Thinking through risk adjustment and selection

I’ve been reading risk adjustment papers this week. Some of them are technical reports while most of them are analysis of population health programs that show that there is significant selection occurring in voluntary programs because the business case incentives lean towards game playing instead of taking care of people. I want to do some writing to think through the issue. All numbers are completely made up. The graph below is for a hypothetical disease where costs vary between $50,000 and $250,000 with a uniform probability of any one individuals’ costs being anywhere in that range. All of these assumptions are pragmatically problematic, but they allow for simple toy thought models to be constructed.

Let’s assume that there is a population health model in place for these 100 people. The model is voluntary and it pays average costs (~$159,000).

Let us also assume that the potential risk bearers (individual doctors, physician groups, hospital systems, insurers etc) have some ability to fairly cheaply and reasonably accurately and fully legally guess whether or not the person that they could voluntarily treat will be in one of two categories: CHEAP or EXPENSIVE.

The screening protocol is not perfect as someone who is expected to be CHEAP could be wicked EXPENSIVE but it is pretty good.

So what happens?

The model is cream-skimmed to death.

The business case here is really straightforward — the only people who are treated in the model that pays ~$159,000 are people who are expected to be CHEAP.  This leads to very profitable patient selection as the average payment is way above average actual costs within the model.  People who are expected to be EXPENSIVE are only going to have a positive business case for treatment within the model when the preliminary screening program wrongly classifies them as CHEAP instead of their actual state of  EXPENSIVE.  These individuals will get their care either outside of the model or not get care at all.

Why does this matter?

We’re moving towards more population health bundles even in the context of significant individual level variation of health status and expected costs even after group risk stratification.  If we’re not risk adjusting bundles to appropriately align payment with roughly expected average costs for an individual then we’re creating strong incentives for risk bearing clinical groups to figure out ways to identify CHEAP vs EXPENSIVE individuals and target only CHEAP individuals within the bundle payment mechanism.