Modeling number of days in a clinic

Hello guys,

I wish to model the number of days patients stayed in a clinic. Some of my predictors included: type of injury/disease, age, insurance, etc. Since days are count data and my data is over-dispersed, negative binomial models seem to be the appropriate technique. However, I know that when using count models the exposure time should be fixed (ex. days absent in a school during a year) or at least control for the exposure time (in Stata you can control this by adding the exposure option). Below are my questions.

1. Is it valid to model days using negative binomial models?
2. How can I control for the exposure time? Is this necessary?
3. Can you recommend good reads that cover this topic, using number of days as the dependent variable?
4. Any other ideas?

Thank you in advance,


Omega Contributor
Can you just look at admission date and time as well as discharge time and date, then model LOS as a continuous variable?
What do you mean? subtract discharged date/time minus admission date/time? I have this variable as the dependent variable? I did this by round to days. ex. some patients have 3, 6, 90 days in the clinic.
Can you elaborate?


Omega Contributor
What I was getting at was formatting LOS as continuous and using a linear regression model. Plus if mean LOS is around 8 days it approximates normal.

Why is exposure varying. So exposure is not well defined ( meaning there are more than one version, say doses)?

Also, you may want to log LOS if positively skewed. I have a feeling our vocational roles may overlap at times. I ran a model with logLOS as DV, and lost no sleep over it.