Likelihood comparisons in bounded outcome score analysis must be internally consistent.

Area of Science:

• Pharmacometrics
• Clinical Trial Analysis
• Statistical Modeling

Background:

• Clinical trial endpoints frequently utilize bounded outcome scores (BOS), which are variables with restricted values within finite intervals.
• Existing analysis methods often categorize BOS data as continuous, categorical, or a hybrid, leading to potential confusion.
• The dual nature of BOS data complicates pharmacometric model evaluation and likelihood comparisons.

Purpose of the Study:

• To clarify fundamental issues in the pharmacometric analysis of bounded outcome scores (BOS).
• To guide appropriate model evaluation and data likelihood comparisons for BOS data in clinical trials.
• To reduce confusion in pharmacometric analyses involving variables with restricted values.

Main Methods:

• This commentary reviews common pharmacometric approaches for analyzing bounded outcome scores.
• It discusses the implications of treating BOS data as continuous, categorical, or mixed.
• The text clarifies the appropriate domains for model evaluation and conditions for comparing data likelihoods.

Main Results:

• Bounded outcome scores (BOS) exhibit characteristics of both continuous and categorical variables.
• Misinterpretation of BOS data can lead to inappropriate model selection and evaluation in pharmacometrics.
• Clearer understanding of BOS data properties is crucial for accurate clinical trial endpoint analysis.

Conclusions:

• Pharmacometric analyses require careful consideration of the bounded nature of outcome scores.
• Appropriate methods for model evaluation and likelihood comparison are essential for valid clinical trial results.
• This work facilitates more rigorous and accurate pharmacometric analyses of bounded outcome scores.