Likelihood based approaches to handling data below the quantification limit using NONMEM VI.

Area of Science:

• Pharmacometrics
• Pharmacokinetics
• Statistical Modeling

Background:

• Handling data below the quantification limit (BQL) is crucial for accurate pharmacokinetic (PK) modeling.
• Likelihood-based methods are commonly used, but their performance with BQL data requires careful evaluation.
• New features in NONMEM VI offer advanced capabilities for analyzing BQL data.

Purpose of the Study:

• To evaluate and compare different likelihood-based methods for handling below-quantification limit (BQL) data in pharmacokinetic modeling.
• To assess the performance of these methods using new features in NONMEM VI.
• To determine the optimal strategy for incorporating BQL data into PK analyses.

Main Methods:

• A two-compartment PK model with first-order absorption was utilized.
• Four methods were compared: discarding BQL data (M1), adjusting likelihood for remaining data (M2), maximizing likelihood above LOQ and censoring BQL (M3), and M3 with conditioning on positive observations (M4).
• Data were simulated using proportional error and truncated normal distributions; parameter estimate bias and precision were assessed.

Main Results:

• Method M3 demonstrated the best overall performance for data simulated with a proportional error model, followed by M2 and M1.
• Methods M3 and M4 produced similar parameter estimates when log transformation was not applied.
• For data simulated from a truncated normal distribution, Method M4 outperformed Method M3.

Conclusions:

• Maximizing the likelihood of data above the limit of quantification (LOQ) while treating below-quantification limit (BQL) data as censored (Method M3) provides the most accurate and precise parameter estimates.
• This approach is recommended for handling BQL data in pharmacokinetic analyses.
• The choice between M3 and M4 may depend on the underlying data distribution and analysis choices like log transformation.