Using logistic regression to compute age-adjusted rates


New Member
Dear All,

Would anybody please give me a breakdown no how to compute "age-adjusted proportions" based on the logistic regression coefficients?

That is specifically, I have CVD status(1 or 0), obesity status (1 or 0), and age in the following model:

(*) g(E CVD_i) = beta_0 + beta_1 * obese_i + beta_2 * age_i,

where g(x)=log(x/(1-x)).

So, what do I do after I have computed betas? (And why?)

TIA for any hints :)


New Member
Dear Hlsmith,

Thank you for your reply! I would like to compute age-adjusted estimates of proportion of CVD-positive subjects in two groups: obese and non-obese. Since I am not sure what they mean by this, hence my question: what exatctly it is?

When you say "to exponentiate the coefficient", would you be very spicific, please?

Thank you! - R


Not a robit
Given the information you provided and the output of a basic logistic regression model:

-all you should have to do is exponentiate the coefficient for obesity to get the odds ratio of obese status compared to a non-obese status for CVD, while controlling for age. You would only have one estimate, since obesity is a binary variable and one group has to be the reference group.

Not sure if you are looking for something else or not, difficult to tell given you description.


New Member
Dear Hlsmith,

Thank you very much, it does make sense to me and it is really very helpful! At the same time I was, in fact, looking to compute adjusted proportion estimate (not the odds ratio). Based on odds ratio we cannot compute the proportions, can we?

Example: in Canadian Inst. for Health Informatics they fit a logistic regression then modify the unadjusted QI results as if each facility served a standard reference population, that is replacing the values of a covariate on the right-hand side of the logistic regression model with an average, as in their publication "What Does 'Adjusted' Mean? A Demonstration of Quality Indicator Calculation in Nursing Homes".

In my example this might mean replacing age on the right-hand-side with an average age over the entire dataset. Yet another source suggested weighted averages. How would you comment on this?