We compared the results of our Fisher’s exact tests to regression-based frameworks. While an exact test is robust for rarer variants, regression methods can incorporate covariates to help to mitigate confounders and are recommended when careful control for confounding cannot be ensured. We performed single-variant association tests across all autosomal variants for 324 Chapter IX binary phenotypes (diseases of the circulatory system; Supplementary Table 29) using SAIGE SPA12 and REGENIE 2.0.2 (ref. 34), including sex, age, sequencing batch and ten principal components as covariates (Supplementary Methods). Fisher’s exact Phred scores (−10 × log10(P values)) were strongly correlated with those from SAIGE SPA (Pearson’s r = 0.95) and REGENIE 2.0.2 (Pearson’s r = 0.94). Fisher’s exact P value statistics were also more conservative for lower frequency variants (MAF ≤ 1%) (Supplementary Table 6). Correlation was higher for signals with a P < 1 × 10−8 in either Fisher’s exact test or SAIGE SPA (Pearson’s r = 0.99) and Fisher’s exact test or REGENIE 2.0.2 (Pearson’s r = 0.99) (Supplementary Figs. 1, 2, Supplementary Table 6). The median lambda inflation factor λGC
