Measuring variability within a system

Does anyone know of a way to measure the amount of variability within a system to create a "metric" of variability that can then be used in a regression?
Here is the problem:
When a group of physicians take care of a patient there are a lot of choices that are made based on physician preference. I would like to determine if the amount of variability between different groups of physicians is associated with outcome (i.e. does it matter if everyone does the same thing, or is it good or bad if different doctors have different preferences).
I have predefined interventions where physician preference is greatest, and a nice clean dichotomous outcome (so a final regression analysis will work fine). The interventions include continuous, dichotomous, and categorical variables and I would like to determine if there is a "metric" that can measure or represent the amount of variability within each group.
Has anyone seen anything like this? I have searched the biomedical literature and have not found anything.