We have the following problem:
We developed what we call a partition roll-up model where subjects are partitioned into subgroups based on hospital diagnostic codes. We then roll-up these partitions in such a way to combine subgroups that are highly enriched for readmission.
In other words, the hospital has 16% readmitted and our model allows us to identify 20% of all patients with a 50% readmission rate. This allows us to intervene on a small number of patients and increase our chance of impacting readmission.
PROBLEM: How do we measure the standard error or some other error around a partition of patients? We don’t want to use Rand since we are bootstrapping several 1000 instances (unless there is a generalized rand index that works with more than 2 partitions).
Any suggestions are welcome. Feel free to email or call (314-704-8725)
Bill Shannon, PhD, MBA
Professor of Biostatistics in Medicine
Washington University School of Medicine
Director, Biostatistics Center