Marijuana use and driver risks: the role of epidemiology and experimentation

Abstract

The paper by Blows and colleagues [1] on the effects of marijuana use on car crash injury and death risks provides a good illustration of the value of the case–control design in risk factor research. Although there is often advocacy for prospective or longitudinal studies of risk, the case–control design has a number of clear advantages in studying risk factors involved in rare but severe outcomes. These advantages include the ability to study sufficient numbers of cases of the outcome and the linking of assessments to a specific incident and set of circumstances, and are clearly evident in the research reported by Blows et al. Their findings add to a growing literature that is beginning to explore the extent to which the use of cannabis has adverse consequences on driver behaviours. Their results show that while the habitual use of cannabis is associated with clear increases in the risk of death or injury from car crashes (OR = 9.5; 95% CI = 2.8–32.3), the association with acute exposure was non-significant (OR = 0.8; 95% CI = 0.2–3.3). Of course, these findings are subject to the usual methodological caveats relating to the use of self-report and possible failure to control confounding. With regard to the analysis of acute effects, there is also a possibility that the association was ‘over-controlled’ as a result of the authors’ controlling factors such as sleepiness and seat belt usage that may have been a consequence of cannabis use. If these factors are not included in the regression adjustment, the results suggest an odds ratio of 3.9 (95% CI = 1.2–12.9) for acute exposure. The principal conclusions that the authors draw from their analysis is that the relationship between cannabis use car crashes is likely to be complex and may be confounded by such factors such as risky driving. They also conclude that targeting high-risk marijuana use groups may be more cost-effective than general population interventions such as random roadside testing. Although these conclusions follow from the authors’ results they beg some important issues. Specifically, there are three key questions to be addressed about marijuana and driving. First, to what extent does the use of marijuana make a causal contribution to motor vehicle accidents and injury risk? Secondly, to what extent can such risks be mitigated by such strategies as random roadside testing? Thirdly, to what extent are these strategies just and cost-effective? The answer to the first question still remains unclear, to the extent that while laboratory studies suggest that the use of marijuana may impair driver skills [2, 3], findings from epidemiological studies on the effects of marijuana on driving under naturalistic conditions have been inconsistent [4-6]. To some extent, Blows et al. pre-empt answers to the latter two questions by suggesting that because of the low rate of cannabis use by drivers (less than 1%) their research suggests limited efficacy for such strategies as random roadside testing. The potential difficulty with such a conclusion is that it may inhibit the development of an effective method of addressing issues relating to marijuana and driving. A better way of resolving such issues is clearly through planned experimentation, which should make direct assessment of the costs and benefits of roadside testing possible. That strategy is currently being investigated by the Victorian State Government in Australia [7]. To date, the results of this experiment have not been promising, with two of the first three individuals screened positive turning out to be false positive on the basis of subsequent laboratory testing [7]. None the less, this type of social experiment will provide an interesting opportunity to look directly at the costs and benefits of random roadside testing for cannabis and other drugs. While epidemiological research can often identify causal linkages, experimental methods offer the best approach for testing causality. In the case of marijuana and motor vehicle accidents, such experiments are possible and should be conducted.