Why Adjusting for Confounders Changes Cannabis Crash-Risk Estimates

Cannabis crash-risk estimates can shrink after researchers account for other differences between drivers. That change does not automatically invalidate the research or prove that cannabis has no effect. It shows why the comparison group and adjustment strategy are central to the estimate.

Crude and adjusted estimates make different comparisons

Hartman and Huestis reviewed epidemiologic studies alongside controlled cannabis experiments. In the observational studies, characteristics such as age, risk-taking and alcohol use could differ between cannabis-exposed drivers and comparison drivers. A crude association includes those differences. An adjusted estimate attempts to account for specified measured factors, although it cannot guarantee that every relevant influence has been removed.

The review reported both retained and lost significance

The reviewed studies did not all respond to adjustment in the same way. Some associations remained statistically significant, while others became nonsignificant. Reporting only the largest unadjusted number would overstate what those analyses established. Reporting only a nonsignificant adjusted finding would also misrepresent a literature that included positive findings. The relevant result belongs to its particular exposure definition, outcome and model.

Exposure classification adds another source of difference

The review noted that urine-based cannabis measurements had not shown the same crash-risk association as some other exposure approaches. Urinary metabolites can reflect prior use beyond the acute-effect period. Delayed blood collection and use of inactive metabolites also complicated interpretation. Thus, disagreement can arise from who is classified as exposed, not solely from the biological effect of THC.

Interpreting the evidence

A useful reading of a crash study names the outcome, reference group, specimen or self-report definition, and factors adjusted for. A population association is not an individual diagnosis of impairment, and an adjusted null result is not proof of universal driving safety.

Related reading: interpreting functional driving evidence.

Research reference

Rebecca L. Hartman; Marilyn A. Huestis (2012). Cannabis Effects on Driving Skills. Clinical chemistry. doi:10.1373/clinchem.2012.194381. Evidence basis: review/commentary abstract and selected explanatory passages.