Graves’ disease liability and prostate cancer: instrument selection
Exposure: genetic liability to Graves’ disease. Outcome: prostate cancer. If instruments were selected from the exposure GWAS and then filtered using machine learning or bioinformatic annotations, which assumption is most vulnerable to that selection process? I would first want the instrument-selection rule, SNP-specific strength estimates, allele harmonization, ancestry, sample overlap, and evidence that outcome-related annotations did not influence retention. Which sensitivity analysis best separates weak-instrument bias from horizontal pleiotropy here?
