Cross-fitted instrument: A blueprint for one-sample Mendelian randomization.

Area of Science:

• Biostatistics
• Genetic Epidemiology
• Causal Inference

Background:

• Weak instruments in instrumental variable regression (IVR) can bias causal effect estimates.
• Mendelian randomization (MR) is a powerful tool for inferring causality but is susceptible to weak instrument bias.

Purpose of the Study:

• To introduce a novel method, Cross-Fitted Instrument (CFI), to mitigate weak instrument bias in IVR.
• To adapt CFI for the MR setting, termed Cross-Fitting for Mendelian Randomization (CFMR), enhancing its applicability and robustness.

Main Methods:

• CFI randomly splits data, estimating instrument-exposure impact in each partition for subsequent IVR.
• CFMR applies CFI to MR, ensuring conservative estimates even with weak instruments and overlapping samples.
• CFMR utilizes all available data for genetic instrument selection, maximizing statistical power.

Main Results:

• CFMR demonstrates conservative bias, tending towards the null, even with weak instruments.
• CFMR outperforms existing methods like MR-RAPS when samples overlap, showing reduced bias.
• CFMR enhances statistical power in meta-analyses and enables cross-ethnic MR analyses, accounting for heterogeneity.

Conclusions:

• CFMR provides a robust, conservative, and powerful one-sample MR approach, overcoming limitations of traditional methods.
• CFMR's ability to handle weak instruments, maximize power, and account for ethnic heterogeneity makes it valuable for large-scale genetic studies.
• CFMR expands the applicability of MR to rare or difficult-to-measure exposures.