There is no universal method for correcting index event bias in disease progression studies. Instead, researchers should choose analytical approaches based on the available data and biological context to improve causal inference.
Background
This study evaluates statistical methods used to address index event bias in Mendelian randomization analyses of disease progression. By comparing existing approaches, the authors aim to improve the reliability of causal inference and support more accurate therapeutic target validation.
Findings
The study found that no single method consistently outperformed the others. Inverse-probability weighting partially reduced bias when individual-level data were available, but it remained prone to false positives, showing meaningful Type 1 error inflation under the null because the selection model could not fully account for unmeasured confounding. Multivariable Mendelian randomization performed well in specific scenarios but was sensitive to pleiotropy. Slope-Hunter performed poorly across all simulation scenarios, even when its underlying assumptions were fully satisfied.