Hello, I am trying to assess the number of complications by doctor experience - defined as the number of patients reviewed by the doctor each year. I would like to perofrm a mixed effects multilevel generalised linear model.
Two variables:
DoctorId variable_id --> 1-400,
Complications ---> binary 0 (no complication) 1(complication)
Patient level factors: age, comordbiditystatus
Question 1 - with regards to the distribution - i'm not sure if i SHOULD be selecting Binomial or poison.
The conditions for poisson seem to be satisfied so- and so - see below
1. Events independent of each other (yes if patient had a complication Y or No)
2. Rate at which event can occur is constant ? i THINK SO
3. Probability of an event occuring is proportion to the interval - Could be...
Any thoughts?
Question 2: Melogit vs glamm
On one webinar they recommended to use glamm, howeverbrowsing on statalist it seems melogit can give equal results.
Is there much of a difference between melogit vs glamm ? Apart from speed on results?
Question 3: Some doctors may review patients in private or public sector (binary variable: institution).
Do you think I should include this as crossed effect...as it is only binary variable
Crossed: institution##DoctorId
Question 4: I wonder if I could use alternatively a Poisson regression model as a generalised linear model, using -glm-.... vce(surgeonid)
But I have been advised to use a multilevel generalised linear model....
Not sure what the best method to go for (poisson is slightly easier for me)
Two variables:
DoctorId variable_id --> 1-400,
Complications ---> binary 0 (no complication) 1(complication)
Patient level factors: age, comordbiditystatus
Question 1 - with regards to the distribution - i'm not sure if i SHOULD be selecting Binomial or poison.
The conditions for poisson seem to be satisfied so- and so - see below
1. Events independent of each other (yes if patient had a complication Y or No)
2. Rate at which event can occur is constant ? i THINK SO
3. Probability of an event occuring is proportion to the interval - Could be...
Any thoughts?
Question 2: Melogit vs glamm
On one webinar they recommended to use glamm, howeverbrowsing on statalist it seems melogit can give equal results.
Is there much of a difference between melogit vs glamm ? Apart from speed on results?
Code:
melogit complications age comordbidtiystatus || DoctorId, family(binomial) vce(robust) or level(95) //random effect based on doctor
Question 3: Some doctors may review patients in private or public sector (binary variable: institution).
Do you think I should include this as crossed effect...as it is only binary variable
Crossed: institution##DoctorId
Question 4: I wonder if I could use alternatively a Poisson regression model as a generalised linear model, using -glm-.... vce(surgeonid)
But I have been advised to use a multilevel generalised linear model....
Not sure what the best method to go for (poisson is slightly easier for me)
