[BUG]: P value and 95% CI doesn't match

[Yvonnehuang1316]

Dear developer,

Recently I ran 'TwosamleMR' and noticed that the mode-based method may have some bugs in reporting P values when the number of IV is small.

For example, I ran an MR with 3 SNPs as IV, here are my results.
id.exposure id.outcome method nsnp b se ci_low ci_upp pval
1 6azTYV iUTOLO Simple mode 3 -5.834308 2.627214 -10.983553 -0.685062833 0.1565154
2 6azTYV iUTOLO Weighted mode 3 -4.327279 2.509336 -9.245486 0.590928338 0.2267644
3 6azTYV iUTOLO Penalised mode 3 -4.327279 2.526705 -9.279529 0.624971024 0.2289150
4 6azTYV iUTOLO Simple mode (NOME) 3 -5.834308 2.278630 -10.300342 -1.368274369 0.1246472
5 6azTYV iUTOLO Weighted mode (NOME) 3 -4.464282 2.275595 -8.924365 -0.004197941 0.1888013

The ci_low and ci_up are all both less than 0, while the pval is not significant.

I went back to the original code, I noticed that the 95% CI is derived from 1000 bootstrap. However, the P value is derived from t-statistics:

"P_Mode <- pt(abs(beta_Mode/se_Mode), df = length(b_exp) - 1, lower.tail = F) * 2"

I guess the problem was the set-up of 'df = length(b_exp) - 1'. If we are estimating p value from bootstrap, should the dr = number of bootstrap -1 ? In my case, I only used 3 SNPs, if df = length(b_exp) - 1, then here the df is 2, which makes the p value really big ...which is inconsistent with 95% CI...

Thank you very much,
Yvonne & Huan