Medicare to Require Prior Authorization Using AI
-
-
I've been working on a project like this for a couple years now involving a private insurer. It's an excellent
application of AI as long as there is a human-centric appeals process. It will make routine approvals instantaneous. My concern would be for the not-so-routine cases which are certain to occur, and I'm not crazy about hiring private companies to do it. The latter concern is due to one tech company I've been working with who has not met my expectations.@Mik said in Medicare to Require Prior Authorization Using AI:
It's an excellent application of AI as long as there is a human-centric appeals process.
Is it safe to assume that by "human-centric," you that an appeal would quickly trigger the involvement of humans?
It will make routine approvals instantaneous. My concern would be for the not-so-routine cases which are certain to occur, and I'm not crazy about hiring private companies to do it.
I think this is the real problem, that private insurance companies have for the most part not acted in good faith and far too often just been an unnecessary road block between a patient and needed/justified medical care.
(Edited to correct to “unnecessary”)
-
The problem is the incentives those humans face. When private companies get a percent of the value of the denials, denials will rise. Many people will not know how to contest or push back, and we can guess who those folks will be.
-
If done well it will take that human incentive out of the picture. I’ve found Medicare, at least traditional, to be very easy to deal with and they already have some automated approval set up. Part C is a different animal and brings the profit incentive back into the equation.
“I refuse to answer that question on the grounds that I don't know the answer”
― Douglas Adams -
Systems like this aren't put in place for your benefit and iirc the opposite of benefit is detriment.
Incidentally, my CVS moved my Schedule IV refill from the 29th day where it has been for a year, to the 29th day, to 30th day, to the 31th day, without notice, and blamed it on the computer system.
I have none on the morning of the 31st day as I should, and have to rearrange my morning whichever day it falls no matter where else I have to be.
Each applicable day I was blamed for having the wrong day after making the effort to be there and told to leave.
One day they say the computer system is, "just the system."
The next they insist they have to defer to the, "system."
They can't do their job with professionalism and compassion.
Being drug enforcement agents after the Doctors refused has led to some very unsavory characters with a sadistic streak.
And people can't understand why I have a visceral dislike of these petit tyrants, with an insufferable attitude, who can't or won't provide federal and state or policy guidelines for what they are doing.
-
If done well it will take that human incentive out of the picture. I’ve found Medicare, at least traditional, to be very easy to deal with and they already have some automated approval set up. Part C is a different animal and brings the profit incentive back into the equation.
-
Because federal computer systems have done little but improve in the last 10 years or so, especially citizen-facing applications. It would, however, be interesting to see the discussions and the written goals leading up to this.
“I refuse to answer that question on the grounds that I don't know the answer”
― Douglas Adams -
About that "if done well" question, @Mik if you don't mind opining more ...
-
For "routine approvals," does AI do better (lower error rate, faster, cheaper) compare to rule-based automated systems?
-
Does the AI also spits out explanation/rationale for its decision to approve or reject an authorization request?
-
-
I think it is able to consider more factors and ask better questions without them being hard-coded than a simple rules-based software, and yes, the stuff I have seen produces a better explanation in plain language. Whether these improvements materialize in this case remains to be seen.
-
Because federal computer systems have done little but improve in the last 10 years or so, especially citizen-facing applications. It would, however, be interesting to see the discussions and the written goals leading up to this.
@Mik said in Medicare to Require Prior Authorization Using AI:
federal computer systems
I don't think "done well" depends on the computer systems (because you're right, the systems are getting better and better). I think it depends on the people involved.
-
Yeah, but remember there's financial incentives for some health care people to order more tests, do procedures and use the higher priced drugs for a kickback - so there is a need on the other side to try and sort that out. The "done right" part for human intervention will be critical. My spouse's bout with cancer has helped to showcase how thoughtless the first round of review can be -whether AI or human.
-
No system will satisfy everyone. We're trying to balance the cost effectiveness of treatment against the "flat of the curve" approach most physicians and patients want. If you're not familiar with the flat of the curve argument, it's an inherent problem in 3rd party payer systems like ours.
Patients pay a co-pay and have out of pocket caps, so once they're done with those the marginal out of pocket cost of more treatment is zero. The incentive, then, is to push till the marginal benefit of treatment is zero as well. That's privately good but socially bad. Society pays the high marginal cost (to doctors, pharmacies, and hospitals) for further -- often useless -- treatments. Doctors and patients are essentially in cahoots against the 3rd party payer (insurers, including the government via medicaid and medicare). This is why we have these reviews. It's an expensive attempt at finding the right balance where the marginal cost of the treatment (to society) is close to the marginal benefit. (to the patient).
A good system should err on the side of treatment, within reason. United Health suggests that the erring is often on the other side, especially if the patient has little free choice (competition) and even less information.
I can see a case for AI in the first round because that should save money by letting a lot of humans whose jobs are mechanical find more productive employment. But the AI is only as good as the incentives of the people programming how it evaluates cases.
And speaking of AI, Google's AI does a very thorough job of explaining the flat of the curve argument in a healthcare system with 3rd party payers...
-
This is my Congressman, Greg Landsman. Ohio is one of the states piloting this. While I do not fully trust him yet I have seen a lot of good signs. If what he says is true about the incentives for the tech company, it's a bad thing.
Link to video
