quite different result from Hedges-Olkin and Simex method

[MeiDongstat]

Dear Author,

Thank you for the great paper and the package. I'm trying to use the package for my own data. My GWAS of incidence is from a meta analysis of 40,000 cases and 70,000 controls. And my survival GWAS includes ~7,000 ceses. I haved pruned my data and harmonized to make sure they're refering to the same alleles. However, I got the following result when I use those two different methods:

**indexevent(match$beta,match$se,match$beta_y,match$se_y)
[1] "Coefficient -0.046"
[1] "Standard error 0"
[1] "95% CI -0.046 -0.046"

indexevent(match$beta,match$se,match$beta_y,match$se_y,method="SIMEX",B=100, seed=2020)
[1] "Coefficient 99.9999443827069"
[1] "Standard error 24.0560512666864"
[1] "95% CI 5.70197744071336 99.9999656266225"**

This result shows quite different coefficients and directions. There is no SNP highly significant with both incidence and prognosis in my data. Could you provide some insight what could be the issue that caused the difference? What else should I do to obtain more "correct" result?

Regards,
Mei

[fdudbridge]

Hi. Sorry for my slow reply. It looks as if the SIMEX has not converged to a proper result. The software searches for a maximum likelihood estimate in the range -100 to 100 and this type of result occurs when no proper maximum is found. I'd suggest first running it again with B=1000 and even B=10000, although this will take a long time to run. The Hedges-Olkin result may be reasonable if the unadjusted coefficient is also negative (this is the b.raw value in the object returned by indexevent). However the lack of a standard error may inflate the significant of the adjusted effects.

We are working on alternative methods to obtain the adjusted coefficient without requiring lengthy simulations.

best wishes, Frank