I wonder how this program’s performance would look if they had recruited candidate members based on a predictive model of future utilization / need rather than focusing on people who were already extremely complex / high cost.
(A big part of my current PhD research involves building predictive models for patients with complex needs, so I grapple with these issues day in and day out, and often on huge scales—millions of patients. Prediction, at least in the usual supervised ML sense, just doesn't cut it for these kinds of problems—you need causal inference.)
Wouldn’t that depend on what benefit you’re trying to achieve? If you want to prevent re-admissions, wouldn’t it be better to focus on people at high risk if future re-admission vs people with high past complexity. (I assume there would be a lot of, but not perfect overlap)