|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-HRT-05-051
Date: October 2005
Crash Cost Estimates by Maximum Police-Reported Injury Severity Within Selected Crash Geometries
PDF Version (424 KB)
PDF files can be viewed with the Acrobat® Reader®
Estimating crash costs requires estimates of the number of people involved in a given crash, the severity of each person's injuries, and the costs of those injuries, as well as associated vehicle damage and travel delay. The following section describes the methodology used to estimate the incidence and severity of crashes for selected geometries and speed limits. The succeeding section explains how the costs of crashes were estimated.
Injury Incidence and Severity Estimation
To estimate injury incidence and severity, procedures developed by Miller and Blincoe (1994)(6) and Miller, Galbraith et al. (1995)(7) and applied in Blincoe, Seay, et al (2002)(8) were followed. The estimates of the average number of people involved in a crash-by-crash geometry, speed limit, and police-reported severity come from National Highway Traffic Safety Administration's (NHTSA() GES and Crashworthiness Data System (CDS).
Crash databases do not accurately describe the severity of crashes. Accordingly, several adjustments, described below, were made to more accurately reflect the severity of crashes.
First, GES provides a sample of U.S. crashes by police-reported severity for all crash types. GES records injury severity by crash victim on the KABCO scale (National Safety Council, 1990)(3) from police crash reports. Police reports in almost every state use KABCO to classify crash victims as K-killed, A-disabling injury, B-evident injury, C-possible injury, or O-no apparent injury. KABCO ratings are coarse and inconsistently coded between states and over time. The codes are selected by police officers without medical training, typically without benefit of a hands-on examination. Some victims are transported from the scene before the police officer who completes the crash report even arrives. Miller, Viner et al. (1991)(5) and Blincoe and Faigin (1992)(9) documented the great diversity in KABCO coding across cases. O'Day (1993)(10) more carefully quantified the great variability in use of the A-injury code between states. Viner and Conley (1994)(11) explained the contribution to this variability of differing state definitions of A-injury. Miller, Whiting, et al. (1987)(12) found police-reported injury counts by KABCO severity systematically varied between states because of differing state crash reporting thresholds (the rules governing which crashes should be reported to the police). Miller and Blincoe (1994)(5) found that state reporting thresholds often changed over time.
Thus, police reports do not accurately describe injuries medically. To minimize the effects of variability in severity definitions between States, reporting thresholds, and police perception of injury severity, NHTSA national data sets were used that included both police-reported KABCO and medical descriptions of injury in the Occupant Injury Coding system (OIC) (American Association of Automotive Medicine(AAAM), 1990; AAAM, 1985).(13,14) OIC codes include AIS score and body region and more detailed type injury descriptors that changed from the 1985 to the 1990 edition. Both 1999-2001 CDS (NHTSA, 2002)(15) and 1982-1986 NASS (NHTSA, 1987)(16) data were used. CDS describes injuries to passenger vehicle occupants involved in tow-away crashes, but not in nontow-away crashes. The 1982-1986 NASS data were used to fill this gap. While not recent, these data provide the most recent medical description available of injuries to other non-CDS crash victims. The NASS data were coded with the 1980 version of AIS, which differs slightly from the 1985 version; but NHTSA made most AIS-85 changes well before their formal adoption. CDS data were coded in AIS-85, then in AIS-90.
The 1999-2001 GES data were used to weight the NASS data so they represent the annual estimated GES injury victim counts in non-CDS crashes. In applying these weights, the data was controlled by crash type, police-reported injury severity, speed limit <=72 km/h (<=45 mi/h) and >=80 km/h (>= 50 mi/h), and restraint use. Weighting the NASS data to GES restraint use levels updates the NASS injury profile to a profile reflecting contemporary belt usage levels. Sample size considerations drove the decision to pool 3 years worth of data. At the completion of the weighting process, a hybrid CDS/NASS file had been developed that included weights that summed to the estimated current annual incidence by police-reported injury severity and other relevant factors.
Crash Cost Estimation
The second step required to estimate average crash costs was to generate estimates of crash costs by severity, as described in this section. To estimate the average costs per crash by geometry, speed limit, and police-reported crash severity, costs per injury by maximum AIS (MAIS), body part, and whether the victim suffered a fracture/dislocation were adapted from the costs in Zaloshnja, Miller, et al.(2) (2002). These costs were merged onto the GES-weighted NASS/CDS file.
Comprehensive costs represent the present value, computed at a 4 percent discount rate, of all costs over the victim's expected life span that result from a crash. The following major categories of costs were included in the calculation of comprehensive costs:
Human capital costs excluded the last item. The following text provides an overview of the bases for each of these cost components.
Medically related costs include ambulance, emergency medical, physician, hospital, rehabilitation, prescription, and related treatment costs, as well as the ancillary costs of crutches, physical therapy, etc. To estimate medical costs, nationally representative samples that use International Classification of Diseases ‑ 9th Revision ‑ Clinical Modification (ICD9-CM) diagnosis codes to describe the injuries of U.S. crash victims were used.(17) The samples were the 1996-1997 National Hospital Discharge Survey (NHDS) for hospitalized victims and 1990-1996 National Health Interview Survey (NHIS) for nonhospitalized victims. The analysis included the following steps, some of which are explained in further detail below:
Cause of Injury Assignment
NHIS explicitly identifies victims of road crashes and NHDS includes data fields where hospitals code injury diagnoses or causes. When all seven fields are used, a cause code is rarely included. Typically, diagnosis codes (which are linked to insurance reimbursement costs) are given priority over cause codes. More than 70 percent of 1996-1997 NHDS cases with less than six diagnoses were cause-coded by age group, sex, and diagnoses for these cases were representative of all injury admissions with less than six diagnoses. For NHDS cases with six or seven diagnoses, causation probabilities by age group, sex, and diagnosis were inferred using data for cases with at least six diagnoses in cause-coded state hospital discharge censuses that had previously been pooled from California, Maryland, Missouri, New York, and Vermont (Lawrence et al., 2000).(18) As a partial check, the resulting firearm injury estimate was compared with a published national surveillance estimate (Annest et al., 1995).(19) The two estimates were less than 5 percent apart.
Estimation of Medical Costs Associated with Each Crash Case in NHDS and NHIS
Except for added tailoring to differentiate the costs of child from adult injury and estimating fatality costs, the methods used were the same as those employed in building the U.S. Consumer Product Safety Commission's (CPSC) injury cost model. These methods are summarized below and documented in detail in Miller et al. (1998),(20) Miller, Romano, and Spicer (2000),(21) Lawrence et al. (2000),(18) and Zaloshnja, Miller, et al (2002)(2)
Although the methods for estimating the costs and consequences associated with each case differed for fatally injured persons, survivors admitted to the hospital, and survivors treated elsewhere, in each case, costs of initial treatment were extracted from nationally representative or statewide data sets. For survivors, diagnosis, aggregate medical follow up, rehabilitation, and long-term costs computed from national data on the percentage of medical costs associated with initial treatment were added. Due to data unavailability, these percentages were less current than the costs for initial treatment.
For hospitalized survivors, medical costs were computed in stages. Maryland and New York were the only States that regulated and tracked the detailed relationships between charges, payments, and actual costs of hospital care in recent years. Moreover, because U.S. health care payers negotiate widely varying, sometimes large discounts from providers, hospital charges bear little relationship to actual hospital costs. Computations were by diagnosis group. Using average cost per day of hospital stay by State as an adjuster (Bureau of the Census, 1999, Table 189),(22) diagnosis-specific hospital costs per day from Maryland in 1994-1995 and New York in 1994 (the last year of that State's cost control) were price-adjusted to national estimates. The costs per day were multiplied by diagnosis times corresponding NHDS lengths of hospital stay. Physician costs estimated from Civilian Health and Medical Program of the Uniformed Services (CHAMPUS) data for 1992-1994 were added to the hospital costs.(23) Costs after hospital discharge were computed from the most recent nationally representative sources available, the 1987 National Medical Expenditure Survey (NMES) and National Council on Compensation Insurance (NCCI) data for 1979-1987.(24,25) Both CHAMPUS and NCCI data report only primary diagnoses at the three-digit ICD level or broader, so mapping was imperfect, especially for brain injury. The NCCI data describe occupational injury; however, following Rice et al. (1989),(26) Miller (1993),(27) and Miller et al. (1995),(7) we assumed the time track of medical care by diagnosis is independent of injury cause. Where the victim was discharged to a nursing home, following Lawrence et al. (2000),(18) nursing home lengths of stay were estimated at two years for burn victims and ten years for other catastrophic injuries, at a cost double the cost of an intermediate care facility. Costs per visit for other nonfatal injuries came from CHAMPUS.
Past studies (e.g., Rice et al., 1989; Miller, 1993; and Miller et al., 1995)(26,27,28) estimated lifetime medical spending due to a child's injury from the all-age average acute care spending shortly after the injury and the longer term recovery pattern of adults or victims of all ages. In this study, the hospitalization cost estimates are age-specific. Using longitudinal 1987-1989 health care claims data from Medstat MarketScan® Databases, diagnosis-specific factors were estimated to adjust all-age and adult estimates of follow up and longer-term care to child-specific treatment patterns.(29) The percentage of medical costs in the first 6 months that resulted from the initial medical visit or hospitalization did not vary with age. After that, children were more resilient; the percentage of their total treatment costs incurred in the first 6 months often was higher, especially for brain injuries. These conclusions were derived from analysis of a random sample of 15,526 episodes of childhood injury and 40,624 episodes of nonoccupational adult injury to victims covered by private health insurance. For each episode, the claims data covered a range of 13-36 months and an average of 24 months after injury. Because it was decided that the diagnostic detail preserved should be maximized, sample size considerations dictated bringing costs forward onto CDS files that represented averages across victims of all ages.
For spinal cord injuries (SCI) and burns, medical costs were not estimated from NHDS and NHIS files because of the limited number of these cases in the files. In addition, long-term SCI costs are not captured in the NHDS and NHIS data. Information from a special study (Berkowitz et al., 1998)(30) was used to estimate first year and annual medical costs for SCI. Costs were estimated by applying the age and gender distribution of SCI victims in the CDS 1993-1999 to a lifetime estimating model with 1997 life expectancy tables adjusted for spinal cord injury mortality rates from Berkowitz et al. (1998).(30) Highway crash-specific costs for burns were adapted from Miller, Brigham, and Cohen et al. (1993),(31) using its regression equations.
Mapping ICD Codes into OIC Codes
ICD-based injury descriptors were mapped to AIS-85 and body part to make the ICD data compatible with CDS and NASS descriptors. AIS-85 was mapped using the ICDmap-85. This map lists AIS by each ICD code up to the five-digit level of detail. For NHIS, which uses almost exclusively three-digit ICDs (85.5 percent of the data set), the lowest AIS within that 3-digit group was selected.
Body part was mapped to AIS from previously collapsed ICD groupings (Miller et al., 1995)(29) and fracture or dislocation was identified with the ICD codes. The ICD/AIS mapping was developed by consensus and contains many assumptions related to the assignment of AIS codes to ICD rubrics (Miller et al. 1995).(28) For multiple-injury NHDS cases, the body part of the maximum AIS injury was assigned. In case of a tie in AIS, the body part defined by the principal diagnosis in discharge records was used. NHIS reports only principal diagnoses.
Inferring Costs for Categories that Appear in CDS Data, but not in the ICD-Based File
Six percent of the AIS/body part/fracture diagnosis categories that appear in CDS crash data did not appear in the ICD-based files. Costs for these categories were assigned as follows: (1) mean costs were estimated for each AIS, (2) based on these averages, incremental cost ratios from one preferably lower AIS to another were estimated. Lower AIS was preferred because it offered larger case counts. Finally, (3) costs for empty ICD-based cells were assigned by multiplying costs from adjacent cells by this ratio. For instance, if the mean medical costs for AIS-2 and AIS-3 were $500 and $1,000, respectively, then the incremental ratio for AIS-2 to AIS-3 was set to: 1,000/500 = 2. Then, the cost for an empty AIS-3 cell was estimated by multiplying the body part/fracture-specific cost for AIS-2 times the ratio. For body parts with no cost estimates available for any AIS, a general average cost for the appropriate AIS was assigned.
Emergency Services Costs
This cost category includes police and fire services. Fire and police costs were computed from assumed response patterns by crash severity and vehicle involvement, constrained by data on total responses. For fatal, injury, and PDO crashes, time spent per police cruiser responding came from ten jurisdictions with automated police time-tracking systems. A single officer was assumed to have responded to a PDO crash and one officer per injury to other crashes. Time spent per fire truck responding came from nine large fire departments. It was assumed that the fire personnel would respond to the following:
Property Damage Cost
This includes the cost to repair or replace damaged vehicles, cargo, and other property, including the costs of damage compensation. Property damage costs are from Blincoe, Seay, et al (2002).(8)
Lost Productivity Cost
Lost productivity costs include wages, benefits, and household work lost by the injured, as well as the costs of processing productivity loss compensation claims. It also includes productivity loss by those stuck in crash-related traffic jams and by coworkers and supervisors investigating crashes, recruiting and training replacements for disabled workers, and repairing damaged company vehicles. Excluded are earnings lost by family and friends caring for the injured and the value of schoolwork lost. The productivity loss resulting from traffic delay is part of the total productivity lost.
Future work-loss costs were estimated using methods that parallel the Consumer Product Safety Council (CPSC) Injury Cost Model. These methods are summarized below and documented in detail in Miller et al. (1998),(20)Lawrence et al. (2000),(18) Miller, Romano, and Spicer (2000),(21) Blincoe, Seay, et al (2002),(8) and Zaloshnja, Miller, et al. (2002).(2)
For nonfatal injuries, the work loss cost is the sum of the lifetime loss due to permanent disability (averaged across permanently disabling and nondisabling cases) plus the loss due to temporary disability. Lifetime wage and household work losses due to a death or permanent total disability were computed and then discounted to present value with the standard age-earnings model described in Rice et al. (1989)(26) and in Miller et al. (1998).(20) The inputs to this model were for 1997-2000. They include, by age group and sex, survival probabilities from National Vital Statistics Reports (1999);(33) weighted estimates of annual earnings tabulated from the 2001 Current Population Survey, a nationally representative sample; and the value of household work performed from Expectancy Data (1999).(33)
For survivors, NCCI probabilities that an occupational injury will result in permanent partial or total disability and the NCCI percentage of earning power lost to partial disability were applied to compute both the number of permanently disabled victims and the percentage of lifetime work lost. These data are listed by diagnosis group and whether injuries resulted in hospitalization. The ICD maps were used to assign 1985 and 1990 OIC injury codes or code groups to each category.
Diagnosis-specific probabilities of injuries to employed people causing wage work loss came from CDS 1993-1999. The days of work loss per person losing work were estimated from the 1999 Survey of Occupational Injury and Illness of the U.S. Bureau of Labor Statistics; this survey contains employer reports of work losses for more than 600,000 workplace injuries coded in a system akin to the OIC but with less diagnostic detail. According to a survey of 10,000 households, injured people lose housework on 90 percent of the days they lose wage work (S. Marquis, 1992).(34) Thus, it was possible to compute the days of household work lost from the days of wage work lost. Household work was valued based on the cost of hiring people to perform household tasks (e.g., cooking, cleaning, and yard work) and the hours typically devoted to each task from Expectancy Data (1999).(33) Lost productivity for repairing vehicles involved in crashes was updated from Miller et al. (1991)(5) and included in the lost household productivity.
For temporary disability, it was assumed that an adult caregiver would lose the same number of days of wage work or housework because of a child's temporarily disabling injury as an adult would lose when suffering the same injury. Since the adult with the lowest salary often stays home as the caregiver, caregiver wages were estimated as the mean hourly earnings for nonsupervisory employees in private nonagricultural industries. These assumptions may provide a small overestimate because the caregiver may be able to do some work at home. Conversely, the analysis may underestimate the losses because it ignored the work loss of other individuals who visit a hospitalized child or rush to the child's bedside shortly after an injury and any temporary wage work or household work loss by adolescents.
Legal and insurance administration costs per crash victim were derived from the medical and work loss costs, using models developed by Miller (1997).(1) Legal costs include the legal fees and court costs associated with civil litigation resulting from motor vehicle crashes. In estimating these costs, the probability of losing work, the percentage of victims who filed claims, the percentage of claimants who hired an attorney, estimated plaintiff's attorney fees, and the ratio of legal costs over plaintiff's attorney fees was taken into consideration. Insurance administration costs include the administrative costs associated with processing insurance claims resulting from motor vehicle crashes and defense attorney fees. In estimating these costs, medical expense claims, liability claims, disability insurance, Worker's Compensation, welfare payments, sick leave, property damage, and life insurance were estimated.
Following Blincoe, Seay, et al (2002)(8) and Zaloshnja, Miller et al. (2002),(2) travel delay was computed with three refinements. First, using a newer and broader survey of five police departments, the hours-of-delay ratio was updated to 49:86:233 for the delays due to PDO, injury, and fatal crashes, respectively. Second, to extract delay per person from delay per crash, data on the average number of people killed or injured in a crash were used. Finally, it was conservatively assumed that only police-reported crashes delay traffic, based on the premise that any substantial impact on traffic would attract the attention of the police.
Monetized Quality-Adjusted Life Years (QALYs)
Monetary lossesassociated with medical care, other resources used, and lost work do not fully capture the burden of injuries. Injuries also cost victims and families by reducing their quality of life. The good health lost when someone suffers an injury or dies can be accounted for by estimating QALYs lost. A QALY is a health outcome measure that assigns a value of one to a year of perfect health and zero to death (Gold et al., 1996).(35) QALY loss is determined by the duration and severity of the health problem. To compute it, this analysis followed Miller (1993)(31) and used diagnosis and age-group specific estimates from Miller et al. (1995)(7) of the fraction of perfect health lost during each year that a victim is recovering from a health problem or living with a residual disability. Such an impairment fraction was estimated by body part, AIS-85, and fracture/dislocation. The resulting estimates in AIS-85 were applied to NHDS and NHIS cases and the respective AIS90 estimates were computed from the diagnosis specific AIS90 ratings. The monetary value of a QALY ($91,752) was derived by dividing the value of statistical life by the number of years in the person's life span. The value of statistical life used in this study came from a systematic review found in Miller (1990)(36) and lies midway the values in two recent meta-analyses (Miller, 2000; Mrozek and Taylor, 2002).(37,38) As with the other components of cost, QALY losses in future years are discounted to present value at a 3 percent discount rate (Gold et al., 1996; Cropper et al, 1991; Viscusi and Moore, 1989).(35,39,40)
Crash Cost Variance Estimation
In addition to estimates of average human capital and comprehensive crash cost for the different crash types and police-reported severity levels, this analysis also attempted to produce an estimate of the standard deviation and the 95 percent confidence intervals for each average cost. Here, the procedure "svymean" in the software STATA® 7 is designed specifically to estimate standard errors and confidence intervals for complex survey data. It takes into account the stratification (strata) and clustering (Primary Sampling Units (PSUs)) used in the survey.
It was not possible to measure the variance of unit cost elements like medical costs, property damage, emergency service, travel delay, insurance administration, etc. Therefore standard errors represent the variance in crash costs caused by differences in the number of people involved in crashes of the same type, the severity of injuries suffered (as described by AIS, body part, and fracture status of the injury), and the age and sex of the victims (very important for the magnitude of lost productivity and QALYs).
Topics: research, safety, stop red light running program
Keywords: research, safety, crash geometries, red light running
TRT Terms: traffic accidents, accident data, cost estimating