"In Reply.-Dr Amoroso et al suggest that tobacco use, not available in our study, may have confounded the association between benzodiazepine use and the risk of motor vehicle crash in the elderly. Given the lack of an effect with short-acting benzodiazepines, and the fact that tobacco use is unlikely to be differentially distributed according to the type of benzodiazepine (long- vs short-acting), confounding by tobacco use is unlikely to account for our results. Drs Green and Wintfeld suggest that insufficient information was available to control for confounding by health status. We used the chronic disease score to obtain a measure of health status based on patterns of drugs dispensed in the year prior to the index date. [1] Thus, cardiovascular disorders, diabetes, and other chronic conditions and their severity are taken into account. Moreover, exposure to other drugs with central nervous system effects, including antidepressants, were controlled for separately in the analysis. It is unlikely that, after controlling for these accurate measures of health status, confounding would still be present. Green and Wintfeld question the validity of both prescription drug data for exposure and police reports for the outcome. Our outcome, which is the driver in a crash that caused injury, is not based on reports made by the subjects for compensation purposes but rather on direct observations made by police present at the crash site. We used the presence of injury specifically to avoid reporting bias from less serious outcomes. Moreover, since it is unlikely that the accuracy of police reports would be differential with respect to benzodiazepine half-life, the lack of an effect from short-acting benzodiazepines provides support for the validity of our results. While we agree that reduced compliance with benzodiazepine treatment and inaccurate dose and duration data could have led to exposure misclassification, this would imply that we in fact underestimated the real risk for both short- and long-acting benzodiazepines. On the other hand, the inaccuracy of 31% for dose and duration of the prescription drug data is explained by the way pharmacies dispensed the prescriptions, such as "a 60- or 90-day prescription split into 2 to 3 thirty-day supplies." [2] This does not influence our exposure definition, since the definition was based on summing the durations of successive prescriptions. Nevertheless, the higher risk observed for the initial 7 days of treatment (the risk was lower after 7 days) for long-acting benzodiazepines-there is no such risk profile for the short-acting ones-suggests that any residual misclassification would be nondifferential and, as such, would only attenuate the difference between the risk estimates of the short- and long-acting agents. We disagree with the implication of Green and Wintfeld that crashes with minor injury are of less relevance and significance. Given the reduced capacity for recovery among the injured elderly, all crashes, regardless of their severity, must be avoided. Finally, we concur with the views of Drs Pomara et al and Drs Soderstrom et al, on the basis of the new information they provide, that further research is needed in other settings to better understand this problem. Brenda Hemmelgarn, PhD Samy Suissa, PhD Royal Victoria Hospital; McGill University; Montreal, Quebec REFERENCES 1. Von Korff M, Wagner EH, Saunders K. A chronic disease score from automated pharmacy data. J Clin Epidemiol. 1992;45:197-203. 2. Tamblyn R, Lavoje G, Petrelia L, Monette J. The use of prescription claims databases in pharmacoepidemiological research: the accuracy and comprehensiveness of the prescription claims database in Quebec. J Clin Epidemiol. 1995;48:999-1009." (Author/publisher) See also ST 972255 fo
Samenvatting