Accident Analysis and Prevention 36 (2004) 1099–1103

Long term medical costs of motor vehicle casualties in Alberta (1999):a population – based, incidence approach

Philip Jacobsa,∗, Douglas Lierb, Donald Schopflocherc

a Institute of Health Economics, University of Alberta #1200–10405 Jasper Avenue, Edmonton, Alberta T5J 3N4, Alta., Canadab Cross Cancer Institute, 11560 University Avenue, NW Edmonton, Alberta T6G 1Z2, Alta., Canada

c Surveillance, Alberta Health and Wellness, 24th Floor, Telus Plaza North Tower, 10025 Jasper Avenue, Edmonton, Alberta T5J 1S6, Alta., Canada

Received in revised form 22 March 2004

Abstract

The purpose of this paper is to estimate the long term medical costs attributable to motor vehicle accidents (MVAs) for all personsin Alberta, Canada in 1999, primarily using observational data. Injury claims with personal identifiers for 1999 were reported by theautomobile insurance companies. These records were linked to the provincial health registry which covers the entire population. Theregistry is linked to databases which identify all inpatient and outpatient (including emergency room) visits, physician services, and otherhealth records. Utilization and costs were derived for all casualties who were admitted to hospital or seen in an emergency room, and for alarge sample of other (low severity) cases; a sample of matched controls was derived and their costs were also estimated. Actual costs wereobtained for 3 years, and longer term costs were projected for subsequent years. Total costs attributable to MVAs were estimated at over$117 million for 1999. Average net costs per casualty, reported by severity group, were: $22.9 thousand for hospital cases; $3.6 thousandfor emergency room – only cases; and $157 for other cases. Long term costs were 65% of first year costs for hospitalized cases and 250%for emergency room cases. Overall, aggregate costs for all non-hospital cases exceeded those for hospitalized cases.© 2004 Elsevier Ltd. All rights reserved.

Keywords: Costs; Motor vehicle accidents; Canada

Alberta, Canada is a province with a population of aboutthree million, with an annual level of collisions exceed-ing 100,000 since 2000.Alberta Transportation (2001)esti-mated, from traffic accident reports, that there were 26,868MVA casualties during 2000 in Alberta. The long term med-ical costs for motor vehicle accidents (MVA) have drawnconsiderable attention. The economic magnitude of motorvehicle accidents has been underscored by SMARTRISK,which estimated that, in 1997, the direct aggregate med-ical costs of MV accidents in Alberta were $115 million(SMARTRISK, 2002).

An ideal direct measure of the economic impact of suchaccidents requires that the costs be population – based,incidence – based, attributable to motor vehicle accidentcasualties, inclusive of downstream events and based ondirect observation. By population – based, we mean that themeasure should capture information on casualties for theentire population. By incidence – based we mean the sumof a stream of medical costs that are expected to result from

∗ Corresponding author. Tel.:+780 448 4881; fax:+780 448 0018.E-mail address: pjacobs@ihe.ab.ca (P. Jacobs).

injuries that occur during one year. By attributable, we meanthat it should identify those costs that resulted from the ac-cident, net of costs that would have arisen anyway. Finally,to the extent that costs can be measured directly, withoutincurring undue costs, rather than modeled, direct measure-ment will clearly be more accurate. In this last criteria, it isimplied that an estimate which directly captures inpatientand outpatient services is better than one which only directlymeasures inpatient hospitalization and indirectly other costs.Cost estimates over the very long-term, however, must al-most always be modeled because of limitations on observeddata. Also, working within project budgets often dictatesinferring follow-up costs for some cases from a sample.

Virtually all health – related long range cost estimates arederived by combining observational and modeling methods(Etzioni et al., 2001; Jacobs et al., 2003). The literature onmotor vehicle accident costs has been reviewed by (Albertand Cloutier, 1999). Previous estimates of motor vehicle ac-cident costs have been conducted primarily through model-ing methods, with a combination of data bases (Miller et al.,1993; Al-Masaeid et al., 1999; Ryan et al., 1998). Recentstudies (Miller et al., 2001; Zaloshnja et al., 2000) have also

0001-4575/$ – see front matter © 2004 Elsevier Ltd. All rights reserved.doi:10.1016/j.aap.2004.05.001

1100 P. Jacobs et al. / Accident Analysis and Prevention 36 (2004) 1099–1103

conducted national MVA medical cost estimates using sucha method. The authors obtained epidemiologic informationon the number of casualties using a national hospital dis-charge survey (for inpatient cases), and a national health in-terview survey (for other cases). The authors provided a de-tailed algorithm to estimate the total lifetime medical costsfor collision casualties. These studies estimated a broadrange of costs, including the value of productivity lost dueto disability. SMARTRISK, a Toronto – based accident pre-vention association, conducted a number of studies for dif-ferent Canadian provinces of costs for unintentional injuries(including MVAs). (SMARTRISK, 1999, 2002). The au-thors based the observational portion of their analyses on theinitial hospitalization, and added a modeling component forall other costs (current and future) which was based on esti-mates from United States data (Miller et al., 1995). Meerdinget al. (2002)reported on phase one of the development of amodel, based on methodology developed in the Netherlands(Mulder et al., 1999), which will track a wide range ofinjuries, including MVAs, in ten countries of the EuropeanCommunity. The methodology for that study is incidence –based and will estimate direct medical costs.Lindqvist andBrodin (1996) took an enumeration approach to identifycasualties arising from MVAs during one year, over a de-fined geographical area in Sweden. That study focused onshort-term costs related to inpatient and outpatient hospitalservices, although it included both direct and indirect costs.

We conducted a study of the medical costs of motor vehi-cle accidents in the province of Alberta, Canada for accidentswhich occurred in 1999. We used several sources of datato obtain our estimates including insurance accident claimswhich were obtained from the insurance industry and healthrecords for the entire provincial population. Our analysis in-cluded a linking of these data, from which we developed ananalysis of one year (short-term) and subsequent (long-term)medical costs attributable to MVAs which occurred in Al-berta in 1999. In our analysis, we take an “attributable” ap-proach to the estimation of direct medical costs; this meansthat we have included all direct medical costs for casualtiesless the medical cost of matched controls1.

1. Methods

The government of Alberta provides comprehensive hos-pital and medical services to virtually all residents of theprovince, pharmaceutical services to those over 65 and se-lected other services. Residents are recorded in a central reg-istry, which contains demographic (name, date of birth) and

1 Our cost estimates differ from those for which the insurance industry,under Alberta law, would be responsible for insured - driver liability.The liability of insurers in Alberta is limited to “third-party” claims.Specifically excluded from the liability are the health service costs of “atfault” drivers (the first party) and the costs arising from “contributorynegligence,” for example, the failure to use seatbelts.

geographic information. All hospital (inpatient and outpa-tient, including emergency room) visits and physician ser-vices are recorded in separate data bases, which can belinked to the registry using a unique personal health number(PHN) identifier. Data on specific other services providedby the government (ambulance services, aids to daily living,prescription drugs for persons over 65) are also maintained.For administrative purposes the province at the time of thestudy was divided into 17 regional health authorities whichvaried considerably in size and degree of urbanicity.

1.1. Short term costs

Automobile insurance in Alberta is provided by privatecompanies. According to the Alberta Hospitals Act, in-surance companies are liable for damage done by at-faultinsurees; potential losses include medical costs that areincurred by the province. We used two sources to identifyMVA casualties, including fatalities. In February 2002, theInsurance Bureau of Canada (IBC), an industry trade as-sociation, canvassed its members to provide the Instituteof Health Economics (IHE) with specific data (name ofclaimant, date of birth, date of accident) on claims for an ac-cident which occurred in 1999. Using this information, weattempted to match the persons who made these claims withpersonal identifiers in the provincial health registry, usingthe first name, last name and date of birth of the claimant.

In order to supplement the MVA casualties identifiedfrom IBC data we utilized a second source. Both provin-cial hospital and emergency room data bases were searchedfor any admission or visit, during 1999, that was identifiedas being MVA – related, as identified by the InternationalClassification of Disease – 9th edition (Clinically Modified)(ICD9CM) diagnosis codes ( codes E810 through E819 ).The PHNs from the resulting records were obtained and theprovincial health data bases were searched for any other uti-lization encounter for those persons who were identified withMVA – related ICD9-CM codes during the period. The firstsuch code during 1999 for each unique PHN was deemedto signify the initial (index) medical event.

The use of these two sources provided virtually the entireenumeration of persons who had an MVA – related emer-gency visit or inpatient admission in 1999. These cases com-prised the two highest severity levels in the study. However,only the first source (i.e., IBC) identified injury – relatedclaims which neither resulted in an inpatient admission noran emergency department visit, following the collision. Wewere unable to identify all 1999 low severity casualties, sothe identified individuals represented only a sample.

A matched control (no casualty) for each MVA – relatedcasualty was obtained. We randomly matched each personwith one of identical age, sex and region of residence, whodid not have an MVA in 1999.

Data for all hospital and medical services for each casu-alty and match were obtained for one year beyond the acci-dent or index date. Costs were assigned to each service using

P. Jacobs et al. / Accident Analysis and Prevention 36 (2004) 1099–1103 1101

1999 hospital – specific per diem costs. The facility com-ponent of hospital outpatient visits, including emergencyroom visits, was evaluated using standard 1999 provincialcosts per visit. All hospital costs were measured by ratesdeveloped in conjunction with the Interprovincial HealthInsurance Agreements Coordinating Committee of HealthCanada. All other costs, including those for all physicianservices within and out of the hospital, were evaluated withprovincial fees (Alberta Health and Wellness, 1999). A costfor each person, casualty and match, was obtained for theyear by summing the dollar values for services used by eachperson. The difference between the costs for the two groups(casualty and matched sample) were deemed to be short term(ST) costs that were “attributable” to the MVAs. These weredivided into three severity groups: hospitalized, emergencyroom only, and neither hospital nor emergency room.

1.2. Long term costs

Long term cost estimates were based on a combinationof observation and modeling techniques. In order to obtaina sufficient length of time for the observed data series (12quarters following the date of accident) we selected casual-ties and matched controls from 1997, by a method similar tothe second source inSection 1.1. Emergency room data wasnot collected before 1997, and so this was the longest ob-servational period available to us. We calculated a long termcost factor (LTCF) for each study and severity group (e.g.,hospital MVA casualties), defined as the health service costfor years 2 through 50 divided by year 1 costs. We used theLTCF to forecast average long term cost (LTC) per personas the product of the average STC and the LTCF. For eachof the hospital and emergency-room severity groups the at-tributable average LTC was equal to the difference in costsbetween the casualty and control group costs.

In calculating the LTCF, the expected cost for each yearwas equal to the product of the average cost per person andcumulative survival. For some pairs, the survival time forthe control exceeded that of corresponding casualty. Thisis appropriate when measuring costs from the attributableperspective (Etzioni et al., 2001).

Observed data was used to calculate average LTC for years1 to 3, whereas, modeling was used for years 4 through 50.Average cost for each of the casualty hospital and emer-gency room severity groups was forecast using a double-logpolynomial regression on group-level data, with grouped av-erage cost per quarter as the dependent variable and time(in quarters) as the independent variable. For the controls,average cost for years 4 through 50 was assumed to equalthe average of the first three years. All cost data used inthe long-term analysis were expressed in constant 1999 dol-lars and converted to 1999 present value using a 3% realdiscount rate. The choice of discount rate is an importantdecision since a small variation in this rate will result in asubstantially different present value. Our choice of the rateis based on several academic sources (Gold, 1996; Miller

et al., 2001; Drummond et al., 1997), as well as a recentanalysis of Canadian long-term rates of return and inflation(Bruce et al., 2003).

For years 1–3, observed survival was used for casualties;however, for controls the general population age-sex specificsurvival rates (from the 2000 Alberta Life Tables) were usedto estimate cumulative survival. For years 4 through 50,for both casualty and control groups, cumulative survivalwas forecast using age-sex specific rates from the generalpopulation life tables.

1.3. Combined short term and long term cost

Total cost for each severity group was estimated as theproduct of the number of casualties and the average total costper person, defined as the sum of STC and LTC per person.Although we have largely complete reporting for the first twogroups, reporting for the least severe group (identified onlyby claims) was less than complete. Using industry premiumsdata for reporting and non – reporting companies, and the%of reported claims which were matched, we projected thetotal number of expected non-severe claimants, and appliedthe non severe unit cost to this statistic.

2. Results

Our health insurance company sample included 23 com-panies with combined premiums of $1.39 billion (85.3%of total premiums) in 1999, out of a provincial total of51 companies with premiums of $1.63 billion. Reasons fornon-reporting and percentages according to premium valuesincluded refusal to participate (2.7%), responding too late(4.4%), smaller companies that were not canvassed (4.2%)and no response (3.3%). In total, there were 25,789 uniqueclaims from reporting companies, providing 20,338 (78.9%of unique claims) complete records. The difference betweenunique claims and complete records were due to incompletedata. We were able to match 14,873 claims (73.1% of com-plete records) to the Personal Health Number (PHN).

The attributable average cost for cases by level of sever-ity is shown inTable 1. This includes observed costs for thefirst post-accident year, and observed and modeled costs forsubsequent years, using the LTCF. In all cases, costs for per-sons in the matched sample were deducted from the costsfor the casualties. Net costs per casualty decline quicklyover time as show inTable 1. Most of the attributable costsare incurred by the end of the tenth year for both the hospi-talized and emergency only casualties. A casualty who washospitalized incurred net medical costs of $14,211 in thefirst year and an additional $8751 (a 61.5% long term fac-tor) in subsequent years. Those with emergency room careonly incurred costs of $1024 in the first year and $2542 insubsequent years. Low severity cases incurred net costs of$157 in the first year; we assumed that there were no longterm costs.

1102 P. Jacobs et al. / Accident Analysis and Prevention 36 (2004) 1099–1103

Table 1Summary statistics of frequencies and costs by level of injury, Alberta, 1999 casualties

Severity level Hospitalized Emergency room only Other Total

Unduplicated count of casualties 2237 (5.4%) 17,686 (43.0%) 21,161 (51.5%) 41,084 (100%)Average net cost year 1 14,211 1024 157 –Averagea net Cost Years 2–3 3430 1073 0 –Averagea net cost years 4–10 3851 1377 0 –Averagea net cost years 11–50 1470 92 0 –All years 22,962 3566 157 –

Total 51,366,000 (43.6%) 63,068,000 (53.5%) 3,322,000 (2.8%) 117,756,000 (100%)

a Average net cost per casualty for all years in the range.

In total, cases in which there was an MVA – related hos-pitalization accounted for 43.6% of the total direct expensesand cases with an MVA – related ER visit accounted for53.5% of all expenses. In the first-year of care, using ourshort-term sample, the most important cost categories forthe two most severe types of cases (hospitalized and emer-gency only) were as follows: inpatient care, 62%; outpa-tient hospital facility costs, 16%; clinical practitioners, 18%;other (health aids, drugs, and ambulance), 4%. In the longterm sample the input breakdown was only available forthree years post-accident. In this period hospital inpatientexpenses fell from 63% in the first year to about 50% inthe next two years. Outpatient hospital expenses rose fromabout 15% to approximately 18% and practitioner expensesrose from about 19% to 25% in the next two years.

3. Discussion

In this paper we report on the long-term medical costs at-tributable to motor vehicle accidents in Alberta. Our analysiswas based on insurance company claims data and popula-tion level data obtained from provincial records up to threeyears beyond the accident, and we forecast costs beyondthis period. Observed data (as distinguished from modeledestimates) made up about 75% of our entire estimate. Ourresults show that short term (first year) costs were over $14thousand per hospitalized case, and $1 thousand per emer-gency room – only case. Long term costs represented a sub-stantial addition to first-year costs. Further, because of thelarge numbers of cases in the emergency room – only cate-gory, the costs for these cases exceeded those of hospitalizedcases in total. The total costs for those who neither went tohospital nor emergency room were considerably smaller.

In some other studies on MVA direct medical costs(Langley et al., 1993; SMARTRISK, 2002), hospitalizedevents have been the main type of service that was directlyobserved. Our results indicate that the hospital componentof costs of the hospitalized cohort account for a small por-tion of total direct medical costs for motor vehicle accidents.An estimate of these costs is $20 million out of a total of$117 million for all medical costs, or about 17%. This isbecause hospitalization costs amounted to about 62% of to-

tal first year costs of hospitalized cases, and these first yearcosts in turn were about 66% of total costs. In addition,costs for hospitalized casualties were only 43% of the totaldirect medical costs. This estimate was made without tak-ing into account the fact that the average person who washospitalized was admitted an average of 1.48 times duringthe year. As a result, the proportion of total direct medicalcosts that were due to the initial hospitalization may havebeen even less. These numbers highlight the importance ofhaving directly observed estimates of medical costs otherthan those of the initial hospitalization.

Our results can be compared with those conducted bySMARTRISK for Alberta (SMARTRISK, 2002). Overall,the aggregate medical costs in our study are very similar tothose of SMARTRISK – ours were $117 million for 1999and those of SMARTRISK for 1997 were $115 million.There was very little inflation in health care in those years,so that inflation will account for little of the difference. How-ever, the severity distribution of the costs in the two studiesis quite different. A relevant statistic is not reported in theAlberta SMARTRISK study. In the Ontario SMARTRISKstudy, which uses a similar methodology, the comparativestatistic of costs for hospitalized versus non-hospitalizedcases is over 60%. In the present study, medical costs forhospitalized (i.e., severe) cases account for 43% of medicalcosts for all cases. There is not enough detail in the SMAR-TRISK study to determine the source of the differences.

There are several limitations of our analysis. First, severalcategories of expenditures were not reported in the provin-cial data bases. These categories include long term care,outpatient prescription drugs and non-hospitalized physio-therapy. The inclusion of long term care would increase thelong term cost factor, but we do not know by how much.Second, although our analysis is based primarily on obser-vational data, those long term costs that were beyond thethird year were modeled. These modeled costs account forabout 25% of the total costs. Since our analysis is population– based, most of any estimating error would occur in ourforecast of the tail of the estimating curve. The goodness offit of the estimate was quite high in the hospital estimatingequations (0.9873), and lower (0.8561) in the emergencyroom – only population. We therefore, have greater confi-dence in the predicted component of the long run cost factor

P. Jacobs et al. / Accident Analysis and Prevention 36 (2004) 1099–1103 1103

for the hospitalized patients than the emergency room –only patients. Third, we have estimated, on the basis of alarge sample, that about 1.3% of MVA casualties, includedin our data set, were involved in more than one collisionduring the first year following the index accident. However,we believe that the presence of such a small proportion ofmultiple-accident casualties resulted in a minor overestima-tion in the health service costs attributable to 1999 MVAs.

The usefulness of our results lies in the identificationof costs, by category, on a per-person basis. Although ourcategories of injury are not nearly as detailed as those inother studies, for example,Miller et al. (2001), the form inwhich they are presented can be of use to planners and per-sons conducting research in the area of accident prevention.In Alberta, as in much of Canada, accidents are routinelyreported by police on the basis of whether or not a hospi-talization is expected. Planners can use our per casualty es-timates, in conjunction with accident trend data, to forecastmedical costs due to accidents. As well, researchers who areconducting studies on the economic value of accident pre-vention can apply the per-person costs that were reported inthis study to the number of officially reported accidents pre-vented in order to obtain an estimate of the savings due tovarious interventions that are related to accident prevention.

Acknowledgements

This study is based on a previous analysis that was con-ducted for a report prepared for the Insurance Bureau ofCanada and Alberta Health and Wellness, Third Party Lia-bility. The authors are solely responsible for its contents.

References

Al-Masaeid, H.R., Al-Mashakbeh, A.A., Qudab, A.M., 1999. Economiccosts of traffic, accidents in Jordan. Accid. Anal. Prev. 31 (4), 347–357.

Albert T., Cloutier E., 1999. The economic burden of unintentional injuryin Canada. SMARTRISK. Toronto.

Alberta Health Care Insurance Plan. Schedule of Medical Benefits. AlbertaHealth and Wellness 1999, Edmonton.

Alberta transportation. Alberta Traffic Collision Statistics 2001, AlbertaTransportation 2001, Edmonton.

Bruce, C., Aldridge, D., Beesley, S., Rathje, K., 2003. Selecting thediscount rate - an update, Spring 8 (1), The Expert Witness Newsletter(Economica Ltd.).

Drummond, M.F., O’Brien, B., Stoddart, G.L., Torrance, G.W., Methodsfor the Economic Evaluation of Health Care Programmes, second ed.Oxford Medical Publications, Toronto.

Etzioni, R., Ramsey, S.D., Berry, K., et al., 2001. The impact of includingfuture medical care costs when estimating the costs attributable to thedisease: a colorectal cancer case study. Health Econ. 10, 245–256.

Gold, M.R., Siegel, J.E., Russell, L.B., Weinstein, M.C., (Eds.),Cost-effectiveness in Health and Medicine, Oxford University Press,New York.

Jacobs, P., Golmohammadi, K., Longobardi, T., 2003. Lifetime costs formedical services: a methodological review. Int. J. Technol. Assess.Health Care 19, 278–286.

Langley, J.D., Phillips, D., Marshall, S.W., 1993. Inpatient costs of injurydue to motor vehicle traffic crashes in New Zealand. Accid. Anal.Prev. 25 (5), 585–592.

Lindqvist, K.S., Brodin, H., 1996. One-year economic consequences ofaccidents in a Swedish municipality. Accid. Anal. Prev. 28 (2), 209–219.

Meerding, W.J., Toet, H., Mulder, S.,van Beeck, E., Mitchell, C.,2002. A surveillance based assessment of medical cost of injury inEurope: phase 1, Consumer Safety Institute, Erasmus University, TheNetherlands.

Miller, T.R., Pindus, N.M., Douglass, J.B., 1993. Medically related motorvehicle injury costs by body region and severity. J. Trauma 34, 270–275.

Miller, T.R., Pindus, N., Douglas, J., Rossman, S., 1995. Databook onNon-Fatal Injury - Incidence Costs and Consequences, The UrbanInstitute Press, Washington, DC.

Miller, T.R., Romano, E., Zoloshnja, E., et al. 2001. Harm 2000: Crashcost and consequences data for the new millennium. In: Proceedings ofthe 45th Annual Proceedings of the Association for the Advancementof Automotive Medicine, 24–26 September 2001, pp. 159–184.

Mulder, S., van Beeck, E., Meerding, W.J., 1999. New directions ininjury surveillance Development of a model for continuous monitoringof direct medical costs. Int. J. Consumer Product Safety. 6 (l) 1l–23(E.45.06 S).

Ryan, G.A., Hendrie, D., Lyle, G., 1998. The use of an injury cost databaseto evaluate vehicle design changes in Australia. In: Proceedings of the42nd Annual proceedings of the Association for the Advancement ofAutomotive Medicine, 5–7 October, pp. 45–56.

SMARTRISK. The economic burden of unintentional injury in Ontario.SMARTRISK, Toronto.

SMARTRISK, 2002. The economic burden of unintentional injury inAlberta. Alberta Centre for Injury Control and Research, Edmonton.

Zaloshnja, E., Miller, T., Romano, E., Spicer, R., 2000. Crashcosts by body part injured, fracture involvement, and threat-to-lifeseverity, Accident Analysis and Prevention 2003, United States, (inpress).