Mixture models for continuous data in dose-response studies when some animals are unaffected by treatment

Area of Science:

• Biostatistics
• Pharmacology
• Toxicology

Background:

• Dose-response studies are crucial for evaluating treatment efficacy and safety.
• A common challenge is identifying individuals within a study population who do not respond to a given treatment.
• Existing models may not adequately account for non-responder populations in continuous outcome data.

Purpose of the Study:

• To present a novel mixture model for analyzing dose-response data with non-responsive subjects.
• To provide a statistical framework for simultaneously estimating treatment response probability and outcome among responders.
• To demonstrate the model's utility with real-world data examples.

Main Methods:

• Development of a mixture model combining logistic regression (for response probability) and linear regression (for responder mean).
• Application of the Expectation-Maximization (EM) algorithm for maximum likelihood estimation of model parameters.
• Utilization of likelihood ratio tests for model comparison and parameter reduction.

Main Results:

• The proposed mixture model effectively accommodates dose-response data where a subset of subjects shows no treatment effect.
• The EM algorithm provides a robust method for parameter estimation.
• Likelihood ratio tests successfully differentiate between the full model and simplified versions, aiding interpretation.

Conclusions:

• The described mixture model offers a flexible and powerful tool for dose-response analysis in the presence of non-responders.
• This approach enhances the accuracy of treatment effect estimation by explicitly modeling non-response.
• The model is applicable across various scientific fields employing dose-response studies, including pharmacology and toxicology.