Abstract
BACKGROUND:
Culpability studies, a common study design in the cannabis crash risk literature, typically report odds-ratios (OR) indicating the raised risks of a culpable accident. This parameter is of unclear policy relevance, and is frequently misinterpreted as an estimate of the increased crash risk, a practice that introduces a substantial “interpretational bias”.
METHODS:
A Bayesian statistical model for culpability study counts is developed to provide inference for both culpable and total crash risks, with a hierarchical effect specification to allow for meta-analysis across studies with potentially heterogeneous risk parameter values. The model is assessed in a bootstrap study and applied to data from 13 published culpability studies.
RESULTS:
The model outperforms the culpability OR in bootstrap analyses. Used on actual study data, the average increase in crash risk is estimated at 1.28 (1.16-1.40). The pooled increased risk of a culpable crash is estimated as 1.42 (95% credibility interval 1.11-1.75), which is similar to pooled estimates using traditional ORs (1.46, 95% CI: 1.24-1.72). The attributable risk fraction of cannabis impaired driving is estimated to lie below 2% for all but two of the included studies.
CONCLUSIONS:
Culpability ORs exaggerate risk increases and parameter uncertainty when misinterpreted as total crash ORs. The increased crash risk associated with THC-positive drivers in culpability studies is low.
Copyright © 2018 The Author. Published by Elsevier Ltd.. All rights reserved.
KEYWORDS:
Bayesian inference; Cannabis; Crash risks; Culpability study; Meta-analysis
- PMID: 30468948
- DOI: 10.1016/j.aap.2018.11.011
