Variance within a single risk adjustment category

This is going to be a very geeky post building on a few previous Exchange  risk adjustment  and strategy posts.  

The trend on Exchange is for low cost insurers with either Medicaid plus rates or Medicare-like rates to dominate the membership, and more importantly, dominate the healthy membership counts.  The question is how do other insurers compete.  The easiest way of competing is narrowing networks, lowering provider payments and exerting more explicit and implicit care veto points and systems of No on the patient.  The other approach is to redefine the problem and not attempt to compete on price.  Instead, compete on quality with a higher premium where enrollment will be low, enrollment will be sicker than average and the revenue from the premiums will be supplemented by net risk adjustment inflow.

Risk adjustment based strategies are a plausible path forward. High cost insurers could offer disease specific plans and make their money by offering good chronic disease control while getting risk adjustment inflows from the low cost plans. The problem with this strategy is the risk adjustment inflow is calculated based on average premiums in the region. The low cost plans bring down regional average premium which means the risk adjustment transfer payment does not fully compensate the high cost plan’s provider reimbursement.

The Exchange risk adjustment model is a modified Medicare HCC model.  It is diagnosis based.  An insurer who receives a valid claim with a valid diagnosis for a given category gets credit for the entire HCC category.  The category credit is an average expected incremental relative cost of treating that condition for a year.  Over a large enough population, the category credit is close enough.

But as you can see there is a problem.  That problem is variance.  Some people with a condition are much cheaper to treat than others with the same HCC profile.  That means there is an opportunity for an aggressive data mining low cost insurers to cherry pick the healthier sick people in order to minimize their net risk adjustment transfers

Let’s work through an example.

Hemophilia (HHS_HCC066_14 ) has a risk adjustment factor of roughly 46 (varies by metal band (p.106)).  That means on average, an individual with hemophilia diagnosis history will cost an insurer roughly 46 times the average state premium to treat compared to the general population.  One hemophiliac, from a risk adjustment point of view, is worth 38 diabetics.

The risk adjustment factor of 46 can be broken into two components.  The first is  prophylactic  treatment in order to minimize internal damage and minimize the number and severity of crisis bleeds.  This is a cost that will be fairly common across any population of hemophiliacs.  It is not an exploitable data hook for an insurance company with deep claims data.

The second major component of the 46 factor is the crisis/emergency treatment.  A major  bleed can very quickly run up multi-million dollar claims.

Major bleeds are rare.  Most individuals with hemophilia are able to keep their condition under control in a given year.  They use up significantly less than a factor of 46 of average premiums for their own care.  The few individuals who have a major bleed have individual costs that are a factor 100, 200, 300 of the average premium.

So what does an insurer do to exploit risk adjustment by exploiting in group cost variance?

If an insurer has a deep claims history on a population of individuals with hemophilia, they can figure out which individuals are medication compliant, which individuals are treatment compliant and then throw in some special sauce to create predictive systems to identify which individuals are highly likely to be well maintained in the next policy year and which individuals have a significant risk factor of having a major bleed.  From there, policies can be crafted either by network design, benefit configuration (coagulating factors with no cost sharing etc), marketing that aim to grab a disproportionate share of low cost hemophiliacs while shifting the highly likely to be high cost individuals to someone else.

This would work because the cherry picking insurer gets a full factor 46 credit for each individual even if their expected average costs according to internal calculations would be a factor 33.

This type of strategy is probably first viable for the 2018 plan year as that is when the insurers will have enough data to make good predictions and enough time to actually model out those types of interactions for hemophilia.  This type of strategy is probably in play for low cost, common chronic conditions (like diabetes where advertising for diabetes specialty plans could be focused at the gym or at farmers markets etc) .