Individualized risk adjustment and IBNR

This is a follow-up from yesterday’s post on the pragmatic barriers to effective and equitable individualized risk adjusted subsidies in a risk-rated individual market. There is another major problem, claims and diagnoses don’t come in instantly. Individual providers/hospitals/physician groups do not have a universal timeline to file claims. My kids’ pediatrician files claims within two days and gets paid within ten days of the service.  My PCP files claims on the 1st and 15th of the month.  My wife has one provider who files weekly.  The other provider has a claim from several months ago that has yet to hit payables.

In 2013,as CMS was setting up the ACO model, this is what a consultant stated typical Medicare claims completion would look like:

The longer the run out period, the more accurate the data will be, which is why CMS waits until later in the year following the measurement period in which to determine shared savings.  A shorter run out period could possibly indicate higher savings than were achieved in actuality.

According to CMS’s proposed ACO rule, based on an analysis of historical claims trends, a 3-month run-out typically results in an approximate completion rate of 98.5% for Part B services and 98% for Part A services.  In comparison, a 6-month run-out would result in an approximate completion rate of 99.5% for Part B services and 99% for Part A services.

Zero-months run out data is useless as less than half the claims are normally in.  1 month of run-out points an insurer in the right direction and three months of run out is pretty damn good.  Data stabilizes at almost full completion by twelve months.  There are odd cases where claims that are for services performed a year or more in the past will not be finalized but most of the time, those are isolated and odd events.

Insurers can handle run-out data gaps because they have millions of claims to work with, they have long periods of coherent data and they have actuaries and statistical analysts who spend all day, every day building and running models that produce reasonably accurate approximations that can be updated as new payable cycles eat claims from the previous months.  A good analytics team should be able to look at their claim universe and project with a reasonable degree of accuracy the projected incremental risk adjustment value that an incomplete month will ultimately produce.  That projection will get better as more run-out occurs, but the combination of expertise, consistent patterns and huge data sets allows for a useful prediction.

Individuals don’t have that luxury of projecting their incurred but not reported (IBNR) claims for their subsidy. If they received a new qualifying diagnosis that triggers an incremental subsidy increase in November and the claim is not processed until January, they are out of luck if they have to make a policy decision during Open Enrollment that ends in the middle of December.  Their actual subsidy will be significantly less than their entitled subsidy while insurers can use the provider notes or medical history to underwrite against them.

A possible solution would be to have a condition where insurers could not underwrite against the last six months of medical history while individuals waited for their claims to complete and an updated and accurate risk score is calculated.  This means that for a January 1 start date, the risk score would be valid as of July 1 and underwriting could only be done against medical history as of July 1.  This starts getting really messy and there will still be edge cases whose claims are not being completed with six months of run-out.  There will still be strong incentives for insurers to find ways to tilt their product and marketing profiles to avoid high cost conditions that were incurred after the data freeze date.