Individual risk adjustment issues

The Healthcare Blog has a great interview with Mark Pauly.   One part made me laugh, all of it made me think long and hard through multiple states on my drive back to North Carolina.   There is one part that left me scratching my head that I want to work through a bit here.

Here is the piece that has me scratching my head.  Dr. Pauly is arguing that community rating is an effective but very inefficient way to make sure that high cost individuals get the effective coverage that they need.

What’s your prescription so that the high-risk aren’t financially fleeced?

MVP: If insurers are permitted to risk rate, the premiums for the healthy will fall and more healthy people will voluntarily buy insurance – let’s agree that’s a healthy outcome. The premiums for the high-risk will surely rise. My proposal– let the insurers risk rate but let us subsidize the premiums for the high-risk.

He clarifies later on that by risk, he means chronic condition risk.  There are two ways this can be done.  The first is to carve a set of chronic, high cost conditions out of the general risk pool and into a high cost, high risk pool.  That high cost, high risk pool would then need to be financed by general taxation so that the people with chronic conditions are not decimated by only high cost people paying for each others’ care.  This would be 5% to 10% of the total covered population and forty to fifty percent of the total costs.  There are other high cost individuals but they are transient high cost individuals due to cats trying to assassinate them, four corner bed impalement or other one-off events.  John Cole is personally a chronic risk.  John Cole is statistically not a chronic risk.

That is one way to do what Dr. Pauly suggests.  The other is to do a risk adjusted individual subsidy.  This is where I scratch my head.  It makes elegant theoretical sense but the implementation hurdles are large.

I spent the last three years at my previous job working on risk adjustment.  Risk adjustment is needed and it is fundamentally ugly.

Most current risk adjustment systems rely on medical and/or pharmacy claims within a defined period of time to create a creditable diagnosis category.  If a claim with a particular diagnosis is not submitted in the current study period, the risk adjustment system will drop the category from the risk score.    One of the last projects I worked on before I left UPMC was looking through procedure code data to identify high cost non-reversible, non-curative surgeries that should be coded every year but were not coded.  A fairly significant number of cases were found.  These cases were worth quite a bit of money.

For instance, an individual can have a status diagnosis Z94.1 (heart transplant status) in 2012, 2013, 2014 and 2015 for a transplant that happened in 2011. In 2016, they got a new PCP and while a history of heart transplant was placed in the notes, the diagnosis code was never entered on a claim.  Unless there is allowable retrospective chart review, the risk adjustment system will no longer consider that person to be a heart transplant recipient and they would lose the subsidy money.  Retrospective chart review is where quite a bit of the Medicare Advantage questionable coding comes in.

So this system would rely on providers to continually and correctly code for diagnoses for patients.  The risk adjustment algorithm would then bucket people into fairly wide categorical buckets where the dollar value of the bucket is usually an average of expected costs for a wide population of people with similar conditions.  There are a couple of major operational challenges to individual subsidies based on risk adjusted conditions.

The first major concern is that this relies on continual coding.  This could be somewhat addressed by carving out a subset of diagnosis and procedure codes that indicate true lifelong conditions and then marking those conditions as permanent.  Amputations, transplants, genetic diseases and the like would be good candidates for carve-outs.  The carve out codes could be up-coded based on clinical observation but they would always be part of a person’s baseline score along with age and demographics.  Secondly the conditions have to be correctly coded.  When I worked with risk adjustment, we basically assumed that anything coming out of the Emergency Room or an Urgent Care for diagnosis was useless or next to useless.  Their incentive is to treat, stabilize and move the patient along.  An ER doc who sees a patient for more than a day is in a very unusual case so they’ll code idiosyncratically.  This is how I have “gout” on my history.  

The next major operational issue is figuring out how to actually collect, store and disseminate this data.  My first thought is that CMS would need to have a national data clearinghouse with individual level category check boxes for each year.  The data feeds would be from every insurer for every line of business, every provider as they treat people who are not insured (including Direct Primary Care) and then an attestation process for international care episodes and other unusual cases.  It would need to be curated, it would need to be searchable, it would need to be a massive vaccuum of data from systems that don’t talk to each other now.  Could this be done?  Sure!  Will it be done easily or quickly?  No.

Now let’s look at an incentive problem.   If the condition has a tight range, insurer cherry picking incentives will be minimized.  However if a condition has a wide variance cherry picking incentives will be strong.  Insurers, especially insurers that are already covering an individual with a chronic condition, can use recent claims history to determine if an individual in a risk adjusted category is a profit center after risk adjustment revenue or a loss even after accounting for the increased risk adjustment.  Insurers can’t kick people off their plans but they can nudge certain people off of their plans during open enrollments.  These strategies range from dropping providers, upping the cost-sharing on certain drugs that are indicators of higher than risk adjusted average spending, not sending reminder mailers/phone calls or any number of things.  Insurers will compete to attract the members who have high risk adjusted premiums but fairly low projected costs.  Yes, guarantee issue will counteract this to some extent, perhaps 105-E will do so to, but this incentive will be real and pervasive for a Gresham Law scenario.  

The next big issue I have with this is a health equity issue.  Two individuals who live next door to each other with identical objective medical situations will get very different subsidies and thus treatment possibilities if one individual is well connected to the medical system and has doctors who aggressively and legally code every diagnosis while the other individual either does not see medical professionals on a regular basis or sees clinicians who have a propensity to code very lightly.  Well connected individuals with sophisticated skills to manipulate the medical system in order to maximize their allowed subsidy will doctor shop and there will be a small consulting industry to assist in this.  Individuals without upfront capital resources or sophisticated system navigation skills will rely on chance to get an appropriate subsidy.

These are some of the issues that I can think of when I think of individualized risk adjusted subsidies.  And until I see good answers to these concerns plus others that I can’t think of, I like the idea in theory but hate it in practice.