G. Anthony Ryan
Casualties involved in road crashes often sustain multiple injuries, yet very little is known about the cost of single injury compared with multiple injury cases. The most common method of dealing with multiple injury cases in road safety research is to allocate a primary injury to casualties on the basis of the injury with the highest severity level. Subsequent analysis of the distribution and cost of road injury is then based on the classification of the primary injury.
The purpose of this study was to explore the cost of road trauma involving single and multiple injury cases. More specifically, the study was conducted to examine the marginal cost of each additional injury sustained by casualties in road crashes. The work is currently in progress, and this paper presents some preliminary results. Information relating to the impact of second and subsequent injuries on the cost of a single injury is to be used in cost-effectiveness analyses of interventions such as vehicle design changes, which may prevent the occurrence of some, but not all, injuries sustained by motor vehicle occupants in a crash.
The cost data used in this study was obtained from the Road Injury Cost Database. This database was constructed primarily from the unit records of personal injury claims paid to road crash casualties in New South Wales, Australia between July 1989 and June 1996. Finalized payments were available for 49,755 claimants. Personal injury insurance payments to individuals in the claims database were recorded for the following cost categories: legal and investigation, long term and home care, home and vehicle modifications, aids and appliances, economic loss and general damages. Methods were developed to allocate the following person-based costs of road injury to claimant records: medical, hospital, rehabilitation, ambulance, future unpaid earnings, losses to non-victims and personal injury insurance administration. Crash-based costs – namely property damage, travel delay and motor vehicle insurance – are included in the Road Injury Cost Database but were not used for the analyses in this study.
Each claimant in the Road Injury Cost Database has up to five injuries coded on the basis of the 1985 revision of the Abbreviated Injury Scale (AIS), with lower extremity injuries divided into lower and upper leg injuries. A primary injury was allocated to claimants on the basis of the injury with the highest severity, and this classification was used for some comparisons of the cost of single and multiple injury cases. If a claimant sustained two or more injuries of the same maximum injury severity level, priority was assigned on the basis of the following ordering: head, spine, lower extremities, thorax, abdomen, upper extremities, neck, face and external (MacKenzie, Shapiro and Siegel, 1988).
Approximately one third of cases in the Road Injury Cost Database had a single injury (n=16 603) and two thirds had multiple injuries (n=33 152). The most commonly occurring single injuries were a minor injury (i.e. AIS 1) to the neck (n=11 021), the spine (n=8 151) and the external body regions (n=3 683). On the basis of the primary injury classification, the most frequently occurring multiple injuries were a minor injury to the spine (n=7 283), the upper leg (n=2 764) and the neck (n=2 462).
A more detailed analysis of the multiple injury cases, based on all injuries sustained rather than the primary injury, indicated that the most commonly occurring injury combinations were as follows: neck AIS 1, spine AIS 1 (n=5 497), neck AIS 1, upper extremity AIS 1 (n=1 901); neck AIS 1, external AIS 1 (n=1 709); neck AIS 1, spine AIS 1, upper extremity AIS 1 (n=989); neck AIS 1, thorax AIS 1 (834); external AIS 1, upper extremity AIS 1(n=678); neck AIS 1, spine AIS 1, external AIS 1 (n=622); external AIS 1; upper leg AIS 1 (n=550); neck AIS 1, upper leg AIS 1 (n=534); and upper extremity AIS 2, external AIS 1 (n=480).
Tables 1 and 2 show the average cost by injury severity level for cases with single and multiple injuries. For both single and multiple injury cases, costs increased with the level of injury severity. For minor, moderate and serious injuries, the cost of cases with multiple injuries was between 50%-80% greater than the cost of cases with a single injury. However for severe and critical injury, cases with a single injury had a higher cost than cases with multiple injuries. This is a surprising result, which needs to be further explored.
|Number of Injuries||Minor
Cost differences by body region between single and multiple injury cases were generally in the expected direction, with the exception of spinal injuries, where cases with multiple injuries had a slightly lower average cost than those with a single injury. Body regions for which cases with multiple injuries had considerably greater costs than those with a single injury were the abdomen (3.3 times higher), the thorax (3.1 times higher) and the face (2.4 times higher).
|Body Region||Single Injury||Multiple Injuries||All Cases|
Simple Additive/Non-Additive Model
A simple model of the cost of single and multiple injury cases was developed using the cost of single injury cases as the basis to derive the expected cost of related multiple injury cases. For example, this model predicts the cost of cases with injuries of minor severity to the neck and spine from the actual cost of (i) cases with a single injury of minor severity to the neck and (ii) cases with a single injury of minor severity to the spine.
In this simple model, some cost categories – the additive costs – were attributed to each individual injury sustained by multiple injury cases. For these additive costs, the relevant costs for single injury cases were added across the different injuries sustained by multiple injury cases to give an aggregate of the additive cost categories. The following cost categories were designated as additive: hospital, medical, rehabilitation, home and vehicle modifications, and aids and appliances.
The remaining cost categories were designated as non-additive. For each non-additive cost category, the largest cost for each body region of injury for single injury cases was assumed to be the appropriate cost for multiple injury cases. For example, if economic loss was greater for single injury cases with a minor spinal injury than for single injury cases with a minor neck injury, then the economic loss cost category for single injury cases with a spinal injury was attributed as the appropriate cost for multiple injury cases with injuries of minor severity to the neck and spine. Figure 1 presents the general equation for the simple additive/non-additive model and a specific example.
General Equation of Additive/Non-Additive Model:
cost of road crash casualties with multiple injuries (i.e. n injuries)=
[sum of additive cost categories for single injury cases with injury 1] +
[sum of additive cost categories for single injury cases with injury 2] +
etc., etc., ………… + [sum of additive cost categories for single injury cases with injury n] +
[maximum of each non-additive cost category for single injury cases with injury 1, injury 2, ..… injury n]
cost of a road crash casualty with a moderate head injury and a minor spinal injury =
[sum of additive cost categories for single injury cases with a moderate head injury] +
[sum of additive cost categories for single injury cases with a minor spinal injury] +
[maximum economic loss for moderate head injury and minor spinal injury cases] +
[maximum general damages for moderate head injury and minor spinal injury cases] +
……. etc., etc.
The second approach to exploring the costs of single and multiple injury cases involved developing a generalized linear model (GLM) of road injury costs (the dependent variable) based on the body regions and injury severity levels of injuries sustained (the independent variables). The body regions of injury were entered in the model as class variables, with indicators used to denote different injury severity levels. The model did not impose any order on the indicators. Figure 2 presents the general equation for the model, together with part of the fitted model and a specific example. The fitted model explained 36 percent of the variation in the total cost of injuries in the Road Injury Cost Database.
| General form of the
cost of road crash casualties with multiple injuries
Sl = minor SPINAL injury
|The fitted model ($000):|
|Specific example =
Fitted model of the cost of a road crash casualty with a moderate head injury and a minor spinal injury = x+ H2 + S1 = $11 600 + $19 760 + $9 920 = $30 870
Table 3 compares costs for selected multiple injury combinations derived from the two models with the equivalent actual costs in the Road Injury Cost Database. For example, the actual average cost of cases with a minor head and minor neck injury was $18,920, while the costs predicted by the simple additive/non-additive model and the GLM model were $21,640 and $17,990 respectively. In general, the GLM costs were closer to the actual costs in the Road Injury Cost Database than those predicted by the simple additive/non-additive model, although this was not always the case (e.g. the simple model gave a closer fit for minor thorax/minor neck multiple injury cases).
|Multiple Injury Combinations||Cost Based on Simple Additive/Non-additive Model||Cost Based
|Actual Cost from Road Injury Cost Database|
|Minor head/minor neck||$21,640||$17,990||$18,920|
|Moderate head/minor neck||$39,510||$33,930||$33,860|
|Moderate head/minor external||$43,690||$31,720||$27,110|
|Minor spine/minor neck||$21,210||$24,100||$23,040|
|Moderate spine/minor neck||$68,990||$66,940||$60,470|
|Moderate spine/minor external||$73,170||$64,730||$70,610|
|Minor upper leg/minor external||$15,930||$17,910||$18,660|
|Minor thorax/minor neck||$17,770||$14,650||$17,580|
This paper has presented some preliminary results of the relationship between single injury and multiple injury road trauma cases. These results suggest that further work is warranted in developing the generalized linear model of the cost of injury based on the body region and injury severity level of injuries sustained. Further work will involve including interaction terms in the model as well as possibly some other refinements. In addition, the Road Injury Cost Database is being extended by the addition of unit records between July 1996 and June 1999. This latter data set will be used to test the relevance of the refined Generalized Linear Model of the cost of road injuries.
Willem Jan Meerding
Ed van Beeck
Information on costs is important for setting priorities for research and prevention in the field of accident-related injuries. This kind of information forms an important addition to the data on the incidence and mortality rate of accident-related injuries that is already recorded.
Hence, a project was launched with the explicit aim of developing a computerized model which could calculate at any given moment the direct medical costs (i.e. the costs within the healthcare sector) of injuries suffered in the Netherlands. One of the conditions was that a uniform method be applied to calculate the costs of the various injury categories (traffic, occupational, home and leisure, sports, violence and self-mutilation). The model should also be able to estimate the costs of specific injury scenarios (e.g. one-sided bicycle accidents, falls among the elderly, horse-riding accidents, facial injuries due to violence etc.).
The methodology was developed by the Working Group on the Costs of Injury. The European Consumer Safety Association (ECOSA) organized several conferences on the burden of injuries. During one such conference it was recommended to establish an international working group which should specify standardized methodologies for gathering appropriate information. ECOSA launched such a Working Group in 1995, in which nine countries (twelve members) participated. The composition of the Working Group is multidisciplinary, including health economists, medical doctors, and epidemiologists. The aim of the Working Group is to develop a tool to assess the societal costs of injuries in Europe as an aid to priority setting and risk management decision making in injury control in particular.
Three reports have been produced to date by the Working Group: a glossary, a state-of-the art report, and a bibliography (all available on http://www.ecosa.org/csi/ecosa.nsf). The Working Group also drafted a proposal on the application of the model for the direct costs of injury within the European Union. This proposal was submitted to the Injury Prevention Programme of the European Commission (to be decided in August 2000).
The calculation model estimates the costs of all LIS-registered injuries, i.e. injuries which are treated at the Emergency Departments of various hospitals in the Netherlands (incidence-based approach). The model was applied to data from 1997. The registered accident-related injuries were assigned to patient groups that were differentiated according to certain characteristics that determine the use and costs of healthcare. These characteristics were: admittance/non-admittance to hospital, type of injury, injury severity, age, and gender.
The model calculates the following cost elements: follow-up and aftercare by the GP, emergency transport, Emergency Department treatment, other outpatient care, daytime nursing, clinical nursing, clinical/therapeutic intervention, rehabilitation, nursing home care, outpatient physiotherapy, home care, medication.
The information used by the model was derived from LIS, standard care registrations, a supplementary survey of a random sample of LIS patients and various sources of cost-price information.
The calculation method works as follows. For each cost element a LIS patient is assigned to a specific patient group. The average costs per patient group are then calculated for each cost element and the total costs of all the elements are calculated for each LIS patient. Lastly, the total costs are determined by adding up the costs of all the patients. This estimate of total costs can be made for any selection of LIS patients.
In 1997 the total direct costs of direct accident-related injuries came to 2.2 billion guilders and accounted for 3.4% of the total healthcare budget for 1997. In the same year 1.1 million injury patients were treated at Emergency Departments and 104,000 patients were admitted to hospital.
Costs Borne by the Healthcare Sector
Almost three quarters (1.6 billion guilders) of the costs of injuries are incurred in hospitals and rehabilitation hospitals. 50% of the total costs are attributable to hospital and nursing home care, 30% to care in outpatient departments (half of which is emergency assistance) and 20% to extramural and other care. Compared with other types of illness the total costs spent on the out-patient care of injuries is disproportionately high.
Costs According to Age and Sex
Though women sustain only 40% of the total number of injuries, they account for 55% of the costs. This is mainly because many of the injuries suffered by older women require a high level of care. Women account for a larger proportion of hospital, nursing home and home care costs (65%) than men (35%).
The costs per injury patient increase with age. This increase is already discernible among young adults and, together with a high incidence rate reaches a clear peak in total costs for men in the 25-44 age group. The upward trend in costs per injury patient becomes exponential in older age groups. Hence, people aged 75 and over account for as much as 32% of the total medical costs of injuries (702 million guilders) while they represent only 5% of the total number of injuries sustained.
Costs by Injury
No less than 53% of all injuries are due to home and leisure accidents (580,000 patients). Their share of the costs is even higher: 59% (1.3 billion guilders). A large proportion of these injuries are hip fractures, which account for as much as one third of the total cost of home and leisure accidents.
Traffic accidents are responsible for 13% of injuries and 19% of the medical costs (410 million guilders). Sports accidents take third place at 10% (206 million guilders). These are followed by occupational injuries (112 million guilders, 5%), violence (53 million guilders, 2,5%) and self-mutilation (45 million guilders, 2%).
Costs by Type of Injury
Injuries to the lower extremities result in medical costs of almost 1 billion guilders (45% of the total costs). This is high considering that they account for only 17% of the total number of injuries. The costs per patient are relatively low for superficial injuries, but an exceptionally high incidence (30% of all injuries) leads to overall costs of 230 million guilders (11% of the total). The greatest strain on the budget are hip fractures (468 million, 22%); followed by superficial injuries (230 million, 11%); open wounds (149 million, 7%); fractures of the knee and lower leg (137 million, 6%); ankle fractures (96 million); and skull and brain injuries, with the exception of concussion (85 million, 4%).
Costs by Accident Scenario
The model must be able to calculate the costs of any selection of LIS patients. This report contains a few examples of ‘accident scenarios’ which provide insight into the costs of injury accruing from car accidents (101 million guilders); cycling accidents (165 million guilders); one-sided cycling accidents (107 million guilders); poisoning incidents involving young children (3 million guilders); fall-related hip fractures among the elderly (75+) (333 million guilders); do-it-yourself activities (24 million guilders); outdoor football accidents (48 million guilders); horse-riding accidents (14 million guilders; open wounds due to occupational accidents (18 million guilders); facial injuries due to violence (10 million guilders); and suicide attempts by poisoning (33 million guilders).
When used in combination with LIS the cost model provides a coherent picture of incidence, the healthcare use and the costs of acute physical injuries in the Netherlands. As a concept, costs provide an easy-to-interpret public health indicator that enables injury-patient data from diverse sources to be combined and integrated under one common denominator: guilders.
The cost model can provide policy information for:
As far as policy-making is concerned, the added value of this cost model rests primarily in the opportunities it provides for ongoing and detailed monitoring of accident-related injuries. For example, it allows distinctions to be made according to healthcare sector, age, gender, accident category and type of injury. Policy can also be supported by detailed estimates of incidence, healthcare use and the direct medical costs of specific accident scenarios (e.g. one-sided cycling accidents). Thanks to a uniform methodology the cost estimates can be effectively compared for all accident categories and healthcare sectors. Hence, the cost model forms a sound basis for the evaluation of the costs and effects of specific preventive measures.
However, when interpreting the results, it is important to take account of certain limitations connected mainly with the availability of data. The model will be refined and optimised in the future when additional and/or better data becomes available. Accordingly, a maintenance plan has been compiled to ensure that the model is kept up-to-date.
It is also the intention to further develop the model with estimates of the indirect costs of injury due to sick leave and occupational disability, and estimates of lost years of life and loss of quality of life due to limitations and handicaps.
In this cost model, data from various sources is integrated and connected in such a way that it results in combined information on injury prevention, healthcare use and medical costs, and facilitates internal comparisons. Hence, the model forms a new source of information for priority setting in injury prevention policy and for the evaluation of policy measures.
David J. Ball
In some sectors of the United Kingdom (UK) economy, proposals for safety interventions are subject to cost-benefit tests prior to implementation, which in turn is reliant upon monetary values being places upon lost lives and injuries. In the transport sector, the currently used breakdown of costs per casualty, based on a 1996 valuation of the benefits of the prevention of road accidents and casualties is set out in Table 1. As can be seen, the largest component of cost is in all cases that ascribed to ‘pain, grief and suffering.’ Valuation of this component has, for many years, been achieved by use of contingent valuation.
In 1997 the Consumer Affairs and Competition Policy Directorate of the UK Department of Trade and Industry (DTI) commissioned its own research on injury valuation in the context of the management of risks in relation to consumer product safety. The outcome of this research was to suggest an alternative strategy for decision making which was later accepted.2 Essentially, it was decided to adopt a range from £1million to £10 million for a statistical life (for most purposes a narrower range of £2 million to £4 million would be used) as a rule of thumb for guiding decisions about safety interventions.
At about the same time, several UK government departments, led by the UK Health & Safety Executive (HSE), commissioned a further ambitious study of the valuation of safety in different contexts, some of the results of which have now been reported.3
This paper provides a brief account and commentary on these developments.
For many years the lead on injury valuation in Britain has been taken by the Department of Transport (now the Department of the Environment, Transport and the Regions - DETR). A great deal of pioneering work has been sponsored by them using contingent valuation (CV) as the reference technique.4 Much of this work has been carried out using increasingly sophisticated and specialist economic tools without, it is probably fair to say, much opportunity for cross-disciplinary or public scrutiny.
Elsewhere, in the broader field of risk management, the emphasis has been moving rapidly away (in theory at least, but no doubt practice will follow) from one of ‘the expert knows best’ towards a much more open and consultative process in which stakeholders are encouraged to contribute directly in decision making. Now, it has been argued that contingent valuation is, indirectly, a consultative technique since it relies upon asking consumers how much they would be willing to pay to avoid the ‘pain, grief and suffering’ of injury and death. What, on the face of it, could be fairer than that? Unfortunately, this argument breaks down should willingness to pay (WTP) fail in some way to measure consumer desires.5 In such circumstances it could even subvert the democratic decision process by masquerading as truth.
In the case of the DTI, it was clear that any use of safety valuation would have to stand a reasonable chance of being acceptable and plausible to the exceedingly diverse group of stakeholders with interests in consumer products and services. Furthermore, consumer safety decision making has tended historically to be a more open and consultative process than has typically been the case in the road safety sector. The implication of this was that room was always going to be necessary for negotiation.
Thus, a critical review was carried out for DTI of WTP and other techniques for injury valuation from a multi-disciplinary perspective. This review concluded that, so far as CV studies were concerned, there remained many reservations about the technique, despite the years of research. In summary, these included the following: people may not have clear pre-formed preferences for non-market goods, and survey responses may therefore not be an accurate measure of true preferences; the CV task may be too complex for respondents; the scenario against which the CV task is performed may inadequately encapsulate the issue with which the respondent is concerned; ‘embedding’ effects may occur if the respondent does not clearly distinguish between subsets of a good and a good in its entirety; people may ‘construct’ their preferences using the information provided; respondents may be insensitive to the size of hypothetical risk reductions; biases may be generated by the survey methodology; and so on.6 To this list should be added the task of reconstructing a societal value of safety from numerous individual WTP responses, itself a process requiring additional non-trivial assumptions and value judgements.
Given that CV is usually regarded as the technique with the most going for it as far as valuation of non-market goods is concerned, the above catalogue makes depressing reading. Or does it? The answer to the question really depends on how accurate you think the answer could be anyway, or is needed to be.
Regarding accuracy, I am reminded of my early days as a physicist when we learnt that there was an absolute limit to the precision with which the position and momentum of sub-atomic particles (electrons) could be determined, irrespective of any future developments in measurement techniques. Electrons apparently were inherently fuzzy objects and, like it or not, we would have to live with it. The analogy for safety valuation is that, apart from methodological problems, the variance observed in results from CV studies must also contain a component attributable to the inherent imprecision or fuzziness of the concept for human beings. On reflection, it is inconceivable that people could hold anything approaching precise valuations of such entities as pain and suffering. This fuzziness is tantamount to random background noise that is impenetrable by further research, however cleverly conducted.
Thus, an alternative way of thinking about the value of a statistical life (VOSL) would be to ask how precise a quantity it would be if a perfect research instrument existed. The answer to such a question would then provide some guidance on what further research was warranted. In the UK there has been a little speculation about what the limit might be. In 1993 Jones-Lee, our most eminent safety economist, suggested the limit might be an order of magnitude.7 More recently, in a report following a UK inter-departmental consultative process on the setting of safety standards, the limit was placed at ± 50%,8 both of which values are of course quite significant. Furthermore, a highly complex backdrop of both personal and societal contextual factors that are constantly being reassessed by those concerned compounds this ‘noise.’ Given the diversity of interest groups in the consumer sector, any suggestion that an attempt to identify anything approaching a high-precision valuation of safety would be a lost cause.
So what should be done? The choice recommended to the DTI, and subsequently accepted, was to bite the bullet and acknowledge that VOSL was not a well-defined quantity, and therefore to specify it as a range rather than a point value. Although this created a ripple of surprise in some circles, this choice in fact had significant advantages. First, it was more ‘scientific’ to be open about uncertainty and hence more defensible; second, admission of uncertainty was likely to make the whole concept more plausible so far as consumer groups were concerned (cf. the 4 significant figures assigned to VOSL in Table 1); third, specifying VOSL as a range is one way of providing decision makers with room to manoeuvre in safety decisions. The pressing need for space to negotiate and show sensitivity to stakeholder concerns is one of many lessons which effective risk managers have learnt over past decades.9,10 Fourth, the upper limit of the range could be used to screen out rogue decisions, which is arguably what the cost of safety debate has largely been about anyway. So, paradoxically, defeat was turned at least in small measure into victory.
There still remained the question of what the range for VOSL should be. The review indicated that so far as human costs (pain, grief and suffering) were concerned, the likely range was £0.5 million to £10 million. This was rounded up to £1 million to £10 million to allow for lost output and other costs. It was suggested that for most decisions DTI might use a more restricted range of from £2 million to £4 million. Again, this caused a ripple of shock in some quarters because it was felt to be ‘high’ compared with currently-accepted valuations like that of £902,500 in Table 1. In defence of the higher value, several factors are pertinent, one of which is the long lead-time before many consumer safety decisions bite. For instance, measures on furniture safety may not fully impact for two decades or more.
Regarding the HSE-led study, only the first phase of this has been reported, so no more than a preliminary account is possible. The overall aim of the study was to produce safety valuations for different hazard contexts, with the first report dealing with road fatalities. The study was singularly noteworthy in the extent to which pre-piloting and other measures were adopted in order to try to deal with some of the known problems associated with CV.
In particular, the problem of lack of sensitivity of respondents to the size of risk reduction available was carefully examined. In each of two phases of piloting, this problem was found to be severe. For example, forty per cent of respondents reported identical WTP for two risk reductions, one of which was three times the other. This was studied using tape recordings of individual interviews and follow-up focus group meetings. It emerged that the likely reasons were: many people found the risk reductions tiny and marginalized them; any safety improvement was seen as a good thing and the actual magnitude of risk reduction was of secondary or no importance; that when considering how much the ‘good thing’ was worth, this was equated to what could easily be afforded – usually £50 to £200 per annum. All of this confirms what has long been said about CV.
In order to try to overcome these problems the authors resorted to various sophisticated measures. This included the use of a combination of CV for a non-fatal injury, a variant of standard gamble to elicit willingness to trade off risk of non-fatal injury against death, and the linkage of these results to infer the marginal rate of substitution for death. Even so, it is apparent from the text that many heroic assumptions and value judgements were necessary in order to turn the results into a value of safety for the avoidance of road traffic fatalities. The final conclusion being that a range was appropriate, in fact any figure from £750,000 to £1,250,000 could be regarded as ‘broadly acceptable,’ a finding which of course is consistent with that in Table 1. However, with this kind of research, in which it is customary to discard data that does not fit the model, one encounters the danger of simply regenerating previously held ideas. Such a thing has happened in physics. At one time there was a remarkable consistency in measurements by different scientists of the velocity of light. This continued until a radically different result was reported, at which point all further measurements clustered around the new value.
Apart from the methodological value judgements, other assumptions are entrained within the procedure that would have substantial effects on the outcome. To mention just one, those who are exposed to road traffic risks are presumed to care only for themselves and not at all for others. Jones-Lee et al.11 have suggested in the past that were an altruism factor included, it alone might boost WTP for avoidance of a fatality by about £0.5 million (1983 prices). While belief in altruism may these days be at a low ebb in CV circles, it is not universally so. As was written to The Times of London in 1999 by the president of The Pedestrians Association (UK):
“But for most of us what counts is funerals per year – how many of our sons, daughters, spouses, parents and neighbours die on the roads. We should, furthermore, forecast these deaths. Let us assume that it will be possible to cut road deaths by 5 per cent a year over the next decade. Between 1999 and 2009 that would still give us 24,000 funerals. Those deaths would be family tragedies.”12
It is hard indeed to believe that individuals could be so uncaring as to place zero value on others, and evidently this is not a position shared by the Pedestrians Association. Given that the annual risk of death from road traffic is about 1:8000 per year in industrialized countries, this suggests that all of us will have a family member or close friend killed during our lifetime. As remarked by Dorman, himself an economist though perhaps not a member of the mainstream:
“We are all at risk of missing the most fundamental aspects of life, things that should be obvious to us but are not. Academic economists possess this trait in abundance”13
Personally I don’t wish to pick on economists in particular. It is well known that experts of all persuasions have blind spots. That is why research such as this, which can have a profound effect upon public safety, needs to be opened up to as wide a scrutiny as possible.
Ted R. Miller
Thomas R. Ireland
National estimates of lifetime economic burdens associated with injury and illness rely on problematic life table and productivity loss computations. In part, these problems arise because the human capital cost method used to estimate economic costs employs costing methods for work loss that were developed in 1966 and then standardized in the early 1980s (Rice et al. 1966, Hodgson and Meiners, 1982). Since the early 1980s, forensic economists have found flaws in those methods and developed improved methods for use when litigation requires valuing health-related work losses. This paper applies what those economists have learned to suggest improvements in the classic approach for valuing injury or disease burden.
The paper makes 11 points: (1) Most life expectancy tables provide a static picture based on current health status in a population cross-section rather than the life expectancy of population age cohorts. Since health is improving, such tables underestimate actual life expectancy. (2) The true life expectancy for serious injury victims is shorter than for the average population. (3) Work loss costs have an appropriate place in a QALY-based cost-framework, albeit a limited and possibly forced one. (4) There are two basic approaches for projecting earnings loss due to injury or illness. Each approach has limitations and neither produces perfect results. (5) The current methods assume that the current national cross-section of wages by age and sex will be the pattern of earnings for individuals as they age in the future. For sex based and education based reasons, it is likely that the future will be different than the present. (6) Too often, we use factors such as labor force participation and unemployment rates that are dependent on the business cycle but apply the most recent year of data rather than an average across the cycle. (7) For purposes of placing costs on occupational injury, using growth rates based on economy-wide averages may distort an analysis. (8) The merit of including the value of family services lost in a WTP or QALY-based framework is an open question. (9) The services an individual provides within the family extend far beyond simple production relationships. If family services are valued separately, there should be more consideration of the complex nexus of interrelationships that produce care, support, guidance and counsel for all family members. The degree to which different surveys of family services capture parts of this important complexity varies enormously. (10) Death and catastrophic injury disrupt the lives of the surviving family, affecting earnings, educational achievement, family services, and quality of life. They can cause depression and other costly mental health problems. Even QALY loss estimates generally fail to capture these impacts, especially for fatalities. (11) When a worker is killed or permanently disabled, society incurs costs beyond individual work losses because of the hiring, training, and other costs of workers shifting between jobs. Such friction costs have been measured in the Netherlands and should be assessed and properly incorporated into the costing framework elsewhere.
What Life Table Should We Use?
Standard life tables have been cross-sectional. They give historical survival probabilities of the current population by age group, sex, and race. For example, the survival probability for someone age 61 might be the mean survival probability from age 60 to 61 observed among people who now are age 63. This approach implies that life expectancy is constant over time. However, the current cohort that includes 10 year olds will have a very different profile of education, early health behaviors, available health treatments, and environmental exposures when they reach age 85 than does the current cohort that includes people age 85. For this reason, the Board of Governors of the U.S. Social Security Commission recently concluded that future retirement benefits have been substantially underestimated. The Governors urged the Commission to develop and adopt cohort life tables that project the survival probabilities of people currently of a given age and gender, taking into account future developments that can be reasonably anticipated. The cost of illness literature needs to make a similar change to cohort based life expectancy tables. This should be done with caution. Available life insurance actuarial tables are often cohort-based, but they describe the expected life spans of insurance purchasers, a risk adverse group with above-average survival probabilities.
A third type of life table has recently been developed in the U.S. (Robine et al. 1991, 1992; Erickson et al. 1995, Expectancy Data 1999) and is needed elsewhere. This type of life table focuses on expected years of healthy life. For each age group by gender, this table is computed by multiplying the probability of surviving another year from a cross-sectional or cohort life table times the average health of people of the given age group and gender. Average health levels are measured in quality-adjusted life years (QALYs). The QALY data come from a national household survey like ones conducted recently in the U.S., Canada, Netherlands, and the United Kingdom, among others. Correctly estimating QALYs lost to premature death or permanent disability requires a Healthy Life Expectancy (HLE) life table. HLE expectancy will always be smaller than life expectancy due to health impairments associated with the aging process. For example, a baby boy who suffocated had about a 75% chance of living to age 66. That means his death cost an expected 0.75 years of life from age 66 to 67. Men age 66 have an average QALY level of 0.74 (Expectancy Data 1999), so the corresponding QALY loss would be 0.75 * 0.74 = 0.555 years.
Do Catastrophic Injury Victims Have Typical Life Spans?
Injury is the leading cause of death from ages 1 to 45 in the U.S. and accounts for more than half of deaths in some age groups. Injury’s dominant role as a cause of death implies survival probabilities in a life table are sensitive to fatal injury risk, especially at younger ages. The issue thus becomes, do people who die from injury have atypical survival probabilities? We first answer this question for alcohol-involved fatalities. High blood alcohol levels significantly increase the probability of fatal or catastrophic injury, most notably in traffic crashes. High blood alcohol levels also correlate with other often-fatal future alcohol-linked morbidity such as cirrhosis and cancer. Further, heavy drinkers have above-average probabilities of smoking, drugging, driving aggressively, and failing to buckle up. These observations strongly suggest that people who are fatally injured while having high blood alcohol levels are risk-takers with elevated probabilities of premature death. Thus, if these individuals had survived the injuries that killed them, they still would have had life expectancies well below the average. In the United States, about 35% of fatal injuries involve persons with high alcohol blood levels (Smith et al. 1999), even though only 11% of the US population are heavy drinkers. This strongly suggests that the average pre-event life expectancy for people who died from injury is substantially below the average for the population as a whole. Yet we all use average life expectancies to compute lifetime medical costs, productivity losses, and QALY losses. That means we overestimate the losses.
Should Productivity Loss Be Explicit in Willingness to Pay (WTP) and QALY Costs?
This section neither endorses nor disavows the use of WTP, a choice on which the authors are divided. We agree that explicit work loss estimates are informative and valuable regardless of the cost framework used. People want to know about work losses and related tax losses, but including them with QALYs or WTP estimates poses the risk of double counting. A WTP value of life or a QALY estimate inherently includes all sources of value from the life process. Therefore, its use with a separate wage loss estimate implies that wage loss has been double counted. However, using only WTP or QALY estimates of injury burden means that persons uncomfortable with using WTP measures or QALYs will be unable to salvage cost estimates from the burden estimates.
One approach which Arthur (1980) and Miller, Calhoun, and Arthur (1989) claim is appropriate is to subtract the work loss from the WTP estimate and show it as a separate cost component. This allows analysts who use a WTP cost framework to provide information on productivity losses without double counting. (Note that the total cost simply is WTP.) Miller (1993) and subsequent studies took this approach, arbitrarily naming the residual “pain, suffering, and lost quality of life.” Using that terminology, “quality of life” encompasses life quality reductions beyond those resulting from losing wages that funded consumption, plus any residual value for being alive as compared to dead. Similarly and less controversially, one could include the work losses in costs and separately show a non-monetary estimate of QALYs lost. The QALY measure used should exclude impacts on work-related functioning or be adjusted to remove those impacts. Even with an appropriate QALY measure, it is unclear that QALY loss and wage loss can be cleanly separated since quality of life comes, in part, from earnings-dependent consumption.
How Do We Calculate Lifetime Productivity Loss: LPE versus WLE?
Two methods typically are used to compute lifetime productivity losses: LPE (Life-Participation-Employment) and Work Life Expectancy (WLE). LPE and WLE estimates typically differ by about 5%-6%. Choice between these methods can depend on data availability or a strength or weakness in a specific application. LPE has most often been used in national cost of injury estimates. The LPE formula (with the sum to the oldest age the data permit) is:99
where for age i and sex j
The WLE formula is simpler (using all of the same symbols):
S Eij * Wij / [(1+p) / (1+r)]**(i-c+1)
i = c
The advantages and disadvantages of WLE and LPE are mirror images. WLE is much easier to understand. With WLE, it is easier to tailor work life to atypical groups and data more often are available for subgroups, for example, by education or occupation. However, most uses of WLE assume that years are worked consecutively, which is a serious problem when valuing losses for women of childbearing age. Arbitrary methods exist to adjust the work life pattern to reflect such considerations but data have not been compiled for a scientifically sound adjustment. Another disadvantage of WLE tables is that they are updated infrequently, while data for LPE calculations are updated annually.
Several problems arise in work loss estimation. Both LPE and WLE assume the labor force participation probability is independent of prior participation. In other words, they use identical probabilities of future participation for situations in which all victims were employed when they died and in which very few victims were employed when they died. An occasional mistake arises when analysts use data for average earnings and then apply an unemployment percentage to those earnings. In cases where average earnings data already include unemployed persons, this results in at least partial double counting of the probability of being unemployed. A more frequent error is the use of the most recent values for P and E in analyzing lifetime losses. Since these factors vary with the business cycle, P and E values should be based on averages over the business cycle. These two problems are compounded if average earnings data that includes unemployment is combined with current P and E values. The average earnings values reflect the stage of the business cycle in a way that is magnified when combined with contemporaneous participation and employment rates. However, trying to average over the cycle is also dangerous. Uncertainty in the lifetime estimates is large because economies often vary in unprecedented ways. In the U.S. for example, the boom side of the business cycle appears to be lengthening.
Both LPE and WLE assume injured persons are randomly drawn from the population. However, the population of severely injured persons is not randomly selected. Heavy drinking, for example, lowers expected earnings (Rice et al. 1990). Homicide and fatal fire victims are disproportionately low-income people. Low-income babies are less likely to travel in child seats (Mayer and LeClere 1994), meaning babies killed in automobile crashes are likely to be low-income babies. For any of these reasons, averages based on the population as a whole can overestimate true losses. Averages also can underestimate true losses. Drowning victims in swimming pools, for example, are likely to drown at home or in a neighbor’s pool, which means that their families could afford swimming pools. People killed in airline crashes could afford to fly. Side airbags are only in newer, more expensive vehicles, so they are only likely to kill high-income people. Anyone killed at work currently was employed, meaning that she or he was not unemployed at the time. Studies that examine workplace fatalities often focus on one occupation or industry, for example construction where workers earn above-average wages. Unfortunately, estimating group-specific participation rates, employment rates and earnings data is very difficult, but accuracy is lost when this is not done.
Should Family Services and Housework Loss Be Evaluated?
Family services and housework losses certainly belong in an economic cost estimate. These are not included in the Gross Domestic Product (GDP) in most countries, which makes their inclusion in cost comparisons both difficult and uncomfortable. Since there are no standardized methods for valuing these services, cross-country comparisons are also hampered. It is debatable whether family services should be valued separately in a QALY or WTP framework. A strong argument against omitting them is that they discriminate against women, especially full-time homemakers, who perform most of these services.
Empirically, the analyst valuing family services needs to thoughtfully consider what time uses have been defined as “household services” in the underlying time use study being used to measure family services. Sometimes, “household services” are narrowly defined and omit many important educational services by parents for children, for example. If educational services by parents have not been accounted for in other ways, they should be included in cost comparisons as part of projected family services. From our perspective, family services such as care, support, guidance, counsel, and financial management are important losses; they should not be ignored.
Should We Value Caregiver, Survivor and Other Family Costs of Catastrophic Injury?
Especially when a child is the victim, catastrophic injury disrupts earnings, education, and family duties of caregivers, other family, and friends. The amount of needed family services rises significantly. The child’s injury drives family members and friends into depression and other mental illness. It reduces the quality of life for all family members in ways that may be partly captured in willingness to pay measures but almost never is in QALY estimates (for an exception, see Mohide et al. 1988). These losses can be substantial, but we rarely account for them. We should.
What About Friction Costs?
Friction costs (Koopmanschapp et al. 1995) measure employer productivity losses outside the family, specifically the costs that ripple through the economy as employers hire and train permanent and temporary replacement workers and juggle work schedules following a death or disabling injury. They belong in WTP, QALY, and economic cost frameworks when estimating societal costs.
This paper derives from work under National Institute on Occupational Safety and Health grant R01/CCR312179 and Health Resources and Services Administration Children’s Safety Network contract MCJ-240-98-0006. All opinions are the authors’ alone.
Lois A. Fingerhut, M.A.
The International Collaborative Effort (ICE) on Injury Statistics is one of several international activities sponsored by the National Center for Health Statistics (NCHS) for the purpose of improving international comparability and quality of injury data. The ICE serves to provide the data needed to better understand the causes of injury and the most effective means of prevention and to provide a forum for international experts to discuss data related issues. The ICE focuses on injury definitions, data collection methodologies, coding and classification. The ICE is comprised of representatives from about a dozen countries including colleagues primarily from government, academia and injury prevention and control units. The first meeting of the ICE was held in May 1994 and meetings have been held nearly annually since then. ICE receives funding from the National Institute of Health’s National Institute of Child Health and Human Development.
Projects that ICE has been engaged in include:
To learn more about ICECI, see http://www.nhtsa.dot.gov/exit.cfm?link=http://www.ecosa.org./csi/ecosa.nsf/news
ICECI is a classification tool for injury researchers and data collectors that is being designed for data collection in emergency departments (referred to also as accident and emergency or a&e). What makes ICECI unique is that it is multi-axial, allowing one to capture many different dimensions of injury (cause, intent, place, activity, drug use, and others) as separate codes rather than the International Classification of Disease (ICD) which is one-dimensional, forcing one code to capture several dimensions. The final version of ICECI is scheduled to be released in October 2000. Primary responsibility for ICECI is with the Consumer Safety Institute in Amsterdam, The Netherlands.
The core elements of ICECI are:
Modules that are being developed include:
In addition to the full ICECI, a short version is under development in the United States as a routine surveillance tool for use in emergency departments. This version was developed to be fully compatible with the full ICECI and with the ICD-10 external cause of injury code sets. The short ICECI is being developed at the National Center for Injury Prevention and Control at the National Centers for Disease Control and Prevention (CDC) in Atlanta, Georgia.
In many parts of the world, the ICECI and even the short version of the ICECI will not be a feasible tool for data collection. A real need was perceived to develop a minimum set of data elements that could be collected even among the most resource challenged environments. Toward that effort, a multi-national group including representatives from the CDC, Norway, South Africa, Uganda, India and the WHO have been working together to develop a manual to provide injury researchers with the tools for establishing a surveillance system that could be used at the most “basic” level.
In selecting a classification, be it the ICD, the ICECI, the short ICECI or any other one, it is important to decide in what settings will it be used; i.e., clinics, emergency departments, inpatient settings or for death certification. The setting can help determine the most appropriate choice. In addition, the user should fully understand the limitations of whatever classification is chosen – for example, does it rely only a single code for use; is it multi-axial; are there guidelines for use.
Challenges facing the international injury data community include
For more information on the ICE, see http://www.nhtsa.dot.gov/exit.cfm?link=http://www.cdc.gov/nchs/about/otheract/ice/ice.htm