Abstract / Summary
Purpose: To present updated estimates from a nationwide Swedish study comparing dispensed medications with forensic toxicological findings (2006-2013), following re-evaluation of the toxicology dataset to address periods with non-uniform analytical detection and inconsistent testing.
Methods: Observations from time periods with non-uniform detection or inconsistent testing were excluded. Analyses were repeated using the PRE2DUP method to model drug exposure. Agreement between PRE2DUP-modelled exposure and post-mortem toxicology was assessed using sensitivity, specificity, predicted adherence, and Cohen's κ.
Results: After exclusion of inconsistent testing periods, the number of individuals for affected substances was reduced (10 060 vs. 18 627). Proportional changes in sensitivity and specificity were negligible. In contrast, predicted adherence and Cohen's κ improved across several substances. Moderate to substantial proportional increases were observed for selected medications but translated into only small group-level effects (approximately 2%-7%). The largest improvements were observed for cardiovascular drugs, beta-blockers, for which Cohen's κ and predicted adherence increased substantially (approximately 40%-55%) at the class level; notably, only for these substances did Cohen's κ shift interpretation (from fair to moderate agreement). Overall, proportional improvements across all substances were moderate (approximately 10%-15%). Despite these changes, overall agreement patterns remained comparable to the original results.
Conclusions: After exclusion of periods with non-uniform or incomplete toxicological testing, agreement between PRE2DUP-modelled exposure and forensic toxicology improved overall. Moderate increases observed for a few individual medications translated into only small effects at the group level, except for cardiovascular drugs, for which improvements were substantial. These updates do not materially alter the original interpretation but strengthen the validity of register-based drug exposure modelling.