Should I use a random effects model or a sandwich estimator to adjust for inter site variability?

I am looking at the association between community (as defined by zip code of a health clinic) and hospitalization rate. I am afraid that if I used a random effects estimator for each clinic, that would violate the assumption that the individual clinic heterogeneity is uncorrelated with the independent variable.
Can I use a sandwich estimator instead?