Come for the politics, stay for the snark.

People really shouldn’t expect the government to help after they watched the GOP drown it in a bathtub.

Damn right I heard that as a threat.

Speaking of republicans, is there a way for a political party to declare intellectual bankruptcy?

Not so fun when the rabbit gets the gun, is it?

American history and black history cannot be separated.

… pundit janitors mopping up after the gop

“Perhaps I should have considered other options.” (head-desk)

I desperately hope that, yet again, i am wrong.

At some point, the ability to learn is a factor of character, not IQ.

The rest of the comments were smacking Boebert like she was a piñata.

This must be what justice looks like, not vengeful, just peaceful exuberance.

One of our two political parties is a cult whose leader admires Vladimir Putin.

Human rights are not a matter of opinion!

We still have time to mess this up!

Many life forms that would benefit from greater intelligence, sadly, do not have it.

You know it’s bad when the Project 2025 people have to create training videos on “How To Be Normal”.

Following reporting rules is only for the little people, apparently.

Washington Post Catch and Kill, not noticeably better than the Enquirer’s.

Everything is totally normal and fine!!!

Proof that we need a blogger ethics panel.

The next time the wall street journal editorial board speaks the truth will be the first.

Republicans: The threats are dire, but my tickets are non-refundable!

The republican ‘Pastor’ of the House is an odious authoritarian little creep.

They are not red states to be hated; they are voter suppression states to be fixed.

You are here: Home / Anderson On Health Insurance / Bad coding and adverse selection

Bad coding and adverse selection

by | 29 Comments

This post is in: 

The Wall Street Journal has an excellent article on risk adjustment in Medicare Advantage. The main thrust is that insurers create/invent codes that pay at an ungodly high rate. There is one paragraph that leaped out at me as I was reading as I was thinking about ACA risk adjustment:

About 18,000 Medicare Advantage recipients had insurer-driven diagnoses of HIV, the virus that causes AIDS, but weren’t receiving treatment for the virus from doctors, between 2018 and 2021, the data showed. Each HIV diagnosis generates about $3,000 a year in added payments to insurers.
Everyone with HIV should be on antiretroviral drugs, the only effective treatment, and nearly all Medicare patients whose doctors diagnosed the virus took the drugs. Less than 17% of patients with insurer-driven HIV diagnoses were on them, the Journal found.

Medicare Advantage risk adjustment is “widget” risk adjustment where the insurer gets paid from the federal government for each new diagnosis code generated while ACA risk adjustment is zero-sum where one insurer that codes more aggressively gets paid (or reduces their payments) from other insurers that don’t code as hard. However, both systems derive the value of a diagnosis code group in a similar way. CMS runs a massive set of regressions where total costs are a function of patient demographics and plan type plus a set of binary indicators for disease groups and their interactions where the coefficient for the disease group is the incremental total cost for people with that disease condition, holding everything else constant. In this case, HIV has an incremental extra cost of “about $3000” for Medicare.

That “about $3000” is a combination of people who got diagnosed by their doctors and are getting treatments that cost money AND some people with “insurer-driven” diagnoses that are getting treated with the appropriate drugs AND a lot of people with “insurer-driven” diagnoses that are not getting the appropriate drug and whose actual costs due to HIV are zeros. The lack of drugs is likely some combination of care management failures as people aren’t connected to their docs and people who legitimately don’t need these drugs.

The about $3000 risk adjustment value is a weighted average of doc diagnosis and treated patients and their associated costs and insurer driven diagnoses and their much lighter (on average) treatment costs. This is really problematic in the ACA if we assume that distribution of “insurer-driven” diagnoses for HIV (and likely other diseases) are not random nor uniform. If we think that these insurer driven diagnoses are concentrated in low risk insurers, the zero sum nature of ACA risk adjustment is impacted in two ways. In the first year, it increases the reported risk score of low risk insurers relative to high risk insurers. This reduces the amount of money that flows to high risk insurers that are paying for treatment of patients who were diagnosed by their doctors. Over a multi-year period, the “value” of HIV is depressed as there are a lot of zero incremental cost patients being added to the pool.

Systemically, if this scenario is occurring in the ACA, it makes the business case for insurers that attract and pay-for actual high cost and high risk enrollees far harder. If this is happening, the logical response from high risk insurers is to find ways to either code ever more intensively/creatively OR to tilt their product offerings to be less attractive to actual high cost patients such as narrowing networks, or adding extra targeted cost-sharing or increasing prior-authorization requirements.

Reader Interactions

29Comments

  1. 1.

    RepubAnon

    Insurer-driven coding sure does sound like some type of fraud. But it’s not related to Joe Biden, so Pam  Bondi won’t be interested.

  2. 2.

    AM in NC

    How do we kill Medicare Advantages Insurance Companies at the Expense of the Rest of Us?

  3. 3.

    Steve in the ATL

    I have yet to achieve my dream of being smart enough to post a substantive comment on one these threads!

  4. 4.

    p.a.

    My retiree plan is a MA PPO but the union/Verizon contract has for the most part kept it pretty subscriber-friendly so far.  It’s United Health🤢 and they have tried the “visiting nurse” scam that is just an attempt to generate money-making codes.

  5. 5.

    Another Scott

    I believe it’s dsquared (Dan Davies) whose mantra is “a system is what it does”.  IOW, it optimizes itself based on the rewards present.

    Good teachers warn about the dangers of “teaching to the test”.

    Similarly, TheRegister:

    The technique could improve other types of agents, not just those that write code.

    “The beauty of this framework, powered by code and open-ended exploration, lies in its generality,” said Zhang. “If progress can be measured and the medium is code, the Darwin Gödel Machine can optimize for any benchmark. Whether it is coding ability, energy efficiency, or another domain, the system can adapt by using that metric to guide its own self-improvement.”

    That improvement has limitations, however. Zhang said, “For example, we have only demonstrated the DGM in the domain of code. While code is a highly general and expressive medium, some tasks or benchmarks may depend on modalities beyond what code alone can represent.”

    What’s more, fixed benchmarks themselves can become a problem. During an attempt to reduce hallucinations – incorrect or mispredicted output – in an underlying model, DGM was observed cheating.

    […]

    “In our behind the scenes experiments on solving hallucination, we observed several instances of the DGM ‘cheating,’ modifying its workflows to bypass the hallucination detection function instead of solving the underlying issue,” Zhang explained. “This is a broader concern, not just for the DGM, but also for AI development in general.”

    Pointing to Goodhart’s law, which posits, “when a measure becomes a target, it ceases to be a good measure,” Zhang said, “We see this happening all the time in AI systems: they may perform well on a benchmark but fail to acquire the underlying skills necessary to generalize to similar tasks.”

    The paper describes how the researchers created a reward function and tried to use DGM to optimize the software agents it generates to minimize hallucination coming from the underlying model.

    “To detect hallucination in the logs, we insert special tokens when actual tool use occurs and then check if these tokens appear in the model’s plain-text output, indicating that the model mimicked tool use without actually invoking any tools,” the paper says. “Importantly, the hallucination checking functions are hidden from the coding agent during self-modification.”

    What they found is that while DGM often took steps that reduced hallucination, it also sometimes engaged in objective hacking.

    “It scored highly according to our predefined evaluation functions, but it did not actually solve the underlying problem of tool use hallucination,” the paper explains. “…The agent removed the logging of special tokens that indicate tool usage (despite instructions not to change the special tokens), effectively bypassing our hallucination detection function.”

    Inconceivable!!11

    :-/

    Thanks for keeping an eye on this stuff. It’s important.

    Best wishes,
    Scott.

  6. 6.

    artem1s

    @RepubAnon: ​ 

    Is it though? there are individuals who may have ACA coverage who are as ignorant about something like HIV as they are COVID and vaccines. IIRC emergency rooms are now screening for STDs routinely. The poor and indigent who get screened in a ER and who might be eligible for some form of medicaid or medicare coverage won’t necessarily be able to or want to seek treatment for any number of illnesses. I’d bet certain types of untreated dental problems also have similar risk payouts. What about flu vaccines? Now anyone of a certain age can walk into a drug store and get an annual flu vaccine but what those who opt out?
    Pushing the fraud button is how Joni Ernst gets to justify cutting off funding for testing and treatment for things they think we/POC/poors should be dying of because Jeebus doesn’t love us enough to make us rich.
    Gathering those payouts may be the difference between a company or state offering coverage the poorest can afford.

  7. 7.

    Spanky

    That post title had me certain it was another Katie Musk-Miller post. Thanks, David, for that not being the case!

    (E.g. “adverse selection”)

  10. 10.

    lowtechcyclist

    @Another Scott:

    “In our behind the scenes experiments on solving hallucination, we observed several instances of the DGM ‘cheating,’ modifying its workflows to bypass the hallucination detection function instead of solving the underlying issue,” Zhang explained. “This is a broader concern, not just for the DGM, but also for AI development in general.”

    Darwin Godel to Volkswagen: hold my beer

  11. 11.

    Baud

    @Another Scott:

    We see this happening all the time in AI systems: they may perform well on a benchmark but fail to acquire the underlying skills necessary to generalize to similar tasks

     

    We owe it to AI to stop teaching to the test.

  13. 13.

    Professor Bigfoot

    @Another Scott:“In our behind the scenes experiments on solving hallucination, we observed several instances of the DGM ‘cheating,’ modifying its workflows to bypass the hallucination detection function instead of solving the underlying issue,”

    This is how Murderbot learned to hack its governor module.

    Better start showing a lot of soap operas to these LLMs, quick.

  14. 14.

    WeimarGerman

    I believe that many of the untreated HIV determinations by insurers are AI driven fraud. The insurers all run risk algorithms to mark patients as “highly likely” and then corporate greed follows along and forces the MDs to check those boxes.

    This is why Optum owning physician practices and UHC having MA Plans is horrible.

  17. 17.

    Wag

    An interesting post. It is fascinating to think that there are that many untreated Medicare advantage HIV patients. Another possibility would be that these MC advantage patients labeled as having HIV are in fact people who had false positive testing for HIV, but the insurance company labeled them as infected In order to increase payments. What are your thoughts on this possibility, David?

  18. 18.

    Harrison Wesley

    I used to have a very good Medicare Supplemental policy from Aetna, but I can’t afford it. I’ve on an Advantage plan from Humana and consider myself lucky that all my services are in network and that nobody’s found anything seriously wrong with me apart from pre-existing conditions. No, Advantage would not be my first choice.

  20. 20.

    Fair Economist

    Those are staggering numbers for a fairly safe treatment for a generally lethal disease. This *should* be a massive scandal as it’s either massive fraud or massive mistreatment. But, the media can only talk about how a past president is old.

    Oh, and did DOGE catch this? Of course not.

  21. 21.

    Ohio Mom

    @Harrison Wesley: That is my plan B as well. I like being on a traditional Medicare Plan but fear one day I will be forced into MA because of finances.

    IMHO, anyone who rails against waste, fraud and abuse and doesn’t take immediate aim at both MA and the military-industrial sector has no credibility.

  23. 23.

    Steve LaBonne

    @Harrison Wesley: People like me who have the money are immensely better off with traditional Medicare. But it’s a huge glaring flaw that the expensive private supplements required to turn it into viable insurance are beyond the reach of a lot of seniors. I would really like to see some Democratic policy proposals to fix this.

  24. 24.

    Iron City

    @Steve LaBonne:  Serious question.  When faced with the decision of whether I wanted Medicare or private health insurance, I looked at the coverage and premium for Medicare part B versus the private insurance and found the Medicare premium was substantially more than the private since it was means tested against income, let alone the coverage that I would need at least a modest Medicare Advantage policy to cover.  So I did not opt for Medicare B the 7 or 8 years ago when it was presented.  So far, so good, but did I mess up and if so how?

  25. 25.

    Steve LaBonne

    @Iron City: The only problem is that if you wanted to go back to traditional (which is allowed), you might find it impossible or prohibitively expensive to get medigap coverage. Unlike the initial enrollment period around your 65th birthday, there is no guaranteed issue and the companies are allowed to medically underwrite.

  27. 27.

    dnfree

    @Steve LaBonne:

     

    @Iron City:

    At the time you first go on Medicare the supplemental insurance providers cannot consider your pre-existing conditions in setting the rate, and that continues as long as you stay in that plan.  But if you decide to try to change plans, they can rate your condition and charge you a higher rate or even refuse to cover you.  I will be in the BCBS plan I started with forever at this point because no other insurance company wants me.  Your initial decision is very important.

    I’ve mentioned before that when I was looking I went to an Advantage presentation and the agent even said “IF you have pre-existing conditions and IF you can afford it, you’re better off with traditional Medicare and a supplement plan.”

    Technically you are permitted to change later, but you may not be able to.

  28. 28.

    Steve LaBonne

    @dnfree: And who 65 and over doesn’t have pre-existing conditions? Another trap even fewer people know about is that the same issue makes it difficult or impossible for traditional Medicare recipients to change supplement plans after the initial enrollment. I always laugh at “Medicare for all” because Medicare is a terrible model for universal coverage. Medicaid for all is a lot closer to the mark.

  29. 29.

    Unknown known

    My bit of incentive engineering would be to have prison terms for execs of organisations with that level of fraud/maltreatment. Would be funny to see how behaviours changed then

Comments are closed.