Dear Experts,
I want to explore if whether a patient was given a medication or not varies depending on their individual characteristics (level 1) and those of their general practice (level 2). Level 1 covariates will be for e.g. education and age while level 2 covaraites will be practice size and urban/rural location.
I anticipate that medication use will vary depending on whether it was a nurse or GP who prescribed the medication. Ideally, I would include the GPs and Nurses as level 2 nested under their practice at level 3. However, the data I am allowed to access does not include individual prescriber id's. Instead, I will just know it was a GP or Nurse who prescribed the medication. With this limited data I cannot include the GP/Nurse prescriber as a level as GP and Nurse are the only 2 options for prescribers (not a random selection of all the possible prescribers.)
Is there any way for me to include the Nurse/GP prescriber information in my model? For each practice there will be Nurses and GPs prescribing medicatons.
Using a dummy as a fixed effect for the gp/ nurse variable is the best that I think you can do.
You could them allow this fixed effect to be random across practices so that you could thereby ascertain which practices are 'high' for nurses and which 'high' for men. It is easiest to do this if you use separate coding (not the usual dummy coding) in the random part as you can then estimate the unexplained variance at the practice level for Nurse and GPs . See chapter 7 of
Book Developing multilevel models for analysing contextuality, he...
Beside a dummy for NURSE/GP prescriber in the fixed part of the model and a random coefficient both on GP and NURSE (to deal with heteroscedastic random effects between practices) in the random part as suggested, I'd also include practice-level % of prescriptions from nurses (or GPs).
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts