Tutorial on Firth's Logistic Regression Models for Biomarkers in Preclinical Space

Area of Science:

• Biostatistics
• Preclinical Research

Background:

• Preclinical studies utilize diverse data, including biomarkers, genetic, imaging, and clinical information.
• Logistic regression is a common statistical model for binary outcomes in these studies.
• Small datasets in preclinical research can present separation issues, leading to unreliable logistic regression results.

Purpose of the Study:

• To demonstrate the challenges of logistic regression with separation in small preclinical datasets.
• To introduce Firth's logistic regression as a solution for bias reduction in such scenarios.
• To compare the performance of standard logistic regression against Firth's logistic regression.

Main Methods:

• Illustrating complete and quasi-complete separation in logistic regression models.
• Applying Firth's logistic regression to penalized regression for bias reduction.
• Utilizing R code and provided datasets for practical examples.

Main Results:

• Standard logistic regression yields inflated coefficient estimates and standard errors when separation occurs.
• Firth's logistic regression successfully reduces bias in coefficient estimates.
• Demonstration of improved model stability and reliability with Firth's method.

Conclusions:

• Firth's logistic regression is a valuable tool for analyzing small preclinical study data with separation issues.
• This penalized approach offers more accurate and reliable estimates compared to standard logistic regression.
• The provided R code and datasets facilitate the application of Firth's method.