Authors’ reply to Olfson and Schoenbaum, Nardo, Bartlett, Moore, Case, Gøtzsche, and Barber and colleagues

We agree with Olfson and Schoenbaum about the need for more systematic use of cause of injury codes.1 2 Public health efforts to understand and reduce suicide risk depend on accurate data. Indeed, this problem (varying use of these codes across and within health systems over time) motivated us to use a proxy measure for suicide attempts.

Several comments question the sensitivity of this proxy measure (the proportion of suicide attempts by psychotropic drug poisoning).3 4 5

Olfson and Schoenbaum cite US national data that sensitivity is likely to be no more than 50%. The use of these data is, however, problematic given the inconsistent use of cause of injury codes. We examined this question in three of our research network health systems where these codes are used regularly. We found that the proportion of suicide attempts (by cause of injury code) across age groups and health systems ranged from roughly 30% to 60%. A sensitivity of 30% to 60% means that this proxy measure would not be appropriate for estimating prevalence, and reduced sensitivity would reduce statistical power for detecting changes over time, although it would not bias our interrupted time series analysis.

Olfson and Schoenbaum also say that only a …