Ways to fit a PK model with some data below the quantification limit.

Area of Science:

• Pharmacokinetics and Pharmacometric Modeling
• Biopharmaceutical Analysis

Background:

• Pharmacokinetic (PK) data frequently include concentrations below the quantification limit (BQL).
• Handling BQL observations is crucial for accurate pharmacokinetic model fitting.
• Current methods for BQL data involve either discarding or specialized statistical treatment.

Purpose of the Study:

• To evaluate seven distinct methods for managing BQL observations in pharmacokinetic data analysis.
• To compare the statistical performance of different BQL handling techniques using simulated data.

Main Methods:

• Simulation of single-subject and population pharmacokinetic data from a one-compartment model.
• Evaluation of methods including: discarding BQLs, treating them as censored data, imputation with 0, and imputation with half the quantification limit (QL/2).
• Assumed coefficient of variation (CV) for measurements at the quantification limit was ≤20%, aligning with FDA guidance.

Main Results:

• Treating BQL observations as fixed-point censored data demonstrated overall statistical advantages.
• The benefit of censored data methods over simple BQL discarding was modest, particularly when BQL frequency was low.
• Imputation with QL/2 improved population parameter estimation but degraded population variance estimation.

Conclusions:

• Handling BQL data as censored observations is statistically advantageous in pharmacokinetic modeling.
• The choice of method depends on the frequency of BQL observations and the specific modeling goals (e.g., parameter estimation vs. variance estimation).
• Simple imputation methods like using 0 should be avoided, while QL/2 imputation has specific trade-offs for population PK.