Eliminating Survivor Bias in Two-stage Instrumental Variable Estimators

Area of Science:

• Epidemiology
• Statistical Genetics
• Biostatistics

Background:

• Mendelian randomization (MR) studies often analyze elderly populations, increasing susceptibility to survivor bias.
• Survivor bias, a form of selection bias, can invalidate instrumental variable (IV) assumptions in MR analyses of mortality.
• Violations of IV assumptions, even minor ones, can significantly bias MR study results.

Purpose of the Study:

• To develop a novel statistical method to mitigate survivor bias in Mendelian randomization studies of mortality.
• To derive a two-stage instrumental variable estimator that is robust to selection bias in elderly cohorts.

Main Methods:

• The study leverages specific conditions under which instrumental variable assumptions hold within defined risk sets of surviving individuals.
• A two-stage instrumental variable (2SIV) estimator is derived based on additivity assumptions between the instrument, unmeasured confounders, exposure, and mortality hazard.
• The proposed method aims to provide an unbiased estimate of the exposure-mortality association.

Main Results:

• The derived two-stage instrumental variable estimator is theoretically insulated against survivor bias under stated additivity assumptions.
• This method allows for valid inference on the causal effect of an exposure on mortality, even in the presence of selection bias.

Conclusions:

• The proposed two-stage instrumental variable approach offers a robust solution for Mendelian randomization studies investigating mortality in older populations.
• This method enhances the reliability of causal inference by addressing critical biases inherent in observational studies of aging and survival.