Epidemiology Community Health A case-control of childhood ... · Acase-control study ofchildhoodpedestrian injuries in Perth, Western Australia for example, spend equal amounts oftime

26ournal of Epidemiology and Community Health 1996;50:280-287

A case-control study of childhood pedestrianinjuries in Perth, Western Australia

Mark Stevenson, Konrad Jamrozik, Paul Burton

Department ofEpidemiology &Biostatistics,School of PublicHealth,Curtin Universityof Technology,GPO Box U1987,Perth, WesternAustralia 6001M Stevenson

Department ofPublic Health,The University ofWestern Australia,Perth,Western Australia 6907K Jamrozik

TVW TelethonInstitute for ChildHealth Research,PO Box 855,West Perth,Western Australia 6872P Burton

Correspondence to:Dr M Stevenson,National Center for InjuryPrevention and Control,Centers for Disease Controland Prevention,4770 Buford Highway,NE (K63), Atlanta,Georgia 3-0341-3724,USA.

Accepted for publicationDecember 1995

AbstractStudy objectives - To identify the de-terminants of childhood pedestrian in-juries, taking the child's exposure to theroad environment into account.Design - This was a case-control study.Setting and participants - The study wasconducted in Perth, Western Australia be-tween 1991 and 1993. Altogether 100 in-jured and 400 uninjured child pedestriansaged 1 to 14 years were studied. Aspectsof the child's social and physical en-vironments, measures of his or her be-haviour, cognitive skills, and "habitual"exposure to the road environment, as wellas his or her knowledge of road safety,were recorded.Main results - The likelihood of injuryincreased by 12% with each 10 000 vehiclesper day increase in the volume of traffic(odds ratio (OR) 1P12, 95% confidence in-terval (CI) = 1-05, 1.19) on roads most fre-quently crossed. In addition, the presenceof visual obstacles on the verge of thechild's street of residence increased thelikelihood ofinjury by more than 2*6 times(OR 2-68, 95% CI = 1-42, 5 02). In contrast,the absence of footpaths was associatedwith a 52% reduction in the likelihood ofinjury compared with the presence offoot-paths on the child's street ofresidence (OR0-48, 95% CI=0-27, 0.87).Conclusion - The amount of exposure tothe road environment and the nature ofthe road environment to which the childpedestrian was exposed partly influencedthe likelihood of injury in children fromlow socioeconomic areas, male children,and children aged 13 to 14 years. Untilnow, the excess incidence of childhoodpedestrian injuries in these subgroups ofthe population had not been explained be-cause the child's exposure per se had notbeen examined.

(Ji Epidemiol Community Health 1996;50:280-287)

Human, motor vehicle, and environmental riskfactors have become a major public healthconcern for Australian children. For example,road systems and motor vehicles have beendesigned for greater efficiency with little con-sideration for the safety of pedestrians. Con-sequently, motor vehicle related injuries tochildren are a significant health problem inwestern society.'

Each year in Australia approximately 110children die and a further 1800 are admittedto hospital as a result of collisions betweenpedestrians and motor vehicles.2 In WesternAustralia alone (total population 1P7 million),childhood pedestrian injuries account for 1300bed-days per year, with most injuries admittedbeing fractures (40%) and intracranial andother internal injuries (30%).3 Andreassen45estimated the direct costs of a pedestrian injuryin urban or rural environments in Australiato be A$89 000 and A$104 000, respectively.These are estimates for pedestrian morbidityalone. If one then considers the potential yearsof life lost for fatal pedestrian injuries, the costis even greater.The child pedestrian fatality rate for Western

Australian children aged 1 to 14 years of 3-2/100 000 per year' is higher than that for chil-dren aged 1 to 14 years in Australia as a whole(3 1/100 000 per year), and is also higher thanthe rates for the United Kingdom7 (2A4/100 000per year) and the United States of America(2 4/100 000 per year)8 for children of the sameage range. The lack of any significant reductionin the mortality rates from pedestrian injury inWestern Australia over the past decade6 sug-gests that child pedestrian injury will continueto represent a major child health problem ifefforts are not directed at prevention. However,at present there are few well established pre-vention strategies.

Children aged 5 to 9 years and males areover represented in pedestrian injuries. Thereis some disagreement in the published reportsabout explanations for these characteristics.Some researchers suggest that boys aged 5 to9 years have an increased exposure to the roadenvironment9"0 while others" 2 attribute theover representation to the child's behaviourwhen crossing the road. However, few studieshave formally examined whether the over rep-resentation of boys aged 5 to 9 years is actuallya result of the amount of exposure to the roadenvironment. The analytical studies publishedto date""'6 have either not measured the child'sexposure to the road environment or, al-ternatively, they have a number of method-ological constraints. For example, Pless,Verreault and Tenina'3 and Backett and John-son '4 matched control subjects to injured ped-estrians by sex. Consequently then, sex cannotbe assessed in terms of its relationship to injuryin either of these studies.The usual descriptive statistics for child ped-

estrian injuries are crude measures that reston an untested assumption that children ofdifferent ages, sexes, and areas of residence,

280

on March 3, 2021 by guest. P

rotected by copyright.http://jech.bm

j.com/

J Epidem

iol Com

munity H

ealth: first published as 10.1136/jech.50.3.280 on 1 June 1996. Dow

nloaded from

http://jech.bmj.com/

A case-control study of childhood pedestrian injuries in Perth, Western Australia

for example, spend equal amounts of time inequally dangerous traffic conditions. Con-sequently, research is needed in order toidentify, firstly, those features of the descriptiveepidemiology that are artefacts arising fromdifferences in children's exposure in the roadenvironment and, secondly, those featureswithin the individual and the physical en-vironment that have a significant influence onthe likelihood of a collision. Without such re-search, preventive strategies will be based in-appropriately on the somewhat unsophisticateddescriptive studies currently available.The race ofthe child, in terms ofwhite versus

non-white, is also associated with differencesin rates of pedestrian injury.`-20 So, too, is apattern of socioeconomic disadvantage as-sociated with higher rates of pedestrianinjury."72"22 However, the family's or child'ssocioeconomic status has seldom been com-prehensively measured by researchers. Fre-quently only one variable (such as the father'soccupation) has been used as a measure ofsocioeconomic status.23 Further research in--corporating a comprehensive assessment of thechild's or family's socioeconomic environmentis needed to determine to what extent socio-economic factors are associated with childhoodpedestrian injury.A scarcity of controlled studies examining

the relationship between developmental abil-ities and child pedestrian injury is evident inthe published reports. If one considers thatthe learning processes of children are likely toremain the same, while the number of motorvehicles in the community increases, it is likelythat the pedestrian task will become furthercomplicated in the future. Consequently, inorder to initiate appropriate prevention pro-grammes, it is necessary to determine whether,in fact, the distribution of pedestrian injuriesis related to the cognitive abilities and de-velopmental stage of children.

Associations between childhood pedestrianinjuries and each of the volume and speed oftraffic, and the absence of play areas have alsobeen reported.2425 However, the research todate has not applied sufficiently reliable andvalid methods of measurement or includedsufficient numbers of subjects to determine thecontribution made by relevant traffic factors tothe risk of pedestrian injury. Furthermore, fewstudies have compared the prevalence of spe-cific physical environmental features such asthe presence of traffic control devices in theneighbourhoods of injured children, with theprevalence of such features in the neigh-bourhoods of uninjured children. More re-search is required to increase our under-standing of the influence of these factors andto identify aspects of the traffic environmentthat can be modified to reduce pedestrian in-juries among children.The present case-control study was con-

ducted in the metropolitan area of Perth (totalpopulation 1 2 million), Western Australia be-tween December 1991 and December 1993,and was carefully designed to address themethodological flaws found in previous studies.

MethodsCase subjects were children aged 1 to 14 yearswho had sustained an injury in a collision witha motor vehicle while walking, running, or, inthe case of infants, crawling on a roadway, roadverge, or footpath in the Perth metropolitanarea. Cases were identified prospectively fromDecember 1991 to December 1993 through anumber of monitoring systems established atthe only paediatric hospital in Perth (PrincessMargaret Hospital for Children), the HealthDepartment of Western Australia, and theWestern Australian Police Department. Whena case subject was identified, the parent orguardian and the injured child were invited byletter to participate. Children injured whileriding bicycles or skateboards or injured by anon-motorised vehicle such as a bicycle wereexcluded from the study.Of the 214 children injured as pedestrians,

15% (n = 32) could not be contacted. Of theremaining 182 children that could be con-tacted, 30% (n=65) proved to be ineligible.One hundred of the remaining 117 were re-cruited, a corrected response rate of 85%.Four hundred non-injured controls were se-

lected from either a list of all public and privateprimary and secondary schools in the Perthmetropolitan area or the Midwives NotificationSystem. Schools were selected with a samplingprobability that was proportional to the numberof children on the school roll, and which re-flected the ratio of public to private schools of2-4: 1. Controls were frequency matched tocases, with age (three strata) and sex as fre-quency matching variables. Preschool age (1to 4 years) control subjects were randomlyselected from the Midwives Notification Sys-tem (one of only two population based registersof preschool age) and frequency matched tocases of preschool age. Selected controls ofpreschool age were traced via immunisationrecords, telephone directories, and electoralenrolments. Letters were then sent invitingthe parents or guardians and their child toparticipate in the study. Of the 499 households(77%) with which contact was made, 16 provedineligible to participate. The response rate forthe remaining 483 households was 83% (n=400).For both cases and controls, face to face

structured interviews were undertaken withboth parents or guardians and the child. Theinterview covered six well defined sets of char-acteristics of the child. Firstly, a comprehensivesocial index, as described by Osborn and Mor-nis,26 was used to rank families along a con-tinuum of social and economic inequality.Secondly, measures of the child's exposure tothe road environment, such as the number ofwalking trips usually undertaken in a period ofseven days, the number of roads crossed, andthe volume of traffic on the roads crossed (vol-ume was obtained from the local governmentengineers) were recorded. The volume of traffic-was recorded for the road(s) most frequentlycrossed during the day, or, if the frequency ofcrossing was the same for each road, the roadwith the greatest volume oftraffic was recorded.The child's habitual exposure to the road en-

281



j.com/

J Epidem

iol Com

munity H


nloaded from


Stevenson, _Jamrozik, Burton

vironment was recorded rather than the ex-posure of the child to the road environment inthe week preceding the interview as the lattermight have been affected by the occurrence ofthe injury. Furthermore, recording the case orcontrol subject's habitual exposure to the roadenvironment overcame the problem of re-cording a particular week that happened tobe atypical and thus either overestimating orunderestimating the child's exposure to theroad environment.The third aspect of the interview measured

the behaviour of cases and controls using avalid and reliable instrument developed by Jel-linek and Murphy,27 the pediatric symptomchecklist (PSC). The PSC identifies childrenwho have sufficient behavioural problems towarrant referral to a physician for further evalu-ation and treatment. The fourth component ofthe interview schedule measured one aspect ofthe child's development namely, figure-grounddiscrimination, using a subtest of the Illinoistest of psycholinguistic abilities (ITPA).28 Thistest has direct application to the road en-vironment. In it children need to discern mov-ing vehicles which can often be partly obscuredfrom the surrounding environment.The final set of questions elicited both the

child's and parent's or guardian's knowledge ofroad safety. The children and their parentsor guardians were asked to list the necessaryprocedures in order to cross the road safely.Other questions such as, "At what age woulda child be capable of crossing a road safely?"were also asked, but only to the parents orguardians.

Information was also sought from the caseson the date and time of the injury, the exactlocation of the injury, the speed limit and roadhierarchy at the site of injury, and the en-vironmental conditions - for example, theweather prevailing at the time. Road hierarchywas an ordinal variable based on the volumeof traffic on the road namely: cul de sac, res-idential road, sub arterial road, and arterialroad. The case subject's medical record wasalso reviewed in order to classify the severityof injury using the abbreviated injury scale(AIS).29 Direct observation of the physical en-vironment at each case and control subject'sstreet of residence was also undertaken. On-street parking and other features ofthe roadwaysuch as the road type, posted speed limit, thenumber of lanes in the street, the presence offootpaths, whether the street had specific trafficcontrol devices, and whether there were visualobstacles in the street which could obscure thechild's or driver's vision were recorded.

STATISTICAL ANALYSISUnivariate statistics were computed from thefinal data set using the StatisticalAnalysis System(SAS) software.30 Proportions were presentedas percentages of the respective denominator(n). Variations in proportions between casesand controls were initially assessed using thestandard X2 test for association, with continuitycorrection where appropriate. Odds ratios(ORs) and 95% confidence intervals (95% CI)

were derived using the standard methods forcase-control studies as outlined by Breslow andDay.3' The independent contributions to theoverall risk of pedestrian injury in children ofvariables shown to be significant in the uni-variate analyses were assessed using un-conditional multiple logistic regression onEGRET software.32 Two models were con-structed.The first model used data for respondents

aged 4 to 14 years (n = 457, cases, n = 97,controls n = 360). The reason for limiting theage group was the restrictions placed on theuse of the PSC which has been validated onlyfor children aged 4 to 14 years. Consequently,the first model included the score on the PSCand 41 other variables. The second model usedthe variables found to explain the risk of child-hood pedestrian injury in the first model, butwas restricted to the age group 4 to 10 years(n = 289) in order to assess the only othervariable in the data set which had an age re-striction, namely the figure-ground dis-crimination test (ITPA).28One of the objectives in this study was to

determine the extent to which differences inthe amount of exposure can explain the ageand sex pattern of child pedestrians who areinjured most frequently. However, becausecoarse frequency matching by age and sex wasundertaken in this study, it meant that theeffects of age and sex could not be interpreteddirectly. The effects of frequency matchingwere, therefore, overcome by applying an "off-set" which contrasted the difference betweenthe probability of a child in the general popu-lation being in a specific age and sex categorycompared with the equivalent probability in acontrol subject (see Appendix).

ETHICAL APPROVALThe study was approved by the research andethics committees of all organisations involvedin the project. In accordance with the guidelinesof the National Health and Medical ResearchCouncil,33 parents or guardians consented totheir own and to their child's participation inthis research.

ResultsWithin the case group 65% (n = 65) were malewhilst in the control group, 57% (n = 230) weremale. These proportions are not significantlydifferent (y2=1P79, df=1, p=0.19). Therewas a significant difference in the proportionsbetween cases and controls in the three agestrata used for frequency matching. Spe-cifically, there were fewer control subjects, pro-portionally, in the 1 to 4 year age group andmore in the 10 to 14 year age group (X2=9-86, df=2, p=0001). The mean age for casesubjects was 8 years 5 months (SEM=0-33,median = 9 years) and for controls it was 9 years3 months (SEM=0 14, median=10 years).The relationship between 42 risk factors and

childhood pedestrian injury was examinedusing unconditional logistic regression. Table1 lists the descriptive statistics for respondents

282



j.com/

J Epidem

iol Com

munity H


nloaded from



Table 1 Distribution of risk factors in children aged 4 to 14 years

Factors Cases (%/6) Controls (%)

Socioeconomic status<47 (high) 12 (12) 55 (15)48-50 (middle) 77 (79 283 (79)51 + (low) 8 (9) 22 (6)Mean 49-3 49-1(SEM) (0-15) (0 08)No 97 360

Volume of traffic (vehicles per week Mon-Fni)Mean 41956 21698(SEM) (5381) (2102)Median 29627 9172No 76 238Visual obstaclesPresent 74 (76) 196 (55)FootpathsAbsent 44 (44) 223 (56)

Pediatric symptom checklist (PSC)Mean 17 15(SEM) (0-94) (045)Median 17 15No 95 349

Table 2 Model offactors associated with injuries to child pedestrians aged 4 to 14 years

Variable OR (95% CI) Likelihood ratio test(p value)

Socioeconomic status 1-251 (0 535)[2]High 1 00 -Middle 0 79 (0 35, 1-81)Low 1-34 (0-41, 4-28)

Volume of traffic (per increase of 10 000 vehicles per day)1-12 (1-05, 1-19) 10-44 (0001)

Visual obstaclesAbsent* 1 00 -Prelent 2-68 (1-42, 5 02) 10-18 (0001)FootpathsPresent* 1 00 -

Absent 0-48 (0-27, 0-87) 5-93 (0-01)

Pediatric symptom checklist (per unit change, range 0-70)1 03 (1 00, 1-07) 3-91 (0048)

Deviance = 298-95 df= 292

* Baseline Level.

aged 4 to 14 years (n= 475) for those variablesin the definitive model, while table 2 presentsthe results of the most parsimonious un-

conditional logistic regression model for thesame age range.

It is evident from the statistics in table 1 thatcases crossed roads carrying almost twice (1-93times) the volume of traffic compared with theroads crossed on a weekly basis by the controlsubjects. The proportion of children living instreets with visual obstacles on the verge andwith footpaths also differed between case andcontrol subjects.As indicated in table 2, the volume of traffic

on the roads crossed by the child during hisor her habitual weekday (Monday to Friday)exposure to the road environment, the presenceor absence of visual obstacles on the verge ofthe child's street of residence, the presence or

absence of footpaths on the child's street ofresidence, and his or her score on the PSC,each have an independent relationship with thelikelihood of childhood pedestrian injury. Itseems plausible that each of these explain partof the pre-impact phase of pedestrian injuries.There was no significant evidence of a re-

quirement for interaction or polynomial termsin the model.The odds of injury increased by 12% with

each 10000 vehicles per day increase in the

volume of traffic on the roads crossed betweenMondays and Fridays. The presence of a"visual obstacle", defined as something within2 metres of the roadside that was large enoughto obscure three quarters of the child on theverge of the child's street of residence, wasassociated with an increase in the odds of injuryof more than 2t6 times (OR==2-68, 95% CI=1-42, 5 02) compared with the absence of a"visual obstacle". Furthermore, by restrictingthe analysis to cases who were injured in hisor her street of residence (n = 25), the odds ofinjury associated with visual obstacles on theverge increase (OR=2-86, 95% CI=1-28,7 22).The presence of footpaths on the child's

street of residence was also associated with anincrease in the likelihood of pedestrian injury.Put another way, the absence of footpaths onthe child's street of residence was associatedwith a 52% (OR=0-48, 95% CI=0-27, 0 87)reduction in the likelihood of injury. This poss-ibly surprising result will be discussed later.A high or low PSC score indicated the likely

presence or absence respectively of a be-havioural problem. A positive coefficient wasevident for PSC; thus, the more likely thepresence of a behavioural problem, the greaterthe odds of injury. This seems intuitively reas-onable.Having forced socioeconomic status (SES)

(as SES was part of the research question)into the model, a "U" shaped relationship wasevident between SES ofthe respondent's familyand the odds of pedestrian injury, although thiswas not significant at the 5% level. Respondentsfrom families of low SES had 34% greater oddsof injury compared with families of high SES,while middle SES families, those with a socialindex score between 48 and 51, had a reducedodds ofinjury compared with respondents fromboth high and low SES families.A second model which was restricted to the

age group 4 to 10 years (n=289) in order toassess the only other variable in the data setwhich had an age restriction, namely, the figureground discrimination test (ITPA), was similarto the definitive model for children aged 4 to14 years. Performance on the ITPA was notindependently associated with the likelihood ofinjury (data not shown).

Table 3 reports the most parsimonious un-conditional logistic regression model for re-spondents aged 4 to 14 years (n = 475) when the"offset", reflecting the age-sex specific controlsampling fractions (see Appendix), was applied.It is apparent that, other than age and sex, theinterpretation of the remaining variables in themodel remains the same. Age is a significantpredictor of the likelihood of injury and an "n"shaped relationship exists between the age ofthe child and the odds of pedestrian injury.Boys had non-significantly increased odds ofinjury compared with girls (OR= 1-02, 95%CI=0-54, 1-75).A further model (see table 4, which consisted

solely of age, sex, and SES and in which the"offset" applied enables one to compare theunadjusted ORs for age, sex and socioeconomicstatus with the ORs for age, sex and socio-

283



j.com/

J Epidem

iol Com

munity H


nloaded from


Stevenson, J'amrozik, Burton

Table 3 Model of the effects of age and sex on the risk ofpedestrian injury

Variable OR (95% CI) Likelihood ratio test(p value) [dfl

Age group (y) 13 14 (0-044)[3]4 to 6 1 00 -7 to 9 1-76 (0 85, 3 66)10 to 12 3-48 (1 64, 7-38)13 to 14 0-83 (029, 2-36)SexFemale* 1 00Male 1.02 (0 54, 1 75) 0 005 (0-941)Socioeconomic status 1-240 (0 538)[2]High 1 00 -Middle 0-80 (0 35, 1 82)Low 1 34 (0 42, 4 29)

Volume of traffic (per increase of 10 000 vehicles per day)1-12 (1 05, 1 19) 10-51 (0 001)

Visual obstaclesAbsent* 1 00 -Present 2-68 (1 42, 5 02) 10 19 (0 001)FootpathsPresent* 1 00 -Absent 0-48 (0 27, 0 87) 5-93 (0 015)

Pediatric symptom checklist (per unit change, range 0-70)1-03 (1 00, 1-07) 3-97 (0046)


* Baseline Level.

Table 4 The effects of age, sex and socioeconomic unadjusted for exposure andenvironmental factors

Variable OR (95% CI) Likelihood ratio test(p value)[dfl

Age group (y) 13 29 (0-004)[3]4 to 6 1 00 -7 to 9 1-52 (0 40, 1 60)10 to 12 3-38 (0 75, 3 03)13 to 14 0 99 (0 33, 2 38)

SexFemale* 1 00Male 1-14 (0 69, 2 05) 0-276 (0 599)

Socioeconomic status 1-495 (0-474)[21High 1 00 -Middle 1-32 (0 61, 2 82)Low 1-99 (0-66, 5 98)


economic status when the child's exposure tothe road environment and nature of the roadenvironment to which the child was exposedwere taken into account.

Comparison of the ORs in table 4 with thefully adjusted ORs in table 3 indicated that a

child's exposure to the road environment, andthe nature of that environment partially ex-

plains the likelihood of pedestrian injury in thehigh risk group, namely, children from lowsocioeconomic areas and males. In other words,the over representation of boys and those fromless privileged areas in the injury statistics re-

flects, at least in part, the environments towhich these children are exposed rather thanrisks directly attributable to maleness or lackof wealth. However, the strong "n" shapedrelationship with age group would appear tobe independent of the environment to whichchildren are exposed.

DiscussionA review of the literature highlighted that theavailable information on childhood pedestrianinjuries has been gathered predominantly fromdescriptive epidemiological studies. A lim-

itation of many of these studies has been a lackof comparison between data on injury and dataon exposure to the road environment in thesame community. Furthermore, many of thestudies do not consider all of the contributingfactors simultaneously and most do not con-sider what happened to uninjured children.Consequently, a research approach, such asthis study, which compared data on injuredand uninjured children and which considereddata on exposure and as many as possibleof the contributing factors simultaneously wasrequired.We found that the amount of "habitual"

exposure to the road environment and thenature of the road environment to which thepedestrian was exposed influenced the like-lihood of injury, particularly in boys and inchildren aged 13 to 14 years. These effectswere most noticeable in children from lowsocioeconomic areas, but the relationship ofsocioeconomic environment to childhood ped-estrian injury is a complex one. As the overrepresentation of children from low socio-economic areas and boys in the injury statisticscan be explained, in part, by the child's ex-posure to the road environment, one mightsuggest that restricting children's exposure totraffic would bring about reductions in childpedestrian injuries. However, researchers34have suggested that restricting a child's ex-posure to traffic will merely boost the socio-economic disparity in childhood mortality. Forexample, Roberts34 indicated that poorer famil-ies would be least able to escort their childrento school by car and supervising children wouldbe more difficult for single parent families.Consequently, one might propose that modi-fications be made to the environment insteadof restricting a child's right to mobility.

Either no or weak to moderate associationsbetween maladaptive behaviours and the riskof childhood pedestrian injury have been re-ported in the literature.'3 35-36 The inconsistencyin the published reports may be due to thetypes of injuries studied, which varied betweenstudies, as did the tools for assessing the child'sbehaviour. The findings in the present study,however, indicate that a child's behaviour issignificantly associated with the risk of ped-estrian injury in children aged 4 to 14 years.Furthermore, a dose-response relationship wasevident in that the higher the PSC score, whichindicates the likelihood of a behavioural prob-lem, the greater the likelihood of injury.One should be cautious in looking for a

relationship between a child's general be-haviour and the risk specifically of pedestrianinjury. Notwithstanding this, the questions onthe PSC elicit behavioural characteristics suchas aggressiveness, lack of attentiveness, a childwho acts immaturely or fidgets, and so on. Itis likely that characteristics of this nature maycompete with the child's ability to focus onpotentially hazardous road environments andhence increase that child's potential for injury.Furthermore, it is interesting to note that inthe present study the higher PSC scores weremore prevalent in boys than girls.

284



j.com/

J Epidem

iol Com

munity H


nloaded from



Although a statistical association is evidentbetween the PSC score and childhood ped-estrian injury, it would not be appropriate tofocus solely on identifying "at risk" childrenbased on whether they have a behavioural prob-lem or not. To determine whether a child hasa behavioural problem is a complex task andone that does not merely require the applicationof the PSC in the target population. Fur-thermore, even if "at risk" children could beidentified, they would comprise a small group.For example, only 12% of cases and 7% ofcontrol subjects in this study were identified ashaving a behavioural problem. Consequently,the potential contribution to prevention fromidentifying the "at risk" behavioural group, interms of reducing the incidence of pedestrianinjuries, would be marginal.The volume of traffic on the roads usually

crossed was strongly associated with the like-lihood of injury. In fact, as the volume oftraffic on the roads usually crossed by thechild increased, so too did the child's odds ofpedestrian injury. Although other studies havenot measured the volume of traffic on theroad(s) to which a child is frequently exposed,they do confirm that a high volume of trafficincreases the risk of pedestrian injury when thevolume of traffic is measured at various injuryand control sites.37A number of studies,'5 37, have demonstrated

the potential for reducing childhood pedestrianinjuries through limiting the volume and re-

ducing the speed of vehicular traffic. Usingcomparable before and after periods, Pharaohand Russell38 reported a reduction of 43% inthe number of pedestrian injuries following theimplementation oftraffic calming. Not only are

there reductions in the number of pedestrianinjuries after traffic calming, but also levels ofnoise and air pollution fall, while pedestrianand cycle activity increase.38 Consequently, theestablishment of pedestrian friendly en-vironments has the potential for a more generalbenefit on the public's health.

In contrast to a number of reports,51137 thisstudy did not identify on-street parking as con-

tributing to the risk of pedestrian injury. How-ever, until now few studies had determinedwhether other obstacles to vision, such as semi-permanent or permanent objects on the vergeof the child's residential street were associatedwith the likelihood of pedestrian injury. It isevident from this research that semipermanentobjects such as rubbish bins and permanentobjects such as telephone booths, trees, andpost boxes can hide a child while they are

walking to the roadside. It is possible that a

child's movement from his or her home tothe roadside could be completely obscured byobstacles to vision on the verge. Therefore,relocation ofsemipermanent objects away fromthe verge, ensuring the design features of per-manent structures, minimises the obscuring ofchildren, and regular pruning of street vegeta-tion, should ensure that drivers have a view ofthe children about to cross the road.

In this study, observations of the road en-

vironment were recorded in the street of thechild's residence rather than at the site ofinjury.

Although both sites are equally important, ob-servations undertaken in the child's street ofresidence have important environmental andeducational implications, especially as 25% ofinjuries occurred on the child's street of res-idence. Notwithstanding this, a further study39which conducted observations at the site ofinjury, reported findings complimentary tothose discussed in this study.The only sub-environment of the road re-

serve which was significantly associated withthe risk of pedestrian injury was the presenceor absence of footpaths. The results showed areduced odds of pedestrian injury associatedwith the absence of footpaths in the child'sstreet of residence. In fact, children living onresidential streets which had no footpaths were52% less likely to be injured (OR=0-48, 95%CI = 0-27, 087) than those children with foot-paths in the street where they lived. Likewise,a reduced odds of pedestrian injury was alsoassociated with the absence of footpaths atthe site of injury (OR=009, 95% CI=0-01,°045)-39Mueller et alP have reported similar findings,

but they only considered footpaths at the siteof injury. This finding has not been reportedbefore in relation to the child's street of res-idence. The explanation proposed by Mueller37that the reduced risk associated with the ab-sence of footpaths was due to footpaths beingprovided more frequently on busy roads, andthat their presence there did not compensatefor the increase in risk associated with greatervolumes and speed of vehicular traffic is per-haps open to debate. Since they included intheir analysis the mean volume of traffic andthe posted speed limit at the sites of injury,the reduced risk of injury associated with theabsence offootpaths can not be easily explainedby greater volumes and speed of vehiculartraffic as these have already been taken intoaccount. However, one cannot exclude the pos-sibility that the presence of footpaths may rep-resent residual confounding not controlled forby measuring the volumes and speed ofvehicu-lar traffic. We also adjusted for the speed andvolume of traffic in our statistical model. Itis difficult to determine why the absence offootpaths in the child's street of residenceshould reduce the likelihood of injury. It couldbe that where there are no footpaths the childhas to walk on the road and he or she istherefore more cautious. Alternatively, the childwith a footpath in his or her street may treatthe footpath as an extension of the perceived"safe" play area and behave inappropriately inclose proximity to traffic, and, therefore, it maybe an issue of a child's complacency when ina familiar environment. No matter which theexplanation is, they are all amenable, to somedegree, to appropriate road safety education.The aim of this study was not only to deter-

mine the major factors related to childhoodpedestrian injuries, but also to ascertain to whatextent differences in the amount of exposureto the road environment and the nature ofthe road environment to which a pedestrian isexposed can explain why children of a certainage, sex, and area of residence are at higher

285



j.com/

J Epidem

iol Com

munity H


nloaded from


Stevenson, J7amrozik, Burton

risk. Although past research has described thepattern of child pedestrian injuries, it has beenbased on an untested assumption that childrenof different ages, sexes, and neighbourhoodsspend equal amounts of time in equally dan-gerous road environments. The results fromthe present study indicate that the over rep-

resentation in the injury statistics of childrenfrom low socioeconomic areas and of boyscan be explained, in part, by their pattern ofexposure to the road environment namely,the high volumes of traffic on the roads theycross. Furthermore, the nature of the roadenvironment to which he or she is exposed,such as the absence of footpaths in the streets

where they live and kerbside features whichobstruct either the child's or driver's view, or

both, also affect that child's risk. In short, theresults from this study suggest that a reductionin the incidence of childhood pedestrian in-juries can best be achieved through modi-fications to the road environment to which thechild is exposed.

The authors wish to thank the Westem Australian HealthPromotion Foundation (Healthway) for financial support. Theposition of senior biostatistician at the TVW Telethon Instituteof Child Health Research is funded by the NH & MRC.

Guyer B, Talbot AM, Pless IB. Pedestrian injuries to chil-dren and youth. Pediatr Clin North Am 1985;32:163-74.

2 Australian Bureau of Statistics. Road traffic accidents reportedto the Police Department, Western Australia. Canberra:Aus-tralian Bureau of Statistics, 1987. Cat No 9406.5.

3 Health Department of Western Australia. Hospital morbiditystatistics. Perth: Health Department, 1988.

4 Andreassen DC. Preliminary costs for accident-types: accidentcosts for project planning and evaluation. Victoria: AustralianRoad Research Board; 1992. Report No: ARR 217.

5 Andreassen DC. Preliminary costs for accident-types: accidentcosts for project planning and evaluation. Victoria: AustralianRoad Research Board; 1992. Report No: ARR 218.

6 Australian Bureau of Statistics. Road traffic accidents involvingcasualties admitted to hospital, Australia. Canberra; 1987.Cat No: 9405.0.

7 Office of Population Censuses and Surveys. Mortality stat-

istics: cause. London: HMSO, 1986. Series: DH2 No 13.8 Waller AE, Baker SP, Szocka A. Childhood injury deaths:

national analysis and geographic variations. Am PublicHealth 1989;79:310-15.

9 Jonah BA, Engel GR. Measuring the relative risk of ped-estrian accidents. Accid Anal Prev 1983;15:193-206.

10 Knighting FA, ColborneHY, Grayson GB. A pilot study ofchildpedestrians in a residential area. Crowthorne, Berkshire:Transport and Road Research Laboratory; 1972.

11 Routledge DA, Repetto-Wright R, Howarth CI. The ex-

posure of young children to accident risk as pedestrians.Ergonomics 1974;17:457-80.

12 Tight MR. A study of the accident involvement and exposure

to risk of child pedestrians on joumeys to and from schoolin urban areas. In: Rothengatter T, de Bruin R, eds. Roaduser behaviour: theory and research. Assen/Maastricht: VanGorcum, 1988:185-91.

13 PlessIB, Verreault R, Tenina S. A case-control study ofpedestrian and bicyclist injuries in childhood. Am PublicHealth 1989;79:995-98.

14 Backett EM, Johnston AM. Social patterns of road accidentsto children. BMJ7 1959;i:409-13.

15Joly MF, Foggin PM, PlessIB. A case-control study oftraffic accidents among child pedestrians. Proceedings ofthe International Conference on Traffic Safety. 1991; NewDelhi.

16 Rowntree G. Accidents among children under two years ofage in Great Britain. Journal of Hygiene 1950;48:323-37.

17 Rivara FP. Child pedestrian injuries in the United States:current status of the problem, potential interventions andfuture research needs. Am Dis Child 1990;144:692-96.

18 Roberts I, StreatS, Judson J, Norton R. Critical injuries inpaediatric pedestrians. NZ Med 199 1;104:247-48.

19 Baker SP, O'Neill B, Ginsburg MJ, Li G. The injury factbook. New York: Oxford University Press, 1992.

20 KingWD, Palmisano PA. Racial differences in childhoodhospitalized pedestrian injuries. Pediatr Emerg Care 1992;8:221-24.

21 Rivara FP, Barber M. Demographic analysis of childhoodpedestrian injuries. Pediatrics 1985;76:375-91.

22 Dougherty G, PlessIB, Wilkins R. Social class and theoccurrence of traffic injuries and deaths in urban children.Can Public Health 1990;81:204-09.

23 Preston B. Statistical analysis ofchild pedestrian accidents inManchester and Salford. Accid Anal Prev 1972;4:323-32.

24 Manheimer DI, Mellinger GD. Personality characteristics ofthe child accident repeater. Child Devel 1967;38:491-513.

25 Read JH, Bradley EJ, Morison JD, Lewall D, Clarke DA.The epidemiology and prevention of traffic accidents in-volving child pedestrians. Can Med Assoc J 1963;89:687-701.

26 Osborn AF, Morris TC. The rationale for a composite indexof social class and its evaluation. British tournal ofSociology1979;30:39-60.

27 Jellinek MS, Murphy JM. The recognition of psychosocialdisorders in pediatric office practice: the current status ofthe pediatric symptom checklist. JDev Behav Pediatr 1990;11:273-78.

28 Kirk SA, McCarthyJJ, Kirk WD. Illinois test ofpsycholinguisticabilities: examiner manual. Chicago: University of Illinois;1968.

29 Association for the Advancement of Automotive Medicine.The abbreviated injury scale. Des Plaines, Illinois. 1990.

30 Statistical Analysis System. SAS computer program. Cary,North Carolina: SAS Institute Inc, 1990.

31 Breslow NE, DayNE. Statistical methods in cancer research:Volume 1 - The analysis of case-control studies. 1st ed.Lyon: International Agency for Research on Cancer, 1980.

32 Epidemiological Graphics, Estimation and Testing. EGRE7Icomputer program. Seattle, Washington: Statistics and Epi-demiology Research Corporation, 1993.

33 National Health and Medical Research Council. Nationalhealth medical research council statement on human ex-perimentation and supplementary notes. Canberra: NHMRC,1992.

34 Roberts I. Why have child pedestrian death rates fallen?BM3 1993;306:1737-39.

35 Langley JD, Silva PA, Williams SM. Psychosocial factors inunintentional childhood injuries: results from a longi-tudinal study. Journal of Safety Research 1987;18:73-89.

36 Smeed RJ. Some aspects of pedestrian safety. J7ournal ofTransport Economics and Policy 1968;2:255-79.

37 Mueller BA, Rivara FP, Shy-Mine L, Weiss NS. En-vironmental factors and the risk for childhood pedestrian-motor vehicles collision occurrence. AmJf Epidemiol 1990;132:550-60.

38 Pharoah T, Russell J. Traffic calming: policy evaluation inthree European countries. London: Department of Planning,Housing and Development, South Bank Polytechnic;1989. Report No: 2/89.

39 Stevenson M, Jamrozik K, Spittle J. A case-control studyof traffic risk factors and child pedestrian injury. Int JEpidemiol 1995;24:957-64.

40 McCullagh P, Nelder JA. Generalized linear models. Mono-graphs on statistics and applied probability 37. 2nd ed. Oxford:Chapman and Hall, 1989:206.

AppendixConsider a case-control study generated froma base population consisting of Mj cases andNj non-cases at level j of a nominal explanatoryfactor (z) with k levels. Designate the levels1 ... k and define level 1 as the baseline category.The true odds (in the base population) of

being a case among individuals at level j of zis Mj/Nj. If the sampling fraction of cases isone (all cases are ascertained) and the samplingfraction of controls at level j is f, the case-control study will include Mj cases and fNMcontrols at level j and the estimated odds at thislevel will be Mj/(fNj).The estimated odds ratio (TP) for level j

compared to level1 (the baseline category) willbe (Mj/fN1)/(M1/f1N1). In a standard un-matched case control study f =f2= ...f... =fkand the estimated odds ratio for level j comparedto level 1 will therefore be equivalent to (Mj/Nj)/(M1/N1), a consistent estimator of the trueodds ratio between these two levels in thebase population. This is because the commonsampling fraction cancels out. However, fre-quency matching on z makes it likely that thecontrol sampling fractions will vary betweenthe differentlevels ofz and will not thereforecancel. In consequence,Tj does not then pro-vide a consistent estimate of the true odds ratioin which we are interested. This is the principalreason why one would not generally choose tomatch on an explanatory variable in which one

was primarily interested.

286



j.com/

J Epidem

iol Com

munity H


nloaded from



In a standard logistic regression model, theanalysis is carried out on the scale of loge(odds)and if Pj estimates the loge(odds ratio) for levelj compared to level 1 of z:

I3i=loge,MJJJ) = log(, _)log fJThe second term on the right hand side of theequation is the loge of the ratio of the samplingfractions at levels j and 1 of z. Having par-titioned off the influence of the varying controlsampling fractions in this manner, the first termnow provides a consistent estimate of the trueloge (odds ratio) we wish to estimate. Thisestimate remains consistent if the samplingfraction of cases is less than one. However, ifthe case sampling fraction also varies with z,full adjustment requires additional componentsto be introduced into the second term to reflectthis variation.

Let us now turn to the present study.Knowing the age-sex structure of the base

population (which we did) and knowing the

age-sex structure of the controls in our study,the age-sex specific control sampling frac-tions(fj) could be estimated directly. A vectorOFFSET could therefore be generated suchthat OFFSETg (the g'" element of OFFSET)took the value -loge(fh/f,) when the gff in-dividual in the study was at level h of z.

In order to adjust for the variation of controlsampling fractions with age and sex the vectorOFFSET was entered into the logistic re-gression equation as an offset.40 An offset maybe viewed as an explanatory variable with aregression coefficient which is forced to be one.This takes up the impact of the variation insampling fraction introduced by the frequencymatching (the second term on the right handside of the equation above) and leaves f3 toestimate the first term which provides a con-sistent estimate ofthe true odds ratio of interestin the base population. In this particular study,there was no evidence that the response ofcases varied markedly with either age or sexand there was therefore no need to modify theelements of OFFSET to reflect case samplingfractions.

287



j.com/

J Epidem

iol Com

munity H


nloaded from


Documents

Epidemiology Community Health A case-control of childhood ... · Acase-control study ofchildhoodpedestrian injuries in Perth, Western Australia for example, spend equal amounts oftime