That Care Management Savings Number Is Probably Regression to the Mean
A vendor deck says the care management program saved $2.4 million. Before anyone celebrates, ask one question: compared to whom?
If the answer is "compared to the same members' costs before the program," the number is mostly an artifact. High-cost members are selected because they're at a cost peak, and cost peaks come down on their own — hospitalizations resolve, episodes end. In one simulation shaped like a real medical population, members selected at their peak dropped roughly 20% in cost the next year with no program at all. That's not savings. That's arithmetic. Six lessons from building an instrument that tests these claims.
1. The design is the number. A before-and-after comparison cannot support a savings claim — full stop. A matched comparison group can, imperfectly. The study design isn't a methods-section detail; it's the difference between a measurement and an accident.
2. The denominator decides everything. Savings measured across only the members who engaged compares people who answered the phone with people who didn't — and phone-answerers differ in ways no adjustment removes. The denominator has to be everyone identified, including everyone never reached.
3. Build the claim from its drivers before you test it. Readmissions avoided, ED visits diverted, admissions prevented — each has a published unit cost and a published best-case effect size. The pooled evidence across 49 trials: about a 12% relative reduction in readmissions, at low certainty. A claim that needs 40% isn't ambitious; it's unpublishable. And when the stated dollars and the drivers don't reconcile, ask for the workpaper.
4. Beware the healthy-adherer effect. Medication adherence "savings" are the classic trap: adherent patients do everything else right too. The canonical demonstration — adherence to placebo predicts halved mortality. Price adherence from a causal anchor, or report it as a quality outcome, not a savings driver.
5. Care gaps closed are revenue, not medical savings. Closing gaps raises near-term utilization — more visits, more screening. That's the point of closing them. The financial value is real, but it runs through quality bonus revenue. Put it on its own line under its own name.
6. Ratios need a denominator too. "One care manager per 78 members" means nothing until you weight the mix — a caseload of complex-care members is not a caseload of well-check reminders. And a savings claim that assumes more engaged members than the staffing can actually carry is telling you something about itself.
None of this means care management doesn't work. It means the measurement is genuinely hard, the published record is sobering, and a consulting practice owes its clients the version that survives an actuary's review — not the version that survives a steering committee.
We built these tests into an instrument that takes a savings claim as given and asks what survives: design, denominator, drivers, reversion exposure, caseload arithmetic — every threshold cited to the study it came from. Browser-only; nothing entered leaves the machine.
A savings number without its design is a rumor with decimals.
What's the strongest savings evidence you've seen a program actually produce — and what design was under it?