Simultaneous confidence bands for nonlinear regression models with application to population pharmacokinetic analyses

Area of Science:

• Biostatistics
• Pharmacometrics
• Statistical Modeling

Background:

• Nonlinear regression models are vital in biostatistics, particularly for population pharmacokinetic and pharmacodynamic (PK/PD) modeling.
• Pointwise confidence intervals are insufficient for simultaneous inference across the entire covariate region.
• Existing methods can lead to reduced coverage probability when assessing the whole covariate range.

Purpose of the Study:

• To develop and evaluate methods for constructing simultaneous confidence bands for mean profiles in nonlinear mixed-effects models.
• To address the limitations of pointwise confidence intervals for comprehensive covariate region assessment.
• To provide tools for reliable inference in dose-response and biosimilarity applications.

Main Methods:

• Proposed two large-sample methods for simultaneous confidence band construction.
• Method 1: Based on the Schwarz inequality and asymptotic chi-squared distribution.
• Method 2: Utilizes simulation from a multivariate normal distribution.

Main Results:

• The study illustrates the application of these methods using theophylline pharmacokinetics data.
• An extensive simulation study was conducted to assess the operating characteristics of the proposed methods.
• The methods were extended for simultaneous confidence bands of model differences and for assessing model equivalence in biosimilarity.

Conclusions:

• The developed methods provide a statistically sound approach for simultaneous inference over the covariate region in nonlinear mixed-effects models.
• These methods enhance the reliability of analyses in pharmacometrics and related biostatistical fields.
• The extensions facilitate robust comparisons and equivalence assessments between models, particularly relevant for biosimilarity evaluations.