I am interested in checking how (and if at all) predictive is the position of genetic mutations in a given disease gene on the severity of the disease phenotype (so, I only have affected individuals (N=120). I have about 5 phenotypic measurements (continuous variables) and 5 events (e.g. syncope, death etc - binary variables) that I am interested in including in the model, alongside age (continuous) and gender (binary). I am using R to do this, and I am wondering which would be the correct approach. The way in which I would proceed is the following:

  1. Build one regression model per phenotypic variable / event (so, some binomial and some linear), in the form:

                           all_other_pheno_or_events, ...)


                        all_other_pheno_or_events, ..., family=binomial)
  2. Perform backward variable selection to check which independent variables have an influence on the predicted phenotype. Here I am not sure whether I should use AIC or BIC. I know the fundamental differences between the two, and maybe I'd go for AIC aiming for "inclusivity" (given it lower penalty for additional predictors), to then:

  3. Check the significance of each predictor retained by the backward variable selection procedure using nested models comparison (by likelihood ratio test) e.g.:

          glm(binary_variable~pred[1]+pred[2]+...+pred[N]),   test="LRT")

I suspect this approach may be unnecessarily complex, but I hope for this idea to be correct. I'd really appreciate some advice from others more expert than me in statistics and regression.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.