Strategies of Modelling Incident Outcomes Using Cox Regression to Estimate the Population Attributable Risk

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Cox models are widely used for survival analysis, but assumptions like risk proportionality are often violated.
• Accurate estimation of population attributable risk (PAR) is essential for public health interventions but is frequently oversimplified.

Purpose of the Study:

• To investigate how different Cox modelling strategies impact the validity of model assumptions.
• To evaluate the effect of these strategies on the estimation of population attributable risk (PAR).

Main Methods:

• Utilized Cox regression models with time-to-event and age-to-event as underlying time variables.
• Assessed risk proportionality using interaction tests and model fit with Akaike Information Criteria (AIC).
• Investigated mutually adjusted models and incorporated age and sex strata variables.

Main Results:

• Models using age as the time variable, with age and sex strata, demonstrated improved risk proportionality and better model fit.
• Mutually adjusted models allowed for more accurate PAR estimation for individual modifiable risk factors.
• The accuracy of PAR estimation decreased when dealing with correlated modifiable risk factors.

Conclusions:

• Strategic selection of the Cox model's time metric and structure is vital for assumption validity and reliable survival analysis.
• Accurate PAR estimation, particularly with mutually adjusted models, enhances the basis for informed public health policy decisions.
• Care must be taken when estimating PAR for correlated risk factors.