Entering edit mode

16 months ago

MatthewP
▴
920

Hello, I know their is multivariate cox regression. In my case I will first scan all variables separately use univariate model, then pick up those `p < 0.05`

to perform multivariate cox regression. My main question is how to adjust p value from univariate cox regression. I use R and do:

```
# allP is p values vector for all variables
adjP <- p.adjust(allP, method = "BH")
```

This artical said _Bonferroni method_ is not good enough. I want to know how you guys adjust p values usually.

There is many things wrong with this approach. At first, selection of variables in this way (significant in a univeriate model) is quite a bad practice. At second, you use Benjamini-Hochberg correction, not Bonferroni. At third, you don't need to perform multiple test correction if p < 0.05 is your selection criterion. Just use this https://web.stanford.edu/~hastie/glmnet/glmnet_alpha.html

Thanks, I do apply lasso cox model later. But I want to filter out some genes(hundres genes in my case) so I use univariate cox model first.

Non significant predictor in a univariate model can easily be very significant in multivariate.