There is increasing evidence that cannabis users who drive while intoxicated put themselves and others at increased risk of motor vehicle crashes. Cannabis produces dose-related cognitive and behavioural impairments in laboratory and simulator studies [1–3]; cannabis users in surveys are more likely to report being involved in accidents than drivers who do not use the drug (e.g. [4,5]), and cannabis is the illicit drug detected most often in drivers who have been killed in motor vehicle crashes (see [6] for a review).

Older studies that measured inactive metabolites of cannabis did not show whether drivers were impaired at the time of the accident [6,7]. More recent studies have provided better evidence that cannabis-affected drivers are at a higher risk of being involved in crashes. Gerberich et al. [8] found that current cannabis users had a higher rate of hospitalization for accidental injury in a cohort of 64 657 patients from a health maintenance organization (HMO) (RR = 1.96). Mura et al. [9] found a similar relationship in a case–control study of Δ-9-tetrahydrocannabinol (THC) in the serum of 900 people hospitalized for injuries in motor vehicle accidents and 900 controls of the same age and sex admitted to the same French hospitals for reasons other than trauma (OR = 2.5). Drummer et al. [10] found an increased culpability in cannabis users (OR = 2.5) in 1420 Australian drivers killed in accidents and a dose–response relationship between blood THC level and culpability.

Cannabis use appears to increase the risk of motor vehicle crashes by two to three times [1], a much lower risk than alcohol (from six to 15 times). Given the lower risk and lower prevalence of cannabis than alcohol use, the proportion of accidents attributable to cannabis is much lower than that for alcohol (an estimated 2.5% of fatal accidents in France compared to 29% for alcohol [11]).

Is there now sufficient evidence to discourage cannabis users from driving by conducting roadside drug testing? Any such policy requires specification of a level of THC in blood that provides per se evidence of impaired driving. Grotenhermen and colleagues, in this issue [12], have derived a provisional definition of a per se level using epidemiological evidence and a meta-analysis of laboratory and simulator studies. Sensibly, they have erred in the direction of setting a high level that may require downward adjustment in the light of further research.

Australian state governments have not waited for the development of a per se level. Victoria introduced random roadside saliva testing for cannabis and other drugs in an analogous way to breath testing in December 2004 and other states and territories are following suit [13]. If Australian legislators had restricted themselves to saliva testing when drivers had had an accident or displayed evidence of impairment, the law would have been uncontroversial but ineffective, because similar testing regimens have limited impacts on drink driving [14]. Instead, Australian legislators have assumed that this policy will produce the substantial reductions in road crashes that random breath testing did when combined with widespread publicity and highly visible and sustained enforcement [15,16].

Given the limited scientific evidence for a per se level of THC the Australian drug testing regimens lack evidential support. The illegality of cannabis has prompted a ‘zero tolerance’ approach in Australia, with any detectable amount of the drug tested constituting an offence [13]. On this policy, the definition of a per se level is irrelevant because road safety benefits are secondary to enforcement of drug laws.

The introduction of random saliva testing in Australia was not preceded by an extensive public debate about its civil liberties implications or probable deterrent effects. The civil liberties issues need more attention. Should the authorities have the power to force citizens to incriminate themselves when they have not committed a driving offence or been involved in an accident? And what real protections are there in this era of widespread DNA testing and offender profiling to prevent police retaining saliva samples of convicted drug drivers for criminal investigations?

Proponents of these laws argue that random drug testing will save lives, but so far no scientifically persuasive evidence has been produced that these laws have done so. The success of Australian roadside drug testing, accordingly, needs to be evaluated thoroughly to see if it reduces drug driving at an acceptable social and economic cost. If evidence of an impact on drug driving is forthcoming, citizens should have the right to debate whether these public health benefits offset the threats to democratic freedoms. Public debate is essential if random alcohol testing is not to serve as a Trojan horse for the introduction of wider and scientifically questionable laws without adequate public scrutiny.