Data were collected as part of an ongoing prospective study from 1 December 2000 to 31 December 2005 via insurance claim records and a validated telephone survey.5
Inclusion criteria included State Farm‐insured passenger vehicles, model year 1990 and newer (to represent the modern passenger vehicle fleet), with at least one child passenger aged
15 years. The crashes examined occurred in 16 US states and the District of Columbia, representing 3 large regions (East: New York, New Jersey (until November 2001), Pennsylvania, Delaware, Maryland, Virginia, West Virginia, North Carolina and the District of Columbia; Midwest: Ohio, Michigan, Indiana and Illinois; and West: California, Nevada, Arizona and Texas (from June 2003)).
The main outcome variable of interest was injury risk. Injuries were defined as those with Abbreviated Injury Scale scores of
2, including concussions and more serious brain injuries, facial bone fractures, spinal cord injuries, internal organ injuries and extremity fractures.
A stratified cluster sample was designed to select vehicles (the unit of sampling) for conducting telephone surveys with the driver. Probability sampling was based on two criteria: whether the vehicle was towed from the scene or not, 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 children received no medical treatment was also included to maintain the representativeness of the sample.
Figure 1 details the derivation of the study sample from the initial eligible population. Claim representatives correctly identified 98% of eligible vehicles. Of the 240, 379 cases selected: full crash data were obtained with consent for 64% of cases; in 9% of cases policyholders could not be contacted or only partial data were received; and there was a 27% refusal rate. Of the policyholders who consented, 16% were sampled for an interview, 76% of whom were successfully contacted and screened for the full survey. A full survey was obtained for 10
704 crashes. A comparison of the sample with known population values from State Farm claims showed minimal differences for the following factors: geographic region of the insured vehicle, vehicle type, non‐drivability after the crash and mean age of the child occupants.6
When compared with the 2000 United State Census, study participants had a similar distribution of race and ethnicity and family income, and a slightly higher level of education than the driver.7
Figure 1Derivation of study sample from initially eligible population (12 December 2000–31 December 2005). *At the onset of the project, consent was obtained from all eligible subjects before sampling. Beginning in June 2003, (more ...)
To assess differences in the distribution of passenger and driver characteristics by classification of driver, Pearson's χ2 tests were used. The adjusted relative risks of injury for children in crashes comparing the three driver classifications were computed, producing point estimates of the risk with associated 95% CIs. In examining injury risks, we also controlled for variables that have previously been shown to predict injury to child passengers. These included age (0–8, 9–12 and 13–15 years), seating row (front vs rear), restraint status (yes or no), gender of driver, and vehicle type (passenger car, cargo van, pickup truck, sport utility vehicle and minivan).
As sampling was based on the likelihood 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 the stratification of subjects by medical treatment, towaway status of the vehicle, clustering of subjects by vehicle and the disproportional probability of selection, Taylor Series linearization estimates of the logistic regression parameter variance were calculated using SAS‐callable SUDAAN V.9.0 (Research Triangle Institute, Research Triangle Park, North Carolina, USA, 2006). Because injury is a relatively rare event, the odds ratio (OR) can be interpreted as a good estimate of relative risk.
All protocols were approved by the institutional review boards of The Children's Hospital of Philadelphia and the University of Pennsylvania.