Article Text

Download PDFPDF
Matched cohort analysis in traffic injury epidemiology: including adults when estimating exposure risks for children
  1. Michael R Elliott1,2
  1. 1Department of Biostatistics, School of Public Health, University of Michigan, Ann Arbor, Michigan, USA
  2. 2Program in Survey Methodology, Institute for Social Research, University of Michigan, Ann Arbor, Michigan, USA
  1. Correspondence to Professor Michael R Elliott, Department of Biostatistics, University of Michigan School of Public Health, M4041, SPH II 1420 Washington Heights, Ann Arbor, MI 48109, USA; mrelliot{at}


Background Matched cohort analyses and their extension to conditional logistic and Poisson regression are powerful tools for assessing risk and protective factors in automobile crashes. However, these analyses rely on assumptions about latent (unobserved) ‘crash severity’ risk measures for subjects in the vehicle being common for all such subjects if confounding with crash severity is present. The assumptions may be questionable if adults are being matched to children.

Methods Simulations were conducted to evaluate conditional Poisson regression in settings where different types of subjects may have different underlying baseline summary risk measures in a given crash—for example, adults and children. Situations were considered where baseline summary risk measures and protective factors are confounded with each other and where the adult and child baseline risk measures are either highly correlated or weakly correlated.

Results When the risk or protective factor is confounded with crash severity, baseline summary risk measures due to crash severity must be perfectly correlated for all subjects in the crash for the factor to be consistently estimated. Relative bias estimates ranged from 5% to 50% depending on the degree of correlation of the risk measure and the number of matched subjects in a crash.

Conclusions Matched cohort analysis and its regression extensions are useful tools in the injury epidemiology toolkit, but require assumptions about high correlation of baseline summary risk measures among adults and children to accurately account for confounding between risk or protective factors of interest by crash severity. Conservative coverage intervals can preserve correct nominal coverage as long as the adult–child risk factors are highly correlated.

  • Conditional logistic regression
  • conditional Poisson regression
  • matched pairs analysis
  • double pair comparison
  • biostatistics
  • child
  • methods
  • MVTC
  • restraint

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


  • See Commentary, p 363

  • Funding State Farm Mutual Automobile Insurance Company.

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Linked Articles