Skip to main content
Annual Proceedings / Association for the Advancement of Automotive Medicine logoLink to Annual Proceedings / Association for the Advancement of Automotive Medicine
. 2003;47:495–506.

Driver Mortality in Paired Side Impact Collisions Due to Incompatible Vehicle Types

CS Crandall 1
PMCID: PMC3217542  PMID: 12941243

Abstract

Using a matched case control design, this study measured the mortality associated with paired passenger car-sport utility vehicle side impact (‘T-bone’) collisions using FARS data. Survival versus fatal outcome within the matched crash pairs was measured with matched pair odds ratios. Conditional logistic regression adjusted for multiple effects. Overall, passenger car drivers experienced greater mortality than did SUV drivers, regardless if they were in the struck or striking vehicle (odds ratio: 10.0; 95% confidence interval: 7.9, 12.5). Differential mortality persisted after adjustment for confounders. Efforts should be sought to improve passenger car side impact crashworthiness and to reduce SUV aggressivity.


Light truck vehicles (LTV) include sport utility vehicles (SUV), minivans and pickups and have gained enormous popularity throughout the United States. LTVs now account for over one-third of the US fleet [Runge, 2003]. With their increased popularity have come numerous concerns, including fuel consumption and safety, particularly with SUVs. Safety concerns include both their greater rollover risk compared to passenger cars and the risks that SUVs pose to drivers and passengers of smaller vehicles. This study was undertaken to explore the differential mortality experiences of drivers in passenger car-sport utility vehicle crashes. Due to incompatibility in vehicle mass, geometry, and stiffness, side impact collisions between passenger cars and sport utility vehicles are thought to produce a significant threat to passenger car occupants [Insurance Institute for Highway Safety, 1999; Summers, Prasad, Hollowell, 2001].

This study was designed to isolate the effects that vehicle incompatibility has on mortality in paired side impact passenger car-SUV crashes.

METHODS

A matched case control study design was used to address the research question. This design has been previously used to evaluate the efficacy of air bags and seat belts to reduce mortality in paired passenger car crashes [Crandall, Olson, Sklar, 2001]. Paired side impact collisions between one passenger car and one sport utility vehicle created crash pairs. Driver mortality differences attributed to vehicle type were estimated from these paired two vehicle collisions.

DATA SOURCES

Data for the analysis came from the US Department of Transportation, National Highway Traffic Safety Administration (NHTSA) Fatality Analysis Reporting System (FARS) data for 1999–2001.

SUBJECTS

Mortality estimates derive from the 1,289 side impact collisions between one passenger car and one SUV and involved 2,578 drivers during the three year study period. Case and controls came from the population of drivers in paired collisions. All left front seat positioned drivers involved in side impact collisions of only two vehicles were selected. Crash pairs were restricted to: (1) vehicles with FARS variable BODY_TYP values 1–11 (passenger cars) or 11–19 (sport utility vehicles), (2) the most harmful event (M_HARM) was a collision with a vehicle in transport, and (3) both the initial and principal impact points (IMPACT1 and IMPACT2) were at 3 or 9 o’clock. Injury severity (INJ_SEV) defined outcome and thus case and control status. If the driver died in the collision, they became a case. If the driver lived, they became a control.

Exclusions

Drivers less than 16 years old were not included in the analysis.

OBSERVATIONS

Survival versus fatal outcome within the matched crash pairs was measured with matched pair odds ratios and 95% confidence intervals. Matched pair odds ratios estimate the odds of fatal outcome in the driver who had the exposure of interest (e.g., driver in an SUV) compared to the odds of survival in the drivers who did not have the exposure of interest (e.g., driver in a passenger car). Standard errors were calculated using Wolff’s method [Schlesselman, 1982].

Figure 1 presents a typical analytic framework for calculating matched pair odds ratios. Case and control status were determined from the outcome, cases being defined by circumstances where the driver of that vehicle died and controls being defined by circumstances where the driver of the paired vehicle survived. Each of the symbols in the cells of this 2×2 table (a, b, c, and d) represent two vehicles (each of the crash pairs).

Figure 1.

Figure 1

Schematic representation of crash pairs used in an analysis of passenger cars versus SUVs. Each of the cell values represents two vehicles (the crash pair).

Figure 2 presents all of the possible match pair combinations for an analysis of near side impact collisions between passenger cars and SUVs involving one driver fatality. Smiling faces represent the drivers. An ‘X’ over the driver denotes a fatality. Note that while both near (driver) side and far (passenger) side impacts were included in the analyses, only the near side impacts are noted in the figure for clarity. The large rectangle denotes pairs used in the analysis. Only the ‘b’ and ‘c’ type pairs are used to calculate the point estimate and the standard error of the matched pair odds ratio. Neither the estimates of the odds ratio nor its standard error depend upon pairs where both of the drivers have the exposure (cell a) nor where both drivers do not have the exposure (cell d).

Figure 2.

Figure 2

Representation of all possible match pair combinations for passenger cars and SUVs in near side impact collisions with one driver fatality.

Additional variables that were considered as potential confounders included vehicle and driver characteristics. Vehicle characteristics included vehicle rollover, vehicle mass, and the difference in model year age between the crash pair. Vehicle mass was derived from the VIN weight (VIN_WGT) and was converted to kilograms. Vehicle weights were divided into 6 categories that corresponded to NHTSA’s classification [Tessner, 2002]. The difference in model year age was calculated within each crash pair. A value of zero was given to the newer vehicle and the number of years of age difference was assigned to the older vehicle. Vehicles of the same year were both assigned values of zero.

Driver characteristics included gender, age, and ejection from the vehicle. Age categories were used in both the crude and adjusted analyses. In the crude analyses, age was grouped into 10 year intervals. Based upon the results of the crude analysis, driver age was divided into two discrete categories for the adjusted analyses (16 to 64 years and 65 years and over). Drivers who were partially ejected were considered ejected. Air bag presence and deployment were also included (AIRBAG). Any air bag deployment was considered, whether side or frontal air bags, although the vast majority of air bag deployments (over 95%) were from the frontal air bag. Restraint use data came from FARS variable REST_USE. Because of the similar effect measures in the crude analyses, any seat belt arrangement, except shoulder belt only and improper use, were collapsed into one category (any belt use vs. no/improper use) for the adjusted analyses.

Mean differences for continuous variables were tested against the t-distribution. Differences between two proportions were tested against the chi square distribution using Yates continuity correction. [Berry and Armitage, 1984]

To adjust for confounding and multiple effects, conditional logistic regression was used to calculate the matched pair odds ratios and 95% confidence intervals. Potential confounding variables were entered in a stepwise fashion, using both statistical significance and the effect that each variable had on the main effect of interest (mortality difference between cars and SUVs) as guidance for inclusion or exclusion in the final model.

For all analyses, a two tailed Type I error rate of 5% was used to determine statistical significance.

SAS/STAT software [SAS Institute, 2001] was used throughout. The SAS/STAT procedure PROC PHREG was used to conduct the conditional logistic regression.

The institutional review board at the University of New Mexico Health Sciences Center certified this study as exempt.

RESULTS

There were 3,778 paired side impact collisions involving passenger cars and/or SUVs. Among these crashes, 2,452 collisions (64.9%) involved two passenger cars, 1,289 collisions (34.1%) involved a passenger car and an SUV, and 37 collisions (1.0%) involved two SUVs. The remaining analyses concentrate only on the 1,289 collisions between passenger cars and SUVs.

Table 1 lists the characteristics of the crash, vehicles, drivers, and restraint use for the passenger car-SUV side impact collisions. Vehicle rollover was more common among SUVs than for cars (8.5% vs. 2.8%, p<0.0001). Concomitant with their more recent popularity, SUVs tended to be newer than passenger cars.

Table 1.

Crash, vehicle, driver, and restraint use characteristics from passenger car-SUV side impact collisions, United States, 1999–2001, FARS.

Total
Passenger Ca
SUV
N (%) N (%) N (%)
Total 2,578 1,289 1,289
Crash and vehicle characteristics
 Vehicle rollover 146 ( 5.7) 36 ( 2.8) 110 (8.5)
 Vehicle weight (kg)
  Class 1 and Class 2 (<1114 kg) 290 (11.2) 285 (22.1) 5 (0.4)
  Class 3 (<1341 kg) 441 (17.1) 418 (32.4) 23 (1.8)
  Class 4 (<1568 kg) 453 (17.6) 372 (28.9) 81 (6.3)
  Class 5 (<1795 kg) 294 (11.4) 101 ( 7.8) 193 (15.0)
  Class 6 (>1795 kg) 440 (17.1) 45 ( 3.5) 395 (30.6)
  Unknown 660 (25.6) 68 ( 5.3) 592 (45.9)
 Difference in vehicle age
  0 years 1,360 (52.8) 499 (38.7) 861 (66.8)
  1 to 3 years 414 (16.1) 237 (18.4) 177 (13.7)
  4 to 6 years 335 (13.0) 206 (16.0) 129 (10.0)
  7 to 9 years 221 ( 8.6) 162 (12.6) 59 (4.6)
  10 or more years 248 ( 9.6) 185 (14.4) 63 (4.9)
Driver characteristics
 Injury Severity
  Fatal 909 (35.3) 818 (63.5) 91 (7.1)
  Incapacitating 356 (13.8) 177 (13.7) 179 (13.9)
  Non-incapacitating 512 (19.9) 144 (11.2) 368 (28.5)
  Other 801 (31.1) 150 (11.6) 651 (50.5)
 Male gender 1,477 (57.3) 694 (53.8) 783 (60.7)
 Age group (in years)
  Less than 25 617 (23.9) 304 (23.6) 313 (24.3)
  25 - 34 494 (19.2) 156 (12.1) 338 (26.2)
  35 - 44 454 (17.6) 143 (11.1) 311 (24.1)
  45 - 54 322 (12.5) 134 (10.4) 188 (14.6)
  55 - 64 218 ( 8.5) 127 ( 9.9) 91 (7.1)
  65 and older 473 (18.3) 425 (33.0) 48 (3.7)
 Driver ejection 106 ( 4.1) 50 ( 3.9) 56 (4.3)
 Driver only occupant 1,289 (50.0) 622 (48.3) 667 (51.7)
Restraint system use
 Air bag present 1,294 (50.2) 602 (46.7) 692 (53.7)
 Air bag deployed 701 (27.2) 289 (22.4) 412 (32.0)
 Any seat belt restraint 1,846 (71.6) 857 (66.5) 989 (76.7)
  Lap belt only 28 (1.1) 16 ( 1.2) 12 (0.9)
  Shoulder belt only 12 ( 0.5) 10 ( 0.8) 2 (0.2)
  Lap and shoulder belt 1,712 (66.4) 792 (61.4) 920 (71.4)
  Belt used, type unknown 94 ( 3.6) 39 ( 3.0) 55 (4.3)
 No or improper use 538 (20.9) 339 (26.3) 199 (15.4)
 Unknown or missing 194 (7.5) 93 (7.2) 101 (7.8)

Compared to SUV drivers, passenger car drivers were more often killed compared to SUV drivers (63.5% vs. 7.1%, p<0.0001). SUV drivers were more likely to be male (p=0.0005). Passenger car drivers were significantly older than SUV drivers (car: (mean±SD) 49.1±24.0 years, median 47 years; SUV: 36.2±14.4 years, median 34 years, p<0.0001).

As SUVs tended to be comparatively newer than their paired passenger car, more SUVs were equipped with air bags than were the passenger cars (p=0.0005). SUV drivers were also more likely to have been wearing their seat belts (p<0.0001).

Table 2 presents the crude matched pair odds ratios from the 1,289 passenger car-SUV side impact collisions. There were 889 crash pairs where either one (but not both) of the drivers died. Overall, car drivers experienced substantially greater mortality than did SUV drivers, regardless if they were in the struck or striking vehicle (odds ratio (OR): 10.0; 95% confidence interval (CI): 7.9, 12.5). Drivers who were struck had substantially greater mortality than the driver in the striking vehicle (OR: 25.3), especially when struck near side (OR: 37.6) compared to far side (OR: 11.4).

Table 2.

Crude matched pair odds ratios and 95% confidence intervals for driver mortality by crash, vehicle, driver and restraint use among passenger car-SUV side impact collisions, United States, 1999–2001, FARS.

No. of pairs
Case exposure + Control exposure − Case exposure − Control exposure + Odds ratio 95% confidence interval
Crash and vehicle characteristics
 Car vs. SUV 808 81 10.0 7.9 12.5
 Struck versus striking vehicle 811 32 25.3 17.8, 36.1
  Struck near side 640 17 37.6 23.3, 60.9
  Struck far side 171 15 11.4 6.7, 19.3
  Car struck near side 598 6 99.7 44.6, 222.7
  Car struck far side 153 6 25.5 11.3, 57.6
 Vehicle rollover 49 35 1.4 0.9, 2.2
 Vehicle weight (kg)
  Class 4+ (>1567 kg) 1.0
  Class 3 (<1341 kg) 166 12 13.8 7.7, 24.9
  Class 2 (<1114 kg) 83 9 9.2 4.6, 18.3
  Class 1 (<888 kg) 8 0
 Wheelbase (cm)
  Long (423+) 1.0
  Medium (394-422) 187 109 1.7 1.4, 2.2
  Short (<394) 82 11 7.5 4.0, 14.0
 Difference in vehicle age
  0 years 1.0
  1 to 3 years 167 120 1.4 1.1, 1.8
  4 to 6 years 135 97 1.4 1.1, 1.8
  7 to 9 years 110 45 2.4 1.7, 3.5
  10 or more years 125 39 3.2 2.2, 4.6
Driver characteristics
 Male vs. female gender 180 237 0.76 0.63, 0.92
 Age group (in years)
  Less than 35 1.0
  35-44 58 53 1.1 0.75, 1.6
  45-54 46 30 1.5 1.0, 2.4
  55-64 62 15 4.1 2.4, 7.3
  65 and over 170 5 34.0 14.0, 82.7
 Driver ejected 75 7 10.7 4.9, 23.2
 Driver only occupant 197 175 1.1 0.9, 1.4
Restraint system use
 Air bag deployed 87 209 0.42 0.32, 0.53
 Any seat belt combination* 53 231 0.23 0.17, 0.31
  Lap belt only 4 6 0.67 0.19, 2.4
  Shoulder belt only 6 0
  Lap and shoulder belt 59 213 0.28 0.21, 0.37
  Belt used, type unknown 4 22 0.18 0.06, 0.53
 Any belt and air bag deployed 67 192 0.35 0.26, 0.46
*

Any belt combination includes lap, lap and shoulder, or used (type unknown). Shoulder belt only assigned to unbelted. Throughout the table, reference groups are denoted by an odds ratio of 1.0 for variables with more than 2 levels, otherwise the reference group is understood to be those without the listed characteristic.

Passenger car drivers struck near side by an SUV experienced substantially higher mortality compared to the SUV driver (OR: 99.7; CI: 44.6, 222.7), as did passenger car drivers struck far side by an SUV (OR: 25.5; CI: 11.3, 57.6).

Assuming that for any given model year that a newer vehicle has better crashworthiness than older vehicles in protecting their occupants, the model year age difference was calculated and expressed in years. Drivers in newer cars fared better than drivers in older cars, especially as the number of years of model age increased between the crash pairs. The greatest mortality difference was observed in vehicles 7 to 9 years (OR: 2.4; CI: 1.7, 3.5) and 10 or more years (OR: 3.2; CI: 2.2, 4.6) of model year age compared to crashes between vehicles of the same model year.

Male drivers had lower mortality compared to female drivers (OR: 0.76; CI: 0.63, 0.92). The mortality risk for older drivers increased exponentially with increasing age compared to the youngest drivers. Compared to drivers less than 35 years of age, drivers 65 years of age and older had the highest risk (OR: 34.0, CI: 14.0, 82.7). Drivers ejected from their vehicles died more frequently than drivers who were not ejected (OR: 10.7; CI: 4.9, 23.2).

Proper restraint use conferred significant reductions in mortality. Although very small in number, shoulder belt use only conferred no risk reduction and was therefore assigned to an unbelted or improper use for higher level analyses. Both air bag deployment (OR: 0.42; CI: 0.32, 0.53) and seat belt use in nearly all combinations (OR: 0.23; CI: 0.17, 0.31) significantly reduced mortality. Using any belt combination and having an air bag that deployed reduced overall mortality by about 65% (OR: 0.35; CI: 0.26, 0.46).

Since several factors that were related to vehicle type (e.g., SUV drivers tended to be younger, to have a newer vehicles, to have worn their seat belts and have had a vehicle equipped with an air bag) also had a strong effect on mortality, higher level analyses were conducted to isolate the effect of vehicle incompatibility on driver mortality. A conditional logistic regression model was used to adjust for these confounding effects.

Table 3 lists the results of the final model. Model 1 includes only those variables with significant independent associations. The differential mortality experienced by passenger car drivers persisted after adjustment for being in the striking versus struck vehicle, vehicle rollover, driver ejection, and driver age. Airbag deployment and difference in vehicle age were no longer significant. Passenger car drivers were about six times as likely to die compared to SUV drivers, regardless of whether they were in the striking or struck vehicle (OR: 5.6, CI: 3.3, 9.5).

Table 3.

Conditional logistic regression model estimates for paired side impact collisions between passenger cars and sport utility vehicles, adjusted for selected crash, vehicle, and driver effects, United States, 1999–2001, FARS.

Model number
(1)
(2)
Model covariates Odds ratio 95% Confidence interval Odds ratio 95% Confidence interval
Passenger car vs. SUV 5.6 3.3 – 9.5 1.4 0.27 – 7.3
Struck vehicle vs. striking vehicle 15.1 8.3 – 27.7 57.6 9.0 – 367.8
Vehicle rollover 5.8 1.7 – 20.3 90.7 3.4 – 2,400.3
Any belt configuation* 0.30 0.13 – 0.71 0.23 0.03 – 1.59
Driver ejection 7.8 2.3 – 26.8 1.3 0.26 – 6.8
Older drivers vs. younger drivers** 16.9 5.0 – 57.8 65.1 5.2 – 815.1
Vehicle mass (per 100 kg) 0.79 0.67 – 0.93
*

Shoulder belt only and improper use assigned to the no use category.

**

Drivers 65 years of age and older compared to younger drivers.

Addition of the striking vehicle’s traveling speed did not affect the mortality estimates and was not significant.

Addition of vehicle weight to the conditional logistic model explained nearly all of the differential mortality between cars and SUVs (Table 3, Model 2).

An analysis that was restricted to only late model year vehicles (model years 1995 and later) did not substantially change the mortality risk estimates.

DISCUSSION

This study provides relatively precise estimates of the mortality risk differences between mismatched/incompatible vehicle types involved in side impact collisions using observational crash data. The popularity of SUVs and their increasing presence on US roadways pose a substantial risk to drivers in smaller cars while conferring a tremendous safety advantage to drivers in SUVs. This study estimates that a passenger car driver in a side impact collision is ten times more likely to be killed than the driver of the SUV, regardless of the vehicle’s role. Drivers in passenger cars struck near side have a nearly 100-fold increased odds of death compared to the driver of the striking SUV.

The odds ratio estimates from this study are greater than those previously reported, ranging from about 10 to 30 [Insurance Institute for Highway Safety, 1998; Summers et al., 2001; Runge, 2003; NHTSA, 1998]. Most of these studies compared the relative mortality rates for occupants in struck vehicles versus striking vehicles based upon vehicle registration data. This study used a different methodology and estimated the mortality difference within the crash pairs. This methodology produces odds ratio estimates which are generally good estimates of the relative risk, but are always more extreme than the relative risk. The results from this study would suggest that the risk to passenger car drivers is more substantial than that previously reported.

The danger that larger vehicles pose to occupants of smaller vehicles has been measured and described by the term vehicle aggressivity [Summers et al., 2001; Joksch HC, 2000]. Differences between vehicle aggressivity can be attributed to several factors, including driver behavior and vehicle incompatibility. Vehicle crash incompatibility can be expressed in terms of mass, geometry, and stiffness [NHTSA, 1998]. Due to limitations of the FARS data set, only vehicle mass differences were examined in this study. Differences in vehicle geometry and stiffness cannot readily be obtained from FARS and were not included in this analysis.

Vehicle mass explained much, but not all, of the mortality differences observed. As the US fleet demographics change and the percentage of SUVs increases among the fleet, it will become even more imperative to reduce the aggressivity of the SUV and to improve passenger car side impact crashworthiness.

This study illustrates the strengths of using a matched case control analysis to measure mortality differences. A matched pair analysis can overcome FARS data limitations and has the power to consider multiple confounding factors. An often remarked limitation of FARS is the exclusion of crash data without fatalities. This limitation makes traditional risk estimation impossible. Since the matched pair odds ratio depends upon differential outcome and exposure, data from nonfatal crashes are not necessary, and this limitation of FARS data is obviated.

Matching on the same crash pair has certain additional advantages, including elimination of confounders that are equivalent within the pair, such as roadway, lighting, and weather conditions at the time of the crash and availability of emergency medical services. Use of conditional logistic regression affords the ability to control for other confounding factors simultaneously. In this study, several confounding factors were present, including seat belt and air bag use, driver age, and vehicle rollover. Conditional logistic regression has become increasingly available and offers an efficient and powerful technique for multiple risk factor adjustment.

LIMITATIONS

Estimates are based upon collisions involving only two vehicles with relatively limited crash data. The effects of more complex crash dynamics (e.g., those involving multiple vehicles or collisions with fixed objects) are unknown. Only side impact collisions at 3 and 9 o’clock were examined. Inclusion of side impact crashes off these clock points may change mortality estimates. Head-on, offset, and rear-end collisions will likely have different mortality odds ratios. These types of crashes will also require different design solutions due to different incompatibility issues. FARS data include relatively little information on the crash mechanics, including deltaV and vehicle geometry.

This study looked only at driver mortality. This study should be replicated on crash data with more extensive information on occupant injury, such as NHTSA’s NASS GES databases.

CONCLUSIONS

Passenger car drivers in side impact collisions experience substantially higher mortality compared to SUV drivers, especially when struck near side. Most of the mortality difference can be attributed to differences in vehicle mass. Efforts to mitigate the differences in mortality between vehicle types should include incentives to narrow the weight differences between vehicle types and improving the crashworthiness of smaller vehicles, particularly by strengthening the vehicle to withstand a lateral impact.

REFERENCES

  1. Berry G, Armitage P. Statistical Methods in Medical Research. Blackwell Science, Inc; Aug, 1994. [Google Scholar]
  2. Crandall CS, Olson LM, Sklar DP. Mortality reduction with air bag and seat belt use in head-on passenger car collisions. Am J Epidemiol. 2001 Feb 1;153(3):219–24. doi: 10.1093/aje/153.3.219. [DOI] [PubMed] [Google Scholar]
  3. Insurance Institute for Highway Safety. Special issue: crash compatibility. Status Report. 1998 February 14;33(1) [Google Scholar]
  4. Insurance Institute for Highway Safety. Special issue: Vehicle compatibility in crashes. Status Report. 1999 October 30;34(9) [Google Scholar]
  5. Joksch HC. Vehicle Design versus Aggressivity. University of Michigan Transportation Research Institute. DOT Report HS 809 184; April 2000. [Google Scholar]
  6. NHTSA. Overview of Vehicle Compatibility/LTV Issues. Technical Report. [Accessed March 14, 2003]. February 1998. http://www.nhtsa.dot.gov/cars/problems/studies/LTV.
  7. Runge JW. Statement to the Committee on Commerce on SUV Safety. Congressional testimony. [Text accessed on 3/5/03]. http://www.nhtsa.dot.gov/nhtsa/announce/testimony/SUVtestimony02-26-03.htm, February 26, 2003.
  8. SAS Institute. SAS/STAT (Statistical Analysis Software, statisitcs module) software package, version 8.2. Cary, NC: 2001. [Google Scholar]
  9. Schlesselman JJ, Stolley PD. Case Control Studies: Design, Conduct and Analysis. New York: Oxford University Press; 1982. [Google Scholar]
  10. Summers SM, Prasad A, Hollowell WT. NHTSA’s Research Program for Vehicle Aggressivity and Fleet Compatibility. Proceedings of the Seventeenth International Conference on Enhanced Safety of Vehicles, Paper No. 249; Amsterdam, Netherlands. June 2001. [Google Scholar]
  11. Tessner JM. FARS Analytic Reference Guide 1975 to 2002. US Department of Transportation, National Highway Traffic Safety Administration; Washington, DC. August 2002. [Google Scholar]

Articles from Annual Proceedings / Association for the Advancement of Automotive Medicine are provided here courtesy of Association for the Advancement of Automotive Medicine

RESOURCES