Better Dosing Through Better Error: Residual Error as a Hidden Lever in Model-Informed Precision Dosing
Résumé
Background: Model-informed precision dosing is increasingly integrated into therapeutic drug monitoring to optimize individualized pharmacotherapy, with maximum a posteriori Bayesian estimation serving as a key tool for pharmacokinetic parameter estimation. However, the impact of residual error specification on its precision remains underexplored. We hypothesized that reducing residual error would decrease the imprecision of area under the concentration–time curve predictions. Methods: Using rich pharmacokinetics datasets comprising 321 profiles for tacrolimus, iohexol, and mycophenolate mofetil, maximum a posteriori Bayesian estimation was applied based on sparse sampling (3 time points) using published population pharmacokinetics models. Proportional residual error settings were varied as near-zero (10 −8 %), low (1%, “Flat1”), and the original published model. Results: A low proportional error reduced the overall root mean square error of individual area under the concentration–time curve predictions by 30%–40% compared with the original residual error configuration. For tacrolimus, the root mean square error decreased from 28.5% under the published error model to 16.3% with a 1% error setting, whereas for iohexol, a near-zero residual error achieved up to a 40% reduction. The near-zero error scenario yielded the most accurate area under the concentration–time curve estimates for 45%–62% of patients across drug models. Conclusions: Reducing residual errors in maximum a posteriori Bayesian estimation significantly strengthens the influence of observed data on posterior calculations, improving the precision area under the concentration–time curve and dosing accuracy without requiring additional data acquisition or model redevelopment.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |
