A novel quantification of information for longitudinal data analyzed by mixed-effects modeling

Area of Science:

• Pharmacometrics
• Statistical Modeling

Background:

• Nonlinear mixed-effects (NLME) modeling is crucial for longitudinal data analysis, particularly with sparse sampling.
• Assessing global model information using Fisher information matrix determinants is common, but parameter-specific information is vital in clinical studies.

Purpose of the Study:

• To introduce a novel, interpretable metric, "Relative Information" (RI), for quantifying parameter-specific information in NLME models.
• To establish the relationship between interindividual variability and parameter estimator variance.

Main Methods:

• Developed the "Relative Information" (RI) metric, ranging from 0% to 100%.
• Established theoretical convergence properties of RI and parameter estimator variance under ideal experimental conditions.
• Validated the metric through extensive simulations and real-world dataset analyses.

Main Results:

• The proposed RI metric effectively characterizes information for specific parameters in NLME models.
• Demonstrated RI's convergence to 100% and parameter variance convergence under ideal experimental conditions.
• RI provides an easy-to-interpret measure of data informativeness for model parameters.

Conclusions:

• The novel Relative Information (RI) metric offers a valuable tool for assessing parameter-specific data informativeness in NLME models.
• RI can enhance the design and diagnostic capabilities of pharmacokinetic and pharmacodynamic studies.
• This metric aids researchers in understanding and optimizing the information content of their data for specific model parameters.