Sport utility vehicles (SUVs) have been the fastest growing segment of the motor vehicle market in the USA through 2007. Between model years 1998 and 2007, sales of SUVs increased 67%, and they accounted for 29% of new passenger vehicle sales during the 2007 model year.1 This increase in popularity has been due, in part, to their increasing use as family vehicles, because they contain more storage space and seating positions than passenger cars. Much of the previous research on SUV safety has focused on their increased tendency to rollover in a crash, resulting in relatively high fatality rates compared with other vehicles.2^–5 Recent research focusing on the safety of children in SUVs showed that the risk of non-fatal injury to child passengers in SUVs is similar to that in passenger cars in crashes.6 However, children in crashes in SUVs experience a nearly 50% lower fatal injury risk than children in passenger cars in crashes in which someone dies.7

Minivans are an alternative choice as a family vehicle, with similar storage and seating capacity to that of many midsize and large SUVs. As sales of SUVs have increased, those of minivans decreased by 36% over the same period,1 suggesting that many families choose between these two vehicle types. The objective of this study was to estimate the relative risk of fatal and non-fatal crash injury for children in minivans compared with children in midsize and large SUVs.

METHODS

Crashes involving non-fatal injuries

Study population and data collection

The source of data on non-fatal injury was the Partners for Child Passenger Safety (PCPS) study. PCPS is a child-specific crash surveillance system: insurance claims from State Farm function as the source of subjects, with telephone survey and on-site crash investigations serving as the primary sources of data. A description of the study methods has been published previously.8

Data for this study were collected from 1 March 2000 to 31 December 2006. Vehicles qualifying for inclusion were State Farm-insured, model year 1990 or newer, and involved in a crash with at least one child occupant ⩽15 years of age. Qualifying crashes were limited to those that occurred in 16 states and the District of Columbia, representing three large regions of the USA (East: New York, New Jersey (through November 2001), Pennsylvania, Delaware, Maryland, Virginia, West Virginia, North Carolina, District of Columbia; Midwest: Ohio, Michigan, Indiana, Illinois; West: California, Nevada, Arizona, Texas (starting June 2003)).

Parent-reported injuries were defined as those with estimated Abbreviated Injury Scale scores of 2 or higher, which includes concussions and more serious brain injuries, facial bone fractures, spinal cord injuries, internal organ injuries, and extremity fractures.9 Child passengers who suffered a fatal injury (n = 13) were excluded from the analysis.

A stratified cluster sample was designed to select vehicles (the unit of sampling) for the conduct of a telephone survey with the driver. Probability sampling was based on two criteria: whether the vehicle was towed from the scene, and the level of medical treatment received by the child passenger(s). If a vehicle was sampled, the cluster of all child passengers in that vehicle was included in the survey. A 2.5% sample of crashes in which no medical treatment was received by children was also included to maintain the representativeness of the sample with respect to the population of all children in crashes in State Farm-insured vehicles. All protocols were approved by the institutional review boards of The Children’s Hospital of Philadelphia and the University of Pennsylvania.

Variable definitions

Vehicle type and size classification was obtained from the vehicle identification number (VIN) using VINDICATOR.10 Vehicles were classified as either minivans or SUVs (limited to midsize and large only). Small SUVs (typically 1701 kg and less) were excluded from this analysis, as were very large SUVs (those typically weighing at least 2609 kg).11 We grouped midsize and large SUVs together after performing a formal test of heterogeneity and not finding a significant difference in adjusted injury risk between the two classes (p = 0.93). Vehicles were restricted to model year 1998 through 2007 to provide a comparable sample in terms of safety technology availability and crashworthiness. Data were missing in ∼0.25% of PCPS cases; for these cases, a single imputation using Imputation and Variance Estimation Software (IVEware) was performed.12

Crashes involving fatalities

Study population and data collection

The sources of data for crashes involving fatalities were two automotive crash surveillance systems operated by the National Highway Traffic Safety Administration: the Fatality Analysis Reporting System (FARS) and the National Automotive Sampling System-Crashworthiness Data System (NASS-CDS). FARS is a census of motor vehicle crashes in the 50 States and the District of Columbia that resulted in the death of at least one person (motorist or non-motorist). NASS-CDS is a representative sample of passenger vehicle crashes selected annually from all police-reported crashes that resulted in at least one vehicle having to be towed from the scene for damage. These two data sources were combined using similar methodology to that described previously13 to create a population-based cohort of children under the age of 16 years involved in police-reported tow-away crashes, with all fatally injured children in the cohort enumerated.

We restricted our analyses to children riding in either SUVs (midsize and large only) or minivans that needed to be towed from the accident scene. Within FARS, we identified 1159 child passenger fatalities in an SUV or minivan between 2000 and 2006. Within NASS-CDS, we identified 1273 child passengers in a minivan or SUV sampled during the same years that were not fatally injured. Because of the complex sample design of NASS-CDS, these 1273 children represented 443 578 children meeting our inclusion criteria.

Variable definitions

As in the PCPS dataset, vehicle type and size classification of the vehicle was obtained from the VIN using VINDICATOR.10 Midsize and large SUVs were grouped together after a formal test of heterogeneity had been performed and no significant difference in adjusted injury risk had been found between the two classes (p = 0.81). Restraint use (child and/or driver) data were missing in ∼0.5% of the NASS-CDS and in more than 9% of the FARS cases. For these cases, multiple imputation using IVEware was performed.12

Data analysis

Because sampling was based on the likelihood or severity of an injury, subjects least likely to be injured were under-represented in the study sample in a manner potentially associated with the predictors of interest. To account for this potential bias, case weights equal to the inverse of the probability of selection and adjusted to known crash totals were used to account for the over-sampling of severe crashes in both datasets. (Case weights in FARS were set to 1, consistent with the fact that the FARS is a census of all fatalities.) To account for the stratification of subjects by medical treatment, clustering of subjects by vehicle, and the disproportional probability of selection, Taylor Series linearization estimates14 of the logistic regression parameter variance were calculated using SAS-callable SUDAAN (Software for the Statistical Analysis of Correlated Data, V9.0.1). Results of logistic regression modeling are expressed as unadjusted and adjusted odds ratios (ORs) with corresponding 95% confidence intervals (CIs). Because fatal and non-fatal injury were relatively rare events in each respective source of data, the odds ratio can be interpreted as a good estimate of relative risk. Adjustments included restraint of the child and driver, seating row of the child (front versus rear), sex of the child and driver, age of the child and driver in years, model year of the vehicle, and whether the vehicle rolled over. For age of child, age of driver, and vehicle model year, we used quadratic splines with linear tails as described by Greenland15 to minimize assumptions about the functional form.

RESULTS

For the PCPS dataset, complete interview data were obtained on 5164 children representing 111 079 child passengers in the study population. A majority of these children were traveling in minivans (55.5%). Injuries occurred in 475 child occupants representing 748 children or 0.67% of the PCPS study population (table 1). As compared with children in SUVs, child passengers in minivans were less often driven by someone <25 years of age (3.4% vs 6.3%), in a model year 2002 or newer vehicle (30.7% vs 44.3%), or involved in a rollover crash (1.3% vs 3.2%).

In the combined NASS-CDS and FARS cohort, we identified tow-away crashes involving 2432 children representing 444 737 child occupants. In this cohort, just over half the child passengers were in minivans (50.2%). There were 1159 child occupant deaths, representing 0.26% of the combined NASS-CDS and FARS cohort. As seen in table 2, drivers of SUVs in crashes were younger (52.5% were <35 years of age) than the drivers of minivans in crashes (25.3% were <35 years of age).

The proportion of child passengers in this cohort that were in a vehicle that rolled over was much greater (11.0% in minivans and 21.9% in SUVs) than for those in the PCPS study population. This was expected because of the higher average crash severity in the NASS-CDS and FARS cohort, which is restricted to tow-away crashes. For SUVs only, there was a greater proportion of front row seating in the NASS-CDS and FARS cohort (23.4%) than in the PCPS study population (13.4%); this was due in part to the former cohort of children being somewhat older.

In the PCPS data (table 3), there was a 45% crude injury risk reduction for children in minivans compared with children in SUVs (unadjusted OR = 0.55). After adjustment for child and driver restraint, seating row of the child, age of the child and the driver, and model year of the vehicle, PCPS child occupants in minivans still had a 44% lower risk of injury than child occupants in SUVs (adjusted OR = 0.56, 95% CI 0.38 to 0.82). A further adjustment for vehicle rollover explained only a small fraction of the remaining difference in injury risk between minivans and SUVs (adjusted OR = 0.65, 95% CI 0.43 to 0.97).

For the combined NASS-CDS and FARS cohort (table 3), the crude relative risk of death was 42% lower (unadjusted OR = 0.58) in minivans than SUVs, similar to the crude risk of injury reduction in the PCPS data. Adjustment for the set of occupant, driver, and vehicle factors mentioned above (except rollover) accounted for a portion of the difference in fatality risk between the vehicle types (adjusted OR = 0.76, 95% CI 0.51 to 1.13). A final adjustment for vehicle rollover resulted in a non-significant association between vehicle type and risk of death (adjusted OR = 1.19, 95% CI 0.73 to 1.94).

DISCUSSION

Results of this study indicate that children seated in minivans during a crash have a significantly lower crude risk of fatal and non-fatal injury than children seated in SUVs. The differences in risk of non-fatal injury are associated with vehicle type even after adjustment for other child and driver characteristics, vehicle model year, and rollover. The increased proportion of SUV crashes involving rollover explained most of the difference in risk of fatal injury for child occupants in minivans versus those in SUVs.

A previous report by the National Highway Traffic Safety Administration showed a lower risk of fatality for child passengers in crashes of light truck vehicles including SUVs, minivans, and pickup trucks compared with passenger cars.7 In that study, however, no differentiation was made between specific vehicle types in the light truck vehicle class. The present research extends this previous work by looking at non-fatal injury (PCPS population) and fatality (NASS-CDS and FARS cohort) risk differences between minivans and SUVs specifically.

Previous research has shown that occupants in rollover crashes are more likely to be injured or killed than those in non-rollover crashes.216^–18 In addition, although there has been a reduction in rollover risk among newer model year SUVs due in part to both the increased presence of electronic stability control1920 and unibody construction (emphasizing a lower center of gravity),421 SUVs are still at higher risk of rollover than minivans of the same model year.17 Given the relatively low rollover risk in the PCPS sample (1.3% for minivans and 3.2% for SUVs), and the potentially wide range of scenarios by which children can be injured in non-fatal crashes, rollover crashes accounted for only a small proportion of all injuries (9% of minivan injuries and 19% of SUV injuries). Hence, after a variety of passenger and vehicle characteristics had been controlled for, the final adjustment for vehicle rollover had a minimal effect in explaining the difference in non-fatal injury risk between minivans and SUVs in the PCPS population (adjusted OR changed from 0.56 to 0.65). The remaining association between vehicle type and non-fatal injury risk may be explained by mediators such as other vehicle characteristics (eg, vehicle size, weight, occupant space) as well as unmeasured confounders such as differences in how the vehicles were driven. In particular, previous research hypothesized that differences in occupant space, seating, and restraint geometry would lead to reductions in risk of injury in minivans.22 The rear rows in minivans may contain captain chairs, which provide a seat and seat belt geometry that may provide better protection than traditional bench seating.23

Given the more severe nature of the crashes in the NASS-CDS and FARS cohort (all were tow-away crashes compared with less than one-third of PCPS crashes), we observed a much higher proportion of crashes of both vehicles that involved a rollover (11.0% for minivans and 21.9% for SUV) than were observed in the PCPS population, which were not restricted to tow-away crashes. The rollover crashes accounted for 37% of the minivan fatalities and 66% of the SUV fatalities, indicating that the primary causation scenario for fatal injuries in SUVs involves rollover. Consequently, adjustment for vehicle rollover in the NASS-CDS and FARS cohort explained a large portion of the association between fatality risk and vehicle type (minivan versus SUV adjusted OR climbed from 0.76 to 1.19 after adjustment for rollover).

Our analyses had limitations. We were not able to study or estimate the relative risk of being in a crash, but rather the relative risk of injury given that a crash had occurred. Specific changes in the SUV and minivan fleets over time are addressed globally in the multivariate analyses by adjusting for model year. The specific model year in which particular “cross-over” SUVs that are constructed more like a passenger car than a light truck were introduced was not explicitly captured in this approach. Individual comparisons of specific SUV models with specific minivan models may result in different findings. The PCPS study obtained nearly all of its data via telephone interview with the driver/parent of the child and is therefore subject to potential misclassification, particularly the over-reporting of restraint use by children who were, in fact, unrestrained. Sensitivity analyses determined that over 70% of the unrestrained SUV or minivan passengers would have to be misclassified in order for the difference in risk between SUV and minivan passengers to be explained by misclassification alone. This does not appear plausible. In addition, ongoing comparisons of driver reports of child restraint use and seating positions with evidence from crash investigations, performed as part of the PCPS research project, show a high degree of agreement (κ = 0.99 for seat row; κ = 0.74 for restraint use). This degree of agreement is similar to that found in previous research of NASS-CDS data comparing restraint use reported by CDS investigators with that reported by the police.23 Surveillance data of the nature presented in this study cannot detect precise injury mechanisms. Therefore, more detailed information on the nature and severity of the injuries is needed to further understand how vehicle type and crash characteristics such as rollover lead to specific types of injuries that may be prevented. Finally, this study measured non-fatal and fatal crash injury risk to occupants of the two vehicle types and did not attempt to measure the additional risk of either vehicle type to other vulnerable road users such as pedestrians, cyclists, and occupants of other vehicles involved in the crash. Accounting for these other sources of risk would influence the evaluation of the overall relative safety between these two vehicle types. This was outside the scope of this research, but should be considered in future analyses.

CONCLUSIONS

Consumers choosing between an SUV and a minivan as a family vehicle should consider the safety of child occupants in addition to other parameters such as fuel economy, storage, and seating configurations. Our results suggest that there may be important differences in the safety of children in these two vehicle types, leading to a decreased risk of non-fatal injury to children seated in minivans compared with SUVs. The increased tendency for SUVs to rollover contributes to a higher overall risk of death of children in these vehicles during more serious crashes. Continued manufacturer improvements that reduce the risk of rollover in SUV crashes, such as electronic stability control and other crash prevention technology, should result in fewer child injuries and deaths.

• Download figure
• Open in new tab
• Download powerpoint