I have a dataset that contains 100 different patients over 5 year’s period. Every patient is examined each month with regard to particular illness and marked as healthy or ill (0 or 1). Every person appears 60 times in my sample (5 * 12 = 60).

Every month patient provides A = Average blood pressure in that month, B = Average daily exercise hours and C = Average number of Cigarettes smoked in that month.

The layout of the dataset is as follows:

ID (Unique Patient Identifier)
Month (1 to 60)
A (Average blood pressure in that month)
B (Average daily exercise hours)
C (Average number of Cigarettes smoked in that month)
Ill (Yes, No)

I was thinking of using Logistic Regression which uses information from last three months and gives a probability for patient to be flagged as Ill in next 2 months.

My problem is that logistic regression assumes that observations are independent whereas in my case they are obviously not.

What should I do? Should I use something like GEE or GLMM or something else?


You could always come up with a set of transformed variables that aggregate the data from 3 months into one observation for each patient (e.g., average blood pressure across the prior 3 months, 3-month exercise hours/cigarette, etc.). Then you have independent observations (1 per patient), and you could build the model. This kind of defeats the purpose of having monthly granularity, but you could evaluate a logistic regression like this against a legitimate longitudinal method.

| cite | improve this answer | |
  • $\begingroup$ Welcome to the site, @Chris. It is fine to answer a question years after it was asked. The actual purpose here is to build up a repository of information for posterity, not quite to help the OP themselves (that's just a byproduct), as counter-intuitive as that may seem. $\endgroup$ – gung - Reinstate Monica Jul 17 '15 at 16:22

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.