The Insurer Is Paid by Diagnosis, So Diagnoses Multiply

Why Risk Adjustment Has to Exist

Let's start with the problem this whole system solves because the payment mechanism seems indefensible without it. Imagine a government that pays private health plans a set amount for each member they enroll no matter who that member actually is. Each plan would then compete to hire healthy people and stay away from sick ones. That's a much easier way to make money than competing on quality of care. And the sickest patients the exact people the plan covers.program they would become the least sought after customers in the entire market

Economists call this adverse selection. If left alone it would hollow out the entire program from within. Risk adjustment It's the solution insurers and regulators decided on: pay each plan according to the expected cost of the specific people it actually enrolls. A plan that enrolls a diabetic member with heart failure receives a substantially higher payment than a plan that enrolls a healthy retiree who plays golf three times a week. I think this part of the design is really good policy. Without it insurers would have every reason to avoid exactly the patients for whom Medicare Advantage exists

How the Payment Is Actually Calculated

In Medicare Advantage each member gets a risk scoreThe formula CMS uses groups those diagnoses into what it calls hierarchical condition categories or HCCs. Each HCC has a fixed weight the weights that apply to a given member are added together and the total becomes that member's risk score. The score then multiplies a county-level benchmark payment to establish what the plan is paid per member permonth

A member with a risk score of 1.0 generates the benchmark amount regardless of what happens in that county. A member with a risk score of 1.6 generates sixty percent more than that every month for the entire following year. Small differences in score become real money once multiplied over a year and across a large membership

That last row sums up the whole thing in a single sentence. A condition a member has doesn't actually generate income unless someone writes it down on a claim or in a medical record that the plan can refer to later. Therefore income doesn't really depend on how sick a member is. It's a function of how completely that illness was documented. Those two things typically go together. They don't have to and the gap between them is where this entire article lies

A Worked Example: What Adding a Diagnosis Does to Revenue

Let me concretize the mechanism with round numbers because a risk score is abstract until it is actually multiplied. None of the figures below are actual numbers from CMS. I choose them because they are easy to follow not because they match any specific county plan or model year

Suppose that a county benchmark payment that is the amount a plan receives for a member with an exactly average risk score is $1,000 per month. Suppose that a particular 72-year-old member's demographic factors that is his or her age sex and Medicaid eligibility status put his or her initial risk score at 0.80 before any diagnosis is counted. With that score his or her plan receives a payment of 1,000 times0.80 that is $800 per month which is equivalent to $9,600 for a full year

Now let's assume that this member actually has type 2 diabetes with a documented complication congestive heart failure and major depressive disorder three conditions that are common in a Medicare-age population and entirely real if a doctor bothers to look for them. Call the illustrative weights for those three conditions 0.15 0.20 and 0.25. If a chart review or home assessment finds and documents all three his risk score becomes0.80 plus 0.15 plus 0.20 plus 0.25 or 1.40

With a score of 1.40 the same plan now receives 1,000 times 1.40 or $1,400 a month which equals $16,800 a year. That's a $7,200-a-year increase in income generated entirely by documentation for a member whose actual health didn't change at all between the two figures. He had diabetes heart failure and depression before anyone wrote them down. The only thing that moved was thepaperwork

Expand on that and the arithmetic gets interesting quickly. Suppose a plan has 100,000 members and such a documentation push reaches only 10 percent of them or 10,000 people each generating a comparable annual increase of $7,200. That's 10,000 times 7,200 or $72,000,000 a year in additional income without spending oran extra dollar in treatment and without any member getting sicker. I want to be clear that this does not mean that the underlying diagnoses are false. Depression and heart failure could be completely real. What the arithmetic shows is how much income depends on the act of finding and recording a diagnosis regardless of anything that happens to the patient afterwards

The Predictable Response

Given all this it is not surprising that plans have invested heavily in ensuring that conditions are actually documented. There are a few main channels they use

In home health assessments. sending a doctor to a member's residence often once a year and often at no cost to the member for a comprehensive examination that may reveal conditions that no one had noted before. Retrospective Chart Review is different. It employs coders not physicians to review existing medical records for documented conditions that were never actually presented in a claim. Provider education and incentive programs push physicians to code with the highest level of specificity that their own existing records already support

Each of these has a completely defensible version. Undiagnosed conditions in an elderly population are real and common. Home visits detect genuine problems that a rushed office visit would miss. Doctors actually do it with little documentation because coding to the highest level of specificity requires time they don't always have. But each of these channels also has a version where the activity produces a payment and nothing more

Case Study: HealthCare Partners and the One Directional Review

The clearest real-world example of where this can go wrong predates most of the recent headlines about Medicare Advantage. It involves HealthCare Partners a large California medical group that DaVita later acquired and renamed DaVita Medical Group

A former employee named James Swoben filed a whistleblower lawsuit alleging that HealthCare Partners performed what has since become known in this industry as a one-way chart review. The company's coders reviewed medical records for documented conditions that had not been submitted for payment which is exactly the legitimate retrospective review process described above. But according to the allegations when those same reviews revealed filed diagnoses that were not actually supported by the record that is conditions that should have been removed because they lowered the chart score.risk the company did not act on those findings the same way it acted on those that added revenue. The diagnoses flowed in only one direction. In not out

The government's theory which is the same theory that has appeared in most risk adjustment applications since then is that a review process created to only add codes that increase revenue and never fix ones that decrease revenue is not really a documentation quality program. It's a one-way filter with a dollar sign attached to it. HealthCare Partners reached a settlement with the Department of Justice worth about $270 million to resolve the allegations. Settlements like this generally do not involve admissions of liability andThat applies in this case too. I'm not claiming that the underlying claims have been proven in court because a settlement is not a verdict. What the case is useful for regardless of how the specific allegations are weighed is to show exactly what regulators mean when they draw the line between a documentation program and an encryption program. The mechanism the whistleblower described a review process that runs in one direction is the same mechanism worth paying attention to in any plan's documentation strategy.current

What Oversight Has Found

Government auditors primarily the HHS Office of Inspector General and the Government Accountability Office have examined this pattern repeatedly and their reports continue to reach a similar finding. A significant proportion of diagnoses that generated payment came solely from a health risk assessment or medical record review with no other clinical services provided to that member throughout the year. Certain categories of conditions also appear much more frequently in Medicare Advantage than in traditional fee-for-service Medicare for populations that should be seensimilar on paper

This was followed by law enforcement primarily under the False Claims Act based on the theory that submitting a diagnosis code that the plan knows is not supported by the medical record is tantamount to billing the government for something it did not actually provide. Several large settlements and a number of ongoing cases involve exactly this allegation and HealthCare Partners is one of the first. The legal question in these cases is typically not whether a documentation program is allowed to run. It is allowed. The more specific question is whether a specific code submittedwas actually supported by the record and whether the plan had reason to know that it was not and acted on that knowledge anyway

The Audit Mechanism and Why It Was Contested

CMS verifies this through what is called risk adjustment data validation audit or RADV. Auditors take a sample of a plan's members request the medical records behind the submitted diagnosis codes and check to see if the documentation actually supports what was billed. Pretty straightforward so far

The contested part is extrapolation that is taking the error rate found in that sample and applying it to all plan members. A small sample finding say an error rate measured on a few hundred records can become a payment obligation covering hundreds of thousands of members once it is extrapolated to the entire book. CMS finalized rules in 2023 that allow this type of extrapolation for payment years going back to 2018 and the industryThe gap between recovering errors found in a sample and extrapolating that error rate to an entire membership is worth billions of dollars industry-wide and would always end up in court

Where the Insurer's Side of This Is Genuinely Strong

So far I've laid out the arguments against aggressive documentation pretty clearly so let me make the case for the other side because it deserves a real hearing rather than a token paragraph

Start with the most basic fact. Medicare Advantage members are on the whole not much different in terms of underlying health from fee-for-service Medicare beneficiaries and treating a genuinely sicker population actually costs more money. If a plan enrolls a real diabetic with real heart failure that member will generate real medical claims and the plan needs real revenue to cover them. No one not even the harshest critics of Medicare Advantage argues that risk adjustment itself should go away. The real debate revolves entirely around whetherwhat proportion of the measured difference in risk scores is due to genuine disease versus applied coding effort

Second traditional fee-for-service Medicare has its own problem of insufficient documentation and I think this part is not valued enough. You pay a fee for the office visit to the on-duty doctor regardless of how many diagnoses you write down so you have almost no financial reason to code every condition a patient has with the highest level of specificity. A patient with diabetes and early-stage kidney disease could simply be coded as diabetic in a fee-for-service setting because coding the kidney complication more accuratelyit doesn't change what the doctor is paid that day. Comparisons between Medicare Advantage risk scores and fee-for-service risk scores for supposedly similar populations compare a system with a coding incentive to a system that has almost none and part of the resulting gap is simply fee-for-service under coding that appears as if it were a Medicare Advantage difference

Third health risk assessments actually find real previously undiagnosed conditions in older patients and I don't think this should be dismissed as a technicality. A home visit that identifies undiagnosed diabetic neuropathy or an early-stage kidney problem is doing something clinically useful even if the plan never bills for a single dollar of additional treatment that same year. Finding a condition is itself valuable information for that patient's future care. Require that treatment be performed inThe same twelve-month period in which the diagnosis was made is I admit a somewhat arbitrary barrier to judging whether a diagnosis was legitimate. An actual fraction of the gap in coding intensity that appears in regulators' reports is genuine clinical discovery not fabricated revenue. Honestly my problem is that I don't know how to clearly separate that fraction from the rest and I don't think anyone else does either

Reading the Sector

For anyone who actually analyzes these companies some revelations carry more weight than the headline numbers. Year-over-year risk score growth that runs well above the rate at which the underlying population ages is the first sign worth watching. It doesn't prove anything on its own but it raises the obvious question of whether the increase reflects better documentation rather than genuinely sicker membership

the medical loss ratio that is the proportion of premium money that is actually spent on member care is the natural counterbalance to consider. A plan that increases revenue through coding intensity without an equivalent increase in actual care costs will eventually show a declining medical loss rate because revenue grows faster than claims. The reserves set aside for government audits and disclosed investigations indicate how the company itself is pricing its own tail risk. And changes to the risk model themselves matterMore than most investors realize. Regulators periodically eliminate categories of conditions that are considered too easy to influence through documentation alone and a change like that can shift revenue to an entire industry in a single model year without a single plan changing the way it actually operates on a day-to-day basis

How I'd Actually Read a Risk Adjustment Disclosure

If I were actually subscribing to one of these companies instead of just reading up on the mechanism this is the order I would work through it and it's different from how most retail investors seem to approach it

I'd start with risk score growth relative to the plan's historical trend not relative to some industry average because different plans serve really different populations and company-to-company comparison muddies the read. If a plan's risk scores have been rising 2 to 3 percent annually for a decade and suddenly jump to 6 or 7 percent that's the kind of change I'd like an explanation for before trusting the earnings growth behind it.behind him

I would then go directly to the trend of the medical loss rate over the same period because that is the check on the history of the risk score. If revenue per member is rising thanks to risk score growth and the medical loss rate is falling at the same time that combination is the one I find most difficult to explain as a coincidence. Treating the sickest members should cost more. A rising risk score combined with a falling cost ratio is a pattern that at the very least deserves a closer look even ifit is ultimately explained by something benign elsewhere in the business

I also carefully read the legal procedures section of the 10K rather than skimming it because that's where a company has to describe its own exposure to the RADV audit and any ongoing investigation in its own words under a legal obligation to be accurate. My honest reading is that a company being investigated is not automatically doing anything wrong. Investigations of large Medicare Advantage plans have become almost routine in this industry and in fact I would be more suspicious of a company without disclosed regulatory scrutiny in this area than one that managesa competently active RADV audit. What I see is a change in language: a company going from describing an audit as immaterial to reserving a specific reservation against it because that change usually means that its own lawyers think the exposure became real

I was wrong the first time I looked at the Medicare Advantage 10K for what it's worth. I consider a rising risk score to be good news since revenue growth is revenue growth and it took me a while to realize that in this specific business the source of the growth matters almost as much as the number itself. A dollar of premium coming from a genuinely sicker population and a dollar of premium coming from more extensive coding look identical on the income statement. They are not the same dollar if youtries to determine if the growth is lasting

The Bottom Line

Risk adjustment exists for a genuinely good reason. Without it insurers would compete to avoid the sick patients that Medicare Advantage is supposed to cover and the program would fail in its real purpose. That part is not up for discussion and it shouldn't be. What the program bought into however is an incentive structure in which a plan's revenue depends on a variable over which the plan itself has enormous influence: how thoroughly a member's conditions are noted. The worked example above shows how mechanical it is.actually that leverage. A few documented conditions can move annual revenue per member by thousands of dollars while the patient's underlying health remains exactly the same. The HealthCare Partners case shows what regulators believe crosses the line: a review process that only adds revenue and never subtracts from it. And the counterargument is equally important. Part of the measured gap between Medicare Advantage risk scores and fee-for-service is a real disease that fee-for-service simply never had a financial reason to document in the first place.place.Determining what percentage of any individual plan's risk score growth is legitimate discovery and what percentage is coding pressure is honestly one of the most difficult analytical problems in this industry and I don't think there is a clear formula for it. The closest thing to one is to look at how the medical loss ratio moves compared to the risk score and read legal disclosures as if they were written by people trying to tell you something true in the driest language possible because they usually are