I want to derive a risk prediction equation for disease D based on Pa and Pb, where Pa and Pb are concentrations of circulating proteins in the blood.

  • Disease D has two sub types: Da and Db.
  • In Da, concentrations of Pa are raised and Pb unaffected.
  • In Db, concentrations of Pb are raised and Pa unaffected.

Can I still model this using the below or is this invalid given that Pa and Pb are not consistently related to disease D (i.e. it depends on the subtype)?

glm(D ~ Pa+Pb,family="binomial",data=data)

1 Answer 1


If half your cases are ‘Da’ and the other half ‘Db’, then you will mask the association between proteins and diseases by combining diseases.

If you already know the disease-protein relationship, why use regression?

  • $\begingroup$ Thanks. The problem is that not all people with high levels of Pa and / or Pb have disease D. I want to use regression with Pa and Pb as well as other predictors of the disease (smoking, body mass index, age etc.) to be able to more accurately predict an individuals risk of disease D. $\endgroup$
    – John
    Feb 12, 2019 at 16:17
  • $\begingroup$ Regressing in the same equation implies the diseases are indistinguishable for your approach. You can consider a multinomial logistic regression if you want to model absence of disease or presence of disease A or presence of disease B. $\endgroup$
    – Todd D
    Feb 12, 2019 at 17:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.