What analytic approach would folk recommend for analyzing the difference between hospital readmission events in two groups? Exposed vs. non-exposed groups - data available at patient level from multiple hospitals - some patients will be discharged and readmitted several times (I would like to capture this if possible after controlling for risk) - data on risk for readmission available (age, gender, multiple chronic conditions, etc.).
I would suggest as a first approach to do a linear multiple regression of no. of readmissions (of individual patient) against coded individual risk factor levels. Do for each hospital seperately, and also pooled. you could then include a hospital code as an independant variable to compare hospitals..
Repeated time to event analysis with censoring would seem to me be a possible option.
I respectfully disagree with James Leigh. Analyzing the "number of readmissions" in a linear regression is inappropriate unless you have EXACTLY equal follow-up time. I see two viable options:
1) As noted by Timothy Goggin, you could perform repeated time-to-event analysis, described nicely here:
http://www.ats.ucla.edu/stat/sas/faq/survival_repeated_events.htm
2) The second option would be a poisson or negative-binomial regression model, using the count of readmissions with an offset term for follow-up time.
I would prefer the first option, but the second could also work. Both of these would allow assessment of risk in the exposed vs. unexposed group with adjustment for confounding factors. Better still, you might also be able to account for changes in exposure status (analyzing the periods of time "exposed" and "unexposed" for each patient).
- Badges
- Science topic
- Similar topics
- Chemistry