We found evidence for an increased risk of being responsible for a road traffic crash for users of prescribed medicines defined as presenting a level 2 or level 3 risk of driving impairment according to the French medication classification system. The fraction of road traffic crashes attributable to levels 2 and 3 medicine use was 3.3% [2.7%–3.9%].
The study protocol planned for the inclusion of a large range of descriptive variables related to the crash and to the drivers involved. In particular, we were able to determine the responsibility status of the driver in the crash and to adjust for key confounding factors. The responsibility analysis is a real strength of the study as it allows for the comparisons of cases and controls that share the same characteristic of being drivers. In a previous study on the impact of illegal drug consumption, using the same national police database but limited to fatal crashes 
, the same method used to determine responsibility was approved by an independent expert evaluation of responsibility. Furthermore, because the responsibility analysis relies on the assumption that nonresponsible drivers are representative of the driving population, the authors of the previous study validated the comparison of a subset of the nonresponsible individuals with the driving population in France 
. Finally, the strong dose-effect relationship found in our study between alcohol level and responsibility is a further indirect validation of the method. Importantly, responsibility levels were calculated independently of alcohol and illicit drug use because of their potential interactions with medicine use.
Medicine exposure was ascertained from computerized records of reimbursed prescriptions filled at the pharmacy. These data were not subject to underreporting, a major problem encountered when medicine exposure data is self-reported 
. On the other hand, it is one of the study limitations that dispensing dates were considered in this study as a surrogate for actual consumption. We did not know whether the medicines were actually ingested or not. Noncompliance, which we were not able to check, would therefore result in exposure misclassification. Other studies using patient-derived data and the same dispensation database showed that the health care insurance data are reliable indicators of actual exposure for medicines used over a long time frame, less so for episodically used medicines 
. We assumed that the exposure period started on the day after dispensing, as medicine dispensation on the day of crash may have been a consequence of the crash. Another limitation was that exposure to nonprescribed drugs can also not be estimated from the health care insurance database. However, less than 15% of medicines sold in France correspond to nonreimbursable medicines and most of these products have either no or negligible influence on the ability to drive.
The comparison between included drivers by means of their national ID and nonincluded drivers showed that injury severity was associated with the probability of being part of the study. Thus severely injured drivers were more likely to be included than slightly injured drivers. Killed drivers and uninjured drivers still had lower inclusion rates. This finding can be explained by the fact that injured drivers were more likely to be admitted to hospital so their health care number was more frequently noted in the police report. Thus, our study sample slightly overrepresented drivers injured in more severe crashes.
After adjustment for crash and individual variables, including exposure to other medicines, the risk of being responsible estimate was reduced for level 3 medicines, but the association did remain significant (from 1.56 [1.42–1.71] to 1.25 [1.12–1.40]). The crude risk of being responsible measured for level 3 medicines was thus partly related to these crash and individual variables and particularly due to a co-consumption of alcohol and level 2 medicines.
The protective effect of level 0 medicines could be explained by the treatment of those minor acute diseases that might lead to an increased risk of being responsible for the crash. Indeed, a number of specific physical and/or psychological conditions are likely to influence driving ability.
Surprisingly, we found no interaction between alcohol level, as reported by police forces, and medicine use, although alcohol is known to potentiate the effects of some medicines. It should be noted, however, that as the presence of alcohol is not always tested for in drivers involved in slight-injury crashes, this variable might be underestimated. Moreover, drivers who had a negative breath test were not tested for blood alcohol concentration (the legal limit in France is less than 0.5g/l). Information about illicit drug use was not available in any database. The analysis was also unable to adjust for driving exposure. Whilst on medication, some people may drive less to compensate for a perceived risk. They may also reduce their speed, pay more attention, or alter the road types that they use. The present study therefore estimated the impact of actual consumption and driving behaviors on the risk of road crash among active drivers.
According to our results, the French risk classification seems relevant regarding medicines classified as levels 2 and 3 of risk for road traffic crashes. Even if the risk for level 2 and 3 medications is similar, we believe that it is useful to differentiate these two levels. The effects of level 2 medicines on driving abilities depends both on the pharmacodynamics of the drug and on individual susceptibility; medical advice is therefore needed to weigh the potential risk for each individual. Various medicines are classified as level 2. The risks found for psycholeptics (mainly anxiolytics) and psychoanaleptics (mainly antidepressants) are concordant with others studies 
. The results for antiepileptics and other nervous system drugs (in particular medicines used to treat opioid dependence) are of interest and deserve further investigation. For some of the ATC classes in level 2, the association in the responsibility analysis was not significant; however, the number of drivers exposed to antihypertensives, muscle relaxants, anti-Parkinson drugs, and antihistamines for systemic use was small. On the other hand, despite a relatively large number of individuals exposed to analgesics (including opioid analgesics), we found no association with the risk of being responsible for a crash. With level 3 medicines, the pharmacodynamic effect is predominant so all users are advised not to drive. The effects of level 1 medicines may be so dependent on individual susceptibility that an effect on driving abilities might be a rare event. Therefore, the relevance of labeling level 1 medicines is questionable.
The respective roles of disease and the medicines used to treat disease are difficult to disentangle. After adjustment for the presence of a long-term chronic disease, results from the responsibility analysis did not suggest an important confounding effect of disease. In the case-crossover method, each individual is his or her own control and confounding due to individual factors is therefore eliminated, including fixed characteristics such as long-term chronic diseases. Other studies have used this approach to examine the relationship between medicines and the risk of injury 
. The use of level 3 medicines was found to be associated with an increased risk of road traffic crash both in the responsibility analysis and in the case-crossover analysis. However, the risk associated with level 2 medicines in the responsibility analysis (OR
1.31 [1.24–1.40]) disappeared in the case-crossover analysis (OR
1.00 [0.95–1.05]). The risk of road traffic crashes associated with chronic exposure to level 2 medicines cannot be assessed by a case-crossover design. Indeed, an individual using a medicine throughout the study period would be exposed on the crash date and on the control day. Our results on level 2 medicines are therefore likely to be related to the impact of chronic medicine consumption, i.e., mainly drugs used in diabetes, opioids, antiepileptics, anxiolytics, and antidepressants. On the other hand, hypnotics and sedatives, mainly representing level 3 medicines, can be used on an acute basis and their impact on road traffic crashes are detected with the case-crossover analysis.
Our study provides evidence of the contribution of medicines to the risk of road traffic crashes. Improving driver behaviour is one of the challenges for improving road safety. Providing patients with proper information on the potential effect of medicines on their ability to drive is the main objective of drug and risk classifications such as the French framework. The European Union is currently aiming to harmonise drug classification systems, using a reliable methodology based on scientific evidence. This epidemiological study provides sound evidence for consideration in such an endeavour. A follow-up study is now needed to evaluate the effect of the French medication labeling system on the prevention of road traffic crashes.