Physicians increasingly using AI to battle insurers

A genius way for docs to focus on their actual jobs of practicing medicine, but one that could give way to an AI arms race

If you ever want to upset a physician beyond belief, all it takes is two words - prior authorization. From patients to medical students to physicians alike, we have all seen the phenomenon. The patient comes to clinic with a health issue, the physician develops a treatment plan based on the most recent guidelines and literature, the insurance company denies the prior authorization, the physician is left to fight the insurer, and now everyone is back to square one.

The prior authorization process was originally intended to lower costs through mitigating the broad use of clinically unnecessary treatments. But what it has really become is an endless mound of paperwork and fighting for physicians and their staff to obtain approval for even the most standard treatments for their patients.

According to a recent survey by the American Medical Association, physicians and their staff spend on average, a ridiculous 12 hours per week on prior authorizations alone. Their practices complete an average of 43 prior authorizations per physician, per week, with the overwhelming majority of respondents claiming that this process is exhaustive and a large contributor to burnout.

In recent years, the physician war against insurance looked to be a losing one with an increasing number treatments and procedures that required prior authorization. But now, it appears that artificial intelligence chatbots have offered a brief respite.

A recent article in the New York Times documented a number of physicians who are integrating AI into their prior authorization workflow to churn out quicker and better response letters to insurers. By simply entering the purpose of the letter and the studies to reference, large language models such as ChatGPT can create a draft in seconds. One of the physicians interviewed in the article, Dr. Azlan Tariq, even found a HIPAA-compliant bot known as Doximity GPT which can pull information from a patient’s chart as well as cross-check requests and response letters against an insurer’s coverage requirements.

Dr. Tariq reported not only a decrease in the time he has had to devote to prior authorizations, but also said that this method is helping him win, claiming that with AI, he has been able to obtain approval for 90% of his requests compared to 10% previously.

The future role of AI models in medicine has been a subject of debate and merits its own post, but the results so far have been undeniable. ChatGPT has already been able to pass the USMLE Step 1, Step 2, and Step 3 exams, along with several post-graduate specialty board exams. Now they are being used as tools to help physicians get through the grunt work of prior authorizations and spend more time doing what they actually went through 8+ years of training for - seeing and caring for patients. This is ingenious and to me, a great introduction of AI into routine clinical practice.

However, this advancement is not without its caveats.

Insurers have developed their own algorithms that swiftly “evaluate” and deny large batches of claims at once. This essentially reduces the evaluation of potentially lifesaving treatment for patients to a computer that likely spends a few seconds per claim. The use of these algorithms have been the subject of several class-action lawsuits arguing that they breach state and federal law.

It will not be long before AI is incorporated by insurers into their evaluations, if it has not been already. Since we already know by now that AI is adept at achieving its given goals, this will inevitably aid insurers in denying even larger numbers of claims in a faster amount of time.

While AI use in prior authorization has provided a temporary win for physicians, I fear that this will devolve into an AI arms race where clinic AI bots are battling insurance AI bots and the losers will once again be the physicians straddled with too little time and more importantly, the patients who will once again see their claims denied.

This is the wrong way that I would want to see the eventual AI conflict be waged. Since AI tries to accomplish a goal that is given by humans, we need to change the task from proving or denying a claim to perhaps building a new system of treatment approval altogether. But that would require physicians and insurers (and their respective AIs) to actually work together for once - so that idea may not be anything more than wishful thinking.