tient diagnostic codes for MS (ICD-9-CM 340) or one inpatient code during the indexperiod (1/1/2004-6/30/2008) were included. Continuous enrollment required 6 mospre-index and 12 mos post-index. A sub-group of newly treated patients, defined asMS patients free of DMT claims for 6 mos prior to initial MS diagnosis, were alsoanalyzed for frequencies of relapses. Using a claims-based definition, severe re-lapses were defined as an inpatient hospitalization with a primary diagnosis of MS.Moderate relapses were defined as an outpatient visit with a diagnosis of MS incombination with a pharmacy or medical claim for a corticosteroid within 7 daysfollowing the outpatient visit. Additional data collected included relapse-relatedcosts and length of inpatient stay. All variables were analyzed descriptively.RESULTS: Among the 23,503 MS patients identified, mean age was 47 years and18,027 (77%) were female. Overall, 1,479 (6.3%) patients had a relapse. 940 patients(4.0%) had a severe relapse, with an average length of inpatient stay of 5.6 days.Among, the sub group of newly treated MS patients (NTMS) (N� 15,059), 1,059 (7.0%)were classified as having a relapse, 710 (4.7%) were severe. Mean cost of relapse[NTMS] was $12,558 [$11,485] for severe, and $1,561 [$1,844] for moderate.CONCLUSIONS: This descriptive analysis provides an updated estimate of the fre-quency and direct medical costs of both severe and moderate relapses amongmultiple sclerosis patients. The ranges of health care services used to manage arelapse reflect severity level. Further database analyses evaluating the impact ofdisease modifying treatment on the rates and costs of relapses is warranted.

PND55USE OF TIME-TO-EVENT ANALYSES TO DEFINE EPISODES OF CARE IN SICKLECELL DISEASEEworuke E, Kauf TUniversity of Florida, Gainesville, FL, USA

OBJECTIVES: With inter- and intra-patient variation in cost and health resourceutilization, standard approaches to defining an episode of care (EOC) may performpoorly. This study takes into account such variation and uses time-to-event anal-ysis within an EOC framework to examine length and cost of care for vaso-occlu-sive crisis (VOC) in sickle cell disease (SCD). METHODS: Florida Medicaid adminis-trative data from 2001-2005 were used to examine EOCs. Enrollees under the age of65 with �1 inpatient or 2 outpatient claims �30 days apart containing a primarydiagnosis of SCD (ICD-9 282.xx) and �6 months of continuous eligibility were in-cluded in the study. Episodes began with the first VOC-related claim. Parametricsurvival analysis was used to calculate episode length as the number of days bywhich adjudicated payments returned to each patient’s pre-diagnosis averagecharge. Episode costs were calculated as the difference between pre- and post-diagnosis payments. EOCs were calculated by subgroups according to age (pediatricand adult), gender, and presence of significant co-morbidity. RESULTS: Among2,543 individuals included in the study, mean age was 14.4 years (standard devia-tion [SD]�11.9) and 48.2% were male. Mean episode length was 11.6 days (95%confidence interval [CI]: 10.9-12.3). Pediatric patients had shorter episodes com-pared to adults (10.5 versus 16.4 days, respectively), but there were no differencesby gender. The presence of acute chest syndrome secondary to VOC increasedepisode length to 13.4 days. The incremental cost of VOC was $327 (95% CI: $310-$343). Costs were slightly higher for males versus females and for adults versuspediatric patients. CONCLUSIONS: Episode length as determined by parametricsurvival analysis was consistent with the clinical presentation of VOC. Our analysissuggests that cost may vary among patient groups with similar episode length.Future work aims to quantify differences between standard and parameteric-based EOC approaches.

PND56COMPARISON OF THE HEALTH CARE UTILIZATION AND COSTS OF MEDICALLYMANAGED PATIENTS VERSUS DEEP BRAIN STUMULATION PATIENTS WITHPARKINSON’S DISEASEWu CF1, Bockstedt L2, Halseth MJ21University of Minnesota, Minneapolis, MN, USA, 2Medtronic, Inc., Minneapolis, MN, USA

OBJECTIVES: Parkinson’s disease (PD) is a degenerative brain disorder of adultonset that impairs motor function. As PD progresses, patients experience reduc-tions in physical functioning and overall quality of life. Deep brain stimulation(DBS) is an implanted medical device that delivers electrical stimulation to specificarea(s) of the brain that control movement and muscle functioning. DBS has beenshown in clinical trials to significantly improve motor functioning, reduce disabil-ity and improve the quality of life in patients and caregivers of PD patients. Theobjective of this study is to compare the healthcare utilization and costs of PDpatients receiving DBS versus medically management. METHODS: Patients with PDbetween 2006 and 2008 were identified the MarketScan Commercial Claims andEncounters database. PD patients were segmented into medically managed pa-tients and deep brain stimulation (DBS) patients. Medically managed PD patientswere required to be on at least one anti-parkinsonian medication. DBS patientswere identified using surgical procedure codes for lead implantation; an indexedlook-back period was used to ensure DBS patients were newly implanted. All pa-tients were followed for one year. Annual healthcare utilization and per patienthealthcare costs were compared. RESULTS: We found that DBS patients had sig-nificantly fewer inpatient hospitalizations (1.6 v. 2.4) and shorter lengths of stay(2.8 v. 6.2) compared to medically managed PD patients (p� 0.01). However, DBSpatients had significantly higher inpatient costs, resulting from the costs associ-ated with DBS implantation. DBS patients had significantly lower annual pharma-ceutical costs than PD patients ($7,105 v $11,735). The differences may be underes-timated as DBS patients typically have more advanced PD than medicallymanaged. CONCLUSIONS: Our results demonstrate that medically managed PD

patients accrued significantly higher pharmaceutical costs and inpatient hospital-ization stays compared to DBS patients. These results suggest DBS patients mayrequire fewer medications to manage PD symptoms.

PND57TEMPORAL TRENDS AND GEOGRAPHIC DISCREPANCIES IN PUBLICEXPENDITURES WITH MULTIPLE SCLEROSIS DRUG TREATMENT IN BRAZILTakemoto MMS, Takemoto ML, Fernandes RA, Duarte GGF, Tolentino ACM, Santos PML,Moretti AIPANOVA - Knowledge Translation, Rio de Janeiro, Brazil

OBJECTIVES: This study aims to describe temporal trends and geographic discrep-ancies on public pharmaceutical expenditures with multiple sclerosis (MS) treat-ment in Brazil. METHODS: Longitudinal analysis of Brazilian MS pharmacy claimsas reported in the Brazilian Ambulatory Information System Database. Analyseswere based on aggregate data from the 27 Brazilian states, observed annually forthe period 2006–2009. The total MS-related expenditures were segmented by state,drug (Glatiramer 20mg, Betainterferon 22mcg, Betainterferon 44mcg, Betainter-feron 30 mcg, and Betainterferon 300mcg), and year. Per capita calculations werealso performed using Brazilian 2009 population. RESULTS: MS public expenditureswith MS drugs had a significant rising from 30,423,930BRL in 2006 to 214,405,349BRLin 2009, a more than 7-fold increasing. Total investments in MS drugs totaled629,917,685BRL. Federal funding transfers for the state of Roraima presented thehighest growth rate (1,497%) rising from 10,549BRL in 2006 to 157,973BRL in 2009.São Paulo (41,384,039 inhabitants, 1/5 of Brazilian population) was responsible forthe higher absolute investment (13,130,526BRL and 88,948,997BRL in 2006 and 2009,respectively) and the higher total expenditure for the period (259,591,564BRL, 1/3 ofnational expenditures). The lowest spending was observed for states in the NorthRegion – Amapá had any MS claim in the 4-year period and Acre had the lowestvalue (184,754BRL). Both states had similar projected 2009 population and are lo-cated at regions with very similar characteristics (both ethnically and geographi-cally). Betainterferon 44mcg represented the highest expenditures in 2009 and hadthe highest annual cost per patient (54,288BRL). Conversely, Betainterferon 22mcgwas the only drug with decrement in the period, decreasing from 34,774,157BRL in2007 to 32,304,159BRL in 2009. CONCLUSIONS: Our findings highlighted geographicdiscrepancies within the Brazilian healthcare system in terms of MS treatmentfunding even when demographic aspects were considered. Nevertheless a con-stant increasing in public expenditures was observed across most states.

PND58INPATIENT HEALTH RESOURCE UTILIZATION AMONG MULTIPLE SCLEROSISPATIENTS IN THE BRAZILIAN PUBLIC HEALTH CARE SYSTEMTakemoto MMS, Takemoto ML, Fernandes RA, Tolentino ACM, Santos PML, Duarte GGF,Moretti AIPANOVA - Knowledge Translation, Rio de Janeiro, Brazil

OBJECTIVES: To describe inpatient resource use among multiple sclerosis (MS)patients within the Brazilian public healthcare system (BPHS). METHODS: Inpa-tient admissions were obtained from the Brazilian Hospital Information SystemDatabase for the period 2006-2009. Records were included if an ICD-10 code G35(MS) appeared as primary or secondary reason for hospitalization. The followingvariables were collected: procedure code for the inpatient admission (relapse man-agement versus other indications), mean length of stay, mean Intensive Care Unit(ICU) days, in-hospital mortality, mean cost per hospitalization. Hospitalizationrates were calculated using the estimated MS population under treatment in theBPHS (estimated from pharmacy claims). Relapse-related admissions were sepa-rately analyzed. RESULTS: Annual hospitalizations due to MS were 2,142 in 2006,2,268 in 2007, 1,648 in 2008, and 1,689 in 2009. The hospitalization rates among MSpatients attending BPHS facilities decreased from 0.60 in 2006 to 0.28 in 2009.Among all admissions, 91.7% were relapse-related in 2006 with similar proportionsin 2007 and 2008, and a slightly decrement in 2009 (88.0%). The in-hospital mortal-ity rate is generally low, with the highest value in 2008 (3.09%) and the lowest in2009 (1.6%). There were no differences between relapse-related and non-relapse-related admissions in terms of in-hospital mortality. The mean length of stay was8.9 days for all MS-related hospitalizations and 7.7 days for those to manage MS.ICU claims were rare in this sample. In 2009, the observed average cost per hospi-talization was 675 BRL and the total expenditure with MS inpatient admission was1,140,418BRL (corresponding to 0.53% of pharmaceutical expenditures due to MS).CONCLUSIONS: Relapses are responsible for over 90% of all inpatient admissions ofMS patients. Although higher health resource consumption is expected in the out-patient setting, therapeutic strategies directed at reducing the incidence of re-lapses can potentially lead to savings within the BPHS.

PND59MEDICATION TREATMENT PATTERNS FOR MULTIPLE SCLEROSIS PATIENTS INTHE BRAZILIAN PUBLIC HEALTH CARE SYSTEMTakemoto MMS, Takemoto ML, Fernandes RA, Duarte GGF, Moretti AIP, Santos PML,Tolentino ACMANOVA - Knowledge Translation, Rio de Janeiro, Brazil

OBJECTIVES: This study aimed to investigate the patterns of medication treatmentamong patients with Multiple Sclerosis (MS) treated in the Brazilian Public Health-care System (BPHS) and examine possible temporal and spatial trends in thosepatterns. METHODS: The study was a longitudinal analysis of Brazilian MS phar-macy claims as reported in the Brazilian Ambulatory Information System Databasefor the period 2006-2009. Five different drugs currently recommended at BPHS MSguidelines were included in the analysis (Glatiramer 20mg, Betainterferon 22mcg,Betainterferon 44mcg, Betainterferon 30mcg, Betainterferon 300mcg) and the pro-

A212 V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) A 1 – A 2 1 4

portion of patients receiving each drug was calculated using defined daily doses.Changes in this distribution over time across all 27 Brazilian states were analyzed.RESULTS: The estimated yearly number of MS patients treated within the BPHSwas 3,569, 4,894, 5,800, and 6,099 from 2006 to 2009. The probability of use ofGlatiramer 20mg increased from 9.5% in 2006 to 20.1% in 2007, maintaining similarproportions in the following years (20.7% and 21.28%). The lower proportion oftreated patients was observed for Betainterferon 22mcg, with a constant decre-ment from 2006 (17.26%) to 2009 (11.39%). The other three available Betainterferonformulations represented stable distribution of 20-25% each across all years with aslightly greater probability of use of the 30mcg intramuscular formulation (admin-istered weekly). Between-states differences were also analyzed and revealed im-portant spatial discrepancies. For instance, the probability of use of Glatiramer in2009 was 32% in Bahia (Northeast Region) and 13% in Santa Catarina (South Region)and the proportion of MS patients taking Betainterferon 30mcg was 36% in RioGrande do Sul (South Region) and 15% in Bahia. CONCLUSIONS: Although a pro-portional distribution of Glatiramer and Betainterferon formulations is coherentwith the BPHS guideline which stated that their efficacy should be consideredsimilar, significant between-states discrepancies in terms of treatment patternswere identified.

Neurological Disorders – Research on Methods

PND60TESTING A COMPUTERIZED METHOD OF ASSESSING MOVEMENT ACCURACYAllen DD1, Ng M1, Greenwald CP2

1University of California, San Francisco/San Francisco State University, San Francisco, CA, USA,2John Muir Hospital, Walnut Creek, CA, USAOBJECTIVES: Computerized games challenge players to move quickly and accu-rately, but result in game-specific scores rather than objective data. Newly devel-oped AccuTrak software with Wii technology can challenge players to move acursor to an on-screen target using hands or feet, but also record data that mighthelp document coordination deficits in people with disabilities. Our purpose was toexamine the reliability and validity of this method. METHODS: Thirty-six healthyadults and nine people with multiple sclerosis (MS) used a Wii device to move acursor and click on the center of a target appearing on a computer screen. AccuTraksoftware provided feedback after each set of 6 trials regarding a) the time fromtarget appearance to click, and b) distance from the target center. Each hand per-formed 30 trials; each foot performed 36 trials. Time and distance data were com-pared left-right and hand-foot, and to self-reported movement accuracy from theMovement Ability Measure as analyzed using item response theory methods.RESULTS: Intraclass correlation coefficients (ICCs) time and distance data were0.66 and 0.90, respectively, between dominant and non-dominant hands; betweenhand and foot performances ICCs were 0.70 and 0.94. The correlation betweenobjective data and logit estimates of self-reported ability to move accurately was�0.53 for time and 0.21 for distance. The time-distance relationship fit a powertrendline with R2 equal to 0.80 for healthy volunteers; volunteers with MS did notfit the same trendline. CONCLUSIONS: Reliability was better for distance data;concurrent validity with self-report was better with time data. Construct validitywas supported in the observance of Fitts’ Law with speed-accuracy trade-off inhealthy volunteers; people with MS tended to have increased error even with lon-ger times. The methodology shows promise for use when investigating coordina-tion deficits such as those in people with multiple sclerosis.

PND61EVALUATION OF CONSISTENCY BETWEEN MULTIPLE SCLEROSIS REGISTRYPUBLICATIONSLion M, Murphy D, Hettle R, Pietri G, Lock K, Moorcroft EHERON Evidence Development Ltd, Luton, UKOBJECTIVES: Multiple Sclerosis (MS) registries collect patient-level longitudinaldata with the aim of improving our understanding of MS. These databases haveprovided extensive studies on the natural progression of MS. The purpose of thisstudy is to assess how disease progression has been estimated and assess theconsistency in methodology and results. METHODS: The publications of 10 majorMS Registry websites were searched, followed up with keyword searches in Embaseand Pubmed. Population-based natural history studies on disease progression wereincluded. Primary and non-primary endpoints, such as Expanded Disability StatusScale (EDSS) outcomes as a measure of progression, were extracted in addition tostatistical methodologies. We evaluated the consistency between studies in termsof endpoints and statistical methodology. RESULTS: Our search identified 23 stud-ies, of which nine met the inclusion criteria. The majority of papers utilized theKaplan-Meier Survival technique to estimate time until disease endpoints and Coxproportional hazard models to determine prognostic factors. Lack of standards inreporting results prevented a global comparison. The most commonly reportedendpoint was median time to EDSS six for MS or relapsing-remitting MS patients,enabling comparison between six studies. Values were reported between 11.9 to27.9 years. Excluding studies prior to 1999, the median time to EDSS six was be-tween 20-27.9 years. Results reported from the European based Lyon and LORSEPregistries were consistent (20-24 years), while studies in Canada and Germanyshowed greater disparity. Common prognostic factors included age and type of MSat onset. CONCLUSIONS: Lack of international standards for reporting outcomes innatural history progression hinders comparisons between studies, especially be-tween the United States and Europe. While there is some consensus between reg-istries regarding prognostic factors for progression, not all agree. Heterogeneity inunderlying characteristics of the populations, evolution in best supportive care andaccess to treatments could all contribute.

PND62MEDICATION POSSESSION RATIO (MPR): A COMPARISON OF TRADITIONAL MPRAND MODIFIED MPR FOR MULTIPLE SCLEROSIS (MS) PATIENTS PRESCRIBEDDISEASE MODIFYING DRUGS (DMDS)Dickson M1, Kozma C2, Phillips A3, Meletiche DM3

1University of South Carolina, Columbia, SC, USA, 2Independent Research Consultant/AdjunctProfessor, University of South Carolina, St. Helena Island, SC, USA, 3EMD Serono, Inc., Rockland,MA, USAOBJECTIVES: The objective of this analysis was to compare two commonly usedadherence calculations for multiple sclerosis patients prescribed disease modify-ing drugs (DMDs) in a national managed care population. METHODS: Patients withpharmacy claims for self-injectable DMDs were selected from 2007-2008 ThomsonMedStat data. Adherence was calculated across all DMDs for 12 months after thefirst DMD claim (i.e., index date) using two different adherence calculation meth-ods. Traditional medication possession ratio (TMPR) was calculated by summingthe days supply from the first to the last prescription and dividing by the timebetween the last prescription date plus the days supply on the last prescription andthe first prescription date. The MMPR used the same numerator but divided by the365 days of the follow-up period. The TMPR was calculated based on available data(no eligibility requirements) while MMPR requires continuous eligibility for the12-month follow-up period. RESULTS: The mean adherence value for TMPR(n�3,405) was 90.5% whereas the mean adherence value for MMPR (n�2,145) was78.0%. Based on TMPR, 82.3% of patients were considered adherent (�80%) whilethis value decreased to 63.7% for MMPR. CONCLUSIONS: The MPR is often used todescribe patient adherence. This adherence measure, however, can be calculatedusing different time periods in the denominator. These results demonstrate theimportance of understanding the adherence calculation method and the popula-tion in which the measure is generated, and the potential implications to patientcare.

PND63LITTLE OR NO TREATMENT EFFECT? APPLICATION OF GROWTH MIXTUREMODELS TO EXPLORE UNKNOWN SUBGROUPS OF DIFFERENTIAL RESPONDERSTO TREATMENTStull DE1, Houghton K2, Wesnes K3

1United BioSource Corporation, Bethesda, MD, USA, 2United BioSource Corporation, London, UK,3United BioSource Corporation, Goring-on-Thames, UKOBJECTIVES: To seek to identify unknown subgroups of patients who exhibit treat-ment effects using growth mixture model (GMM) analyses in studies where con-ventional analysis failed to find overall treatment effects. METHODS: Analyseswere performed on data from two 26-week clinical trials assessing the effects ofD-Cycloserine on cognitive function using the CDR System in patients with Alzhei-mer’s disease. Both trials contained three dose-ranging treatment groups and oneplacebo group. Latent growth model (LGM) analyses explored composite scores ofattention (Power of Attention - PowAtt) and speed of retrieval of information heldin memory (SpeedMem), while GMM analyses were used to identify latent classesof differential responders. RESULTS: While no overall treatment benefit was iden-tified, variability in PowAtt and SpeedMem was found suggesting possible subsetsof patients with differential response to treatment. GMM analyses identified twodistinct subsets of patients, one comprising 70% and the other 30% of the trialsample. The larger subset had low baseline SpeedMem scores compared to thesmaller subset (range: 6 – 8 seconds total response time vs. 16-20 seconds) andexhibited no change over time. Within the smaller subset, the low/medium doseand placebo arms showed deterioration in SpeedMem scores of 3.4 to 9 seconds at26 weeks. The low/medium dose patients showed the greatest deteriorationwhereas the high dose patients showed little deterioration at 26 weeks (1.5seconds). CONCLUSIONS: Original analyses indicated no treatment benefit whileLGMs showed all treatment arms to deteriorate over the trial. The GMMs, however,indicated a subset of patients on high dose (7% of trial sample) who showed littledeterioration whereas all other treatment arms showed substantial slowing onmemory tests. The GMMs thus have the potential to identify subsets of patientswho may benefit from treatment even when the main trial findings are inconclu-sive.

PND64APPLICATION OF PREDICTIVE MODELING TO CLASSIFY FREQUENT SNORINGIDENTIFIED FROM ROUTINE MEDICAL EXAMINATIONS USING THE NHANESDATABASERow BUniversity of Louisville, Louisville, KY, USAOBJECTIVES: Increased upper airway resistance during sleep, or snoring, is a riskfactor for sleep disordered breathing, and has been implicated in the developmentof adverse cardiovascular outcomes as well as components of the metabolic syn-drome such as obesity, insulin resistance, and dyslipidemia. Despite increasingawareness of the health risks associated with frequent snoring, many patientsremain untreated and may be unaware of snoring, especially if living alone. For thisreason, we examined whether data available in routinely administered physiolog-ical and laboratory exams would prove useful in developing predictive models ofhabitual snoring. METHODS: A total of 10,482 respondents from the 2005-2008National Health and Nutrition Examination Survey (NHANES), for which individualsleep survey, demographic, and physiological data were available, and who werenot previously diagnosed with sleep apnea, were classified as frequent snorers (5 ormore nights per week, n� 3222), or control (n�7260). Sample data were partitionedinto training (45%), validation (35%), and testing (30%) data sets using an equalstratification criterion for development of logistic regression, decision tree, andneural network predictive models using SAS Enterprise Miner. RESULTS: All three

A213V A L U E I N H E A L T H 1 4 ( 2 0 1 1 ) A 1 – A 2 1 4