I'm working on a medical problem, where I want to analyze the effect of taking cholesterol medications on the occurrence of heart attack. Once a medication with a specific dosage is prescribed, it'll be used by a patient for $k$ consecutive periods (i.e., refills periods).

Moreover, I have a retrospective data of several patients, where each patient

  1. is prescribed with different dosages,
  2. has the fixed $k$ periods of refill,
  3. may have some instances of heart attack during those $k$ periods, and
  4. has other risk factors (as confounding factors).

Now, if a patient takes a specific dosage of that medication for $k$ consecutive periods (in a prospective sense), how can I predict the risk of heart attack at any time during those $k$ periods? I guess there might be some relevance to Cox survival analysis, but I'm not quite sure. So, any comment/help is really appreciated!



First thought that comes to mind is using a binary logistic regression. You can predict probability of a heart attack happening after fitting your dataset to a logistic model.

Here is the formula for generating the probability of $Y_i = 1$

$ P(Y_i = 1|\, x_{i1},\ldots,x_{ip}) = \frac{e^{\beta_0 + \beta_1 x_{i1} + \ldots + \beta_p x_{ip}}}{1+e^{\beta_0 + \beta_1 x_{i1} + \ldots + \beta_p x_{ip} }} = \pi_i $

  • $\begingroup$ I'm familiar with the concept of logistic regression. The problem with this model is that it gives me the probability of a heart attack at a fixed time. However, as I explained above, I want a model that takes the data of $k$ periods and tells me the probability of this event happening once (or at least once) during the $k$ periods. $\endgroup$ – Alex Jul 25 '16 at 19:09
  • $\begingroup$ @Alex I might be having trouble understanding your problem or the way your dataset looks but I wonder if it would be possible to take the mean of the predicted probabilities across $k$ periods. $\endgroup$ – dylanjm Jul 25 '16 at 20:14
  • $\begingroup$ Thanks, but that does not work for my case. $\endgroup$ – Alex Jul 25 '16 at 22:39

Your Answer

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

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