Testing for mechanistic interactions in long-term follow-up studies.

Area of Science:

• Epidemiology
• Biostatistics
• Survival Analysis

Background:

• Multiplicative interaction terms in Cox models are often misinterpreted as mechanistic interactions.
• Existing indices for mechanistic interaction (RERI, PRISM) have limitations, particularly with censored data.
• Accurate assessment of mechanistic interactions is crucial in epidemiological and clinical research.

Purpose of the Study:

• To propose a novel 'mechanistic interaction test' (MIT) specifically designed for censored data.
• To evaluate the performance of MIT compared to modified RERI and PRISM tests.
• To provide a statistically robust method for testing genuine mechanistic interactions in survival studies.

Main Methods:

• Development of a new mechanistic interaction test (MIT) for censored data.
• Utilizing Monte-Carlo simulations to assess Type I error rates and statistical power.
• Comparison of MIT's performance against modified RERI and PRISM tests under various hazard curve scenarios.

Main Results:

• The proposed MIT maintains accurate Type I error rates for censored data, regardless of hazard curve proportionality.
• MIT demonstrates significantly greater statistical power than modified RERI and PRISM tests.
• Simulations confirm MIT's reliability across proportional, non-proportional, and crossing hazard curves.

Conclusions:

• The novel mechanistic interaction test (MIT) is a reliable and powerful tool for analyzing censored data.
• MIT overcomes the limitations of existing methods, offering accurate assessment of genuine mechanistic interactions.
• Researchers are recommended to use MIT for testing mechanistic interactions in studies with censored survival data due to its superior statistical properties.