Adjusting to Risk Adjustment

Removing loss vs. understanding variation

Ordinary manufacturing takes uniform inputs through a uniform process to a uniform outcome. Healthcare is the opposite: varied patients go through varied treatment with a uniform good outcome still the goal. Population risk-adjustment models try to tame that by grouping people to balance financial risk. Successive generations have improved, but current models still account for only 10–15% of the variation. That may be adequate for underwriting, but with 80%+ unaccounted for, providers are rightly skeptical that these models control for the risk they’re being asked to accept.

The opportunity is in the variation

Improving care and value takes more than reducing the chance of loss; it takes understanding the uncertainty. Areas of greater variation are opportunities to learn, to match the level of care to the patient’s real severity. A model should encapsulate your best current understanding and point at where it fails, because that’s where the next improvement is.

Choose informative outcomes, not convenient ones

A big source of noise is choosing convenient outcomes over informative ones. Annual expenditure is convenient for contracting but a poor measure where one year is too short to capture the course of an illness. Consider a population with a 25% annual chance of hospitalization: those hospitalized cost roughly an order of magnitude more, and over four years most of the group takes its turn distorting the mean. Modeling the risk of hospitalization (or re-hospitalization) over a longer horizon tells you far more than a single year of cost.

The opportunity

We’ll keep reducing the chance of loss, that’s the heart of traditional risk adjustment. But gains in value, health state, and accountability come from understanding the variation in patient needs. Smarter models (with better outcomes, time horizons, and predictors that reveal that variation) turn uncertainty into improved care at lower cost. That’s the real competitive edge.