective   of   this   thesis   is   to   estimate   health-­‐selective  migration   in   SLE  

and  Sjogren’s  patients  using  Quebec  administrative  databases.  However,  because  health-­‐

selective  migration   is   a   broad   and   complex   topic   the   overall   objective   will   be   broken  

down  into  three  more  specific  objectives:    

1) To  estimate  health-­‐selective  migration  in  SLE  and  Sjogren’s  patients  in  Montreal,  

Quebec  relative  to  the  general  population  using  Quebec  administrative  databases.  

2) To   estimate   health-­‐selective   migration   in   SLE   and   Sjogren’s   patients   in   the  

province  of  Quebec  relative  to  their  own  pre-­‐diagnosis  migration  rates.  

3) To   estimate   health-­‐selective  migration   on   a   regional   scale   in   SLE   and   Sjogren’s  

patients  in  the  province  of  Quebec  relative  to  their  pre-­‐diagnosis  migration  rates.  

These   objectives   will   be   realized   using   two   different   comparison   groups   and  

estimating  migration  on  two  geographic  scales.  Together,  they  will  address  the  questions  

of   whether   health-­‐selective   migration   exists   among   SLE   and   Sjogren’s   patients   and  

whether  migration  differs  by  comparison  group  and  geographic  scale.    

 

Page 19: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 18  

1.4  Thesis  structure    

This  thesis  is  divided  into  five  chapters.  This  introductory  chapter  has  briefly  

discussed  health-­‐selective  migration,  SLE  and  Sjogren’s  disease,  and  set  out  the  

objectives  of  this  thesis.  Chapter  2  provides  a  literature  review  of  previous  related  

research,  focusing  on  health-­‐selective  migration.  A  description  of  the  data  and  methods  

used  to  meet  the  three  main  objectives  enumerated  above  follows  in  Chapter  3.  The  

fourth  chapter  will  report  the  main  results  from  the  analyses  carried  out  in  this  thesis.  

The  fifth  and  final  chapter  will  discuss  the  results  and  provide  overall  conclusions.  

 

Page 20: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  19  

2  Literature  review  

 

 

This   chapter   presents   a   critical   review   of   the   literature   pertaining   to   health-­‐

selective  migration.  It  will  begin  with  a  description  of  the  search  strategy  used  to  identify  

the   literature   on   health-­‐selective   migration.   The   review   itself   will   be   divided   into  

historical   research,   modern   research   and   a   section   on   theoretical   papers   on   how  

selective-­‐migration  can  affect  estimates  of  association.  The  section  on  modern  research  

will   be   further   subdivided   into   three   subsections   according   to   how   the   research   was  

performed  and  will  by  followed  by  a  short  summary.    

Although   much   work   has   been   done   regarding   health-­‐selected   international  

immigration   (Borràs   et   al.,   1995;  Gushulak  &  MacPherson,   2006;  Harding  &  Balarajan,  

1996;  Newbold,  2005),  given  the  objectives  of  this  thesis,  this  literature  review  will  focus  

on  internal  migration,  i.e.  migration  within  national,  provincial  or  state  borders.  There  is  

also  a  sizable  literature  on  how  health-­‐selective  mobility  (whether  social  or  geographic)  

can  reinforce  inequitable  distributions  of  SES  or  deprivation  (Boyle  et  al.,  2009;  Curtis  et  

al.,   2009;   O'Reilly,   1994;   Richardson   et   al.,   2009).   This   topic   will   also   not   be   covered  

except  where  it  directly  pertains  to  the  topic  at  hand.  

The  chapter  concludes  with  a  short  discussion  of  the  limitations  of  administrative  

databases,   and  why   health-­‐selective  migration   is   of   particular   interest  within   SLE   and  

Sjogren’s  patients.  

Page 21: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 20  

 

2.1  Search  strategy  

Searches   were   conducted   in   MEDLINE   and   EMBASE,   covering   the   years   from  

1946   to  2011.   Search   terms  used  were:  migration,   residential  mobility  or   immigration  

and  emigration  AND  health  and  health  status.  Selected  articles  were  hand-­‐searched  for  

additional  references.  Only  English-­‐language  papers  on  health-­‐selective  migration   from  

North   American   and   Europe   were   included.   After   screening   titles   and   abstracts   and  

hand-­‐searching  articles,  50  references  were  selected  for  the   literature  review  spanning  

from  1871  to  2011.    

 

2.2  Historical  research  (1871-­1970)  

The  earliest  mention  of  health–selective  migration,  though  not  by  that  name,  was  

by  Thomas  A.  Welton  (1871)  who  remarked  “that  the  mortality  happening  in  London  is  

diminished,  ‘because  domestic  servants,  shopwomen  and  milliners,’  who  have  come  from  

the   country,   retire  when  health   fails   them   ‘to   their  native   air'."  He   found  his   efforts   to  

analyze  patterns  of  mortality   in  England  and  Wales   frustrating,   because   “[i]t   is   vain,   it  

would   seem   to   look   (at   least   in   this   country)   for  any  quiet   self   contained  place,  where  

population   remains   undisturbed   by   migrations,   and   where   statistics   of   mortality   can  

therefore  be  obtained,  requiring  no  rectification.”  He  observed,  more  generally,  that  sick  

people  migrate  away  from  hazards  and  toward  care.  

In   response   to   a   comment   from   a   founding   father   of   epidemiology   Dr.  William  

Farr,   “that   migration   appeared   to   go   on   without   any   definite   law,”   Ernest   George  

Page 22: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  21  

Ravenstein  published  two  seminal  papers  entitled  “The  Laws  of  Migration”  (1885;  1889)  

wherein   he   laid   out,   for   the   first   time,   rules   governing   the   flow   of   migrants.   Though  

Ravenstein’s  work  focused  on  differential  migration  with  respect  to  sex  and  urbanicity,  it  

was  the  basis  for  much  of  the  work  on  health-­‐selective  migration  in  the  first  half  of  the  

20th  century,  leading  to  research  on  how  physical  fitness,  mental  illness  and  intelligence  

might  influence  migration  patterns  (Thomas,  1938).  After  the  second  world  war,  the  U.S.  

census   started   taking   an   interest   in   what   caused   people   to   move   (Bureau   of   the   U.S.  

Census,   1947),   but   research   on   health-­‐selective  migration  was   sparse   until   after   1970  

when  patterns  of  migration   in   the  elderly    became  of   interest.  Up  to   that  point,  health-­‐

selective   migration   was   often   seen   as   unimportant.   In   fact,   Lee   (1966)   even   failed   to  

mention  health  as  an  obstacle  or  motive  for  migration  in  his  own  work  despite  the  fact  

that  Haenszel   and  Dawson   had   published   the   year   before   (Haenszel  &  Dawson,   1965)  

that  estimates  of  the  prevalence  of  difference  health  conditions  within  a  region  differed,  

depending  on  whether  one  considered  all  or  only  long-­‐term  residents  of  a  given  region.  

 

2.3  Modern  research  (1970-­2011)  

Research   into   health-­‐selective  migration   regained   importance   after   1970  when  

researchers  interested  in  patterns  of  elderly  migration  began  understanding  health  as  a  

determinant.   Many   different   methods   have   been   employed   to   investigate   the  

relationship   between   health   and   migration.   Often,   the   nature   of   a   data   set   required  

researchers   to   use   one  method   or   another,  with   each   study   having   different   strengths  

and  weaknesses.  The  research  can  be  grouped  into  three  main  categories.  The  first  uses  

Page 23: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 22  

migration  as  the  outcome  and  measures  of  health  as  a  predictor.  The  second  compares  

health   in   migrants   and   non-­‐migrants   using   crude   comparisons,   standardization   or  

multivariate   regression   modeling.   The   last   uses   surveys   to   directly   elicit   reasons   for  

migration  from  migrants.    Each  of  these  areas  will  now  be  summarized  in  turn.  

 

2.3.1  Migration  as  the  outcome  

This   section  will   group   studies   that   have  migration   as   the   outcome.   This   is   the  

method  used  in  this  thesis  and  will  therefore  be  most  pertinent.  Most  such  studies  rely  

on   survey   data   although   health   administrative   data   (e.g.   physician   billing   and  

hospitalization)  and  census  data  have  also  been  used.  Analytical  methods  include  crude  

comparisons,  matching  and  logistic  or  multinomial  regression.  

The  most  widely  used  dataset  on  health-­‐selective  migration  is  the  US  Longitudinal  

Study  on  Aging.  It  is  composed  of  data  on  Americans  70  years  and  older  from  the  1984  

Supplement  on  Aging  to  the  National  Health  Interview  Survey  and  a  follow-­‐up  surveys  in  

1986,  1988  and  1990.  It  is  worth  elaborating  on  these  articles  because  they  use  slightly  

different  measures   and  methods   for   the   same   dataset,   allowing   us   to   investigate   how  

robust   results   are   to   these   differences.   The   1984   survey   contained   questions   about  

health,  disability,   functional  ability,   family  relationships,   living  arrangements  and  social  

support   (Speare   et   al.,   1991).   The   follow-­‐up   surveys   repeated   questions   on   health,  

disability   and   living   arrangements   as  well   as  whether   any  moves   had   occurred   in   the  

past  two  years.  The  measure  of  health  used  in  these  papers  is  the  number  of  activities  of  

daily   living   (ADLs;   activities   considered   essential   to   daily   living   such   as   self-­‐feeding,  

Page 24: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  23  

personal   hygiene,   bowel   and   bladder   control,   etc.)   or   instrumental   activities   of   daily  

living  (IADLs;  activities  that  aren’t  considered  essential  but  allow  for  independence  such  

as   housework,   shopping,   using   the   telephone,   etc.)  with  which   a   person   has   difficulty.  

One  important  caveat  of  this  data  is  that  disability  at  baseline  can  be  said  with  certainty  

to  have  occurred  before  migration   events   reported   in   follow-­‐up   survey.    However,   the  

same   cannot   be   said   for   change   in   health   or   disability   after   baseline,   because  

respondents   did   not   report   the   order   in   which   migration   events   and   changes   in  

disabilities   occurred.   Also,   each   paper   uses   slightly   different   inclusion   criteria   for  

subjects  and  different  model  selection  methods.    

Speare  et  al.  (1991)  looked  at  how  the  sum  of  the  number  of  ADLs  and  IADLs  with  

which  a  person  had  difficulty   is   related   to   three  outcomes:  migration   to  an   institution,  

any   migration,   and   a   change   in   living   arrangement   that   involves   living   with   an   adult  

other   than   a   spouse.   They   found   that   baseline   disability   is   related   to   a   move   to   an  

institution   (OR   1.20,   95%   CI:   1.14-­‐1.26),   and   change   in   living   arrangement   (OR   1.12,  

95%  CI:  1.06-­‐1.19)  but  was  not  necessarily   associated  with  general  mobility   (OR  0.97,  

95%  CI:0.91-­‐1.03).  Longino  et  al.  (1991)  look  specifically  at  IADLs  and  general  mobility  

finding  an  OR  of  1.08  (95%  CI:  0.98-­‐1.19).  Jackson  et  al.  (1991)  found  similar  results  to  

Longino   et   al.  when   controlling   for   a   few  more   potential   confounders   and   including   a  

change  in   living  arrangements  with  migration  as  part  of  the  outcome  (OR  1.09,  95%  CI  

1.01-­‐1.17).  Worobey  and  Angel  (1990)  looked  at  increases  in  dependence  (living  with  an  

adult  or  moving  to  an  institution)  with  polychotomous  ordered  response  regression  and  

found  that  ADLs  were  associated  with  a  small  increase  in  dependence    with  an  OR  1.06  

Page 25: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 24  

for   an   increase   in   dependence   (the   result  was   statistically   significant   but   no   standard  

error  was  reported).  De  Jong  et  al.  (1995)  found  baseline  ADLs  (OR=0.97,  95%  CI  0.92-­‐

1.03)  may  be  related  to  a  decrease  in  migration  using  all  waves  of  the  survey  from  1986-­‐

1990.  Silverstein  and  Zablotsky  (1996)  also  used  the  1984-­‐1990  waves  of  the  LSOA  and  

multinomial   regression   with   different   migration   destinations   as   the   outcomes  

(retirement  community  service-­‐rich,  retirement  community  service  poor,  other  general  

community,  institution).  They  found  that  migration  to  retirement  communities,  whether  

service-­‐rich   or   poor,   peaked   at   moderate   levels   of   disability   (OR=2.83,   reported   as   a  

statistically  significant  result  but  no  standard  error  given)  and  were  near  null  for  those  

with  little  or  no  disability  as  well  as  among  those  with  high  levels  of  disability.  Synthesis  

of  all  these  results  is  difficult  to  fully  quantify  due  to  the  lack  of  reported  standard  errors  

or   confidence   intervals   though  most   point   to   an   increase   in  migration   in   elderly  with  

disabilities.  However,  interpretation  of  all  studies  using  this  dataset  should  consider  the  

possible  non-­‐linearity  between  disabilities  and  migration.    

Two   studies   were   done   during   different   time   periods   using   the   Panel   Study   of  

Income  Dynamics.  The  first  only  used  data  on  people  aged  60  and  over  from  1969-­‐1980  

(Henretta,   1986).   The   association   between   the   head   of   the   household   having   a   health  

limitation   and   migration   using   logistic   regression   was   inconclusive   due   to   wide  

confidence   intervals.   A   second   study   used   data   on   all   ages   from   1984-­‐1993   and   also  

made   sure   that   the   self-­‐reported   health  was  measured   before   the  migration   event   for  

each  observation  (Halliday  &  Kimmitt,  2008).  Using  age-­‐stratified  probit  regression,  men  

older  than  60  self-­‐reporting  either  very  good  or  excellent  health  and  those  reporting  fair  

Page 26: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  25  

or  poor  health  were  associated  with  increased  migration  relative  to  those  in  good  health.  

This  non-­‐linearity  is  opposite  to  that  found  in  Silverstein  and  Zablotsky,  where  migration  

peaks  at  the  ends  of  the  health  distribution,  not  in  the  middle,  though  different  measures  

of  health  are  being  used.  This  relationship,  however,  was  not  found  among  women  over  

60   years.   Among   men   younger   than   60,   poorer   health   is   associated   with   a   32-­‐40%  

decrease  in  migration  (a  range  representing  possible  values  depending  on  covariates  and  

model  specifications).  A  smaller,  yet  not  statistically  significant  decrease  of  12%  to  18%  

was  found  among  women.  

The  remaining  studies  use  limiting  long-­‐term  illness  (LLTI)  or  chronic  diseases  as  

a   measure   of   health   though   some   supplement   this   with   other   measures.   A   study   of  

people  50  years  and  older  using  the  British  Household  Panel  Study  looked  at  migration  

rates  stratified  by  ten-­‐year  age  groups  and  by  no/non-­‐limiting/limiting  long-­‐term  illness  

found   that   men   and   women   with   a   LLTI   tended   to   have   higher   migration   rates  

particularly  among   those  older   than  80  years   (Evandrou  et  al.,  2010).  The  relationship  

between   changes   in   self-­‐rated   health   and   migration   was   less   clear.   A   change   in   self-­‐

reported  health  of  two  points  on  the  Likert  scale,  whether  an  improvement  or  worsening  

of  health,  appeared  to  be  associated  with  increased  migration  except  among  those  aged  

70-­‐79  where  a  worsening  of  health  was  associated  with  a  decrease  in  migration  and  an  

improvement  in  health  was  associated  with  an  increase  in  migration.  This  non-­‐linearity  

is  similar  to  Halliday  and  Kimmit  (2008)  where  migration  is  lowest  among  those  in  the  

middle  of  the  health  distribution.  

A  study  in  three  cities  in  Quebec  of  elderly  subjects  over  age  65  used  discriminant  

Page 27: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 26  

analysis   to   determine  whether   a   list   of   health   problems,   including   number   of   chronic  

diseases   and   self-­‐rated  health,   economic   and   social   variables  predicted   either   filing   an  

application  to  move  to  an  institution  or  desire  to  remain  at  home  (Béland,  1984).  Their  

resulting   model   found   that   the   number   of   chronic   diseases   was   positively   associated  

with  filing  an  application  to  move  to  an  institution.  Self-­‐rated  health  and  reduced  activity  

due   to   health  were  not   related   to   either   filing   an   application  or   a   desire   to   stay  home  

although  no  estimates  were  reported,  only  that  they  were  selected  out  of  the  model.    

Findley   (1988)   did   an   exploratory   analysis   using  data   from   the  1979   and  1980  

waves  of  the  National  Health  Interview  Survey  in  the  United  States  finding  that,  in  all  age  

groups  but  particularly  among  people  older  than  45  years,  participants  were  more  likely  

to   experience   a   chronic   illness   after   a  migration   than   vice   versa.   However,   the   survey  

contained  no  direct  data  on  onset  of  chronic  illness  and  the  last  date  a  doctor  was  seen  

for   a   chronic   condition  was   used   as   a   proxy.   The   result   is   that   the   date   of   onset  was  

necessarily  always  moved  forward   in   time   increasing  the   likelihood  that  any  migration  

events  between  the  actual  onset  and  the  interview  would  occur  before  and  not  after  the  

proxy  onset.  This  measurement  error  could  easily  create  a  situation  where  a  change   in  

health  status  precipitated  a  migration  look  like  the  migration  precipitated  the  change  in  

health  status.  The  crude  likelihood  of  migrating,  stratified  by  age,  was  roughly  the  same  

among   those  who   had   not   experience   a   health   event   in   the   past   two   years   as   among  

those  who  had.    The  unknown  magnitude  of  the  measurement  error,  however,  makes  it  

difficult  to  draw  any  reliable  inference  from  this  paper.  

Page 28: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  27  

A  comparison  of  urban  and  rural  movers  and  stayers  in  and  around  Eindhoven  in  

the   Netherlands   using   a   longitudinal   survey   also   suffers   from   methodological   flaws  

rendering   reliable   inferences   difficult   (Verheij   et   al.,   1998).   Two   logistic   regression  

analyses  comparing  movers  to  stayers  and  urban  migrants  to  rural  migrants  both  suffer  

from  mis-­‐specification   because   they   consider   age   interactions   without   including  main  

effects   for   age.   In   bivariate   comparisons,   movers   had   better   perceived   health   and  

reported  less  chronic  conditions  than  stayers.  

An  Australian  longitudinal  survey  investigated  how  health  characteristics  such  as  

self-­‐reported   health,   number   of   chronic   conditions   and   visits   to   a   specialist   could  

subsequently   lead   to   within   postal   code,   between   postal   code   and   rural   to   urban  

migration   among   women   aged   45-­‐50   (Larson   et   al.,   2004).   Only   p-­‐value   inequalities  

(p<0.05,  p<0.01)  were  presented  making  it  difficult  to  correctly  interpret  the  importance  

of   the   results.  Also,   the   likely   strong   collinearity   between  health  measures   included   in  

the  model   make   each   individual   coefficient   difficult   to   interpret.   Having   two   or   more  

chronic   conditions   was   associated   with   an   increased   likelihood   of   moving   within   and  

between  postal  codes  as  well  as  rural  to  urban  moves.  Poor  mental  health  and  smoking  

were   associated  with   increased  migration  within   and   between   postal   codes.   Three   or  

more  visits   to  a  specialist   in   the  previous  year  was  also  associated  with  rural   to  urban  

moves.  

Another   Australian   study   compared   directly   standardized   migration   rates   in   healthy  

people  of  all  ages  and  those  with  a  serious  disease  where  migration  was  a  move  between  

metropolitan,   rural   and   remote   regions   (Moorin   et   al.,   2006).   They   found   that   people  

Page 29: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 28  

with  a  serious  disease  were  less  likely  to  migrate  in  any  direction  (more  or  less  urban).  

People  with  connective  tissue  and  rheumatoid  disorders,  a  category  including  lupus  and  

Sjogren’s,  were   less   likely   to  move  out  of   the  city   than  healthy  controls  (incidence  rate  

ratio:   0.46,   95%   CI   0.27-­‐0.75).   Unfortunately,   results   that   were   not   statistically  

significant  were  not  presented  but  could  have  been  of  interest  if  the  confidence  intervals  

included  values  of  interest.  Though  temporality  could  not  be  established  in  this  case,  it  is  

more   likely   that   diseases   like   lupus   and   Sjogren’s   precipitated   a  migration   event   than  

vice-­‐versa.   Breslow   et   al.   (1998)   looked   at   a   similar   phenomenon,  which   they   termed  

“county   drift”,   in   the   city   of   Albany,   New   York,   the   location   of   a   psychiatric   hospital  

serving  the  entire  county  of  Albany.  They  found  that  patients  of  the  hospital  were  more  

likely  to  move  to  Albany  than  they  were  to  move  away.    

Of   particular   interest,   Yiannakoulias   et   al.   (2007)   has   many   characteristics   in  

common   with   this   thesis.   It   used   administrative   data   to   estimate   health-­‐selective  

migration,   takes   place   in   Canada   and   considers   specific   diseases.   Parkinson’s   disease  

(PS)  and  multiple  sclerosis  (MS)  patients  identified  in  fee-­‐for-­‐service  and  hospitalization  

databases  were  matched  to  members  of  the  general  population  base  on  sex,  age  (within  

five   years),   SES   (based   on   health   insurance   premium   subsidy   level)   and  municipality.  

They   found   that   MS   patients   were   slightly   more   likely   to   migrate   at   least   once   when  

compared   to   controls   (45.9%   in   MS   patients   vs.   43.6%   of   controls)   and   PS   patients  

where  much  more  likely  to  migrate  at  least  once  than  controls  (35.7%  among  PS  patients  

vs.  26.2%  of  controls).  One  implicit  assumption  this  analysis  makes   is  that  the  effect  of  

being  diagnosed  with  MS  or  PS  is  homogeneous  across  both  age  and  disease  duration.  If  

Page 30: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  29  

one   or   both   of   these   assumptions   is   not  met,   and   there   is   literature   that   suggests   this  

could  be  the  case  (Boyle  et  al.,  2002;  Halliday  &  Kimmitt,  2008),  the  estimates  presented  

become   a   summary   estimate   of   heterogeneous   effects   by   age   and   disease   duration  

influenced  by  the  effect  size  in  different  strata  and  the  structure  of  the  population  and  no  

longer   retains   any   meaningful   interpretation.   Working   with   the   knowledge   that   MS  

affects  people  in  mid-­‐life  and  PS  people  of  older  age,  we  could  expect  these  estimates  to  

be  influenced  most  by  rate  ratios  in  those  ages  groups.  

 

2.3.2  Comparative  health  of  migrants  and  non-­migrants  

Many   researchers   have   chosen   to   investigate   health-­‐selective   migration   by  

comparing  measures   of   health   in  migrants   and  non-­‐migrants   using   logistic   regression,  

standardization  or  crude  comparisons.  In  these  studies,  researchers  were  more  likely  to  

stratify   by   age   but   often,   due   to   the   nature   of   the   data   sources,   could   not   resolve   the  

temporal   order   of   health   and   migration   events.   The   primary   health   outcomes  

investigated  are  LLTI,  mortality  and  specific  health  problems.  Mortality,   in  this  context,  

must  be  seen  a  proxy  for  poor  health  because  it  clearly  cannot  occur  before  a  migration  

event.    

Bentham   (1988),   using   the   British   census,   analyzed   the   proportion   of  

‘permanently’  or  ‘temporarily’  sick  among  non-­‐migrants  and  local  and  regional  migrants.  

The  author  found  that  young  local  migrants  were  less  likely  to  be  permanently  sick  when  

compared   to   the   district   average   but   that   older   migrants   were   more   likely   to   be  

permanently  sick.  A  similar  pattern  was  found  among  the  temporarily  sick  although  the  

Page 31: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 30  

effect   was   not   as   pronounced.   For   regional   migrants,   the   proportion   permanently   or  

temporarily  sick  was,  for  the  most  part,  less  than  the  average  of  the  district.  No  measures  

of   uncertainty   were   reported   which   limits   the   interpretability   of   inferences   from   this  

study.    

Brimblecombe,  Dorling  and  Shaw  (2000)  used  data  from  the  first  six  waves  of  the  

British  Household   Panel   Survey   (1991-­‐1996)   and   compared  migrants   to   non-­‐migrants  

stratified  living  in  a  high  or  low-­‐mortality  district.  Men  and  women  moving  from  high  to  

low  mortality  districts  self-­‐reported  being  in  better  health  and  that  health  was  less  likely  

to  limit  their  activities  than  those  remaining  in  high  mortality  districts.  Men  moving  from  

low  to  high  mortality  districts  were  less  likely  to  report  that  health  limits  their  work  and  

more  likely  to  score  poorly  on  a  questionnaire  rating  mental  health.  

 Three  papers   considered   specific  health  problems  as   the  outcome.  Among  men  

40-­‐59  years  old   in   the  UK,   internal  migrants  were  23%  (95%  CI:  6-­‐37%)   less   likely   to  

experience   a   major   ischemic   heart   disease   event   than   non-­‐migrants,   and   had   lower  

systolic   (3.0   mmHg,   95%   CI   2.0-­‐4.0)   and   diastolic   (1.2   mmHg,   95%   CI   0.6-­‐1.8)   blood  

pressure  (Elford  et  al.,  1990;  1989).  In  response,  Martyn  et  al.  (Martyn  et  al.,  1993)  claim  

that   this   pattern   may   be   explained,   in   part   at   least,   by   selective   migration   by   birth  

weight.  Using  hospital  records,  they  found  that  men  that  migrated  away  had,  on  average,  

higher   birth   weight   than   those   who   stayed.   Among   women   the   difference   was  

inconclusive.  A  cross-­‐sectional  study  in  Croatia,  where  health  outcomes  were  measured  

by   nurses,   found   that   migrants   (excluding   employment-­‐related   migrants)   in   all   age  

groups   tended   to  have   lower  prevalence  of  hypertension,  overweight  and  obesity   than  

Page 32: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  31  

non-­‐migrants  (Kolcić  &  Polasek,  2009).  Exceptions  were   that  hypertension  and  obesity  

in  those  aged  18-­‐34  was  higher  in  migrants  and  overweight  in  those  aged  50-­‐64  was  high  

in  migrants.  

  LLTIs,  particularly  in  the  UK,  are  a  popular  measure  of  health.  A  study  using  the  

1991  British  Census,  found  long  distance  migrants  (>50km)  in  England  and  Wales  18-­‐64  

years  old  had  lower  odds    (OR  0.86,  95%  CI:0.78-­‐0.95)  of  reporting  a  LLTI  than  non  and  

short-­‐distance  migrants  (Boyle  et  al.,  2001).  This  study  was  repeated   in  Scotland  using  

the   same   data   source   but   a   different   cut   point   for   short   and   long   distance   migrants.    

Short-­‐distance  migrants  (<10km)  were  more  likely  to  have  a  LLTI  than  non-­‐migrants  and  

long-­‐distance  migrants  less  likely  to  migrate  than  non-­‐migrant  between  30-­‐65  years  old  

(Boyle  et  al.,  2002).  After  retirement,  long-­‐distance  migrants  had  similar  levels  of  LLTI  as  

non-­‐migrants,   whereas   short-­‐distance   migrants   had   higher   levels.   This   study  

demonstrates  some  heterogeneity  by  age  but  includes  no  confidence  intervals  nor  does  it  

include  the  coefficients  from  the  model.    

In  the  Netherlands,  a  national  survey  was  used  to  study  the  relationship  between  

three   measures   of   health   (self-­‐reported   health,   long-­‐standing   health   problems,   and  

disabilities)   and   migration   stratified   by   deprived   and   non-­‐deprived   areas   (Jongeneel-­‐

Grimen  et  al.,  2011).  They   found   that  people  moving   to  non-­‐deprived  areas  had  worse  

self-­‐reported   health   and  more   disabilities   than   the   people   already   living   there.   It   was  

also   found   that   people   moving   to   deprived   areas   had   better   health   than   the   people  

already   there.   This   may   suggest   that   the   area   a   person   moves   to   might   be   an   effect  

measure  modifier.  Wiggins  et  al.  (2002)  found  no  association  between  LLTI  and  internal  

Page 33: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 32  

migration   in   English   and   Welsh   women   aged   35-­‐65   using   the   Office   for   National  

Statistics’   Longitudinal   Survey   (ONS-­‐LS)   though   the   estimate   and   confidence   intervals  

were  not  reported  because  the  result  was  not  statistically  significant.  Also,  migration  was  

only  considered  between  1971-­‐1981  and  presence  of  an  LLTI  was  assessed  in  1991.  The  

ten-­‐year  gap  between  exposure  and  outcome  assessment  likely  weakens  the  association.  

Another   study   used   the   same   data   source   as  Wiggins   et   al.   but   over   a   different   time  

period  (1981-­‐2001)  and  with  mortality  as   the  outcome  found  different  results  (Riva  et  

al.,  2011).  Using  logistic  regression  stratified  by  age  (<65,  65+),  time  period  (1981-­‐2001,  

1991-­‐2001)   and   migrants   type   (urban   outmigrant,   rural   outmigrant,   long-­‐term   rural  

resident,   long-­‐term   urban   resident),   they   found,   among   people   65   years   and   over,  

migrating  out  of  an  urban  area  between  1981-­‐2001  was  associated  with   lower  odds  of  

death  between  2001-­‐2005  (OR  0.86,  95%  CI:  0.77-­‐0.94).  Among  people  younger  than  65  

years,  conclusively  lower  odds  of  death  were  found  when  limiting  the  migration  period  

to   1991-­‐2001   though   the   confidence   intervals   included   many   low   OR   values   when  

migration  was  measured  between  1981-­‐2001.  Heterogeneity  in  ORs  by  age  is  suggested  

in  this  study  but  the  confidence  intervals  overlap  slightly.  

Two  studies  were  found  comparing  residents  to  outmigrants  of  an  area  in  order  

to   assess   the   impact   of   health-­‐selective   migration   on   their   research   questions.  

Tousignant  et  al.  (1994)  found  that  people  who  migrated  out  of  an  area  before  a  cross-­‐

sectional   study   on   the   health   effects   of   point   source   pollution  were   less   likely   to   have  

heart  disease   (RR:  0.17,  95%  CI  0.10-­‐0.29),  hypertension   (RR:  0.25,  95%  CI  0.17-­‐0.36)  

and   migraines   (RR:   0.59,   95%   CI   0.43-­‐0.81).   Only   1.3%   of   migrants,   when   asked,  

Page 34: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  33  

reported  moving  away  for  health  reasons.  Matthews  et  al.  (2004)    found  that  people  65  

and  over  who  moved  away  between   sample   selection  and  baseline   interview  were  1.8  

times   (1.3-­‐2.4)  more   likely   to   die  within   two   years.   Those  who  moved   away   between  

successive  waves  of  a  survey  were  more   likely   to  be  smokers,  demented  or  depressed.  

(Larson  et  al.,  2004).  

A  number  of   studies  have  used  standardized  mortality   ratios   to  compare  health  

with   migration   in   ecological   studies.   Indirect   standardization   may   lead   to   spurious  

results   in   this   context   because   it   is   unlikely   that   the   assumption   of   similar   population  

composition  (Tsai  &  Wen,  1986)  is  met  between  migrant  and  non-­‐migrant  populations.  

Therefore,  all  studies  using  indirect  standardization  should  be  interpreted  with  caution.  

Fox  et  al.  (1982)  used  SMRs  to  compare  migrants  to  the  entire  population  of  England  and  

Wales.  They  found  that  within-­‐county  migrants  had  an  SMR  of  1.10  (95%  CI:  1.07-­‐1.13)  

and   between-­‐county   and   between-­‐region  migrations   had   SMRs   of   0.93   (95%  CI:   0.86-­‐

1.00)  and  0.91  (95%  CI:  0.84-­‐0.98)  respectively.  Therefore,  short  distance  migrants  were  

less  healthy  than   long  distance  migrants.  One  study  Norman  et  al.   (2005)  was  the  only  

other   paper   to   use   SMRs   of   general  mortality.  Migrants   and   non-­‐migrants   had   similar  

mortality  except  in  the  60-­‐79  age  group  where  migrants  had  lower  mortality.  They  found  

that  among  people  between  40-­‐59  years  old,  migrants  had  less  LLTI  than  non-­‐migrants  

but  the  relationship  was  less  clear  for  older  age  groups.    

A  number  of  studies  looked  at  mortality  due  to  a  specific  cause.  Using  U.K.  census  

data  linked  to  the  death  registry,  Strachan  et  al.  (1995)  compared  SMRs  adjusted  for  sex,  

age,   calendar   period,   housing   tenure,   and   car   ownership   among   non-­‐migrants   and  

Page 35: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 34  

migrants  using  Poisson  regression.  They  found  that  migrants  had  lower  death  rates  than  

non-­‐migrants   for  both   ischemic  heart  disease  (RR:  0.90,  95%  CI:  0.87-­‐0.94)  and  stroke  

(RR:  0.94,  95%  CI:  0.88-­‐0.99).  The  average  age  is  not  reported  but  the  information  given  

allows  the  calculation  that  participants  were  at   least  32  and  the  distribution  of  person-­‐

years  is  likely  mostly  above  65  years  of  age.    

Two   American   studies   found   that   the   SMRs   of   migrants   differed   both   in  

magnitude   and   direction   depending   on   the   state.   Kliewer   (1992)   compared   colon   and  

stomach   cancer  mortality   rates   among  migrants   and   non-­‐migrants   in   11  western   U.S.  

states  using  SMRs  and  found  that  they  differed  from  state  to  state.    Lanska  and  Peterson  

(1995)  found  similar  result  in  stroke  mortality  looking  at  all  the  coterminous  U.S.  states.  

 

2.3.3  Surveys  on  reasons  for  migration  

Six  surveys  where  health  related  migration  was  investigated  were  found.  Four  of  

these  sampled  exclusively  elderly  people  and  two  looked  at  people  from  a  variety  of  age  

groups.    

A  prospective  cohort  of  Rhode  Islanders  who  had  moved  since  the  inception  of  a  

cohort  12  years  earlier  were  asked   the  reason   for   their  move  (Meyer  &  Speare,  1985).  

The  proportion  of  migrants,  who  moved   for   health   reasons,   either   closer   to   friends  or  

relatives  for  assistance  reasons  or  entry  into  a  nursing  home,  remained  roughly  constant  

around  7%  between  27-­‐59  years  old  but   increased   to  18.5%  and  34.5%  among  60-­‐69  

and  70+   year   olds   respectively   indicating   that   health-­‐selective  migration   exists   among  

younger   people   but   is   particularly   important   among   the   elderly.   Meyer   and   Speare,  

Page 36: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  35  

however,   only   report   interviews   with   migrants   and   therefore   could   not   address   the  

question   of  whether   some  potential  migrants  were   prevented   from  migrating   by   poor  

health.   If   an   equal   number   of   people   were   prevented   from   moving   because   of   their  

health,  health-­‐selective  would  be  difficult  to  observe.  The  representativeness  among  this  

cohort   is   also   called   into   question   because,   though   the   cohort   was   validated   as   being  

representative  of  the  Rhode  Island  population  at  its  inception,  28%  of  those  alive  at  the  

end  of  the  cohort  were  lost  to  follow  up.    Those  lost  to  follow  up  were  less   likely  to  be  

married,   less   likely   to   be   employed,   in   poorer   health   and   more   likely   to   be   renters  

meaning  that  proportions  of  health-­‐related  migrants  are  likely  underestimates.  Analysis  

of  variance  was  performed  on  demographic  characteristics  of  different  types  of  movers  

(including   non-­‐movers).   Some   interesting   patterns   are   observed   such   as   important  

differences   between   local   and   out   of   state   health-­‐related   migrants   in   terms   of   age,  

education   and   income   but   no   variances   estimates   are   given   making   it   impossible   to  

evaluate   the   importance   of   the   differences   between   groups   of  movers.   Of   31  Alaskans  

(ages  ranged  from  21  to  64)  who  had  moved  from  a  rural  to  urban  area  that  participated  

in   a   qualitative   survey,   over   half   gave   a   medical   reason   for   the   move   (Driscoll   et   al.,  

2010).  

The  remaining  surveys  only   looked  at   retired   individuals,   finding  mixed  results.  

Twenty  four  percent  of  92  recent  migrants  to  retirement  community  in  Arizona  reported  

moving   there   for  health   reasons   (Gober  &  Zonn,   1983).   In   a   city   in  upstate  New  York,  

60%   of   senior   citizens   surveyed   reported   considering   a   move.   Interestingly,   health  

condition  and   the  presence  of  health  problems  were  not  associated  with  considering  a  

Page 37: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 36  

move   but   an   inability   to   perform   certain   activities   of   daily   living   was   negatively  

associated   with   considering   a   move.   This   is   evidence,   though   tenuous,   against   the  

hypothesis  that  disabilities  cause  moves  through  requiring  special  care  and  contradicts  

research  from  previous  sections  (Nelson  &  Winter,  1975).  

Silverstein  and  Angelello  (Silverstein  &  Angelelli,  1998)  used  the  Asset  and  Health  

Dynamics  of   the  Oldest  Old,  a   representative  national   sample  of  Americans  over  70,   to  

investigate   characteristics   that   make   the   elderly   consider   migration.   People   were  

considered  at  risk  of  a  move  if  they  reported  a  great  than  20%  chance  of  moving  in  the  

next   five   years.   Self-­‐rated   health   was   positively   related   to   considering   a   move,   while  

difficulty  with  IADLs  and  cognitive  impairment  were  inversely  related  with  considering  a  

move.  Therefore,  poorer  health   led   to   lower  odds  of   considering  a  move.  Among   those  

with  children,  self-­‐rated  health  was  not  related  to  the  expectation  of  moving  closer  to  or  

in   with   one   of   their   children   unless   they   lived   alone.     In   this   case,   it   was   strongly  

positively   related,   suggesting   living   alone   is   an   effect  measure  modifier   of   the   health-­‐

migration  relationship.  

Sergeant   and   Ekerdt   (2008)   performed   a   qualitative   survey   of   30  movers   in   a  

Mid-­‐western  US  state  aged  60-­‐87.  Many  cited  health  problems  as  a  reason  to  move  but  

saw   health   problems   as   related   to   a   diagnosis   as   distinct   from   functional   limitation,  

which  they  saw  as  related  to  effect  of  aging.    

 

 

 

Page 38: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  37  

2.3.4  How  selective  migration  can  affect  estimates  

This  section  will  present  simulation  studies  of  health-­‐selective  migration  and  how  

it  affects  estimates  of  prevalence  or  associations  with  exposures.  

A  number  of   studies  have   looked  at  how  health-­‐selective  migration  might  affect  

estimates   of   the   relationship   between   place   and   health   with   the   methods   employed  

increasing   in   complexity.   One   of   the   earliest   studies   estimated   how   non-­‐differential  

migration  with  regards  to  health  could  influence  relative  excess  risk  of  cancer  incidence  

using  U.S.  data  on  continuous  residence  and  cancer  incidence  rates  (1980).  Depending  on  

different   assumptions   about   scale   of   analysis   (place,   county,   state)   and   latency  period,  

migration   could   reduce   the   attributable   risk   of   an   environmental   exposure   anywhere  

from  8%  to  73%.  Cancers  with  increased  latency  periods  and  analyses  using  small-­‐scale  

unites  were  more  prone  to  bias.    

In  Sydney,  Nova  Scotia,  Veugelers  and  Guernsey  (1999)  performed  a    sensitivity  

analysis  to  investigate  how  health-­‐selective  migration  could  influence  life  expectancy  in  

men.   They   assumed   outmigrants   were   either   slightly   healthier   (had   a   life   expectancy  

equal   to   the  average  Canadian  male)  or  much  healthier   (had  a   life  expectancy  equal   to  

the   average   Canadian   female)   than   the   Sydney   population   and   found   that   health-­‐

selective   outmigration   from   Sydney   might   be   responsible   for   a   decrease   in   life  

expectancy  of  up  to  three  years  from  1950-­‐1995.      

Over   twenty  years   later,   a   similar  approach   to   the  Polissar  model   found  similar  

results   (Rogerson   &   Han,   2002).   Migration   probabilities   were   based   on   disease  

presence/absence.   Using   data   on   migration   between   counties   in   New   York   State   and  

Page 39: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 38  

looking  at  lung  cancer  as  the  target  disease,  they  found  that  a  county  with  high  mobility  

would   retain   about   half   of   its   excess   risk   while   a   county   with   relatively   low  mobility  

retained  nearly  all  of  it  (95%).  

More   recently,   Tiefelsdorf   (2007)   incorporated   an   explicit   spatial   model   of  

prostate   cancer   rates  by  State  Economic  Area   (SEA)   to   correct   for  migration   (which   is  

assumed   to   be   non-­‐differential   with   respect   to   health).   He   finds   that   an   unadjusted  

model   trying   to  predict  prostate   cancer   rates  by   SEA   finds   radon   levels   to  be   a   strong  

explanatory   variable   but   this   effect   disappears   when   the   data   are   corrected   for  

migration.    

 

2.4  Summary  of  health-­selective  migration  

Overall,   though   much   of   the   evidence   has   weaknesses,   it   appears   that   young  

migrants  are  more  likely  to  be  healthy  while  the  evidence  for  health-­‐selective  migration  

in   older   people   is   mixed.   Three   studies   showed   evidence   of   a   non-­‐linear   relationship  

between   health   and   migration   with   two   finding   that   migration   peaks   at   poor   and  

excellent   health   and   another   finding   it   peaks   at   intermediate   health.   There   is   also  

evidence   that   the   distance   migrated   may   be   qualitatively   related   to   health.   Short-­‐

distance   migrants   may   be   in   poorer   health   than   non-­‐migrants   and   long-­‐distance  

migrants  may  be   in  better  health.  These   conclusions  must   remain   tentative  due   to   the  

numerous   problems   with   these   studies   including   model   selection,   model  

misspecification,   use   of   indirect   standardization,   lack   of   variance   estimates   and   an  

inability  to  properly  order  the  health  and  migration  events.  

Page 40: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  39  

The  majority  of  the  research  has  taken  place  in  Europe  and  the  United  States.  Only  

three   studies   of   Canadian   health-­‐selective   migration   have   been   carried   out   to   date  

(Béland,  1984;  Tousignant  et  al.,  1994;  Yiannakoulias  et  al.,  2007).    

 

2.5  Administrative  databases  

  The   use   of   administrative   databases   in   epidemiological   research   is   increasing  

(Leal  &  Laupland,  2009;  Schneeweiss  &  Avorn,  2005;  Tricco  et  al.,  2008),  in  part  because  

they  provide  access  to  large  amounts  of  data  without  the  costs  associated  with  primary  

data   collection.  However,  because  administrative  data  are   collected   for  purposes  other  

than  research  they  can  be  prone  to  misclassification,  and  prevalence  and  other  estimates  

can  substantially  depend  on  the  methods  used  (Ladouceur  et  al.,  2006).  

Administrative  databases  have  been  widely  used   in   rheumatic  diseases.  A  2012  

systematic  review  (Widdifield  et  al.,  2011)  identified  23  studies  that  validated  the  use  of  

administrative   databases   to   identify   rheumatic   diseases.   Sensitivities   ranged   from   20-­‐

100%  and  specificities  ranged  from  74-­‐97.1%  depending  on  the  algorithm  and  database  

used.   Only   two   studies   reported   either   the   sensitivity   or   specificity   of   algorithms   to  

identify   SLE   or   Sjogren’s   cases   in   administrative   databases.   No   other   study   validated  

Sjogren’s   diagnosis   using   administrative   data.   Katz   et   al.   (1997)   compared   34   SLE  

patients   identified   by   chart   review   to   SLE   cases   in   Medicare   physician   claims   data  

identified  as  anyone  who  had  received  at  least  one  SLE  diagnosis  from  a  rheumatologist.  

They   found  a   sensitivity  of  85%  (95%  CI:  73-­‐0.97%).  Bernatsky  et   al.   (2011)  used   the  

same  algorithm  as  in  this  thesis  in  the  Canadian  province  of  Nova  Scotia.  Validation  was  

Page 41: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 40  

done   using   chart   review   of   patients   at   an   arthritis   centre   in   Halifax.   A   sensitivity   of  

98.2%  (95%  CI  95.5-­‐99.3%)  and  a  specificity  of  72.5%  (95%  CI  68.7-­‐75.9%)  were  found.  

The   low   specificity  may   have   due   to   the   control   group   that   consisted   of   patients  with  

similar   rheumatic   diseases   at   the   same   arthritis   centre.   Bernatsky   et   al.   also   used   the  

same  methodology  for  Sjogren’s  patients  and  found  a  sensitivity  of  95.5%  (95%  CI  88.8-­‐

98.2)  and  a  specificity  of  95.8%  (95%  CI  94.1-­‐97.0).      However,  even  a  specificity  near  

96%  could  mean  that  the  majority  of  subjects  identified  as  having  the  condition  will  not  

in  fact  have  it.    For  example,  if  a  condition  occurs  in  1%  of  the  population,  even  if  all  such  

cases  are   found,  a   specificity  of  95%  means   that   there  will  be   five   times  as  many  non-­‐

cases   as   true   cases   in   the   supposedly   case   population.     This   obviously   casts   a   large  

shadow  of  doubt  over  any   research  using  administrative  databases   that  has  either  not  

adjusted   for   this   misclassification   or   verified   that   sensitivity   and   specificity   are   high  

enough  to  avoid  large  misclassification  errors.  

This   literature  demonstrates   that   for  SLE  and  Sjogren’s,  administrative  data  can  

be   prone   to   misclassification.   Therefore,   it   must   be   kept   in   mind   that   any   study  

identifying   SLE  or   Sjogren’s  patients  using   administrative  data,   including   this   one,  will  

incur  some  level  of  misclassification.  

 

2.6  SLE,  Sjogren’s  and  health-­selective  migration  

Health-­‐selective  migration  has  never  been   investigated  specifically   in  either  SLE  

or  Sjogren’s.  Only  one  study,  to  our  knowledge,  has  looked  at  health-­‐selective  migration  

in  patients  with  rheumatic  diseases,  finding  that  these  patients  were  less  likely  to  move  

Page 42: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  41  

out   of   the   city   than   healthy   controls.   Some   of   the   previously   mentioned   studies   that  

looked  at  disabilities  that  could  be  applied  to  SLE  and  Sjogren’s,  given  that  these  diseases  

are  associated  with  increases  in  disabilities  (P. Katz et al., 2008; Strömbeck et al., 2003).  

Given  that  many  studies  of  prevalence,  incidence  or  mortality  in  rheumatic  diseases  rely  

on   geographical   patterns   abstracted   from   administrative   databases   (Alamanos   et   al.,  

2003;  Andrianakos  et  al.,  2003;  Barnabe  et  al.,  2012;  Gómez-­‐Rubio  &  López-­‐Quílez,  2010;  

Hart   et   al.,   2009;   Kurahara   et   al.,   2007;   Labrecque,   Joseph,   et   al.,   2010;   Labrecque,  

Smargiassi,  et  al.,  2010;  Walsh  &  Gilchrist,  2006;  Ward,  2010)  or  from  sources  prone  to  

health-­‐selective  migration  (Alamanos  et  al.,  2003;  Andrianakos  et  al.,  2003;  Gómez-­‐Rubio  

&  López-­‐Quílez,  2010;  Hart  et  al.,  2009;  Walsh  &  Gilchrist,  2006),  it  is  important  to  have  

an  understanding  of  how  this  process  operates  in  patients  with  rheumatic  diseases.  For  

instance,   Walsh   and   Gilchrist   (2006)   investigate   clustering   at   the   county   level   of   SLE  

mortality   in   the   lower   48   United   States   finding   that   high   SLE   mortality   clusters   are  

related  to  poverty,  Hispanic  ethnicity  and  exposure  to  ultraviolet  B  radiation.  Articles  in  

this   literature  review  provide  evidence   for   the  possibility  of  health-­‐selective  migration,  

particularly  among  the  elderly,  explaining  at  least  part  of  this  clustering  (Gober  &  Zonn,  

1983;   Litwak   &   Longino,   1987;   Sergeant   &   Ekerdt,   2008).   Similarly,   any   urban/rural  

differences   in   rates   of   rheumatic   diseases   may   be   explained   by   increased   migration  

towards  cities  or  a  decreased  tendency  to  move  out  of  them  (Moorin  et  al.,  2006).  

Health-­‐selective   migration   in   either   SLE   or   Sjogren’s   patients   may   also   be  

indicative  of  gaps  in  access  to  health  care.  Increased  migration  may  be  expected  in  these  

diseases   given   Canadian   data   documenting   reduced   access   to   healthcare   specialists  

Page 43: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 42  

among   the   rural   elderly   (Allan   &   Cloutier-­‐Fisher,   2006;   McDonald   &   Conde,   2010).  

Conversely,  reduced  migration  among  SLE  or  Sjogren’s  patients  may  suggest  an  inability  

to   migrate   when   it   would   be   required   or   desired   for   other   reasons.   However,   more  

detailed   data   would   be   required   to   tease   apart   conflicting   reasons   for   migrating   or  

staying  in  the  same  location.  

Page 44: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  43  

3  Methodology  

 

 

This  chapter  is  divided  into  the  three  sections.  The  first  section  is  a  brief  overview  

of   the   methods   and   how   they   relate   to   the   specific   objectives.   The   second   section  

describes  the  data  sources  used.  The  third  section  describes  the  three  analyses  that  were  

conducted  and  the  statistical  methods  used.    The  McGill  University  Ethics  Review  Board  

approved  this  study.    

3.1  Overview    

The  three  specific  objectives,  as  described   in   the   introduction  to   this   thesis,  will  

be  achieved  by  comparing  between-­‐forward  sortation  area  (FSA;  the  first  three  digits  of  

the  postal  code)  or  regional  migration  rates  among  SLE  and  Sjogren’s  patients  to  rates  in  

two   controls   groups.   The   estimates   will   be   derived   from   physician   billing   and  

hospitalization  databases.    

The  first  objective  will  be  met  by  comparing  between-­‐FSA  migration  rates  in  post-­‐

diagnosis  SLE  and  Sjogren’s  patients  to  an  age  and  sex  matched  sample  from  the  general  

population  of  Montreal,  Quebec.  This  will  estimate  health-­‐selective  migration  in  SLE  and  

Sjogren’s   patients   living   in   Montreal   relative   to   the   general   population.   The   second  

objective  will  compare  pre-­‐  and  post-­‐diagnosis  between-­‐FSA  migration  rates  in  SLE  and  

Sjogren’s  patients  from  the  entire  province  of  Quebec.  This  analysis  will  estimate  health-­‐

selective   migration   as   relative   to   pre-­‐SLE   and   pre-­‐Sjogren’s   populations.   The   third  

Page 45: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 44  

objective  will  be  met  using  the  same  comparison  as  the  second  objective  but  substituting  

regional  migration  rates  for  between-­‐FSA  migration  rates.  This  last  analysis  will  estimate  

health-­‐selective  migration  in  SLE  and  Sjogren’s  patients  at  regional  scales  and  will  allow  

comparison  with  between-­‐FSA  health-­‐selective  migration.  

3.2  Data  sources  

The  data  in  this  thesis  come  from  a  variety  of  sources.  Administrative  databases  

are   used   to   measure  migration   events,   to   determine   SLE   and   Sjogren’s   diagnosis   and  

diagnosis   date,   and   demographic   data.   Other   sources   were   used   for   possible  

confounders,  as  described  below.  

 

3.2.1  Administrative  databases    

Quebec   is   a   province   of   7.5   million   people   where   healthcare   is   universally  

available   to   residents,   though  a  3  month  waiting  period   is   required  before  newcomers  

can  enrol.      Administrative  databases  are  kept  in  the  province  recording  all  medical  acts.    

This  study  used  two  of  these  Quebec  administrative  data  sources:  the  Régie  d’assurance  

maladie  du  Québec  (RAMQ)  and  the  Ministry  of  Health’s  Maintenance  et  Exploitation  des  

Données  pour  l’Étude  de  la  Clientèle  Hospitalière  (MEDECHO).    

The   RAMQ   is   a   physician-­‐billing   database   containing   information   on   all  

provincially  reimbursed  physician  services  including  the  primary  diagnosis  assigned  by  

the   physician   (or   his/her   staff)   for   the   patient   (identified   by   the   unique   provincial  

beneficiary  number)  and  the  place  of  treatment.  Each  physician  has  a  unique  identifier,  

Page 46: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  45  

and   the  RAMQ  physician  database   contains   information   about   the   physician,   including  

his  or  her  specialty.  A  diagnostic  code  is  required  for  the  physician  to  receive  payment,  

but  the  code  is  not  validated  and  only  one  code  is  allowed  per  visit.  During  the  years  in  

question  for  this  thesis,  the  RAMQ  used  diagnostic  codes  according  to  the  ninth  version  

of  the  International  Classification  of  Diseases  (ICD-­‐9).  Each  claim  in  the  billing  database  

contains   information   on   the   ICD-­‐9   code   associated   with   the   visit,   the   date   of   the  

encounter,   the   FSA   of   residence   of   the   beneficiary   and   codes   allowing   the   claims  

database  to  be  linked  to  the  RAMQ  physician  database  and  a  beneficiaries  demographics  

database.   It   is   important   to   note   that   FSA  of   residence   of   a   beneficiary   (i.e.   patient)   is  

only   available   in   years   where   a   physician   bills   RAMQ   for   an   act   performed   on   that  

beneficiary.   Another   important   note   is   that   the   FSA   of   residence   associated  with   each  

visit  is  the  FSA  of  residence  on  July  1st  the  year  of  the  visit  and  not  the  FSA  of  residence  

the  date  the  visit  was  made.    Therefore,  only  one  FSA  is  recorded  per  individual  per  year.  

The   beneficiaries’   demographics   database   contains   information   about   all   Quebec  

beneficiaries   of   the   provincial   health   plan   including   age,   sex   and   date   of   death.   These  

databases  can  further  be  linked  to  the  provincial  hospitalization  database,  described  next  

(as  well  as,  for  those  residents  who  qualify  for  provincial  drug  insurance,  to  a  database  of  

all  drugs  dispensed  by  pharmacies).    

  The   province-­‐wide   hospital   database,   MEDECHO,   contains   admission   date,  

diagnosis   date,   a   primary   diagnosis   and   up   to   15   secondary   diagnoses.   However,   the  

diagnoses   in   MEDECHO   are   abstracted   from   a   patient’s   chart   by   medical   clerks   and  

therefore  not  necessarily  identical  to  the  diagnoses  recorded  by  the  RAMQ.  

Page 47: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 46  

  Data  were   obtained   from   two   different   cohorts,   one   composed   of   residents   for  

whom  a  physician  or  hospital   submitted   a   claim  or   a   systemic   autoimmune   rheumatic  

disease  and  a  second  (serving  as  a  proxy  for  general-­‐population  that  uses  the  healthcare  

system)  was  formed  of  all  Quebec  residents  for  whom  a  physician  or  hospital  submitted  

a   claim   for   any   flu-­‐like   illness   or   symptom   (described   in   detail   shortly)   in   the   census  

metropolitan   area   of   Montreal,   Quebec.   A   true   random   sample   of   RAMQ   beneficiaries  

was   not   available,   therefore   the   rationale   for   using   this   as   a   proxy   for   the   general-­‐

population   was   that,   aside   from   chronic   diseases,   this   is   one   of   the   most   common  

reasons,  in  the  general  population,  for  seeking  medical  care,  and  would  presumably  form  

a  cohort  that  resembled  the  general  population  that  accesses  medical  services  in  terms  of  

age,  sex,  and  other  unmeasured  attributes.    

 

3.2.2  Systemic  autoimmune  rheumatic  diseases  cohort  

The   systemic   autoimmune   rheumatic  diseases   cohort   that  was   studied   included  

all  individuals  who  registered  at  least  one  International  Classification  of  Diseases  code  of  

710   between   1989   and   2003   in   either   the   RAMQ   physician   billing   or   MEDECHO  

hospitalization   databases.   These   databases   were   linked   (using   the   unique   provincial  

beneficiary  number)  to  each  other  and  also  to  the  physician  and  provincial  beneficiaries  

RAMQ   databases   allowing   all   patient,   physician   and   diagnosis   information   associated  

with  a  claim  to  be  combined.  For  all  patients  who  had  at  least  one  physician  or  hospital  

visit   for   an   ICD-­‐9   code   of   710,   information   was   available   for   all   medical   contact  

Page 48: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  47  

(physician  visits  and  hospitalizations)  from  January  1,  1989  to  the  earliest  date  between  

the  date  of  death,  emigration  out  of  the  province  or  December  31,  2003.    

Individuals   in   the   systemic   autoimmune   rheumatic   diseases   cohort   were  

considered  SLE  or  Sjogren’s  patients   if   they   fulfilled  at   least  one  of   the   following   three  

criteria:  

1)  Two  or  more  billing  diagnoses  by  any  physician  at  least  eight  weeks  apart  but  

within  two  years  

2)  At  least  one  billing  diagnosis  by  a  rheumatologist  

3)  At  least  one  hospital  discharge  diagnosis  

The   diagnoses   in   question   for   SLE   and   Sjogren’s   were   ICD-­‐9   710.0   and   710.2  

respectively.  The  diagnosis  date  was  considered  to  be  the  first  date  at  least  one  of  these  

three  criteria  was  satisfied.    

Regional  migration  rates  by  five  year  age  groups  in  this  cohort  will  be  compared  

to  estimates  of  regional  migration  from  the  Quebec  government  (Institut  de  la  statistique  

Québec,   2007)   to   check   the   validity   of   using   administrative   data   to   detect   migration  

events.      

 

3.2.3  Montreal  CMA  comparison  cohort  

To   compare  migration   in   SLE   and   Sjogren’s   patients   to   the   general   population,  we  

drew   from   a   general   population   cohort   that   contained   approximately   86%   of   the  

population  of  the  census  metropolitan  area  (CMA)  of  Montreal.  Inclusion  criteria  in  this  

cohort  were:  

Page 49: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 48  

1) Having  lived  at  least  one  year  in  the  census  metropolitan  area  (CMA)  of  Montreal  

(a  list  of  FSAs  considered  part  of  the  CMA  of  Montreal  can  be  found  in  Appendix  

A)  between  1999  and  2003,  inclusively    

2) Having  had  medical  contact  (physician  billing  or  hospitalization)  for  at   least  one  

flu-­‐like  illness  or  symptom  in  RAMQ  or  MEDECHO  during  the  same  time  period.    

Among  the  conditions  considered  under   flu-­‐like   illness  or  symptom  are:   the   flu,   the  

common   cold,   pneumonia,   bronchitis,   sinusitis,   laryngitis,   pharyngitis   and   codes   for  

specific   symptoms   such  as   fever,   chest  pain,   sore   throat   and   cough   (a   full   list   of   ICD-­‐9  

codes  included  in  this  cohort  for  both  RAMQ  and  MEDECHO  databases  can  also  be  found  

in  the  Appendix  A).  All  RAMQ  and  MEDECHO  records  during  the  entire  time  period  were  

obtained  for   individuals  who  satisfied  both  criteria.    The  age  and  sex  distribution  in  all  

people  living  in  the  CMA  of  Montreal  in  2001  in  this  cohort  will  be  compared  to  the  2001  

Canadian  Census  data  population  estimates   for   the  CMA  of  Montreal   to  determine  how  

representative  the  sample  is  (Statistics  Canada,  2001a).    

The   two   datasets   (Sjogren’s/SLE   cohort   and   the   general-­‐population   cohort)   were  

obtained   from   the   province   in   two   separate   data   requests,   with   the   original   unique  

provincial  identifier  removed.  We  could,  therefore,  not  distinguish  to  what  extent  some  

patients  appeared  in  both.  However,  given  the  relatively  small  number  SLE  and  Sjogren’s  

patients   in  Montreal   and   the   large   number   of   people   in   the  Montreal   FSA   comparison  

cohort,   it   is  unlikely   that  more   than  70  SLE  or  Sjogren’s  patients  appear   in   the  50,000  

person  sample.  

 

Page 50: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  49  

3.2.4  Outcomes  

Two   outcomes   will   be   used   in   the   following   analysis:   between-­‐FSA  migration   and  

regional  migration.  A  between-­‐FSA  migration  will  be  considered  a  year  when  the  FSA  of  

residence  has  changed  relative  to  the  FSA  of  residence  of  the  previous  year.  For  example,  

if  the  FSA  recorded  for  an  individual  in  1990  differed  from  the  FSA  recorded  in  1989,  this  

was   considered   a  migration   event.   If   the   two   FSAs  were   the   same,   the   individual  was  

considered  not   to  have  migrated.  Therefore,  within-­‐FSA  migration,   for   the  purposes  of  

this   thesis,   will   not   be   considered   a   migration   event.   Also,   migration   can   only   be  

measured  from  year  to  year  and  not  on  shorter  intervals.  An  individual  migrating  out  of  

their  FSA  and  back  within  one  year  cannot  be  measured.  Another  limitation  this  poses  is  

that  no  specific  date  can  be  attributed  to  the  move  other  than  to  say  it  occurred  at  some  

point  between  July  1st  of  the  previous  and  current  year.  A  consequence  of  this  is  that  it  is  

not   possible   to   tell   whether   a   migration   event   occurring   the   same   year   as   an   SLE   or  

Sjogren’s   diagnosis   happened   before   or   after   the   diagnosis.   Therefore   observations  

occurring  the  year  of  diagnosis  were  excluded.    

The   number   of   migration   events   in   1992   in   the   systemic   autoimmune   rheumatic  

diseases   cohort  was   anomalously   low   in   comparison   to   other   years   (less   than   2%   the  

number  of  migration  events  in  other  years).  The  primary  analyses  were  therefore  done  

excluding  this  year  the  analysis.  Sensitivity  analyses  were  also  done  using  only  data  from  

1993   to   2003,   and   the   results   were   unchanged,   so   only   the   results   from   the   primary  

analyses  are  presented.  

 

Page 51: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 50  

In   the   third  part  of   the  analysis,   regional  migration  will  be  used  as  an  outcome.  

Regional  migration  will  be  defined  similarly  to  between-­‐FSA  migration  but  considering  a  

change  in  administrative  region  instead  of  a  change  in  FSA.  

 

3.2.5  Potential  confounders  

Socioeconomic  status  variables  were  obtained  at  the  FSA  level  from  the  Canadian  

Census,   which   is   performed   every   five   years.   Data   were   available   from   the   2001  

Canadian   census   on  percent   of   households   that   are   renting,   employment   rate,   average  

income   and   a   composite   education   variable   (calculated   as   the   proportion   of   the  

population   with   a   university   degree,   college   degree   or   diploma  minus   the   proportion  

who   have   not   graduated   high   school)   (Statistics   Canada,   2001b).   These   variables   are  

ecological  (at  the  FSA  level)  and  therefore  can  be  both  seen  as  a  proxy  for  the  true  value  

for   an   individual   and   as   a  measure   of   the   neighbourhood   average.   The   SES   covariates  

associated  with  an  observation  are  those  from  the  FSA  from  the  previous  year  in  order  to  

measure  the  association  between  the  previous  location  and  residential  mobility,  not  the  

destination.   For   example,   if   a   move   was   recorded   in   1990   (i.e.   the   FSA   in   1990   of   a  

beneficiary   differed   from   the   FSA   in   1989),   SES   covariates   associated  with   that  move  

would  be  those  from  the  FSA  in  1989  and  not  1990.  

An  FSA  was  considered  in  a  CMA  if  over  50%  of  the  postal  codes  within  the  FSA  

fell   within   one   of   the   six   CMAs   in   Quebec   (Montreal,   Quebec   City,   Sherbrooke,  

Ottawa/Gatineau,   Trois   Rivières   and   the   Saguenay).   This   was   done   using   postal   code  

conversion   files   available   through   the   University   of   Toronto   (Statistics   Canada,   2002).  

Page 52: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  51  

Data   on   which   FSAs   had   rheumatologists   were   obtained   from   a   database   of  

rheumatologists   according   to   the   Quebec   Rheumatologist   Association   (Association   des  

médecins  rhumatologues  du  Québec,  2010).  

 

3.3  Statistical  methods  

The  statistical  methods  sections  will   include  a  description  of  the  descriptive  and  

bivariate   analyses   performed   followed   by   a   detailed   description   of   the   three   main  

comparisons.  

Table  1  describes  the  three  main  comparisons  carried  out.  Each  differs  in  terms  of  

control   group,   years   included,   geographic   extent   and   geographic   scale   of  migration   in  

order   to   address   specifically   each   of   the   three   objectives   of   this   thesis.   In   all   three  

comparisons,   one   observation   will   be   considered   one   person-­‐year   where   the   FSA   of  

residence   of   the   current   year   and   of   the   previous   year   are   known.   Therefore,   each  

individual   can   contribute   multiple   observations:   one   for   each   year   a   migration   could  

have  been  measured.  

 

Table  1-­Summary  of  the  three  comparisons  performed  

Comparison Patients Controls Years Outcome

1) General population comparison

Subset that meet geographic criteria for Montreal CMA comparison cohort

Age and sex frequency-matched from Montreal CMA comparison cohort

1999-2003 Between FSA migration

2) Pre-diagnosis comparison From entire province

People with SLE or Sjogren's before their diagnosis

1990-2003 (except 1992)

Between FSA migration

3) Regional migration comparison From entire province

People with SLE or Sjogren's before their diagnosis

1990-2003 (except 1992)

Interregional migration

 

Page 53: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 52  

3.3.1  Descriptive  and  bivariate  analyses  

The   number   of   moves   and   migration   rate   (with   95%   CI)   are   calculated   for  

controls,   SLE  and  Sjogren’s  patients  within   strata  of  disease  duration  and   ten-­‐year  age  

group  to  ensure  positivity  in  combinations  of  age  and  disease  duration.    

Regional   migration   rates   by   age   in   SLE   and   Sjogren’s   patients   pre-­‐   and   post-­‐

diagnosis  are  compared  to  regional  migration  rates  in  the  Quebec  population  (Institut  de  

la   statistique   Québec,   2007)   allowing   the   assessment   of   validity   of   using   the   RAMQ  

billing  database  to  estimate  migration  rates.  

In  order  to  assess  the  validity  of  using  the  Montreal  CMA  comparison  cohort  as  a  

proxy  for  the  general  population,  the  overall  population  and  age  distribution  of  all  people  

from  the  Montreal  CMA  comparison  cohort   in  2001   is  compared   to   the  2001  Canadian  

census  estimates  for  the  CMA  of  Montreal  (Statistics  Canada,  2001a).  

Finally,  descriptive   statistics   (mean  and   standard  deviation)   are   then   calculated  

for   each   variable   in   the   general   population   comparison   and   pre-­‐diagnosis   comparison  

datasets.   Descriptive   statistics   for   the   regional   migration   analysis   are   not   necessary  

because   it   uses   the   same   data   set   as   the   pre-­‐diagnosis   comparison.   Bivariate   analyses  

were  also  conducted  using  hierarchical   logistic  regression  for  the  same  set  of  variables  

described  previously  in  section  3.2.  Hierarchical  logistic  regression  was  used  to  account  

for  multiple   observations  within   individuals.   In   these   analyses,   the   crude   relationship  

between  disease  duration  and  migration  was  estimated  with   the  disease  main  effect   in  

the   model.   Also,   the   relationship   between   age   and   migration   was   estimated   with   a  

quadratic  age  term.  

Page 54: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  53  

 

3.3.2  General  population  comparison  

The  first  part  of  the  health-­‐selective  migration  analysis,  which  will  henceforth  be  

referred   to   as   the   ‘general  population   comparison’,  will   compare  migration   rates   in   all  

Sjogren’s  and  SLE  patients  that  lived  in  the  CMA  of  Montreal  from  1999-­‐2003  to  50,000  

age   (by   five   year   age   groups)   and   sex   frequency-­‐matched   subjects   from   the   general  

population   cohort   (35,000   matched   to   SLE   patients   and   15,000   matched   to   Sjogren’s  

patients)  using  hierarchical  logistic  regression.  All  observations  in  people  younger  than  

25   years   were   excluded   to   avoid   measuring   migration   associated   with   parental  

migration.  The  date  of  diagnosis  of  SLE  and  Sjogren’s  patients  used  to  calculate  disease  

duration  will  be   calculated  using   the   full   range  of  data   from   the   systemic  autoimmune  

rheumatic   diseases   cohort   (1989-­‐2003)   but   only   observations   after   or   including   1999  

are   included   in   the   regression.   Therefore,   up   to   five   observations,   from  1999   to   2003,  

could  be  included  for  each  individual  in  this  comparison.  

Independent  variables   included   indicator  variables   for  SLE  and   for  Sjogren’s,   as  

well   as   disease   duration   for   each,   age   centered   at   50   years,   and   sex.   The   FSA-­‐level  

variables  included  were  the  presence  of  a  rheumatologist  in  the  FSA,  whether  or  not  the  

FSA   was   in   a   metropolitan   area,   and   four   census   variables:   employment   level   (%),  

average   income   (in   thousands   of   dollars),   average   educational   attainment   (%)   and  

proportion   of   households   that   are   renters   (%).   A   random   intercept   by   individual  was  

required   to   account   for   within-­‐individual   correlations   between   observations.   Three  

models   were   considered,   as   detailed   in   appendix   B.   The   simplest   was   a   model   that  

Page 55: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 54  

included   a   random   intercept,   Sjogren’s   and   SLE   as   indicator   variables,   sex,   age   and   a  

quadratic  age  term.  The  quadratic  age  term  was  added  because  both  provincial  estimates  

of   migration   by   age   group   are   non-­‐linear.   The   second   model   considered   included   all  

terms  in  model  1  as  well  as  an  SLE  by  age  interaction,  Sjogren’s  by  age  interaction,  SLE  

duration  and  Sjogren’s  duration.  The  last  model  included  all  terms  in  model  2  and  all  the  

potential   confounders   listed   in   section   3.2:   whether   a   person   lived   in   an   FSA   with   a  

rheumatologist,   lived   in   a   census   metropolitan   area,   and   FSA-­‐level   average   income,  

percent  of  households  that  are  renters,  a  composite  education  variable,  and  employment  

rate.  

The  presence  of  age  by  disease  interaction  and  disease  duration  was  assessed  by  

whether  its  inclusion  caused  important  differences  in  ORs  of  migration  by  strata  of  age  

and  disease  duration.  An  important  difference  was  considered  to  be  0.10.  Assessment  of  

confounding   was   done   by   examining   correlations   between   possible   confounders   and  

SLE,   Sjogren’s   and   the   outcome   as   well   as   using   a   10%   change   of   estimate   criteria  

(Mickey  &  Greenland,  1989)  in  the  OR.  Odds  ratios  (OR)  and  predicted  probabilities,  with  

95%  credible   intervals   (CrI)  were  calculated   from  the  selected  models.  Most  migration  

rates  in  these  analyses  are  below  ten  percent  so  the  OR  reasonably  approximates  the  risk  

ratio   (Sinclair   &   Bracken,   1994).   Because   the   model   is   hierarchical,   predicted  

probabilities  were  calculated  using  the  mean  and  standard  deviation  of  the  overall  mean  

intercept.    

Page 56: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  55  

For  patients  in  the  systemic  autoimmune  rheumatic  diseases  cohort  that  did  not  

have   medical   contact   in   a   specific   year,   the   FSA   was   unknown   for   that   year.   These  

observations  made  up  less  than  4%  of  observations  in  all  analyses  and  were  excluded.      

All   analyses   were   carried   out   using   Bayesian   inference.   Bayesian   analyses  

calculate   a   joint   posterior   probability   distribution   across   all   parameters   of   interest   by  

updating  a  prior  joint  density  with  the  likelihood  function  via  Bayes’  Theorem.  The  prior  

probabilities  used  in  this  analysis  were  all  diffuse,  meaning  that  they  do  not  contribute  

meaningfully   to   the   posterior   probabilities   of   the   parameters.   Ninety-­‐five   percent  

credible   intervals   for   the   parameters   are   calculated   by   taking   the   2.5%   and   97.5%  

percentiles   of   the   posterior   probability   distribution   for   each   parameter.   Credible  

intervals   can   be   interpreted   as   an   interval   with   a   95%   probability   of   containing   the  

parameter   of   interest   as   opposed   to   confidence   intervals   that   rely   on   long   run  

probabilities   (i.e.   the  method   used   to   calculated   a   confidence   interval  will   contain   the  

parameter   of   interest   95%  of   the   time,   through   application   to   different   problems).   All  

analyses  were   carried   out   using  WinBUGS   1.43   (Lunn   et   al.   2000;   see   appendix   C   for  

exact  WinBUGS  programs  and  the  parameters  of  the  priors  used.).  

This  comparison  gives  insight  on  health-­‐selective  migration  in  SLE  and  Sjogren’s  

patients  relative  to  the  general  population.  

 

3.3.3  Pre-­diagnosis  comparison  

The   second   comparison,   which   will   be   referred   to   as   the   ‘pre-­‐diagnosis  

comparison’,   compares  between-­‐FSA  migration   in  SLE  and  Sjogren’s  patients  province-­‐

Page 57: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 56  

wide  with  migration  rates  in  the  same  population  of  patients  before  their  diagnosis  using  

hierarchical  logistic  regression.  Observations  were  included  from  the  entire  range  of  the  

dataset   from  1990   to  2003   (migration  events   could  not  be   recorded   for  1989  because  

there  was  no  previous  FSA   for  comparison).  Observations  of  people   less   than  25  years  

old  were  excluded.  

The   hierarchical   logistic   regression   and  model   selection  was   carried   out   in   the  

same  manner  as  in  the  general  population  comparison.  

This   second   comparison   will   be   less   prone   to   individual   level   unmeasured  

confounders  because   it  uses  migration   rates   in  SLE  and  Sjogren’s  patients  before   their  

diagnosis.   However,   the   diagnosis   date   is   a   somewhat   artificial   distinction   between  

healthy   and   SLE   or   Sjogren’s   person-­‐time.   Patients   experience   symptoms   before   their  

diagnosis,   therefore,   the   control   group   in   this   comparison   is   likely   less  healthy   than   in  

the  ‘general  population  comparison’.  This  analysis  will  therefore  measure  how  migration  

rates  can  change  following  diagnosis.      

 

3.3.4  Regional  migration  comparison  

The   final   analysis,   the   ‘regional  migration   comparison’   uses   the   same   data   and  

criteria   as   the   pre-­‐diagnosis   comparison   except   the   outcome   is   defined   as   a   move  

between  Quebec’s  administrative  regions  (of  which  there  are  17),  as  opposed  to  between  

FSAs   (which   are   generally   much   smaller   than   the   administrative   regions).   The  

hierarchical  logistic  regression  and  model  selection  was  carried  out  in  the  same  manner  

as  the  other  two  comparisons.  

Page 58: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  57  

This  comparison  will  investigate  health-­‐selective  migration  over  larger  distances.    

 

3.3.5  A  note  on  comparisons  

It  is  important  to  recall  that  these  control  groups  differ  in  key  ways  other  than  the  

descriptors   in   their   names.   Firstly,   patients   from   general   population   cohort   are   likely  

healthier  than  the  SLE  or  Sjogren’s  patients  even  previous  to  their  diagnosis  when  they  

may   suffer   from   pre-­‐clinical   symptoms.   That   is,   the  majority   of   those   seeking  medical  

care   for   a   flu-­‐like   illness   would   presumably   be   those   without   major   chronic   medical  

disease   (the   distribution   would   likely   match   the   population   that   uses   medical   care).  

Secondly,   the   general   population   analysis   only   includes   people   who   lived   in   the  

metropolitan   area   of   Montreal   between   1999   and   2003   whereas   the   pre-­‐diagnosis  

control  analysis  includes  Sjogren’s  and  SLE  patients  from  the  entire  province.  Lastly,  the  

general  population  control  analysis  only  includes  person-­‐time  from  1999-­‐2003  whereas  

the   pre-­‐diagnosis   control   analysis   includes   person-­‐time   from   1989-­‐2003   (excluding  

1992).  

Page 59: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 58  

Page 60: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  59  

 4  Results      

 

This   chapter   begins   with   a   section   describing   the   descriptive   and   bivariate  

analyses  including  a  comparison  of  regional  migration  rates  in  SLE  and  Sjogren’s  patients  

to   provincial   estimates   of   regional   migration   and   a   comparison   of   the   Montreal   CMA  

comparison  cohort  to  the  general  population  of  the  CMA  of  Montreal.  The  following  three  

sections  will  describe  the  results  from  each  of  the  three  comparisons.  

 4.1  Descriptive  and  bivariate  analyses         The   sample   sizes   in   both   datasets   used   in   this   thesis   can   be   found   in   Table   2.  

People  without  at  least  one  observation  either  because  the  FSA  was  missing  or  they  did  

not   have   two   consecutive   years   where   their   FSA   was   known   are   considered   to   have  

incomplete  data  and  were  excluded.  The  general  population  comparison  included  2,713  

people  with  SLE   (85%   female  with  an  average  age  42.7  years   at   SLE  diagnosis),   1,114  

people   with   Sjogren’s   (90%   female   with   an   average   age   52.2   years   at   Sjogren’s  

diagnosis),  and  48,608  (86%  female)  age  and  sex  frequency  matched  general  population  

Table  2—Sample  sizes  for  each  analysis  

Pre-diagnosis comparison*

General population comparison**

SLE Sjogren's SLE Sjogren's Control Met criteria 5182 2605 3184 1306 50000 Incomplete data 102 60 471 192 1392 In analysis 5080 2545 2713 1114 48608 *  total  number  of  people  7010  because  of  overlap,  **  total  number  of  people  52286  because  of  overlap    

Page 61: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 60  

controls.   The   pre-­‐diagnosis   analysis   included   5,080   with   SLE   (85%   female   with   an  

average  age  of  42.6  years  at  SLE  diagnosis)  and  2,545  with  Sjogren’s  (89%  female  with  

an   average   age   of   55.7   years   at   Sjogren’s   diagnosis)   people   for   a   total   of   7,010   (615  

people  met  the  criteria  for  both  SLE  and  Sjogren’s).  Table  3  displays  the  distribution  of  

person-­‐time   and  moves   in   the   general   population   comparison   by   ten-­‐year   age   groups  

and   five-­‐year   periods   of   disease  duration.   In   Sjogren’s   and   SLE  patients   as  well   as   the  

general  population,  moves  are  concentrated  below  the  age  of  85  and  in  disease  duration  

from  0-­‐9  years.    

Table  4A  displays   the  distribution  of  person-­‐time,  regional  moves  and  between-­‐

FSA   moves   in   the   pre-­‐diagnosis   comparison,   by   ten-­‐year   age   groups   and   five-­‐year  

periods   of   disease   duration.   As   with   the   general   population   comparison,   moves   are  

concentrated  below  the  age  of  85  and  between  disease  durations  of  0-­‐9  years.  Regional  

moves  are  particularly  sparse  at  disease  durations  greater  than  ten  years  (Table  4B).    

In   order   to   assess   the   validity   of   using   administrative   data   to   detect  migration  

events,   regional   migration   rates   pre-­‐   and   post-­‐Sjogren’s   and   SLE   diagnosis   were  

compared   to   provincial   estimates   (Figure   1).   Pre-­‐diagnosis   migration   rates   closely  

matched  rates  observed  provincially  among  people  less  than  45  years  old.  After  45,  the  

migration  rates  among  pre-­‐diagnosis  individuals  are  lower  than  the  provincial  estimates,  

tended  to  be  about  5  to  7  moves  per  1,000  people  lower.  The  crude  regional  migration  

rates   in   Sjogren’s   and   SLE   patients   differ   slightly   from   from   provincial   migration  

patterns.  The  difference  is  largest  in  the  25-­‐34  year  age  group  where  migration  is  about  

50%   lower   than   provincial   estimates.   This   difference   decreases   in   35-­‐44   year   olds

Page 62: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  61  

Page 63: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 62  

Page 64: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  63  

 and  increases  again  among  45-­‐54  year  olds.  In  people  over  60,  the  difference  in  regional  

migration  is  relatively  small.    

  In   order   to   get   an   idea   of   the   representativeness   of   the   entire   Montreal   CMA  

comparison  cohort  (not   just  the  50,000  in  the  frequency  matched  sample),   the  age  and  

sex  distribution  is  compared  to  population  estimates  from  the  2001  Canadian  Census  of  

the  CMA  of  Montreal.  Overall,  the  population  in  the  Montreal  CMA  comparison  cohort  is  

86%  the   total  population  of  Montreal   (Table  5).  By  sex,   the  Montreal  CMA  comparison  

 Figure  1—Regional  migration  rates  among  SLE  and  Sjogren’s  patients  compared  to  provincial  estimates    

30 40 50 60 70 80

010

2030

4050

60

Age (years)

Reg

iona

l mig

ratio

n ra

te (m

oves

/100

0 pe

ople

)

Provincial estimatePre-diagnosisSLESjogren's

Page 65: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 64  

cohort  captures  between  86%  and  94%  of  women  in  the  Canadian  Census  and  between  

77%   and   99%   of   men   in   every   twenty-­‐year   age   year   age   group.   The   Montreal   CMA  

comparison  cohort  is  53.1%  female  whereas  the  Canadian  Census  is  51.6%  female.    The  

age   distribution   among   women   in   the   Montreal   CMA   comparison   cohort   is   nearly  

identical   to   the   age   distribution   in   the   2001   Canadian   Census.   The   age   distribution  

among   men   in   the   Montreal   CMA   comparison   cohort   differs   particularly   in   the   age  

groups  below  40.  

  Results  from  the  descriptive  analysis  of  the  independent  variables  can  be  found  in  

Table   6.   In   both   analyses,   Sjogren’s   patients   were   roughly   ten   years   older   than   SLE  

patients  as  would  be  expected,   since  Sjogren’s   tends   to  occur   in  older  people   than  SLE  

 !

Table&5&–&Comparison&between&populations&in&the&full&Montreal&CMA&comparison&cohort&to&the&2001&Canadian&Census.&Proportion&of&the&population&found&in&each&age&group&is&found&in&parentheses&to&allow&comparison&of&age&distribution&by&sex&

Sex Age

Population in Montreal

comparison cohort

Population in Canadian Census

Proportion of the population

captured

0-19 380373 (0.24) 404230 (0.23) 0.94 20-39 449929 (0.29) 511105 (0.29) 0.88 40-59 439614 (0.28) 510155 (0.29) 0.86 60-79 239272 (0.15) 273510 (0.15) 0.87 80+ 64155 (0.04) 69715 (0.04) 0.92

Female

Total 1573343 1768715 0.89

0-19 383531 (0.28) 419850 (0.25) 0.91 20-39 390783 (0.28) 505930 (0.31) 0.77 40-59 389281 (0.28) 483000 (0.29) 0.81 60-79 194410 (0.14) 218520 (0.13) 0.89 80+ 30160 (0.02) 30335 (0.02) 0.99

Male

Total 1388165 1657635 0.84 Entire population 2961508 3426350 0.86

!

Page 66: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  65  

 

Page 67: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 66  

(Gaubitz,  2006).  Of  interest,  the  proportion  of  females  in  the  pre-­‐diagnosis  analysis  was  

lower   in   the  controls   than   in  either  SLE  or  Sjogren’s  patients.  This  may  be  unexpected  

since  they  are  composed  entirely  of  pre-­‐SLE  or  pre-­‐Sjogren’s  patients.  This  result  comes  

about   because   the   proportion   of   females   among   patients  with   no  migration   data   after  

diagnosis   is   approximately   10%   lower   than   the   proportion   of   females   among   patients  

with  migration  data  before  diagnosis.  Percent  renters,  education  level,  employment  rate  

and   income   are   similar   between   all   groups   in   each   analysis.   People   in   the   general  

population  comparison  were  more  likely  to  live  in  a  metropolitan  area  because  this  was  

part   of   the   entry   criteria.   These   values   were   below   100%   because   people   were   only  

required   to   live   in   Montreal   for   one   year   during   this   period.   Controls   in   the   general  

population  comparison  were  less  likely  to  live  in  an  FSA  with  a  rheumatologist.    

Bivariate   logistic   regression   analyses   (Table   7)   suggest   that   SLE   and   Sjogren’s  

both  have  lower  migration  rates  than  the  comparator,  in  all  three  comparisons.  Disease  

duration   in   SLE   patients   is   associated   with   reduced  migration   but   the   relationship   in  

Sjogren’s  patients  is  unclear  (the  95%  credible  intervals  include  a  wide  range  of  values  

on  each  side  of   the  null).  Females  were   less   likely   to  migrate   in   the  general  population  

comparison  but  more  likely  to  migrate  in  the  pre-­‐diagnosis  comparison.  Living  in  a  CMA  

was  associated  with  decreased  odds  of  migration  in  the  general  population  and  regional  

migration  comparisons  and  was  associated  with  increased  odds  of  migration  in  the  pre-­‐

diagnosis  comparison.  This  association  in  the  general  population  comparison  may  be  in  

part  an  artifact  because  of   living   in  the  CMA  of  Montreal  was  an  entry  criterion  for  the  

comparator  cohort.  Living  in  an  FSA  with  a  rheumatologist  was  associated  with      

Page 68: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  67  

 

increased  odds  of  migration   in   the   general   population   comparison,   inconclusive   in   the  

pre-­‐diagnosis   comparison   and   associated   with   decreased   odds   of   migration   in   the  

regional  migration  comparison.  The  pattern  in  the  FSA-­‐level  SES  variables  was  different  

in  all  three  comparisons  except  for  income,  which  was  associated  with  decreased  odds  of  

migration.    

 

4.2  Health-­selective  migration  comparisons  

Model   2,   with   SLE,   Sjogren’s,   disease   duration,   age,   sex,   and   age   by   disease  

interactions,  was  retained   in  each  comparison.  Disease  by  age   interactions  and  disease  

Table  7—Coefficients  (and  95%  CrI)  from  bivariate  hierarchical  logistic  regression  analyses  in  the  healthy  control,  pre-­diagnosis  and  regional  migration  analyses    

Variable General population

comparison Pre-diagnosis comparison

Regional migration comparison

SLE -0.407 (-0.508,-0.309) -0.285 (-0.361,-0.207) -0.400 (-0.535,-0.265)

SJO -0.475 (-0.638,-0.319) -0.385 (-0.494,-0.273) -0.589 (-0.790,-0.386) disease duration in SLE* -0.202 (-0.332,-0.077) -0.171 (-0.265,-0.082) -0.124 (-0.276,0.025)

disease duration in Sjogren's* 0.162 (-0.084,0.402) -0.010 (-0.162,0.142) 0.082 (-0.211,0.369)

female -0.052 (-0.108,-0.002) 0.087 (-0.020,0.203) 0.018 (-0.179,0.229) Census metropolitan area -0.268 (-0.315,-0.220) 0.433 (0.346,0.523) -0.321 (-0.467,-0.179)

Rheumatologist in FSA 0.107 (0.062,0.150) 0.042 (-0.054,0.137) -0.176 (-0.339,-0.013)

Household that are renters (%)** 0.046 (0.038,0.054) 0.071 (0.052,0.089) -0.085 (-0.114,-0.055)

Education (%)** 0.051 (0.034,0.071) -0.088 (-0.124,-0.055) 0.116 (0.053,0.176) Employment rate (%)** -0.071 (-0.093,-0.048) 0.093 (0.043,0.135) 0.178 (0.099,0.259)

Average income (000’s)** -0.061 (-0.071,-0.051) -0.020 (-0.041,0.001) -0.070 (-0.112,-0.024)

age*** -0.044 (-0.044,-0.042) -0.030 (-0.032,-0.027) -0.033 (-0.038,-0.028)

age2*** 1.02E-03 (9.44E-4,1.10E-3) 8.28E-04 (6.82E-4,9.86-4) 6.06E-4 (3.32E-4,8.75E-4) * estimated with disease main effect, coefficient per five year increase ** coefficient per ten unit increase  *** estimated in the same model

 

Page 69: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 68  

duration   both   changed   the   OR   point   estimates   associated   with   SLE   and   Sjogren’s   in  

different   age   groups   and   disease   durations   by   greater   than   0.10.   Model   3   was   not  

retained   because   none   of   the   potential   confounders   was   strongly   related   to   disease  

status  and  their  addition  to  model  2  did  not  change  the  estimates  of  OR  point  estimate  

for   the   effect   of   SLE   or   Sjogren’s   appreciably   (maximum   change   in   ORs   of   2.6%).   The  

presence  of  interactions  between  disease  and  age  indicates  that  measures  of  association  

are   heterogeneous   across   strata   of   age   and   disease   duration.   Therefore,   ORs   are  

presented  for  each  stratum.    

In   these   comparisons,   a   complex   picture   of   health-­‐selective  migration   emerges  

which  involves  the  interplay  between  disease,  disease  duration  and  age.  Odds  ratios  for  

migration  in  SLE  and  Sjogren’s,  relative  to  general  population  controls,  by  strata  of  age  

and  disease  duration,  are  presented  in  Table  8.    

 

Table  8—Odds  ratios  (95%  CrI)  of  migration  relative  to  controls  by  age  and  disease  duration  strata    

Disease Age Disease duration

General population control analysis

Pre-diagnosis comparison

Regional migration comparison

30 2 0.41 (0.28-0.56) 0.55 (0.43-0.69) 0.50 (0.32-0.74) 50 2 0.66 (0.51-0.83) 0.74 (0.63-0.86) 0.62 (0.46-0.82) 70 2 1.09 (0.81-1.42) 1.01 (0.84-1.20) 0.79 (0.55-1.10)

30 10 0.59 (0.40-0.83) 0.61 (0.46-0.79) 0.64 (0.37-1.01) 50 10 0.95 (0.73-1.22) 0.82 (0.67-1.00) 0.79 (0.54-1.11)

Sjogren's

70 10 1.56 (1.18-2.02) 1.11 (0.90-1.37) 0.99 (0.65-1.44) 30 2 0.54 (0.45-0.64) 0.68 (0.60-0.77) 0.58 (0.47-0.72) 50 2 0.83 (0.71-0.96) 0.87 (0.79-0.96) 0.76 (0.64-0.90) 70 2 1.29 (1.04-1.58) 1.12 (0.95-1.31) 1.01 (0.74-1.34)

30 10 0.38 (0.32-0.44) 0.53 (0.46-0.61) 0.50 (0.39-0.63) 50 10 0.59 (0.51-0.67) 0.68 (0.61-0.76) 0.65 (0.53-0.79)

SLE

70 10 0.91 (0.74-1.11) 0.88 (0.73-1.03) 0.86 (0.62-1.15)

 

Page 70: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  69  

4.2.1  General  population  comparison  

In  the  general  population  comparison,  both  SLE  and  Sjogren’s  are  associated  with  

reduced  migration  in  young  people,  regardless  of  disease  duration  (Table  8,  column  1).  

For  example,   in  30  year  olds  with  two  years  of  disease  duration  the  OR  of  migration   is  

0.54   (95%   CrI:   0.45-­‐0.64)   in   SLE   patients   and   0.41   (95%   CrI:   0.28-­‐0.56)   in   Sjogren’s  

patients.  Within  strata  of  disease  duration,  increasing  age  increases  the  OR  towards  the  

null,   and   in   some   cases   crosses   the   null.   For   example,   higher   odds   of   migration   are  

suggested  in  70-­‐year-­‐old  SLE  patients  with  two  years  of  disease  duration  (OR:  1.29,  95%  

CrI:  1.04-­‐1.58)  and  70-­‐year-­‐old  Sjogren’s  patients  with  ten  years  of  disease  duration  (OR:  

1.56,  95%  CrI:  1.18-­‐2.02).    

Disease   duration   among   Sjogren’s   patients   is   associated   with   an   increased  

probability   of  moving   (OR   for   a   five   year   increase   in   disease   duration=1.26,   95%   CrI:  

0.99-­‐1.61).   The   association   between   disease   duration   and   migration   is   qualitatively  

different  among  SLE  patients,  where  the  OR  for  a  five  year  increase  in  disease  duration  is  

0.80   (95%  CrI:  0.71-­‐0.92).  For   this   reason,   the   lowest  ORs   (for   the  effect  of  disease  on  

migration)   are   found   among   younger   SLE   patients   with   longer   disease   duration   and  

younger  Sjogren’s  patients  with  shorter  disease  duration.  Similarly,  the  highest  ORs  (for  

the  effect  of  disease  on  migration)  are  found  in  older  SLE  patients  with  shorter  disease  

duration  and   in  older  Sjogren’s  patients  with   longer  disease  duration.  Only   the  general  

population   comparison   found   females   had   different   odds   of   migration   compared   to  

males  (OR:0.83  95%  CrI:  0.80-­‐0.88).    

  In   individuals  with  either  SLE  or  Sjogren’s,  age  and  disease  duration  necessarily  

Page 71: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 70  

increase  together.  Therefore  it  is  of  interest  to  examine  how  migration  changes  as  both  

disease   duration   and   age   change   together.   Table   9   presents   a   progression   of   ORs   for  

people  diagnosed  with  SLE  or  Sjogren’s  at  30  and  at  50  years  old  (column  1).  Older  ages  

are  not  presented  because  the  few  events  in  these  categories  means  that  error  bars  are  

wide  and  thus  the  results  are  uninformative.  

Migration  generally  decreases  with  age  in  all  groups  (with  the  exception  of  ages  

greater  than  around  70  where  it  increases  slightly  again),  therefore  the  effect  of  Sjogren’s  

duration  tends  to  counteract  the  effect  of  age.  However,  the  effect  of  SLE  duration  works  

in  concert  with  it.  This  can  be  seen  in  the  first  column  of  table  8.  Sjogren’s  is  associated  

with  a  decrease  in  migration  early  after  disease  diagnosis  but,  for  both  people  diagnosed  

at   30   and   50,   the   ORs   trend   back   toward   the   null,   with   increasing   disease   duration.  

Therefore,  the  credible  interval  of  the  Sjogren’s/migration  association  among  older    

Table  9—The  progression  of  odds  ratios  (95%  CrI)  relative  to  controls  for  people  diagnosed  at  30  and  50  years  old  for  all  three  comparisons  

Disease Age Disease Duration

General population comparison

Pre-diagnosis comparison

Regional migration

comparison

32 2 0.43 (0.30-0.58) 0.57 (0.45-0.70) 0.51 (0.33-0.75) 35 5 0.52 (0.40-0.68) 0.61 (0.50-0.73) 0.57 (0.40-0.79)

40 10 0.75 (0.55-1.00) 0.70 (0.55-0.88) 0.71 (0.45-1.04)

52 2 0.70 (0.54-0.87) 0.77 (0.66-0.88) 0.64 (0.47-0.83)

55 5 0.85 (0.73-1.00) 0.83 (0.74-0.93) 0.72 (0.58-0.87)

Sjogren's

60 10 1.22 (0.94-1.54) 0.95 (0.78-1.16) 0.88 (0.60-1.23)

32 2 0.56 (0.47-0.67) 0.70 (0.62-0.78) 0.60 (0.49-0.73)

35 5 0.52 (0.46-0.59) 0.66 (0.6-0.72) 0.59 (0.50-0.69)

40 10 0.47 (0.41-0.54) 0.60 (0.54-0.68) 0.57 (0.46-0.69)

52 2 0.87 (0.74-1.01) 0.89 (0.81-0.99) 0.78 (0.65-0.94)

55 5 0.81 (0.72-0.91) 0.85 (0.77-0.92) 0.77 (0.65-0.90)

SLE

60 10 0.73 (0.62-0.86) 0.77 (0.67-0.88) 0.75 (0.58-0.94)

 

Page 72: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  71  

Sjogren’s   patients   is   consistent   with   anywhere   from   a   slightly   negative   to   a   strongly  

positive   association  with  migration.   For   people   diagnosed  with   Sjogren’s   at   a   younger  

age   the   credible   interval   is   consistent   with   anywhere   from   a   strongly   negative  

association   between   Sjogren’s   and   migration   to   a   null   association.   In   SLE   patients,  

disease  duration  further  decreases  the  ORs  to  low  levels  among  60  year  olds  diagnosed  

at  50  (OR  0.72,  95%  CrI:  0.61-­‐0.84)  and  to  very  low  levels  among  40  year  olds  diagnosed  

at  30  (OR    0.48,  95%  CrI:  0.42-­‐0.55).  

In   order   to   illustrate   absolute   migration   rates,   Figure   2   graphs   the   predicted  

probabilities  of  migration  in  women  diagnosed  with  SLE  or  Sjogren’s  at  30  and  50  years.  

These   results   are   presented   specifically   in   women   because   calculating   predicted  

probabilities  from  logistic  regression  requires  specification  of  all  variables  in  the  model.  

Predicted  probabilities   in  men  were  2-­‐7%  higher  depending  on  the  age  but   the  overall  

ORs  and  conclusions  were  similar.  Migration  rates  in  Sjogren’s  patients  diagnosed  at  30  

tend  to  remain  stable  up  to  age  50  (with  wide  credible  intervals)  at  around  7%.  In  this  

same  age  range,  the  migration  rate  among  general  population  controls  goes  from  17%  to  

7%.   SLE   patients   diagnosed   at   30   see   their   migration   rate   decrease   from   9%   to   3%.  

Among  people  diagnosed  at  50,  predicted  probability  of  migration   remains   lower   than  

controls   in   SLE.   In   Sjogren’s,   the  predicted  probability  of  migration   increases  with  age  

whereas   it  decreases   in   controls.  After  15  years  of  disease  duration,   Sjogren’s  patients  

diagnosed   at   age   50   have   a   higher   predicted   probability   of   migration   than   general  

population  controls.  

 

Page 73: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 72  

4.2.2  Pre-­diagnosis  comparison  

Using  pre-­‐diagnosis  SLE  and  Sjogren’s  patients  as  a  comparator  yields  results  that  

are  similar  to  the  general  population  comparison,  except  the  point  estimates  are  slightly  

closer   to   the   null   (though   the   CrIs   all   overlap).   SLE   and   Sjogren’s   are   uniformly  

associated   with   lower   odds   of   migration   at   younger   ages   (30-­‐50   years)   regardless   of  

   Figure  2   –  Predicted  probabilities  of  migration   in  women  diagnosed  at  30  and  50  years   in   the  general  population  comparison  with  95%  CrI  error  bars  

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

Age (years)

Pre

dict

ed p

roba

bilit

y of

a m

ove

Sjogren'sSLEGeneral population

diagnosed at 30 diagnosed at 50

Page 74: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  73  

disease   duration   (Table   8,   column   2).   The   lowest   and   highest   ORs   among   Sjogren’s  

patients  are  also   found   in  younger  patients  with  short  disease  duration  (OR  0.55,  95%  

CrI:  0.43-­‐0.69)  and  in  older  patients  with  long  disease  duration  (OR  1.11,  95%  CrI:  0.90-­‐

1.37),  respectively.  The  lowest  and  highest  ORs  among  SLE  patients  are  found  in  younger  

patients  with  long  disease  duration  (OR  0.53,  95%  CrI:  0.46-­‐0.61)  and  in  older  patients  

with  short  disease  duration  (OR  1.12,  95%  CrI:  0.95-­‐1.31),  respectively.    

Five  years  of  disease  duration  among  Sjogren’s  patients  was  associated  with  an  

OR   of   1.06   (95%   CrI:   0.91-­‐1.25),   which   is   inconclusive   because   the   CrI   is   wide   and  

includes  important  values  above  and  below  the  null.  In  SLE  patients,  five  years  of  disease  

duration  was  associated  with  an  OR  of  0.89  (95%  CrI:  0.79-­‐0.94).  

The  progression  of  ORs  with  age  and  disease  duration  among  SLE  and  Sjogren’s  

patients   diagnosed   at   30   and   50   (table   8,   column   2)   follows   a   similar   pattern   to   the  

general  population  comparison.  However,   in  Sjogren’s  patients,   the  ORs  tend  to  change  

less  quickly  in  the  pre-­‐diagnosis  comparison  than  in  the  general  population  comparison.  

For  example  when  diagnosed  at  50,  the  ORs  in  Sjogren’s  patients  increases  from  0.77  to  

0.95   in   the   pre-­‐diagnosis   comparison   whereas   it   increases   from   0.70   to   1.22   in   the  

general   population   comparison.   In   SLE   patients,   the   progression   of   ORs  with   age   and  

disease   duration   is   roughly   the   same   in   the   general   population   and   pre-­‐diagnosis  

comparisons  though  the  estimates  tend  to  be  slightly  lower  in  the  former.  

  Predicted  probabilities  of  migration  (Figure  3)  follow  a  similar  pattern  to  that  of  

the  general  population  analysis.  Migration  rates   in  SLE  patients  diagnosed  at  30  and  at  

50  remain  lower  than  migration  rates  of  pre-­‐diagnosis  patients.  Sjogren’s  patients    

Page 75: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 74  

 

diagnosed   at   30  migrate   at   lower   rates,   although   at   15   years   of   disease   duration   the  

credible  intervals  are  wide  and  include  the  null  values.  In  Sjogren’s  patients  diagnosed  at  

50,   there   is  a   suggestion   that  migration  rates  may  be  higher   than  controls  at  a  disease  

   Figure  3  –  Predicted  probabilities  of  migration   in  people  diagnosed  at  30  and  50  years   in   the  pre-­diagnosis  analysis  with  95%  CrI  error  bars.      

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

30 40 50 60

0.00

0.02

0.04

0.06

0.08

0.10

0.12

Age (years)

Pre

dict

ed p

roba

bilit

y of

a m

ove

Sjogren'sSLEPre-diagnosis

diagnosed at 30 diagnosed at 50

Page 76: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  75  

duration  of  15  years,  though,  again,  the  credible  interval  are  very  wide.  

 

4.2.3  Regional  migration  comparison  

In   the   ‘regional   migration   comparison’   interregional   migration   is   used   as   the  

outcome.  The  OR  point  estimates  are  almost  all  lower  than  the  pre-­‐diagnosis  comparison  

(Table  8,  column  3).  The  ORs  estimating  the  effects  of  SLE  and  Sjogren’s  on  migration,  do  

tend  to  follow  the  same  patterns  (for  age  groups  and  disease  durations)  as  was  found  in  

the  previous   two   comparison,  but   the   credible   intervals   are  wider  because   there  were  

fewer  inter-­‐regional  moves  than  between-­‐FSA  moves.    

Regional   migration   in   Sjogren’s   patients   is   lowest   relative   to   pre-­‐diagnosis  

patients  at  younger  ages  and  short  disease  duration  (OR:  0.50,  95%  CrI:  0.32-­‐0.74  among  

30  year  old  patients  with   two  years  of  disease  duration)  and  highest  at  older  ages  and  

longer  disease  duration  (OR:  0.99,  95%  CrI:  0.65-­‐1.44  among  70  year  old  patients  with  

ten   years   of   disease   duration).   In   SLE   patients,   migration   relative   to   pre-­‐diagnosis  

patients   is   lower  at  younger  ages  with   long  disease  duration   (OR:  0.50,  95%  CrI:  0.39-­‐

0.63  among  30  year  old  patients  with  ten  years  of  disease  duration)  and  highest  among  

older  patients  with  short  disease  duration  (OR:  1.01,  95%  CrI:  0.74-­‐1.34  among  70  year  

old  patients  with  two  years  of  disease  duration).    

Five   years  of   disease  duration   is   associated  with   an  OR  of   1.15   (95%  CrI:   0.85-­‐

1.54)   in   Sjogren’s   patients   and   an  OR  of   0.90   (95%  CrI:   0.78-­‐1.05)   in   SLE  patients.   As  

disease   duration   and   age   increase   (Table   9,   column   3)   SLE   patients   migrate   between  

Page 77: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 76  

regions   at   about   the   same   rate   relative   to   controls.   In   Sjogren’s,   health-­‐selective  

migration  decreases  (becomes  closer  to  the  null)  with  age  and  disease  duration.  

In  absolute  terms,  wide  confidence  intervals  make  it  difficult  to  draw  conclusions  

but  for  SLE,  as  in  other  analyses,  patients  migrate  at   lower  rates  than  controls  both  for  

patients  diagnosed  at  age  30  and  age  50  (Figure  4).  Migration  rates  for  Sjogren’s  patients  

approach  the  migration  rates  of  controls,  as  disease  duration  increases.      

Page 78: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  77  

                                                     

       

   Figure  4  –  Predicted  probabilities  of  migration  of  people  diagnosed  at  30  and  50  years  in  the  regional  migration  analysis  with  95%  CrI  error  bars    

30 40 50 60

0.000

0.005

0.010

0.015

0.020

0.025

0.030

30 40 50 60

0.000

0.005

0.010

0.015

0.020

0.025

0.030

30 40 50 60

0.000

0.005

0.010

0.015

0.020

0.025

0.030

30 40 50 60

0.000

0.005

0.010

0.015

0.020

0.025

0.030

30 40 50 60

0.000

0.005

0.010

0.015

0.020

0.025

0.030

30 40 50 60

0.000

0.005

0.010

0.015

0.020

0.025

0.030

30 40 50 60

0.000

0.005

0.010

0.015

0.020

0.025

0.030

Age (years)

Pre

dict

ed p

roba

bilit

y of

a m

ove

Sjogren'sSLEPre-diagnosis

diagnosed at 30 diagnosed at 50

Page 79: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 78  

 

Page 80: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  79  

5  Discussion    

 

Overall,   SLE  and  Sjogren’s  patients   in  Quebec   appear   to  migrate   less  often   than  

either  the  general  population  or  pre-­‐diagnosis  SLE  and  Sjogren’s  patients.  However,  age  

and   disease   duration   are   important   effect   measure   modifiers,   and,   for   some  

combinations  of  age  and  disease  duration,  SLE  and  particularly  Sjogren’s  were  associated  

with  increased  odds  of  the  migration.  Depending  on  age  and  disease  duration,  SLE  can  be  

associated  with   approximately   62%   lower   odds   of  migrating   and   29%   higher   odds   of  

migration  relative  to  general  population  controls,  and  47%  lower  odds  of  migration  and  

possibly  slightly  increased  odds  of  migrating  relative  to  pre-­‐diagnosis  controls.  Similarly,  

Sjogren’s  is  associated  with  between  59%  lower  odds  and  56%  higher  odds  of  migration  

relative  to  general  population  controls,  and  45%  lower  odds  and  slightly  higher  odds  of  

migration   relative   to   pre-­‐diagnosis   controls.   ORs   estimates   for   regional   migration   are  

similar  to  estimates  in  pre-­‐diagnosis  controls.      

As   expected,   estimates   of   health-­‐selective   migration   in   the   general   population  

comparison   are   further   from   the   null   than   the   pre-­‐diagnosis   controls,   though   the  

credible   intervals  overlap.  Also,   the   estimates   for   regional  moves   among  pre-­‐diagnosis  

controls   are   further   from   the   null   than   between-­‐FSA   migration   although,   again,   the  

credible  intervals  overlap.  Disease  duration  appears  to  have  a  different  effect  among  SLE  

and  Sjogren’s  patients,  decreasing  the  odds  of  migration  in  the  former  and  increasing  the  

odds   in   the   latter.  This  may   indicate   that  Sjogren’s  patients   (who  on  average  will  have  

Page 81: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 80  

less   severe  manifestations   than   SLE  patients)   are   better   able   to   adapt   to   their   disease  

while   disease   progression   in   SLE  makes   it   increasingly   difficult   to  migrate.   It  was   not  

expected   that   Sjogren’s   patients   would   have   lower   odds   of   migration   shortly   after  

disease   onset   than   SLE  patients,   though   this   is   reversed  when   looking   among  patients  

with  ten  years  of  disease  duration.  

Comparisons   of   provincial   migration   estimates   to   regional   migration   rates  

estimated  from  administrative  data  demonstrate  that  it  is  feasible  to  estimate  migration  

using  administrative  data.  

 

5.1  How  the  results  compare  to  the  literature  

Findley  (1988)  described  the  concept  of  ‘health-­‐migration  relations  with  dynamic  

effects’,   and   hypothesized   that   at   disease   onset,   a   relatively  minor   condition   (e.g.   like  

Sjogren’s)   could   experience   a   stronger   decrease   in   migration   when   compared   to   the  

serious  condition  (i.e.  SLE)  because  the  lag  between  health  change  and  migration  would  

be   shorter   in   the   former.    As  disease  duration   increase,   this   trend  would   reverse  with  

disease  progression.  This  pattern,  though  based  entirely  on  theory  and  not  on  empirical  

evidence,   precisely   what   is   seen   in   this   thesis.   No   other   study   looked   specifically   at  

disease  duration.    

The   finding   that   age   is   an   effect   measure   modifier   for   the   health-­‐migration  

relationship   is   found   in   a  number  of   other   studies   (Bentham,  1988;  Boyle   et   al.,   2002;  

Halliday   &   Kimmitt,   2008;   Riva   et   al.,   2011).   The   often   observed   pattern   of   young  

migrants  being  healthier  than  non-­‐migrants  and  older  migrants  being  as  or  less  healthy  

Page 82: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  81  

than   non-­‐migrants   found   in   the   literature   can   found   in   this   thesis   as   well.   This   is  

particularly  evident  when  looking  at  the  literature  on  the  relationship  between  disability  

and  migration.  The  qualitative  differences  found  in  short  versus  regional-­‐scale  migration  

are  not  observed  in  this  thesis  (Boyle  et  al.,  2002)  but  this  may  be  because  between  FSA  

migration   is   not   sufficiently   small   scale   to   observe   this   difference.   Methodologically,  

Yiannakoulias  et  al.  (2007)  is  the  most  comparable  study  to  the  results  presented  in  my  

thesis  because  it  also  used  administrative  data  and  looked  at  specific  illnesses.  However,  

Yiannakoulias  et  al.  used  matching  and  waere  thus  unable  to  account  for  age  interactions.  

Also,   Yiannakoulias   et   al.   considered   migration   between   six   digit   postal   codes,   which  

would  include  many  moves  on  a  shorter  scale  (as  compared  to  the  analyses  in  this  thesis,  

which  was  unable  to  examine  moves  with  greater  detail  than  between  FSAs).  

The  finding  that  older  SLE  and  Sjogren’s  patients  may  migrate  more  than  controls  

is  consistent  with  many  of  the  findings  using  the  US  Longitudinal  Study  on  Aging  which  

found  that  both  activities  of  daily   living  and   instrumental  activities  of  daily   living  were  

associated  with  increased  migration  (Longino  et  al.,  1991;  Speare  et  al.,  1991).  SLE  and  

Sjogren’s   are   associated   with   an   increase   in   disability   making   these   references  

particularly  relevant.    

Silverstein   and   Zablotsky   (1996)   found   that   intermediate   disability   levels   are  

associated  with  increased  migration  among  people  70  years  and  older.  Sjogren’s  incurs  

less  disability  than  SLE  but  more  disability  than  a  person  in  good  health,  therefore  that  

finding   is   also   consistent  with   the   results  presented   in   this   thesis.  However,   two  other  

studies   found   the   reverse   relationship   where  migration   peaked   at   poor   and   excellent  

Page 83: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 82  

health   rather   than   at   intermediate   health   (Evandrou   et   al.,   2010;   Halliday   &   Kimmitt,  

2008).  

5.2  Strengths  

Verheij   et   al.   (1998)   state   that,   “to   investigate   [health-­‐]selection   effects,  

longitudinal  data  are  needed  on  large  numbers  of  people,  including  data  on  mobility  and  

(past)  illnesses  and  health  risk  factors.”  Administrative  data  provide  this  type  of  data  for  

mobility   and   illnesses   by   allowing  observation  of   the   same   individual   both  before   and  

after   diagnosis,   in   this   case,   with   SLE   or   Sjogren’s.   This   type   of   data   also   allows   the  

identification   of   the   temporal   order   of   health   and  migration   events.  When   comparing  

migration  pre-­‐  and  post-­‐diagnosis,  this  design  can  help  reduce  confounding  to  a  certain  

extent  for  some  variables,  such  as  race  and  SES,  with  the  notable  exception  of  age,  which  

can  be  controlled  for.    

Administrative  data  have  some  important  advantages  over  surveys,  which  many  

studies  have  used  to  study  health-­‐selective  migration.  Survey  data  can  be  biased  due  to  

non-­‐response  whereas  administrative  data,  particularly  in  Quebec,  includes  all  residents,  

at  least  in  theory.  Surveys  also  rely  on  both  migration  and  health  self-­‐report,  which  can  

be  prone  to  bias  due  to  not  understanding  questions,  difficulty  recalling  the  information  

asked,  and  social  desirability  bias  (Johnson  &  Fendrich,  2005;  Verheij  et  al.,  1998).      Also,  

the  inclusion  of  two  control  groups  with  different  strengths  and  weaknesses  allows  the  

findings  in  this  thesis  to  be  checked  for  robustness.  The  general  population  comparison  

provides   a   comparison   to   the   general   health-­‐care   using   population   but   may   be   more  

prone   to   unmeasured   confounding   because   it   was   not   possible   to  match   on   variables  

Page 84: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  83  

such  as  SES  and  race.  The  pre-­‐diagnosis  and  regional  analyses  use  the  same  population  

pre-­‐  and  post-­‐diagnosis  making  confounding  less  likely.  Despite  the  fact  that  the  control  

groups  have  different  health  profiles,  the  similarity  of  the  results  and  consistency  of  the  

effect  measure  modifiers  points  to  the  robustness  of  the  findings.  

 

5.3  Limitations  

The  results  of  this  thesis  are  must  be  considered  in  the  context  of  some  limitations.  

The  most  important  limitation  of  this  thesis  is  the  accuracy  of  administrative  databases  

with  regard  to  SLE  and  Sjogren’s  diagnosis.  A  systematic  review  of  validation  studies  of  

using   administrative   databases   to   identify   rheumatic   diseases   reported   sensitivities  

ranging  from  20%  to  100%  and  specificities  ranging  from  74%  to  97.1%  (Widdifield  et  

al.,  2011).  Bernatsky  et  al.  (2011)  used  the  same  criteria  as  this  thesis  to  identify  cases  in  

administrative   data   from  Nova   Scotia   and   found   a   sensitivity   of   98.2%   (95%  CI:   95.5-­‐

99.3)   and   specificity   of   72.5%   (95%   CI:   68.7-­‐75.9)   for   SLE   and   a   sensitivity   of   95.5%  

(95%   CI:   88.8-­‐98.2)   and   specificity   of   95.8%   (95%   CI:   94.1-­‐97.0)   for   Sjogren’s.   The  

controls  used  were  patients  with  other  rheumatic  diseases,   therefore,   the  specificity  of  

SLE  diagnosis  is  likely  much  higher  in  the  general  population.  Administrative  databases  

are  also  prone  to  the  same  problems  as  morbidity  measures  as  measured  by  physicians,  

which  is  a  function  not  only  of  ill-­‐health  but  also  of  people’s  propensity  to  visit  the  doctor  

and  of  the  doctor’s  diagnostic  and  recording  habits  (McAlister,  2004).  

Larson  et  al.  (2004)  address  other  limitations  that  are  pertinent  to  this  thesis  such  

as   movers   are   more   likely   to   be   single   and   live   alone.   Data   on   civil   status   were   not  

Page 85: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 84  

available.  However,  additional  analyses  restricted  to  women  and  using  the  ICD  code  for  

childbirth  as  a  proxy  for  civil  status  produced  values  consistent  with  the  findings  of  this  

thesis.   Larson   et   al.   also   point   out   that   the   size   and   shape   of   zones   against   which  

migration  is  measured  may  confound  results.  This  may  be  relevant  to  the  analyses  in  this  

thesis  because  of  the  diversity  of  shapes  and  sizes  of  FSAs  both  on  the  island  of  Montreal  

and   province  wide.   The   area   of   FSAs   varies   from   0.33   km2   for   an   FSA   in  Montreal   to  

nearly  500,000  km2  for  an  FSA  that  covers  a   large  portion  of   the  north  of   the  province  

(Statistics  Canada,  2006).  Further  complications  arise  because  FSAs  in  smaller  cities  and  

towns   can   be   entirely   contained   inside   a   rural   FSA.   Both   these   characteristics   of   FSAs  

mean   that   a   long   distance   move   in   rural   areas   are   less   likely   to   be   captured   in   this  

analysis  whereas   shorter  urban  moves   are  more   likely   to  be   captured.   It   is   difficult   to  

speculate  on  how  this  limitation  might  bias  this  analysis.  

Another   limitation   of   the   data   was   the   reason   for   extremely   low   levels   of  

migration  in  1992  in  the  systemic  autoimmune  rheumatic  diseases  cohort  was  not  clear.  

The   comparison   of   regional   migration   rates   in   the   systemic   autoimmune   rheumatic  

diseases   cohort   to  provincial   estimates   reassures,   to  a   certain  degree,   that   the  data  on  

migration   for   the  remaining  years   is  valid.  Also,   the  number  of  migration  events   in   the  

years  before  and  after  1992  are  consistent  and  close  to  what  would  be  expected.  

Disease  duration  was  measured  with  error  among  cases  that  were  incident  prior  

to  1989.  This  measurement  error  would  be  concentrated  among  older  patients  who  were  

more   likely   to   be   diagnosed   prior   to   1989.   Also,   disease   duration   is   taken   as   starting  

Page 86: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  85  

from   the   date   of   diagnosis   while   pre-­‐clinical   symptoms   of   SLE   and   Sjogren’s   may  

influence  migration  in  the  pre-­‐diagnosis  and  regional  analyses.    

Controlling   for   SES   using   variables   at   the   FSA   level   is   not   ideal   and   implies  

measurement  error.  SES  has  been  found  to  be  related  to  the  incidence,  disease  severity  

and  manifestation  of  SLE  (Demas  &  Costenbader,  2009;  Studenski  et  al.,  1987;  Sutcliffe  et  

al.,   1999)   though   the   relative   causal   role   played   by   genetics   and   social   variables   is   in  

question  (Sule  &  Petri,  2006).  SES,  as  measured  by  education  (Dahl,  2002),  income  (Islam  

&  Choudhury,  1991;  Yankow,  1999)  and  employment  (Hacker,  2000),   is  also  related  to  

migration  making  SES  a  potential  confounder.  However,  a  number  of  studies  looking  at  

income  have  not  found  evidence  that  it  is  either  a  confounder  (Larson  et  al.,  2004;  Riva  et  

al.,  2011)  or  mediator  (Zimmerman  et  al.,  1993).  Nonetheless,   it   is  conceivable  that  the  

lack  of   individual   level   SES  measures   induces  bias   into   the   analysis.  The  direction   and  

magnitude  of  this  bias  would  depend  on  the  direction  and  magnitude  of  the  health-­‐SES  

and  migration-­‐SES  relationships.  Many  SES  variables  are  likely  simultaneously  mediators  

and  confounders.  If  this  is  the  case,  the  only  appropriate  way  to  control  for  them  is  using  

time-­‐series  data  using  appropriate  models  such  as  marginal  structural  models.  

Lastly,  ethnicity  may  be  a  possible  confounder  because  it   is  a  risk  factor  for  SLE  

(Danchenko   et   al.,   2006;   McCarty   et   al.,   1995;   Peschken   &   Esdaile,   2000;   Serdula   &  

Rhoads,  1979)  and  Sjogren’s  (Schein  et  al.,  1999).  Race  is  also  related  to  migration  (Frey  

et   al.,   2005;  Robinson,  1993).  As  with   civil   status,  data  on   race  were  not  available  and  

was,  therefore,  unable  to  be  controlled  for.  

 

Page 87: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 86  

5.4  Overall  Conclusions  

This  thesis  provides  evidence  that  chronic  diseases  can  impact  migration  rates  

and  that  this  impact  can  vary  by  age,  disease  and  disease  duration.  It  is  one  of  the  first  

uses  of  administrative  data  to  measure  health-­‐selective  migration.  

These  results  may  indicate  that  when  people  are  diagnosed  with  either  SLE  or  

Sjogren’s  at  a  young  age,  their  illness  prevents  moves  that  they  would  have  made  had  

they  been  healthy.  The  reason  for  this  may  be  to  remain  near  specialty  care,  family  or  

friends,  or  because  the  disabilities  associated  with  these  illnesses  may  make  it  too  

difficult  to  migrate.  Older  SLE  and  particularly  Sjogren’s  patients  demonstrate  a  

tendency  toward  increased  migration  relative  to  the  general  population  possibly  

indicating    that  migration  may  be  necessary  to  be  closer  to  care  or  because  they  are  

unable  to  function  sufficiently  in  their  own  home.  Both  of  these  possibilities  may  indicate  

shortcomings  in  care  received  by  these  patients.  

This  thesis  also  suggests  that  patterns  in  SLE  prevalence  could  potentially  be  

influenced  by  health-­‐selective  migration.  If  the  migrations  SLE  patients  are  making  or  

not  making  are  of  a  specific  kind,  for  example,  if  they  were  more  inclined  to  move  into  an  

urban  area  or  less  inclined  to  move  out  of  an  urban  area  because  of  the  presence  of  

specialized  care,  this  could  explain,  in  part  at  least,  higher  prevalence  of  SLE  in  urban  

areas.  If  pre-­‐clinical  symptoms  also  cause  this  type  of  health-­‐selective  migration,  it  could  

also  explain  patterns  in  incidence.  In  older  age  groups,  patterns  of  Sjogren’s  prevalence  

could  be  influenced  directly  by  patients  moving  to  specific  areas  for  specific  reasons.    

 

Page 88: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  87  

5.5  Future  Research    

There  are  a  number  of  avenues  for  future  research  in  health-­‐selective  migration  in  

rheumatic  diseases.  Firstly,  health-­‐selective  migration  should  be  estimated  in  different  

rheumatic  diseases  as  different  disabilities  and  symptoms  may  affect  migration  in  

different  ways.  Secondly,  health-­‐selective  migration  post-­‐diagnosis  may  affect  

prevalence  of  a  condition  but  health-­‐selective  migration  may  also  operate  pre-­‐diagnosis  

due  to  pre-­‐clinical  symptoms.  If  health-­‐selective  migration  can  occur  pre-­‐diagnosis,  

regional  estimates  of  incidence  as  well  as  prevalence  would  be  affected.  Thirdly,  research  

should  be  done  on  whether  this  health-­‐selective  migration  has  any  specific  geographic  

destinations  such  as  toward  urban  areas  or  away  from  areas  with  poor  health  services.  

Lastly,  very  little  work  has  been  done  validating  addresses  or  postal  codes  from  

administrative  databases.  Validation  studies  of  addresses  from  administrative  databases  

could  make  estimates  of  health-­‐selective  migration  using  these  databases  more  reliable.      

     

Page 89: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 88  

Page 90: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  89  

 6  Bibliography        Alamanos,  Y.,  Voulgari,  P.  V.,  Siozos,  C.,  Katsimpri,  P.,  et  al.  (2003).  Epidemiology  of  

systemic  lupus  erythematosus  in  northwest  Greece  1982-­‐2001.  The  Journal  of  Rheumatology,  30(4),  731–735.  

Allan,  D.,  &  Cloutier-­‐Fisher,  D.  (2006).  Health  service  utilization  among  older  adults  in  British  Columbia:  making  sense  of  geography.  Canadian  Journal  on  Aging  -­  La  Revue  Canadienne  du  Vieillissement,  25(2),  219–232.  

Andrianakos,  A.,  Trontzas,  P.,  Christoyannis,  F.,  Dantis,  P.,  et  al.  (2003).  Prevalence  of  rheumatic  diseases  in  Greece:  a  cross-­‐sectional  population  based  epidemiological  study.  The  ESORDIG  Study.  The  Journal  of  Rheumatology,  30(7),  1589–1601.    

Association  des  médecins  rhumatologues  du  Québec,  2010.  Trouver  un  rhumatologue.  Retrieved  May  19,  2011:  http://rhumatologie.org/rhumatologie.html  

Avina-­‐Zubieta,  J.,  Sayre,  E.,  Bernatsky,  S.,  Shojania,  K.,  et  al.  (2011).  Adult  Prevalence  Of  Systemic  Autoimmune  Rheumatic  Diseases  (SARDs)  In  British  Columbia,  Canada  (Vol.  63,  pp.  1–2).  Presented  at  the  American  College  of  Rheumatology/Association  of  Rheumatology  Health  Professionals  Annual  Scientific  Meeting,  Chicago,  Illinois.  

Barnabe,  C.,  Joseph,  L.,  Belisle,  P.,  Labrecque,  J.,  et  al.  (2012).  Prevalence  of  autoimmune  inflammatory  myopathy  in  Alberta's  first  nations  population.  Arthritis  Care  &  Research,  E–pub  ahead  of  print.  

Barsby,  S.  L.,  &  Cox,  D.  R.  (1975).  Interstate  migration  of  the  elderly:  an  economic  analysis.  Lexington  Books.  

Bentham,  G.  (1988).  Migration  and  morbidity:  implications  for  geographical  studies  of  disease.  Social  Science  &  Medicine,  26(1),  49–54.  

Bernatsky,  S.,  Joseph,  L.,  Pineau,  C.  A.,  Tamblyn,  R.,  et  al.  (2007).  A  population-­‐based  assessment  of  systemic  lupus  erythematosus  incidence  and  prevalence  results  and  implications  of  using  administrative  data  for  epidemiological  studies.  Rheumatology,  46(12),  1814–1818.  

Bernatsky,  S.,  Linehan,  T.,  &  Hanly,  J.  G.  (2011).  The  Accuracy  of  Administrative  Data  Diagnoses  of  Systemic  Autoimmune  Rheumatic  Diseases.  The  Journal  of  Rheumatology,  38(8),  1612–1616.  

Bertoli,  A.  M.,  Fernandez,  M.,  Alarcón,  G.  S.,  Vila,  L.  M.,  et  al.  (2006).  Systemic  lupus  erythematosus  in  a  multiethnic  US  cohort  LUMINA  (XLI):  factors  predictive  of  self-­‐reported  work  disability.  Annals  of  the  Rheumatic  Diseases,  66(1),  12–17.  

Béland,  F.  (1984).  The  decision  of  elderly  persons  to  leave  their  homes.  The  Gerontologist,  24(2),  179–185.  

Boomsma,  M.  M.,  Bijl,  M.,  Stegeman,  C.  A.,  Kallenberg,  C.  G.  M.,  et  al.  (2002).  Patients'  perceptions  of  the  effects  of  systemic  lupus  erythematosus  on  health,  function,  income,  and  interpersonal  relationships:  a  comparison  with  Wegeners  granulomatosis.  Arthritis  &  Rheumatism,  47(2),  196–201.    

Page 91: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 90  

 Borders,  T.  F.,  Rohrer,  J.  E.,  Hilsenrath,  P.  E.,  &  Ward,  M.  M.  (2000).  Why  rural  residents  

migrate  for  family  physician  care.  The  Journal  of  Rural  Health,  16(4),  337–348.  Borràs,  J.  M.,  Sánchez,  V.,  Moreno,  V.,  Izquierdo,  A.,  et  al.  (1995).  Cervical  cancer:  

incidence  and  survival  in  migrants  within  Spain.  Journal  of  Epidemiology  and  Community  Health,  49(2),  153–157.  

Boyle,  P.  J.,  Gatrell,  A.  C.,  &  Duke-­‐Williams,  O.  (2001).  Do  area-­‐level  population  change,  deprivation  and  variations  in  deprivation  affect  individual-­‐level  self-­‐reported  limiting  long-­‐term  illness?  Social  Science  &  Medicine,  53(6),  795–799.  

Boyle,  P.  J.,  Norman,  P.,  &  Popham,  F.  (2009).  Social  mobility:  Evidence  that  it  can  widen  health  inequalities.  Social  Science  &  Medicine,  68(10),  1835–1842.    

Boyle,  P.  J.,  Norman,  P.,  &  Rees,  P.  (2002).  Does  migration  exaggerate  the  relationship  between  deprivation  and  limiting  long-­‐term  illness?  A  Scottish  analysis.  Social  Science  &  Medicine,  55(1),  21–31.  

Breslow,  R.,  Klinger,  B.,  &  Erickson,  B.  (1998).  County  drift:  A  type  of  geographic  mobility  of  chronic  psychiatric  patients.  General  Hospital  Psychiatry,  20(1),  44–47.  

Brimblecombe,  N.,  Dorling,  D.,  &  Shaw,  M.  (2000).  Migration  and  geographical  inequalities  in  health  in  Britain.  Social  Science  &  Medicine,  50(6),  861–878.  

Buchanan,  R.  J.,  Wang,  S.,  Stuifbergen,  A.,  Chakravorty,  B.  J.,  et  al.  (2006).  Urban/rural  differences  in  the  use  of  physician  services  by  people  with  multiple  sclerosis.  NeuroRehabilitation,  21(3),  177–187.  

Bureau  of  the  U.S.  Census,  (1947).  Postwar  Migration  and  Its  Causes  in  the  United  States:  August,  1945,  to  October,  1946.  Issue  4  of  Current  Population  Reports  population  characteristics  P-­‐20,  no.4.  

Curtis,  S.,  Setia,  M.  S.,  &  Quesnel-­‐Vallee,  A.  (2009).  Socio-­‐geographic  mobility  and  health  status:  A  longitudinal  analysis  using  the  National  Population  Health  Survey  of  Canada.  Social  Science  &  Medicine,  69(12),  1845–1853.    

Dahl,  G.  B.  (2002).  Mobility  and  the  return  to  education:  Testing  a  Roy  model  with  multiple  markets.  Econometrica,  70(6),  2367–2420.    

Danchenko,  N.,  Satia,  J.,  &  Anthony,  M.  (2006).  Epidemiology  of  systemic  lupus  erythematosus:  a  comparison  of  worldwide  disease  burden.  Lupus,  15(5),  308–318.  

De  Jong,  G.  F.,  Wilmoth,  J.  M.,  Angel,  J.  L.,  &  Cornwell,  G.  T.  (1995).  Motive  and  the  geographic  mobility  of  very  old  Americans.  The  Journals  of  Gerontology.  Series  B,  Psychological  Sciences  and  Social  Sciences,  50(6),  S395–404.  

Demas,  K.  L.,  &  Costenbader,  K.  H.  (2009).  Disparities  in  lupus  care  and  outcomes.  Current  Opinion  in  Rheumatology,  21(2),  102–109.  

Driscoll,  D.,  Dotterrer,  B.,  Miller,  J.,  &  Voorhees,  H.  (2010).  Assessing  the  influence  of  health  on  rural  outmigration  in  Alaska.  International  Journal  of  Circumpolar  Health,  69(5),  528–544.  

Elford,  J.,  Phillips,  A.,  Thomson,  A.  G.,  &  Shaper,  A.  G.  (1990).  Migration  and  geographic  variations  in  blood  pressure  in  Britain.  BMJ  (Clinical  research  ed.),  300(6720),  291–295.  

 

Page 92: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  91  

Elford,  J.,  Thomson,  A.,  &  Phillips,  A.  (1989).  Migration  and  geographic  variations  in  ischaemic  heart  disease  in  Great  Britain.  The  Lancet.  1(8634),  343-­‐346.  

Evandrou,  M.,  Falkingham,  J.,  &  Green,  M.  (2010).  Migration  in  later  life:  evidence  from  the  British  Household  Panel  Study.  Population  trends,  (141),  74–91.  

Ferraro,  K.  F.  (1983).  The  health  consequences  of  relocation  among  the  aged  in  the  community.  Journal  of  Gerontology,  38(1),  90–96.  

Findley,  S.  E.  (1988).  The  directionality  and  age  selectivity  of  the  health-­‐migration  relation:  evidence  from  sequences  of  disability  and  mobility  in  the  United  States.  The  International  Migration  Review,  22(3),  4–29.  

Flynn,  C.  (1980).  General  versus  aged  interstate  migration,  1965-­‐1970.  Research  on  Aging.  2(2),  165-­‐176.  

Fox,  A.,  Goldblatt,  P.,  &  Adelstein,  A.  (1982).  Selection  and  mortality  differentials.  Journal  of  Epidemiology  and  Community  Health.  36(2),  69-­‐79.  

Frey,  W.  H.,  Liaw,  K.  L.,  Wright,  R.,  &  White,  M.  J.  (2005).  Migration  within  the  United  States:  Role  of  Race-­‐Ethnicity  [with  Comments].  Brookings-­Wharton  Papers  on  Urban  Affairs.  pp.  207–262.    

Gaubitz,  M.  (2006).  Epidemiology  of  connective  tissue  disorders.  Rheumatology,  45(Supplement  3),  iii3–iii4.  

Gober,  P.,  &  Zonn,  L.  E.  (1983).  Kin  and  elderly  amenity  migration.  The  Gerontologist,  23(3),  288–294.  

Gómez-­‐Rubio,  V.,  &  López-­‐Quílez,  A.  (2010).  Statistical  Methods  for  the  Geographical  Analysis  of  Rare  Diseases.  Advances  in  Experimental  Medicine  and  Biology.  686,  151–171.    

Gran,  J.  T.  (2002).  Diagnosis  and  definition  of  primary  Sjögren's  syndrome.  Scandinavian  Journal  of  Rheumatology,  31(2),  57–59.    

Gushulak,  B.  D.,  &  MacPherson,  D.  W.  (2006).  The  basic  principles  of  migration  health:  population  mobility  and  gaps  in  disease  prevalence.  Emerging  Themes  in  Epidemiology,  3,  3.  

Hacker,  R.  S.  (2000).  Mobility  and  regional  economic  downturns.  Journal  of  Regional  Science,  40(1),  45–65.    

Haenszel,  W.,  &  Dawson,  E.  (1965).  A  note  on  mortality  from  cancer  of  the  colon  and  rectum  in  the  United  States.  Cancer,  18(3),  265–272.    

Halliday,  T.  J.,  &  Kimmitt,  M.  C.  (2008a).  Selective  migration  and  health  in  the  USA,  1984–93.  Population  Studies,  62(3),  321–334.  

Hansen,  B.  U.  (1991).  Primary  Sjögren's  Syndrome:  Its  prevalence  and  association  with  some  autoimmune  diseases  Clinical  and  laboratory  studies  with  special  reference  to  anti-­‐SS-­‐B/La  antibodies.  Scandinavian  Journal  of  Rheumatology,  20(2),  137–137.  

Harding,  S.,  &  Balarajan,  R.  (1996).  Patterns  of  mortality  in  second  generation  Irish  living  in  England  and  Wales:  longitudinal  study.  BMJ  (Clinical  research  ed.),  312(7043),  1389–1392.  

Hart,  J.  E.,  Laden,  F.,  Puett,  R.  C.,  Costenbader,  K.  H.,  et  al.  (2009).  Exposure  to  Traffic  Pollution  and  Increased  Risk  of  Rheumatoid  Arthritis.  Environmental  Health  Perspectives,  117(7),  1065–1069.  

Page 93: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 92  

Henretta,  J.  C.  (1986).  Retirement  and  Residential  Moves  by  Elderly  Households.  Research  on  Aging,  8(1),  23–37.  

Hull,  D.  (1979).  Migration,  adaptation,  and  illness:  a  review.  Social  Science  &  Medicine  Part  A:  Medical  Psychology  &  Medical  Sociology.  13A,  25-­‐36.  

Institut  de  la  statistique  Québec,  (2007).  Moyennes  sur  5  ans  des  migrations  internes  annuelles  1996-­‐2001  et  2001-­‐2006.  Retrieved  July  25,  2011  from:  http://www.stat.gouv.qc.ca/donstat/societe/demographie/migrt_poplt_imigr/migr_interne_5ans.htm    

Islam,  M.  N.,  &  Choudhury,  S.  A.  (1991).  Self-­‐selection  and  intermunicipal  migration  in  Canada.  Regional  Science  and  Urban  Economics,  20(4),  459–472.    

Jackson,  D.  J.,  Longino,  C.  F.,  Zimmerman,  R.  S.,  &  Bradsher,  J.  E.  (1991).  Environmental  Adjustments  to  Declining  Functional  Ability:  Residential  Mobility  and  Living  Arrangements.  Research  on  Aging,  13(3),  289–309.  

Johnson,  T.,  &  Fendrich,  M.  (2005).  Modeling  Sources  of  Self-­‐report  Bias  in  a  Survey  of  Drug  Use  Epidemiology.  Annals  of  Epidemiology,  15(5),  381–389.  

Jongeneel-­‐Grimen,  B.,  Droomers,  M.,  Stronks,  K.,  &  Kunst,  A.  E.  (2011).  Migration  does  not  enlarge  inequalities  in  health  between  rich  and  poor  neighbourhoods  in  The  Netherlands.  Health  &  Place,  17(4),  988–995.    

Kabasakal,  Y.,  Kitapcioglu,  G.,  Turk,  T.,  Öder,  G.,  et  al.  (2006).  The  prevalence  of  Sjögren's  syndrome  in  adult  women.  Scandinavian  Journal  of  Rheumatology,  35(5),  379–383.  

Katz,  J.  N.,  Barrett,  J.,  Liang,  M.  H.,  Bacon,  A.  M.,  et  al.  (1997).  Sensitivity  and  positive  predictive  value  of  Medicare  Part  B  physician  claims  for  rheumatologic  diagnoses  and  procedures.  Arthritis  &  Rheumatism,  40(9),  1594–1600.  

Katz,  P.,  Morris,  A.,  Trupin,  L.,  Yazdany,  J.,  et  al.  (2008).  Disability  in  valued  life  activities  among  individuals  with  systemic  lupus  erythematosus.  Arthritis  &  Rheumatism,  59(4),  465–473.  

Kennedy,  B.  P.,  Kawachi,  I.,  Glass,  R.,  &  Prothrow-­‐Stith,  D.  (1998).  Income  distribution,  socioeconomic  status,  and  self  rated  health  in  the  United  States:  multilevel  analysis.  BMJ  (Clinical  research  ed.),  317(7163),  917–921.    

Kliewer,  E.  V.  (1992).  Influence  of  migrants  on  regional  variations  of  stomach  and  colon  cancer  mortality  in  the  Western  United  States.  International  Journal  of  Epidemiology,  21(3),  442–449.  

Kolcić,  I.,  &  Polasek,  O.  (2009).  Healthy  migrant  effect  within  Croatia.  Collegium  Antropologicum,  33  Suppl  1,  141–145.  

Kurahara,  D.  K.,  Grandinetti,  A.,  Fujii,  L.  L.  A.,  Tokuda,  A.  A.,  et  al.  (2007).  Visiting  consultant  clinics  to  study  prevalence  rates  of  juvenile  rheumatoid  arthritis  and  childhood  systemic  lupus  erythematosus  across  dispersed  geographic  areas.  The  Journal  of  Rheumatology,  34(2),  425–429.  

Labrecque,  J.,  Joseph,  L.,  Smargiassi,  A.,  Hudson,  M.,  et  al.  (2010).  Preliminary  Analyses  of  Spatial  Clustering  of  the  Prevalence  of  Systemic  Autoimmune  Rheumatic  Diseases  in  Montreal,  Quebec.  Canadian  Rheumatology  Association  Annual  Meeting.  Quebec  City,  Canada.  Februrary  3-­‐6,  2010.  

 

Page 94: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  93  

   Labrecque,  J.,  Smargiassi,  A.,  Joseph,  L.,  Hudson,  M.,  et  al.  (2010,  February  3).  A  

Population-­‐Based  Ecological  Investigation  of  the  Relationship  between  Road  Density  and  Systemic  Autoimmune  Rheumatic  Disease  Prevalence.  Canadian  Rheumatology  Association  Annual  Meeting.  Quebec  City,  Canada.  February  3-­‐6,  2010.  

Ladouceur,  M.,  Rahme,  E.,  Pineau,  C.  A.,  &  Joseph,  L.  (2006).  Robustness  of  Prevalence  Estimates  Derived  from  Misclassified  Data  from  Administrative  Databases.  Biometrics,  63(1),  272–279.  

Lanska,  D.  J.,  &  Peterson,  P.  M.  (1995).  Effects  of  Interstate  Migration  on  the  Geographic  Distribution  of  Stroke  Mortality  in  the  United  States.  Stroke,  26(4),  554–561.  

Larson,  A.,  Bell,  M.,  &  Young,  A.  F.  (2004).  Clarifying  the  relationships  between  health  and  residential  mobility.  Social  Science  &  Medicine,  59(10),  2149–2160.  

Leal,  J.  R.,  &  Laupland,  K.  B.  (2009).  Validity  of  ascertainment  of  co-­‐morbid  illness  using  administrative  databases:  a  systematic  review.  Clinical  Microbiology  and  Infection,  16(6),  715–721.  

Lee,  E.  (1966).  A  theory  of  migration.  Demography,  3(1),  47-­‐57.  Lewis,  S.  (2003).  Migration  and  health  impact  assessment.  Public  Health,  117(5),  305–

311.  Litwak,  E.,  &  Longino,  C.  F.  (1987).  Migration  patterns  among  the  elderly:  a  

developmental  perspective.  The  Gerontologist,  27(3),  266–272.  Longino,  C.  F.,  Jackson,  D.  J.,  Zimmerman,  R.  S.,  &  Bradsher,  J.  E.  (1991).  The  second  move:  

health  and  geographic  mobility.  Journal  of  Gerontology,  46(4),  S218–24.  Lunn,  D.  J.,  Thomas,  A.,  Best,  N.,  &  Spiegelhalter,  D.  (2000).  WinBUGS-­‐a  Bayesian  

modelling  framework:  concepts,  structure,  and  extensibility.  Statistics  and  Computing,  10(4),  325–337.    

Martyn,  C.  N.,  Barker,  D.  J.,  &  Osmond,  C.  (1993).  Selective  migration  by  birthweight.  Journal  of  Epidemiology  and  Community  Health,  47(1),  76.  

Matthews,  F.  E.,  Chatfield,  M.,  Freeman,  C.,  McCracken,  C.,  et  al.  (2004).  Attrition  and  bias  in  the  MRC  cognitive  function  and  ageing  study:  an  epidemiological  investigation.  BMC  Public  Health,  4(12),  1-­‐10.  

Mau,  W.,  Listing,  J.,  Huscher,  D.,  Zeidler,  H.,  et  al.  (2005).  Employment  across  chronic  inflammatory  rheumatic  diseases  and  comparison  with  the  general  population.  The  Journal  of  Rheumatology,  32(4),  721–728.  

McAlister,  F.  A.  (2004).  Influence  of  socioeconomic  deprivation  on  the  primary  care  burden  and  treatment  of  patients  with  a  diagnosis  of  heart  failure  in  general  practice  in  Scotland:  population  based  study.  BMJ  (Clinical  research  ed.),  328(7448),  1110.  

McCarty,  D.  J.,  Manzi,  S.,  Medsger,  T.  A.  J.,  Ramsey-­‐Goldman,  R.,  et  al.  (1995).  Incidence  of  systemic  lupus  erythematosus.  Race  and  gender  differences.  Arthritis  &  Rheumatism,  38(9),  1260–1270.  

McDonald,  J.  T.,  &  Conde,  H.  (2010).  Does  Geography  Matter?  The  Health  Service  Use  and  Unmet  Health  Care  Needs  of  Older  Canadians.  Canadian  Journal  on  Aging  -­  La  Revue  Canadienne  du  Vieillissement,  29(01),  23.  

Page 95: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 94  

Mcelhone,  K.,  Abbott,  J.,  &  Teh,  L.  S.  (2006).  A  review  of  health  related  quality  of  life  in  systemic  lupus  erythematosus.  Lupus,  15(10),  633–643.  

Meijer,  J.  M.,  Meiners,  P.  M.,  Hudson,  M.,  Spijkervet,  F.  K.  L.,  et  al.  (2009).  Health-­‐related  quality  of  life,  employment  and  disability  in  patients  with  Sjogren's  syndrome.  Rheumatology,  48(9),  1077–1082.  

Meyer,  J.  W.,  &  Speare,  A.  (1985).  Distinctively  elderly  mobility:    types  and  determinants.  Economic  Geography,  61(1),  79–88.  

Mickey,  R.  M.,  &  Greenland,  S.  (1989).  The  impact  of  confounder  selection  criteria  on  effect  estimation.  American  Journal  of  Epidemiology,  129(1),  125–137.  

Moorin,  R.  E.,  Holman,  C.  D.  J.,  GARFIELD,  C.,  &  Brameld,  K.  J.  (2006).  Health  related  migration:  evidence  of  reduced  "urban-­‐drift".  Health  &  Place,  12(2),  131–140.  

Nelson,  L.  M.,  &  Winter,  M.  (1975).  Life  disruption,  independence,  satisfaction,  and  the  consideration  of  moving.  The  Gerontologist,  15(2),  160–164.  

Newbold,  K.  B.  (2005).  Self-­‐rated  health  within  the  Canadian  immigrant  population:  risk  and  the  healthy  immigrant  effect.  Social  Science  &  Medicine,  60(6),  1359–1370.  

Norman,  P.,  Boyle,  P.,  &  Rees,  P.  (2005).  Selective  migration,  health  and  deprivation:  a  longitudinal  analysis.  Social  Science  &  Medicine,  60(12),  2755–2771.  

O'Reilly,  D.  (1994).  Health  and  social  inequality  in  Europe.  Migration  from  deprived  areas  may  be  a  factor.  BMJ  (Clinical  research  ed.),  309(6946),  57–58.  

O'Reilly,  D.,  &  Stevenson,  M.  (2003).  Selective  migration  from  deprived  areas  in  Northern  Ireland  and  the  spatial  distribution  of  inequalities:  implications  for  monitoring  health  and  inequalities  in  health.  Social  Science  &  Medicine,  57(8),  1455–1462.  

Ocaña-­‐Riola,  R.,  Fernández-­‐Ajuria,  A.,  Mayoral-­‐Cortés,  J.  M.,  Toro-­‐Cárdenas,  S.,  et  al.  (2009).  Uncontrolled  Migrations  as  a  Cause  of  Inequality  in  Health  and  Mortality  in  Small-­‐area  Studies.  Epidemiology,  20(3),  411–418.  

Patrick,  C.  H.  (1980).  Health  and  Migration  of  the  Elderly.  Research  on  Aging,  2(2),  233–241.  

Peschken,  C.  A.,  &  Esdaile,  J.  M.  (2000).  Systemic  lupus  erythematosus  in  North  American  Indians:  a  population  based  study.  The  Journal  of  Rheumatology,  27(8),  1884–1891.  

Pillemer,  S.  R.,  Matteson,  E.  L.,  Jacobsson,  L.  T.  H.,  Martens,  P.  B.,  et  al.  (2001).  Incidence  of  Physician-­‐Diagnosed  Primary  Sjögren  Syndrome  in  Residents  of  Olmsted  County,  Minnesota.  Mayo  Clinic  Proceedings,  76(6),  593–599.  

Polissar,  L.  (1980).  The  effect  of  migration  on  comparison  of  disease  rates  in  geographic  studies  in  the  United  States.  American  Journal  of  Epidemiology,  111(2),  175–182.  

Ravenstein,  E.  (1885).  The  laws  of  migration.  Journal  of  the  Statistical  Society  of  London,  48(2),  167-­‐235.  

Ravenstein,  E.  (1889).  The  laws  of  migration.  Journal  of  the  Royal  Statistical  Society,  52(2),  241-­‐305.  

Richardson,  K.,  Blakely,  T.,  Young,  J.,  Graham,  P.,  et  al.  (2009).  Do  ethnic  and  socio-­‐economic  inequalities  in  mortality  vary  by  region  in  New  Zealand?  An  application  of  hierarchical  Bayesian  modelling.  Social  Science  &  Medicine,  69(8),  1252–1260.    

Riva,  M.,  Curtis,  S.,  &  Norman,  P.  (2011).  Residential  mobility  within  England  and  urban-­‐rural  inequalities  in  mortality.  Social  Science  &  Medicine,  1–32.    

Page 96: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  95  

Robinson,  V.  (1993).  'Race',  gender,  and  internal  migration  within  England  and  Wales.  Environment  and  Planning  A,  25(10),  1453–1465.  

Rogerson,  P.  A.,  &  Han,  D.  (2002).  The  effects  of  migration  on  the  detection  of  geographic  differences  in  disease  risk.  Social  Science  &  Medicine,  55(10),  1817–1828.  

Saag,  K.  G.,  Doebbeling,  B.  N.,  Rohrer,  J.  E.,  Kolluri,  S.,  et  al.  (1998).  Arthritis  health  service  utilization  among  the  elderly:  the  role  of  urban-­‐rural  residence  and  other  utilization  factors.  Arthritis  Care  &  Research,  11(3),  177–185.  

Schein,  O.  D.,  Hochberg,  M.  C.,  Munoz,  B.,  Tielsch,  J.  M.,  et  al.  (1999).  Dry  eye  and  dry  mouth  in  the  elderly:  a  population-­‐based  assessment.  Archives  of  Internal  Medicine,  159(12),  1359–1363.  

Schneeweiss,  S.,  &  Avorn,  J.  (2005).  A  review  of  uses  of  health  care  utilization  databases  for  epidemiologic  research  on  therapeutics.  Journal  of  Clinical  Epidemiology,  58(4),  323–337.  

Serdula,  M.  K.,  &  Rhoads,  G.  G.  (1979).  Frequency  of  systemic  lupus  erythematosus  in  different  ethnic  groups  in  Hawaii.  Arthritis  &  Rheumatism,  22(4),  328–333.  

Sergeant,  J.  F.,  &  Ekerdt,  D.  J.  (2008).  Motives  for  Residential  Mobility  in  Later  Life:  Post-­‐Move  Perspectives  of  Elders  and  Family  Members.  The  International  Journal  of  Aging  and  Human  Development,  66(2),  131–154.  

Silverstein,  M.,  &  Angelelli,  J.  J.  (1998).  Older  parents'  expectations  of  moving  closer  to  their  children.  The  Journals  of  Gerontology.  Series  B,  Psychological  Sciences  and  Social  Sciences,  53(3),  S153–S163.  

Silverstein,  M.,  &  Zablotsky,  D.  L.  (1996).  Health  and  social  precursors  of  later  life  retirement-­‐community  migration.  The  Journals  of  Gerontology.  Series  B,  Psychological  Sciences  and  Social  Sciences,  51(3),  S150–S156.  

Sinclair,  J.  C.,  &  Bracken,  M.  B.  (1994).  Clinically  useful  measures  of  effect  in  binary  analyses  of  randomized  trials.  Journal  of  Clinical  Epidemiology,  47(8),  881–889.  

Speare,  A.,  Avery,  R.,  &  Lawton,  L.  (1991).  Disability,  Residential-­‐Mobility,  and  Changes  in  Living  Arrangements.  Journal  of  Gerontology,  46(3),  S133–S142.  

Statistics  Canada,  (2001a).  Census  of  Population,  Statistics  Canada  (catalogue  no.  95F0486XCB2001004).  Ottawa,  Ontario.  Retrieved  May  25,  2012:    http://www12.statcan.gc.ca/english/census01/products/standard/profiles/Rp-­‐eng.cfm?TABID=1&LANG=E&APATH=1&DETAIL=1&DIM=0&FL=A&FREE=0&GC=0&GK=0&GRP=1&PID=56140&PRID=0&PTYPE=55430,53293,55440,55496,71090&S=0&SHOWALL=0&SUB=0&Temporal=2001&THEME=57&VID=0&VNAMEE=&VNAMEF=  

Statistics  Canada  (2001b).  Profile  of  Income  of  Individuals,  Families  and  Households,  Social  and  Economic  Characteristics  of  Individuals,  Families  and  Households,  Housing  Costs,  and  Religion,  for  Canada,  Provinces,  Territories  and  Forward  Sortation  Areas,  2001  Census  (catalog  no.  95F0492XCB2001003).  Ottawa,  Ontario.  http://www12.statcan.ca/english/census01/Products/standard/themes/DataProducts.cfm?S=1&T=55&ALEVEL=2&FREE=0  

   

Page 97: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 96  

Statistics  Canada  (2002).    2001  Postal  Code  Conversion  File.  Retrieved  June  23,  2011  through  the  University  of  Toronto  Map  &  Data  library:  http://dc1.chass.utoronto.ca/census/2001_fsa_pccf_vjan07.html  

Statistics  Canada  (2006).    Census  of  Canada  2006:  Geographic  Reference  Files-­‐FSA  land  area  file.  Retrieved  May  5,  2012  through  the  University  of  Toronto  Map  &  Data  library:  http://prod.library.utoronto.ca:8090/datalib/codebooks/c/cc06/georef/fsa_area_2006.csv  

Strachan,  D.  P.,  Leon,  D.  A.,  &  Dodgeon,  B.  (1995).  Mortality  from  cardiovascular  disease  among  interregional  migrants  in  England  and  Wales.  BMJ  (Clinical  research  ed.),  310(6977),  423–427.  

Strömbeck,  B.,  Ekdahl,  C.,  Manthorpe,  R.,  &  Jacobsson,  L.  T.  H.  (2003).  Physical  capacity  in  women  with  primary  Sjögren's  syndrome:  A  controlled  study.  Arthritis  &  Rheumatism,  49(5),  681–688.  

Studenski,  S.,  Allen,  N.  B.,  Caldwell,  D.  S.,  Rice,  J.  R.,  et  al.  (1987).  Survival  in  systemic  lupus  erythematosus.  A  multivariate  analysis  of  demographic  factors.  Arthritis  &  Rheumatism,  30(12),  1326–1332.  

Sule,  S.,  &  Petri,  M.  (2006).  Socioeconomic  status  in  systemic  lupus  erythematosus.  Lupus,  15(11),  720–723.  

Sutcliffe,  N.,  Clarke,  A.  E.,  Gordon,  C.,  Farewell,  V.,  et  al.  (1999).  The  association  of  socio-­‐economic  status,  race,  psychosocial  factors  and  outcome  in  patients  with  systemic  lupus  erythematosus.  Rheumatology  (Oxford,  England),  38(11),  1130–1137.  

Tench,  C.,  Bentley,  D.,  Vleck,  V.,  McCurdie,  I.,  et  al.  (2002).  Aerobic  fitness,  fatigue,  and  physical  disability  in  systemic  lupus  erythematosus.  The  Journal  of  Rheumatology,  29(3),  474–481.  

Thomas,  D.  S.  (1938).  Review:  Selective  Migration.  The  Milbank  Memorial  Fund  Quarterly,  16(4),  403-­‐407.  

Tiefelsdorf,  M.  (2007).  Controlling  for  migration  effects  in  ecological  disease  mapping  of  prostate  cancer.  Stochastic  Environmental  Research  and  Risk  Assessment,  21(5),  615–624.  

Tousignant,  P.,  Groome,  P.  A.,  Spitzer,  W.  O.,  Schechter,  M.  T.,  et  al.  (1994).  Outmigrant  ascertainment  for  bias  assessment  in  environmental  epidemiology.  International  Journal  of  Epidemiology,  23(5),  1091–1098.  

Tricco,  A.  C.,  Pham,  B.,  &  Rawson,  N.  S.  B.  (2008).  Manitoba  and  Saskatchewan  administrative  health  care  utilization  databases  are  used  differently  to  answer  epidemiologic  research  questions.  Journal  of  Clinical  Epidemiology,  61(2),  192–197.  

Tsai,  S.  P.,  &  Wen,  C.  P.  (1986).  A  review  of  methodological  issues  of  the  standardized  mortality  ratio  (SMR)  in  occupational  cohort  studies.  International  Journal  of  Epidemiology,  15(1),  8–21.  

Verheij,  R.  A.,  van  de  Mheen,  H.  D.,  de  Bakker,  D.  H.,  Groenewegen,  P.  P.,  et  al.  (1998).  Urban-­‐rural  variations  in  health  in  The  Netherlands:  does  selective  migration  play  a  part?  Journal  of  Epidemiology  and  Community  Health,  52(8),  487–493.  

Veugelers,  P.  J.,  &  Guernsey,  J.  R.  (1999).  Sensitivity  analysis  of  selective  migration  in  

Page 98: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  97  

ecologic  comparisons  of  health.  Epidemiology,  10(6),  784–785.  Walsh,  S.  J.,  &  Gilchrist,  A.  (2006).  Geographical  clustering  of  mortality  from  systemic  

lupus  erythematosus  in  the  United  States:  contributions  of  poverty,  Hispanic  ethnicity  and  solar  radiation.  Lupus,  15(10),  662–670.  

Ward,  M.  M.  (2010).  Access  to  Care  and  the  Incidence  of  Endstage  Renal  Disease  Due  to  Systemic  Lupus  Erythematosus.  The  Journal  of  Rheumatology,  37(6),  1158–1163.  

Welton,  T.  (1871).  On  the  Effect  of  Migrations  in  Disturbing  Local  Rates  of  Mortality,  as  Exemplified  in  the  Statistics  of  London  and  the  Surrounding  Country,  for  the  Years  1851-­‐1860.  Journal  of  the  Institute  of  Actuaries  and  Assurance  Magazine,  16(3),  153-­‐186.  

Widdifield,  J.,  Lix,  L.  M.,  Labrecque,  J.,  Bernatsky,  S.,  et  al.  (2011).    A  systematic  review  to  evaluate  the  quality  and  reporting  of  administrative  database  validation  studies  for  rheumatic  diseases.  Canadian  Rheumatology  Association  &  Mexican-­Canadian  Congress  of  Rhuematology  (CRA-­MCR).  Cancun,  Mexico.  February  11-­‐15,  2012.  

Wiggins,  R.,  Joshi,  H.,  Bartley,  M.,  &  Gleave,  S.  (2002).  Place  and  personal  circumstances  in  a  multilevel  account  of  women's  long-­‐term  illness.  Social  Science  &  Medicine,  54(5),  827-­‐838.  

Worobey,  J.  L.,  &  Angel,  R.  J.  (1990).  Functional  capacity  and  living  arrangements  of  unmarried  elderly  persons.  Journal  of  Gerontology,  45(3),  S95–S101.  

Yankow,  J.  J.  (1999).  The  wage  dynamics  of  internal  migration  within  the  United  States.  Eastern  Economic  Journal,  25(3),  265–278.    

Yelin,  E.,  Trupin,  L.,  Katz,  P.,  Criswell,  L.,  et  al.  (2007).  Work  dynamics  among  persons  with  systemic  lupus  erythematosus.  Arthritis  &  Rheumatism,  57(1),  56–63.  

Yiannakoulias,  N.,  Schopflocher,  D.  R.,  Warren,  S.  A.,  &  Svenson,  L.  W.  (2007).  Parkinson's  disease,  multiple  sclerosis  and  changes  of  residence  in  Alberta.  The  Canadian  Journal  of  Neurological  Sciences  -­  Le  Journal  Canadien  des  Sciences  Neurologiques,  34(3),  343–348.  

Zimmerman,  R.  S.,  Jackson,  D.  J.,  Longino,  C.  F.,  &  Bradsher,  J.  E.  (1993).  Interpersonal  and  Economic  Resources  as  Mediators  of  the  Effects  of  Health  Decline  on  the  Geographic  Mobility  of  the  Elderly.  Journal  of  Aging  and  Health,  5(1),  37–57.  

         

Page 99: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 98  

Page 100: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  99  

 Appendix   A—Forward   sortation   areas   and   ICD-­9   codes   used   in   this  thesis    

 

FSAs  considered  part  of  the  CMA  of  Montreal:  H1A,  H1B,  H1C,  H1E,  H1G,  H1H,  H1J,  H1K,  H1L,  H1M,  H1N,  H1P,  H1R,  H1S,  H1T,  H1V,  H1W,  H1X,  H1Y,  H1Z,  H2A,  H2B,  H2C,  H2E,  H2G,  H2H,  H2J,  H2K,  H2L,  H2M,  H2N,  H2P,  H2R,  H2S,  H2T,  H2V,  H2W,  H2X,  H2Y,  H2Z,  H3A,  H3B,  H3C,  H3E,  H3G,  H3H,  H3J,  H3K,  H3L,  H3M,  H3N,  H3P,  H3R,  H3S,  H3T,  H3V,  H3W,  H3X,  H3Y,  H3Z,  H4A,  H4B,  H4C,  H4E,  H4G,  H4H,  H4J,  H4K,  H4L,  H4M,  H4N,  H4P,  H4R,  H4S,  H4T,  H4V,  H4W,  H4X,  H4Y,  H4Z,  H5A,  H5B,  H7A,  H7B,  H7C,  H7E,  H7G,  H7H,  H7J,  H7K,  H7L,  H7M,  H7N,  H7P,  H7R,  H7S,  H7T,  H7V,  H7W,  H7X,  H7Y,  H8N,  H8P,  H8R,  H8S,  H8T,  H8Y,  H8Z,  H9A,  H9B,  H9C,  H9E,  H9G,  H9H,  H9J,  H9K,  H9P,  H9R,  H9S,  H9W,  H9X,  J0K,  J0L,  J0N,  J0P,  J0R,  J0S,  J0V,  J3E,  J3G,  J3H,  J3L,  J3N,  J3V,  J3X,  J3Y,  J3Z,  J4B,  J4G,  J4H,  J4J,  J4K,  J4L,  J4M,  J4N,  J4P,  J4R,  J4S,  J4T,  J4V,  J4W,  J4X,  J4Y,  J4Z,  J5A,  J5B,  J5C,  J5K,  J5L,  J5M,  J5R,  J5T,  J5W,  J5X,  J5Y,  J5Z,  J6A,  J6J,  J6K,  J6N,  J6R,  J6V,  J6W,  J6X,  J6Y,  J6Z,  J7A,  J7B,  J7C,  J7E,  J7G,  J7H,  J7J,  J7K,  J7L,  J7M,  J7N,  J7P,  J7R,  J7T,  J7V,  J7W,  J7X,  J7Y,  J7Z,  J8H,  J8L        RAMQ  ICD-­9  codes  for  entry  into  the  influenza-­like  illness  cohort:  010,   0100,   0101,   0108,   0109,   011,   0110,   0111,   0112,   0113,   0114,   0115,   0116,   0117,  0118,  0119,  020,  0203,  0204,  0205,  0219,  022,  0221,  024,  0249,  025,  0259,  032,  0320,  0321,  0322,  0323,  0329,  033,  0330,  0331,  0338,  0339,  034,  0340,  0529,  0551,  073,  0739,  0741,   0790,   0793,   0798,   0799,   0830,   1124,   1149,   115,   1150,   1151,   1159,   1309,   1363,  382,   3820,   3824,   3829,   460,   4609,   461,   4618,   4619,   462,   4629,   463,   4639,   464,   4640,  4641,  4642,  4643,  4644,  465,  4650,  4658,  4659,  466,  4660,  4661,  4789,  480,  4800,  4801,  4802,  4808,  4809,  481,  4819,  482,  4820,  4821,  4822,  4823,  4824,  4828,  4829,  483,  4839,  484,  4841,  4843,  4845,  4846,  4847,  4848,  485,  4859,  486,  4869,  487,  4870,  4871,  4878,  490,  4909,  491,  4910,  4911,  4918,  4919,  507,  5070,  5071,  5078,  511,  5110,  5111,  5118,  5119,   513,   5130,   5131,   518,   5180,   5184,   5188,   5192,   7806,   7841,   7860,   7861,   7862,  7865,  7953,  V018                  

Page 101: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 100  

         MEDECHO  ICD-­9  codes  for  entry  into  the  influenza-­like  illness  cohort:  0100,  0101,  0108,  0109,  0110,  0111,  0112,  0113,  0114,  0115,  0116,  0117,  0118,  0119,  0203,  0204,  0205,  0219,  0221,  0249,  0259,  0320,  0321,  0322,  0323,  0329,  0330,  0331,  0338,  0339,  0340,  0529,  0551,  0739,  0741,  0790,  0793,  0798,  0799,  0830,  1124,  1149,  1150,  1151,  1159,  1309,  1363,  3820,  3824,  3829,  4609,  4618,  4619,  4629,  4639,  4640,  4641,  4642,  4643,  4644,  4650,  4658,  4659,  4660,  4661,  4789,  4800,  4801,  4802,  4808,  4809,  4819,  4820,  4821,  4822,  4823,  4824,  4828,  4829,  4839,  4841,  4843,  4845,  4846,  4847,  4848,  4859,  4869,  4870,  4871,  4878,  4909,  4910,  4911,  4918,  4919,  5070,  5071,  5078,  5110,  5111,  5118,  5119,  5130,  5131,  5180,  5184,  5188,  5192,  7806,  7841,  7860,  7861,   7862,   7865,   7953,   V018,   01000,   01001,   01002,   01003,   01004,   01005,   01009,  01010,   01011,   01012,   01013,   01014,   01015,   01019,   01080,   01081,   01082,   01083,  01084,   01085,   01089,   01090,   01091,   01092,   01093,   01094,   01095,   01099,   01100,  01101,   01102,   01103,   01104,   01105,   01109,   01110,   01111,   01112,   01113,   01114,  01115,   01119,   01120,   01121,   01122,   01123,   01124,   01125,   01129,   01130,   01131,  01132,   01133,   01134,   01135,   01139,   01140,   01141,   01142,   01143,   01144,   01145,  01149,   01150,   01151,   01152,   01153,   01154,   01155,   01159,   01160,   01161,   01162,  01163,   01164,   01165,   01169,   01170,   01171,   01172,   01173,   01174,   01175,   01179,  01180,   01181,   01182,   01183,   01184,   01185,   01189,   01190,   01191,   01192,   01193,  01194,   01195,   01199,   05290,   05291,   05297,   05298,   05299,   51881,   51882,   51889,  78650,  78651,  78652,  78659      

Page 102: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  101  

 APPENDIX  B—Models  considered  in  analyses    Each  model  uses  the  notation  of  observation  i  within  individual  j:      Model  1    

logit(moveij ) = α j + β1 * SLEij + +β2 * Sjogren'sij + β3 * ageij + β4 * age2ij + β5 * femaleij

   Model  2  

 

logit(moveij ) = α j + β1 * SLEij + +β2 * Sjogren'sij + β3 * ageij + β4 * age2ij + β5 * femaleij +

β6 * SLEij * ageij + β7 * Sjogren'sij * ageij + β8 * SLE diseasedurationij +

β9 * Sjogren'sdiseasedurationij  

   

Model  3    

logit(moveij ) = α j + β1 * SLEij + +β2 * Sjogren'sij + β3 * ageij + β4 * age2ij + β5 * femaleij +

β6 * SLEij * ageij + β7 * Sjogren'sij * ageij + β8 * SLE diseasedurationij +

β9 * Sjogren'sdiseasedurationij + β10 *educationij + β11 * incomeij + β12 * rentij + β13 *employmentij +

β14 *CMAij + β15 * rheumato logist in FSAij

Page 103: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 102  

 

Page 104: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  103  

 APPENDIX    C  –  Sample  WinBUGS  code    Example  of  WinBUGS  code  for  model  2:    model { for (j in 1:(number of individuals)) # loop 1: over every individual { for (i in index[j]:index2[j]) # loop 2: over every observation within each individual { logit(p[i]) <- alpha[j] + b.SLE*SLE[i] + b.SJO*SJO[i] + b.female*female[i] + b.agec*agec[i] + b.age2*age2[i] + b.SLE.age*SLE.age[i] + b.SJO.age*SJO.age[i] b.dd.SLE*dd.SLE[i] + b.dd.SJO*dd.SJO[i] move[i] ~ dbern(p[i]) # distribution of the outcome } # end loop 1 alpha[j] ~ dnorm(mu,tau) # random intercept } # end loop 2 #priors mu ~ dnorm(0,0.01) tau <- 1/(sigma*sigma) sigma ~ dunif(0,10) b.SLE ~ dnorm(0,0.01) b.SJO ~ dnorm(0,0.01) b.female ~ dnorm(0,0.01) b.agec ~ dnorm(0,0.01) b.age2 ~ dnorm(0,0.01) b.SLE.age ~ dnorm(0,0.01) b.SJO.age ~ dnorm(0,0.01) b.dd.SLE ~ dnorm(0,0.01) b.dd.SJO ~ dnorm(0,0.01) #-----------------------------# #-- Predicted probabilities --# #-----------------------------# # No disease--------------------------------- pp.30.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*-20 + b.age2*400 + b.SLE.age*0*-20 + b.SJO.age*0*-20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.32.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*-18 + b.age2*324 + b.SLE.age*0*-20 + b.SJO.age*0*-20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.35.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*-15 + b.age2*225 + b.SLE.age*0*-15 + b.SJO.age*0*-20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.40.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*-10 + b.age2*100 + b.SLE.age*0*-10 + b.SJO.age*0*-20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.45.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*-5 + b.age2*25 + b.SLE.age*0*-5 + b.SJO.age*0*-20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.50.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*0 + b.age2*0 + b.SLE.age*0*0 + b.SJO.age*0*0 + b.dd.SLE*0 + b.dd.SJO*0))) pp.52.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*2 + b.age2*4 + b.SLE.age*0*0 + b.SJO.age*0*0 + b.dd.SLE*0 + b.dd.SJO*0))) pp.55.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*5 + b.age2*25 + b.SLE.age*0*-15 + b.SJO.age*0*-20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.60.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*10 + b.age2*100 + b.SLE.age*0*-10 + b.SJO.age*0*-20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.65.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*15 + b.age2*225 + b.SLE.age*0*-5 + b.SJO.age*0*-20 + b.dd.SLE*0 + b.dd.SJO*0)))

Page 105: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 104  

pp.70.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*20 + b.age2*400 + b.SLE.age*0*20 + b.SJO.age*0*20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.72.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*22 + b.age2*484 + b.SLE.age*0*20 + b.SJO.age*0*20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.75.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*25 + b.age2*625 + b.SLE.age*0*20 + b.SJO.age*0*20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.80.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*30 + b.age2*900 + b.SLE.age*0*20 + b.SJO.age*0*20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.85.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*1 + b.agec*35 + b.age2*1225+ b.SLE.age*0*20 + b.SJO.age*0*20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.30.M <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*0 + b.agec*-20 + b.age2*400 + b.SLE.age*0*-20 + b.SJO.age*0*-20 + b.dd.SLE*0 + b.dd.SJO*0))) pp.50.M <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*0 + b.agec*0 + b.age2*0 + b.SLE.age*0*0 + b.SJO.age*0*0 + b.dd.SLE*0 + b.dd.SJO*0))) pp.70.M <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*0 + b.female*0 + b.agec*20 + b.age2*400 + b.SLE.age*0*20 + b.SJO.age*0*20 + b.dd.SLE*0 + b.dd.SJO*0))) # SLE------------------------------------- pp.SLE.30.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*-20 + b.age2*400 + b.SLE.age*1*-20 + b.SJO.age*0*-20 + b.dd.SLE*2 + b.dd.SJO*0))) pp.SLE.50.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*0 + b.age2*0 + b.SLE.age*1*0 + b.SJO.age*0*0 + b.dd.SLE*2 + b.dd.SJO*0))) pp.SLE.70.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*20 + b.age2*400 + b.SLE.age*1*20 + b.SJO.age*0*20 + b.dd.SLE*2 + b.dd.SJO*0))) pp.SLE.30.dd2.M <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*0 + b.agec*-20 + b.age2*400 + b.SLE.age*1*-20 + b.SJO.age*0*-20 + b.dd.SLE*2 + b.dd.SJO*0))) pp.SLE.50.dd2.M <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*0 + b.agec*0 + b.age2*0 + b.SLE.age*1*0 + b.SJO.age*0*0 + b.dd.SLE*2 + b.dd.SJO*0))) pp.SLE.70.dd2.M <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*0 + b.agec*20 + b.age2*400 + b.SLE.age*1*20 + b.SJO.age*0*20 + b.dd.SLE*2 + b.dd.SJO*0))) pp.SLE.30.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*-20 + b.age2*400 + b.SLE.age*1*-20 + b.SJO.age*0*-20 + b.dd.SLE*10 + b.dd.SJO*0))) pp.SLE.50.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*0 + b.age2*0 + b.SLE.age*1*0 + b.SJO.age*0*0 + b.dd.SLE*10 + b.dd.SJO*0))) pp.SLE.70.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*20 + b.age2*400 + b.SLE.age*1*20 + b.SJO.age*0*20 + b.dd.SLE*10 + b.dd.SJO*0))) pp.SLE.30.dd10.M <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*0 + b.agec*-20 + b.age2*400 + b.SLE.age*1*-20 + b.SJO.age*0*-20 + b.dd.SLE*10 + b.dd.SJO*0))) pp.SLE.50.dd10.M <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*0 + b.agec*0 + b.age2*0 + b.SLE.age*1*0 + b.SJO.age*0*0 + b.dd.SLE*10 + b.dd.SJO*0))) pp.SLE.70.dd10.M <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*0 + b.agec*20 + b.age2*400 + b.SLE.age*1*20 + b.SJO.age*0*20 + b.dd.SLE*10 + b.dd.SJO*0))) pp.SLE.32.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*-18 + b.age2*324 + b.SLE.age*1*-18 + b.dd.SLE*2 ))) pp.SLE.35.dd5.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*-15 + b.age2*225 + b.SLE.age*1*-15 + b.dd.SLE*5 ))) pp.SLE.40.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*-10 + b.age2*100 + b.SLE.age*1*-10 + b.dd.SLE*10))) pp.SLE.45.dd15.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*-5 + b.age2*25 + b.SLE.age*1*-5 + b.dd.SLE*15))) pp.SLE.52.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*2 + b.age2*4 + b.SLE.age*1*2 + b.dd.SLE*2 ))) pp.SLE.55.dd5.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*5 + b.age2*25 + b.SLE.age*1*5 + b.dd.SLE*5 ))) pp.SLE.60.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*10 + b.age2*100 + b.SLE.age*1*10 + b.dd.SLE*10))) pp.SLE.65.dd15.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*15 + b.age2*225 + b.SLE.age*1*15 + b.dd.SLE*15)))

Page 106: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  105  

pp.SLE.72.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*22 + b.age2*484 + b.SLE.age*1*22 + b.dd.SLE*2 ))) pp.SLE.75.dd5.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*25 + b.age2*625 + b.SLE.age*1*25 + b.dd.SLE*5 ))) pp.SLE.80.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*30 + b.age2*900 + b.SLE.age*1*30 + b.dd.SLE*10))) pp.SLE.85.dd15.F <- 1/(1+exp(-1*(mu + b.SLE*1 + b.SJO*0 + b.female*1 + b.agec*35 + b.age2*1225 + b.SLE.age*1*35 + b.dd.SLE*15))) # SJO -------------------------------------------- pp.SJO.30.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*-20 + b.age2*400 + b.SLE.age*0*-20 + b.SJO.age*1*-20 + b.dd.SLE*0 + b.dd.SJO*2))) pp.SJO.50.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*0 + b.age2*0 + b.SLE.age*0*0 + b.SJO.age*1*0 + b.dd.SLE*0 + b.dd.SJO*2))) pp.SJO.70.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*20 + b.age2*400 + b.SLE.age*0*20 + b.SJO.age*1*20 + b.dd.SLE*0 + b.dd.SJO*2))) pp.SJO.30.dd2.M <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*0 + b.agec*-20 + b.age2*400 + b.SLE.age*0*-20 + b.SJO.age*1*-20 + b.dd.SLE*0 + b.dd.SJO*2))) pp.SJO.50.dd2.M <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*0 + b.agec*0 + b.age2*0 + b.SLE.age*0*0 + b.SJO.age*1*0 + b.dd.SLE*0 + b.dd.SJO*2)) pp.SJO.70.dd2.M <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*0 + b.agec*20 + b.age2*400 + b.SLE.age*0*20 + b.SJO.age*1*20 + b.dd.SLE*0 + b.dd.SJO*2))) pp.SJO.30.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*-20 + b.age2*400 + b.SLE.age*0*-20 + b.SJO.age*1*-20 + b.dd.SLE*0 + b.dd.SJO*10))) pp.SJO.50.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*0 + b.age2*0 + b.SLE.age*0*0 + b.SJO.age*1*0 + b.dd.SLE*0 + b.dd.SJO*10))) pp.SJO.70.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*20 + b.age2*400 + b.SLE.age*0*20 + b.SJO.age*1*20 + b.dd.SLE*0 + b.dd.SJO*10))) pp.SJO.30.dd10.M <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*0 + b.agec*-20 + b.age2*400 + b.SLE.age*0*-20 + b.SJO.age*1*-20 + b.dd.SLE*0 + b.dd.SJO*10))) pp.SJO.50.dd10.M <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*0 + b.agec*0 + b.age2*0 + b.SLE.age*0*0 + b.SJO.age*1*0 + b.dd.SLE*0 + b.dd.SJO*10))) pp.SJO.70.dd10.M <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*0 + b.agec*20 + b.age2*400 + b.SLE.age*0*20 + b.SJO.age*1*20 + b.dd.SLE*0 + b.dd.SJO*10))) pp.SJO.32.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*-18 + b.age2*324 + b.SJO.age*1*-18 + b.dd.SJO*2 ))) pp.SJO.35.dd5.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*-15 + b.age2*225 + b.SJO.age*1*-15 + b.dd.SJO*5 ))) pp.SJO.40.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*-10 + b.age2*100 + b.SJO.age*1*-10 + b.dd.SJO*10))) pp.SJO.45.dd15.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*-5 + b.age2*25 + b.SJO.age*1*-5 + b.dd.SJO*15))) pp.SJO.52.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*2 + b.age2*4 + b.SJO.age*1*2 + b.dd.SJO*2 ))) pp.SJO.55.dd5.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*5 + b.age2*25 + b.SJO.age*1*5 + b.dd.SJO*5 ))) pp.SJO.60.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*10 + b.age2*100 + b.SJO.age*1*10 + b.dd.SJO*10))) pp.SJO.65.dd15.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*15 + b.age2*225 + b.SJO.age*1*15 + b.dd.SJO*15))) pp.SJO.72.dd2.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*22 + b.age2*484 + b.SJO.age*1*22 + b.dd.SJO*2 ))) pp.SJO.75.dd5.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*25 + b.age2*625 + b.SJO.age*1*25 + b.dd.SJO*5 ))) pp.SJO.80.dd10.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*30 + b.age2*900 + b.SJO.age*1*30 + b.dd.SJO*10))) pp.SJO.85.dd15.F <- 1/(1+exp(-1*(mu + b.SLE*0 + b.SJO*1 + b.female*1 + b.agec*35 + b.age2*1225 + b.SJO.age*1*35 + b.dd.SJO*15)))

Page 107: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 106  

#---------------------------------------------------------# # -- ORs -------------------------------------------------# #---------------------------------------------------------# #SLE------------------------------------------------------- or.SLE.30.dd2 <- exp(b.SLE + -20*b.SLE.age + 2*b.dd.SLE) or.SLE.50.dd2 <- exp(b.SLE + 0*b.SLE.age + 2*b.dd.SLE) or.SLE.70.dd2 <- exp(b.SLE + 20*b.SLE.age + 2*b.dd.SLE) or.SLE.30.dd10 <- exp(b.SLE + -20*b.SLE.age + 10*b.dd.SLE) or.SLE.50.dd10 <- exp(b.SLE + 0*b.SLE.age + 10*b.dd.SLE) or.SLE.70.dd10 <- exp(b.SLE + 20*b.SLE.age + 10*b.dd.SLE) or.SLE.32.dd2 <- exp(b.SLE + -18*b.SLE.age + 2*b.dd.SLE) or.SLE.35.dd5 <- exp(b.SLE + -15*b.SLE.age + 5*b.dd.SLE) or.SLE.40.dd10 <- exp(b.SLE + -10*b.SLE.age + 10*b.dd.SLE) or.SLE.45.dd15 <- exp(b.SLE + -5*b.SLE.age + 15*b.dd.SLE) or.SLE.52.dd2 <- exp(b.SLE + 2*b.SLE.age + 2*b.dd.SLE) or.SLE.55.dd5 <- exp(b.SLE + 5*b.SLE.age + 5*b.dd.SLE) or.SLE.60.dd10 <- exp(b.SLE + 10*b.SLE.age + 10*b.dd.SLE) or.SLE.65.dd15 <- exp(b.SLE + 15*b.SLE.age + 15*b.dd.SLE) or.SLE.72.dd2 <- exp(b.SLE + 22*b.SLE.age + 2*b.dd.SLE) or.SLE.75.dd5 <- exp(b.SLE + 25*b.SLE.age + 5*b.dd.SLE) or.SLE.80.dd10 <- exp(b.SLE + 30*b.SLE.age + 10*b.dd.SLE) or.SLE.85.dd15 <- exp(b.SLE + 35*b.SLE.age + 15*b.dd.SLE) #SJO-------------------------------------------------------- or.SJO.30.dd2 <- exp(b.SJO + -20*b.SJO.age + 2*b.dd.SJO) or.SJO.50.dd2 <- exp(b.SJO + 0*b.SJO.age + 2*b.dd.SJO) or.SJO.70.dd2 <- exp(b.SJO + 20*b.SJO.age + 2*b.dd.SJO) or.SJO.30.dd10 <- exp(b.SJO + -20*b.SJO.age + 10*b.dd.SJO) or.SJO.50.dd10 <- exp(b.SJO + 0*b.SJO.age + 10*b.dd.SJO) or.SJO.70.dd10 <- exp(b.SJO + 20*b.SJO.age + 10*b.dd.SJO) or.SJO.32.dd2 <- exp(b.SJO + -18*b.SJO.age + 2*b.dd.SJO) or.SJO.35.dd5 <- exp(b.SJO + -15*b.SJO.age + 5*b.dd.SJO) or.SJO.40.dd10 <- exp(b.SJO + -10*b.SJO.age + 10*b.dd.SJO) or.SJO.45.dd15 <- exp(b.SJO + -5*b.SJO.age + 15*b.dd.SJO) or.SJO.52.dd2 <- exp(b.SJO + 2*b.SJO.age + 2*b.dd.SJO) or.SJO.55.dd5 <- exp(b.SJO + 5*b.SJO.age + 5*b.dd.SJO) or.SJO.60.dd10 <- exp(b.SJO + 10*b.SJO.age + 10*b.dd.SJO) or.SJO.65.dd15 <- exp(b.SJO + 15*b.SJO.age + 15*b.dd.SJO) or.SJO.72.dd2 <- exp(b.SJO + 22*b.SJO.age + 2*b.dd.SJO) or.SJO.75.dd5 <- exp(b.SJO + 25*b.SJO.age + 5*b.dd.SJO) or.SJO.80.dd10 <- exp(b.SJO + 30*b.SJO.age + 10*b.dd.SJO) or.SJO.85.dd15 <- exp(b.SJO + 35*b.SJO.age + 15*b.dd.SJO) } # end model

   

Page 108: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

  107  

APPENDIX  D  –  Coefficients  from  all  models  

   

               

Coefficients  (95%  CrI)  from  model  1:      

Variable General population

comparison Pre-diagnosis comparison

Regional comparison analysis

Intercept -3.522 (-3.916,-3.128) -4.690 (-5.162,-4.218) -4.829 (-5.075,-4.581)

Sjogren's -0.510 (-0.830,-0.190) -0.336 (-0.530,-0.144) -0.366 (-0.570,-0.169)

SLE -0.117 (-0.303,0.071) -0.086 (-0.215,0.038) -0.451 (-0.589,-0.313

age -0.045 (-0.046,-0.043) -0.034 (-0.037,-0.031) -0.032 (-0.037,-0.027)

age2 9.38E-4 (8.64E-4,1.02E-3) 6.61E-4 (5.00E-4,8.22E-4) 5.63E-4 (2.94E-4,8.26E-4)

female -0.129 (-0.182,-0.077) -0.016 (-0.126,0.102) 0.003 (-0.207,0.210)

 

Coefficients  (95%  CrI)  from  model  2  (presented  in  body  of  thesis):    

Variable General population

comparison Pre-diagnosis comparison

Regional comparison analysis

Intercept -2.764 (-2.814,-2.709) -3,059 (-3.204,-2.941) -4.801 (-5.028,-4.565)

Sjogren's -0.511 (-0.835,-0.199) -0.324 (-0.526,-0.131) -0.538 (-0.913,-0.170)

Sjogren's x age 0.025 (0.015,0.035) 0.015 (0.008,0.023) 0.011 (-0.002,0.024)

SLE -0.104 (-0.294,0.090) -0.079 (-0.200,0.043) -0.235 (-0.455,-0.018)

SLE x age 0.022 (0.016,0.028) 0.012 (0.007,0.018) 0.014 (0.004,0.023)

age2 1.01E-3 (9.29E-4,1.08E-3) 7.11E-4 (5.51E-4,8.68E-4) 5.05E-4 (2.33E-4,7.78E-4)

age 0.045 (-0.046,-0.044) -0.034 (-0.037,-0.031) -0.037 (-0.043,-0.031) Sjogren's disease duration 0.045 (-0.001,0.093) 0.012 (-0.019,0.045) 0.028 (-0.033,0.086) SLE disease duration -0.043 (-0.069,-0.018) -0.030 (-0.048,-0.013) 0.020 (-0.051,0.011)

female -0.130 (-0.182,-0.078) -0.002 (-0.116,0.117) -0.048 (-0.240,0.147)

 

Page 109: Estimatinghealth-selectivemigration)in) patients ...digitool.library.mcgill.ca/thesisfile114191.pdf · des! déménagements! chez! des! patients! ayant! le! lupus! érythémateux!

 108  

   

 

Coefficients  (95%  CrI)  from  model  3:    

Variable General population

comparison Pre-diagnosis comparison

Regional comparison analysis

Intercept -3.522 (-3.916,-3.128) -4.690 (-5.162,-4.218) -4.914 (-6.268,-2.884)

Sjogren's -0.510 (-0.830,-0.190) -0.336 (-0.530,-0.144) -0.533 (-0.890,-0.178)

Sjogren's x age 0.023 (0.013,0.033) 0.016 (0.009,0.023) 0.011 (-0.003,0.025)

SLE -0.117 (-0.303,0.071) -0.086 (-0.215,0.038) -0.239 (-0.467,-0.021)

SLE x age 0.021 (0.015,0.027) 0.013 (0.007,0.018) 0.016 (0.006,0.025)

age2 9.38E-4 (8.64E-4,1.02E-3) 6.61E-4 (5.00E-4,8.22E-4) 4.86E-4 (1.99E-4,7.66E-4)

age -0.045 (-0.046,-0.043) -0.034 (-0.037,-0.031) -0.036 (-0.042,0.031) Sjogren's disease duration

0.042 (-0.005,0.090) 0.013 (-0.019,0.043) 0.028 (-0.031,0.085)

SLE disease duration

-0.042 (-0.067,-0.017) -0.031 (-0.049,-0.014) -0.023 (-0.054,0.009)

Education -0.002 (-0.008,0.002) -0.004 (-0.016,0.003) 0.002 (-0.016,0.015)

Employment 0.011 (0.003,0.016) 0.021 (0.006,0.031) 0.019 (0.001,0.033)

female -0.129 (-0.182,-0.077) -0.016 (-0.126,0.102) -0.018 (-0.218,0.0178) FSA with a rheumatologist

0.092 (0.048,0.136) -0.064 (-0.158,0.028) -0.022 (-0.199,0.152)

Income -0.002 (-0.004,0.000) -0.007 (-0.011,-0.003) -0.013 (-0.023,-0.05)

rent 0.009 (0.007,0.011) 0.009 (0.005,0.012) -0.004 (-0.010,0.001)

CMA -0.559 (-0.623,-0.494) 0.188 (0.075,0.305) -0.140 (-0.326,0.037)