Accounting for vehicle, crash, and occupant characteristics in traffic crash studies
- 1Department of Epidemiology, School of Public Health, University of Washington, Seattle, Washington, USA
- 2Harborview Injury Prevention and Research Center, University of Washington, Seattle, Washington, USA
- 3Department of Biostatistics, School of Public Health, University of Washington, Seattle, Washington, USA
- Correspondence to Dr Peter Cummings, 250 Grandview Drive, Bishop, CA 93514, USA;
Contributors Both authors drafted and revised this commentary and approved the final version.
- Accepted 20 August 2010
- Published Online First 5 October 2010
In 2006, Elliott et al1 reported that children age 2–6 years in a crash had a lower risk of death when using a child restraint (car seat or seat belt with booster seat), compared with a seat belt only: risk ratio (RR) 0.79, and 0.72 if restraint misuse was excluded. In 2009, Rice et al2 3 reported that child restraints, compared with seat belts, did little to reduce mortality for children age 2–8 years: RR 0.52 for a child younger than 1 year, 0.51 for age 1, 0.93 for age 2, 1.05 for age 3, 0.89 for 4–5 years, and 1.15 for 6–8 years.
In a letter, Elliott et al4 argued that because Rice et al2 included adults in their study, their estimates might be biased. In response, Rice and Anderson5 estimated that for children age 3 years and younger the RR for death in a safety seat, compared with a seat belt, was 0.85 if the analysis included adults and 0.94 if adults were excluded. In their articles in this issue of Injury Prevention, Elliott (see page 367),372),7 expand on this question: will a study of child restraints be biased if adults are included?
Some methods to reduce confounding bias
To estimate whether restraint use influences the risk of death, we can compare the proportions of restrained and unrestrained children who died in crashes. However, this may be a biased estimate of restraint effects if those restrained and unrestrained differ in regard to other factors related to death. Factors that might produce this bias include vehicle features (model, weight), crash attributes (force due to deceleration, distance to a trauma center), and occupant characteristics (age, seat position). This type of bias is called confounding. Methods to reduce confounding include …