Limited transportability of published acenocoumarol dosing algorithms in an Algerian pharmacogenetic cohort

Published acenocoumarol dose-prediction equations may perform differently across populations. We hypothesized that equations developed outside Algeria would show limited transportability. This study externally validated two pharmacogenetic equations and one matched clinical equation in 119 adults receiving stable acenocoumarol therapy at the University Hospital of Sétif, Algeria. Published coefficients were applied without refitting. Model-specific analyses included 109 participants for the Tong pharmacogenetic equation and 119 for the Tong clinical and Bulgarian pharmacogenetic equations, and direct comparisons used the common 109-participant subset. Performance was assessed using prediction error, prediction R², calibration, and the proportion of estimates within 20% of the observed dose, with 95% confidence intervals obtained from 5000 bootstrap resamples. Mean errors were -10.07, -7.44, and -8.21 mg/week for the Tong pharmacogenetic, Tong clinical, and Bulgarian equations, respectively. Corresponding mean absolute errors were 10.73, 8.96, and 10.48 mg/week, and 19.3%, 31.9%, and 26.9% of predictions were within 20% of the observed dose. All prediction R² values were negative, calibration was inadequate, and performance was poorest in patients requiring more than 21 mg/week. Sensitivity analyses supported the same interpretation. These equations substantially underpredicted stable dose and require formal recalibration and independent prospective validation before use in comparable Algerian patients.