One of the issues that the Centers for Medicare & Medicaid Services (CMS) addressed in its new rule for Medicare Advantage (MA) plans, CMS-4201-F, was the use of artificial intelligence (AI) and automated protocols to deny care. And obviously, their guidance was that this was not acceptable.
Just to clarify, the MA plans were using AI to predict the length of stay for inpatient rehabilitation facility (IRF) admissions, skilled nursing facility (SNF) care, and home health episodes of care. If the AI program said a patient needed eight days in the SNF, they refused to pay at day 9. CMS explained that before discontinuing coverage, the plan must review the patient’s specific clinical needs and compare that to Medicare coverage guidelines, and they must approve continuing care if the patient still meets the Medicare criteria, regardless of the AI prediction.
Now, how does this AI work? Well, they take millions of past admissions and use those patient characteristics and lengths of stay to then predict the length of stay for the current patient.
Well, if you think about it, many hospitals do the exact same thing to their doctors. They have staff who determine the working Diagnosis-Related Group (DRG) for each patient and then use the Medicare geometric mean length of stay (GMLOS) to tell the doctor when they expect their patient to be discharged. But just like these AI programs, Medicare’s GMLOS is based on thousands and thousands of admissions in that DRG, and in no way is it intended to predict the expected length of stay of any one patient. Think about it: to hit the highest-weighted DRG in a triad, the patient needs one major complication or comorbidity (MCC). But does anyone think the length of stay is going to be the same for a patient with one MCC, compared to the patient with two MCCs and three CCs? Of course not. But that’s what happens. So, stop using the GMLOS like this, unless you want to also give your MA plans permission to use their own AI tools.
On another note, as many of you may know, one of the other big topics in healthcare reform, other than Medicare Advantage and price transparency, is prior authorization. No one likes prior authorization, and CMS has proposed rules to rein in insurers. But you may recall that traditional Medicare has its own prior authorization program for specific services performed in the hospital outpatient setting.
When this program was introduced, many were surprised that CMS limited it to the hospital outpatient setting and excluded inpatient surgeries and surgeries performed at surgery centers. Well, it appears that may soon change. In a notice published two weeks ago, CMS announced that they are creating a prior authorization process for ambulatory surgery centers (ASCs) as a demonstration project to “develop improved procedures for the identification, investigation, and prosecution of Medicare fraud occurring in ambulatory surgical centers providing services to Medicare beneficiaries.”
Unfortunately, we have few details of this new process, and no idea when it will begin, nor what surgeries will require prior authorization. But this process will differ from the outpatient hospital process. For hospitals, obtaining a prior authorization is a condition of payment, so if it is not done, the hospital gets paid nothing, with no appeal rights.
But CMS has said that for ASCs, if the prior authorization is not obtained, the Medicare Administrative Contractors (MACs) will simply contact the ASC and request the records to determine if the claim should be paid. Now, why is this different? Well, CMS said they were limited by regulatory and statutory differences between ASCs and hospitals.
As I have reported here, CMS stated that about 20 percent of hospital outpatient prior authorizations are denied at first pass, so it will be interesting to watch how ASCs perform.
I will also note that in the description of the hospital outpatient program, CMS talks about reducing improper payments, but in the CMS notice on the ASC program, they state that their intent is to identify and prosecute fraud. That’s serious talk; was this the intention?
No one knows.
