Hello,

I'm a relatively new Stata user and am working on a project. I'm looking at length of stay (days) which is heavily right skewed and analgesic usage (days), the data is clustered within hospitals so I'm using fixed effects modeling.

Code:
 xtmixed  log_LOS ty_iv_usage opioid_usage keto_usage age_year sex ib1.race ib2.ethnicity ib0.insurance open perf year  ib3.region || hospital_number :, mle variance nostderr
I think this is a fair approach, however the interpretation is not as intuitive as being able to say that change in a drugs usage increases LOS by X days. I'm curious about potentially using Duan's smearing to retransform the coefficient as a potential way to make the interpretation more audience friendly.

One of the issues is the clustering within hospitals, which to me adds a layer of complexity.

Another thought was to leave LOS untransformed and run a median (quantile) regression with clustered bootstrapping.

Code:
bootstrap, cluster(hospital_number) reps(100) seed(5) : qreg length_of_stay post_ty_iv_usage post_opioid_usage post_keto_usage age_year sex ib1.race ib2.ethnicity ib0.insurance open perf year  ib3.region, quantile(.5)
This seems like a reasonable approach to being able to interpret the results as a change in drug use results in a median increase in LOS by X. However, I have not used the boostrap clustered command in Stata and want to verify that it is accurately accounting for patients being clustered within hospitals.

Hopefully I've given any readers enough information. Any suggestions/advice is welcome.

Thank you,