Wald ratio with delta method

[LucilleAugusta]

Hi!

I would like to use Wald ratio and the delta method.
When looking in the code I find both mr_meta_fixed and mr_meta_random muted from themr_method_list- is there a reason for this? [Link to code]

I would like to know if there is anything wrong with the calculations or a better way to do it.

I hope some one has some input to this.
Thanks,
Lucille

[mightyphil2000]

Hi Lucille

The quick answer is that those functions have been replaced with mr_ivw_fe (equivalent to mr_meta_fixed ) and mr_ivw_mre (better version of mr_meta_random). These methods are appropriate when you have multiple independent SNPs available to instrument your exposure.

Longer answer:
The mr_meta_fixed method is equivalent to mr_ivw_fe. They should give the same result. mr_meta_fixed works by taking a weighted average of Wald ratios (using inverse variance of the SNP-outcome effect as the weight). mr_ivw_fe is the slope from a regression model of the SNP-outcome effects regressed on the SNP-exposure effects, with the standard error scaled to reflect a fixed effects model (this is done by removing a correction for over-dispersion, which is implemented by default in the regression model).

mr_meta_random works similarly to mr_meta_fixed, except that the weights additionally incorporate information about heterogeneity between the Wald ratios (the fixed effects method assumes there is no heterogeneity). We have replaced mr_meta_random with mr_ivw_mre (multiplicative random effects). The mr_meta_random method implements an additive random effects model, whereas mr_ivw_mre is a multiplicative random effects model, which we think is a better method (for various reasons that we can discuss further if interested).

bw
Philip

[LucilleAugusta]

Thanks for the quick response!
I'll look into these functions.

/Lucille