The risk of determining risk with multivariable models

Area of Science:

• Medical Statistics
• Epidemiology
• Biostatistics

Background:

• Multivariable statistical methods are increasingly prevalent in medical research.
• Accurate application and reporting of these methods are essential for reliable findings.

Purpose of the Study:

• To review the principles of multivariable analysis.
• To examine the application of multivariable statistical methods in general medical literature.

Main Methods:

• A literature search identified 451 publications using multivariable methods (1985-1989).
• Sixty articles employing logistic regression or proportional hazards analysis were reviewed for adherence to guidelines.

Main Results:

• The use of multivariable methods rose from 10% to 18% between 1985 and 1989.
• 73% of reviewed articles quantified risk estimates, but many violated or omitted methodological guidelines.
• Common issues included overfitting, lack of interaction testing, and unspecified variable selection, potentially leading to inaccurate interpretations.

Conclusions:

• There is a significant need to enhance the reporting and execution of multivariable analyses in medical research.
• Adherence to methodological guidelines is critical for the validity and interpretability of results.