Better Dosing Through Better Error: Residual Error as a Hidden Lever in Model-Informed Precision Dosing.

Area of Science:

• Pharmacokinetics and Pharmacodynamics
• Computational Biology and Bioinformatics
• Clinical Pharmacology

Background:

• Model-informed precision dosing (MIPD) utilizes therapeutic drug monitoring to optimize patient-specific pharmacotherapy.
• Maximum a posteriori (MAP) Bayesian estimation is crucial for pharmacokinetic parameter estimation within MIPD.
• The impact of residual error specification on MAP estimation precision is not well understood.

Purpose of the Study:

• To investigate how residual error specification affects the precision of pharmacokinetic parameter estimation using MAP Bayesian methods.
• To test the hypothesis that reducing residual error improves the accuracy of area under the concentration-time curve (AUC) predictions.

Main Methods:

• Utilized 321 pharmacokinetic profiles for tacrolimus, iohexol, and mycophenolate mofetil.
• Applied MAP Bayesian estimation with sparse sampling (3 time points) based on published population pharmacokinetic models.
• Varied proportional residual error settings: near-zero (0.1%-0.8%), low (1%), and the original published model's error.

Main Results:

• A low proportional error setting (1%) reduced overall root mean square error (RMSE) of AUC predictions by 30%-40% compared to original settings.
• For tacrolimus, RMSE decreased from 28.5% to 16.3% with a 1% error setting; for iohexol, near-zero error reduced RMSE by up to 40%.
• The near-zero error scenario provided the most accurate AUC estimates for 45%-62% of patients across different drug models.

Conclusions:

• Decreasing residual error in MAP Bayesian estimation enhances the impact of observed data on posterior parameter calculations.
• Improved AUC precision and dosing accuracy were achieved without needing additional data or model redevelopment.
• This approach offers a viable strategy for refining individualized pharmacotherapy in clinical practice.