Rich data sets and risk adjustment

I had a lot of fun talking about risk adjustment yesterday. Michael Kalina is an insurance plumber who worked at a co-op and he raised a very good point.

I have long believed that legacy companies are able to use institutional knowledge to "enhance" their results, which creates asymmetries.

— Michael Kalina (@MPKalina) June 1, 2017

I agree completely with him. Deep data sets across multiple lines of business are extremely valuable if they are properly used in risk adjustment projections and risk adjustment revenue optimization.

When I left UPMC Health Plan, I was running the data mining for the Medicaid risk adjustment optimization team. We were active in most lines of business; Medicaid, Medicare Advantage, Medicare SNP, CHIP, Exchange, and employer sponsored insurance. Our optimization program required a diagnosis on a valid claim within a reference period. The data mining looked for highly probable persistent diagnoses that had shown up previously but had not yet shown up in the reference period.

The best use example is a lower leg amputation. Once a leg is gone, it is gone. My data mining looked at Medicaid claims, it looked at Medicare claims, it looked at Exchange claims, it looked at group claims. And if there was strong evidence from a 2005 group claim for a brand new to Medicaid member, a gap would be identified for the 2017 risk adjustment cycle. Similar logic played out for the Exchange and the Medicare Advantage risk adjustment teams.

New insurers would miss that risk adjustment gain. Insurers that only offered coverage in a single line of business like the individual market would miss that potential gain. Insurers with multiple lines of business but small market share in a region would miss a significant proportion of the gains from data mining.

Rich data sets are an extremely valuable asset and competitive advantage for locally significant legacy carriers.

Now what are the policy implications of this fact of life?

Good, widespread information is critical. States that have all payer claims databases should be sending identifiable diagnosis and procedure code data to all risk adjusted product lines for all carriers. This would allow new carriers and tightly focused carriers to get a significant data flow that they would otherwise not have. It would increase the probability that a high probability of being an actual gap is identified as a gap in the risk adjustment revenue chase.

Another approach would be for a state or a collection of states to keep a list of truly persistent conditions for individuals. Any condition on this list would be automatically attributed to the risk adjustment score of the insurer. A diagnosis or a prescription would not be required. For instance, a state could decide that an amputation is a truly persistent condition. If a person was identified as having a leg amputation in a previous year, there would no need for an insurer to see a Z-status code submitted in the current year to get credit. This list of truly persistent conditions would be small (amputations, organ transplant status codes, cystic fibrosis, HIV status codes etc) and it would be an additive list. This again minimizes the inherent big data advantage of long established multi-line carriers. They can still mine their data for the non-obvious but the obvious is common property.