Comparison of two-stage methods for count data in Mendelian randomization: a simulation study.

Area of Science:

• Epidemiology
• Statistical Genetics
• Biostatistics

Background:

• Mendelian randomization (MR) methods, typically for continuous data, are often applied to count data without clear validity.
• Two-stage MR methods like TSPS and TSRI are common but their comparative performance for count exposures/outcomes is unknown.

Purpose of the Study:

• To systematically evaluate the performance of TSPS and TSRI for count data in MR.
• To provide guidance on selecting valid instruments and ensuring reliable causal inference with count data.

Main Methods:

• Simulated MR scenarios with varying instrument strength, confounding, sample size, and invalid instruments.
• Evaluated TSPS and TSRI using Poisson and negative binomial models.
• Assessed performance via bias, RMSE, CI coverage, CI width, and Type I error rate.
• Applied methods to investigate alcohol consumption and gout attacks.

Main Results:

• TSRI with the Poisson model showed superior stability, lower bias, and RMSE compared to TSPS across scenarios.
• TSRI maintained near-nominal CI coverage and narrow widths, with Type I error near 0.05.
• Instrument strength significantly impacted performance; weaker instruments increased bias and RMSE.
• Increasing sample size with invalid/weak instruments exacerbated bias.
• Empirical application suggested a causal link between alcohol consumption and gout attacks, though potentially biased.

Conclusions:

• TSRI is a more robust method than TSPS for MR with count exposures and outcomes.
• The study clarifies the behavior of TSPS and TSRI in count data settings.
• Provides practical recommendations for valid instrument selection and causal inference in MR studies involving count data.