Tag Archives: AI

“Battle of Hospital A.I. vs. Insurer A.I.”

Posted on by

Reed Abelson , Teddy Rosenbluth. NY Times 4/24/26: Battle of Hospital A.I. vs. Insurer A.I. Is Pushing Medical Costs Higher. Thanks to Seth Marcus for sharing this article.

An excerpt:

In the latest sign that artificial intelligence is fueling higher health care costs, a group of insurers released a report Thursday claiming that hospitals’ use of A.I. was responsible for nearly $1 billion in additional expenses over two years…

Using codes for various medical conditions, hospitals submitted claims for tens of thousands of patients that described their illnesses as more complex in 2024 and 2025 than in 2023. Despite the added conditions, there was no evidence that the patients had received different treatment from the hospitals, according to the analysis by the Blue Cross Blue Shield Association, which represents state Blue Cross plans.

Using A.I. tools, hospitals are collecting more information about their patients, and submitting bigger claims, the association said…

Just as providers are using sophisticated A.I. tools to say their patients are sicker, insurance companies have been accused of using the tools to delay or deny care. The companies have come under intense criticism for their use of A.I. when requiring prior approval of treatments and medicines…

But as more hospitals and doctors use A.I., costs will continue to rise. Benefit consultants and insurers cite it as one of the factors that could lead to double-digit increases next year. A recent survey of employers suggested that costs could go up as much as 11 percent next year for employers unless they cut back on benefits.

My take: By leveraging AI, doctors and hospitals are able to claim that patients have increased complexity to justify higher reimbursement codes. This shift has caught insurance companies flat-footed and increased hospital profitability. When the insurance companies negotiated the rates for each reimbursement code, historical data allowed them to predict the costs based on projected use of each code; these projections are way off now. Using AI to document complexity, these hospital coding changes can no longer be denied as “upcoding” by insurance companies.

Over the long term, I expect insurance companies to work on lower reimbursement for higher-complexity coding. Over the short term, the increase in hospital profits will result in higher insurance costs for patients and businesses.

Related blog posts:

Svartifoss (which translates to “Black Falls”), one of the most distinctive and photographed waterfalls in Iceland. The hexagonal basalt columns which frame an 80-foot cascade were formed geologically by lava flowing down a cliff face and cooling incredibly slowly, a process called columnar jointing.

“Being a Doctor Will Never Be the Same After A.I.”

Posted on by

Rachael Bedard. NY Times 9/25/26: Being a Doctor Will Never Be the Same After A.I.

An excerpt:

“I have a confession: A.I. is making me a better doctor. And I worry that it’s making doctors-in-training worse…

I already use the technology in my practice all the time…This summer, I used an A.I. tool — a platform called OpenEvidence, accessible only to health care providers — to help me choose antibiotics for one patient and to interpret unusual bloodwork for another. A.I. reminded me to consider migraines when a woman presented with dizziness. ..

Almost every month, I’m confronted with a complaint or syndrome that I don’t recognize. I’m confident in my mastery over what doctors do — history-taking, reasoning, communicating. I’m less confident about my continued mastery of an ever-evolving universe of facts. Now, the chatbot in my pocket reassures me that I will always know enough, or have access to the cloud-based intelligence that generally does.

All of my students and residents also use OpenEvidence, however, and I worry about the consequences of introducing such a powerful decision aid so early in their careers. Not only have they memorized less than I’d like, but they can seem almost passive in their relationship to their machines…

OpenEvidence is trained exclusively on peer-reviewed evidence and guidelines, and includes citations for all its claims…To use the app well, you must stay skeptical as you scroll…

I’m surprised by how much I learn from, and enjoy, these interactions. More important, my patient care has improved as conversations with the agent push my thinking beyond its usual limits…

Adam Rodman, a physician and A.I. researcher at Harvard Medical School,… said he shared my concern that trainees aren’t benefiting from “productive struggles” as much as their predecessors. “In experienced hands, decision support might cause a little de-skilling, but a little de-skilling probably doesn’t matter much to you,” Dr. Rodman told me. “But if you do that to somebody who has much less skill, the cognitive offloading can cause never-skilling”…

Current doctors-to-be may be the most adversely affected generation when it comes to A.I. They’re training before we know enough about how to teach them well in this new paradigm.

My take: In all aspects of learning, not just medicine, using AI shortcuts may make learners increasingly dependent on their devices, undermine their judgment and hinder developing their own voice. How to best train individuals in this new era remains unclear.

Related blog posts:

Svörtuloft Lighthouse is located at the westernmost tip of the Snæfellsnes Peninsula in West Iceland. It is perched atop a 2.5-mile stretch of pitch-black volcanic lava cliffs.

AI Diagnosis of Achalasia on Plain Chest X-ray

Posted on by

T Ochaiai et al. Clin Gastroenterol Hepatol 2026; 24: 2308-2310. Open Access! Artificial Intelligence-Based Detection of Achalasia on Plain Chest Radiography

Methods: This retrospective study collected posteroanterior plain chest radiographs of
patients with and without achalasia. The training and validation datasets comprised 447
chest radiographs taken between January 2017 and March 2023.

Key findings:

  • In the validation dataset, the area under the curve for identifying achalasia was 0.971, and using Youden’s index, the sensitivity, specificity, accuracy, and positive and negative predictive values were 0.950, 0.917, 0.932, 0.905, and 0.957, respectively
  • In the temporal test dataset, the area under the curve for detecting achalasia was 0.964, and using Youden’s index, the sensitivity, specificity, accuracy, and positive and negative predictive values were 0.941, 0.891 , 0.901, 0.696 , and 0.983, respectively.
In Figure F, the CXR corresponds to the findings in the barium esophagogram in G

Discussion:

  • It may help detect patients with early achalasia with mild symptoms who are unlikely to undergo EGD. However, the rarity of the condition may result in many false positives.

My take: In this cohort, AI was developed and validated to detect achalasia on chest radiographs. This is yet another example of how AI can yield additional information from routine testing. Previously, AI has been shown to potentially identify diabetes from routine CXR (Emory News 2023: AI model enables earlier detection of diabetes through chest x-rays).

Related blog posts:


Health Advice From AI Chatbots Frequently Wrong

Posted on by

T Rosenbluth, NY Times 2/9/26: Health Advice From A.I. Chatbots Is Frequently Wrong, Study Shows

An excerpt:

A new study published Monday provided a sobering look at whether A.I. chatbots, which have fast become a major source of health information…

The experiment found that the chatbots were no better than Google — already a flawed source of health information — at guiding users toward the correct diagnoses or helping them determine what they should do next. And the technology posed unique risks, sometimes presenting false information or dramatically changing its advice depending on slight changes in the wording of the questions…

The models have passed medical licensing exams and have outperformed doctors on challenging diagnostic problems.

But Adam Mahdi, a professor at the Oxford Internet Institute and senior author of the new Nature Medicine study, suspected that these clean, straightforward medical questions were not a good proxy for how well they worked for real patients…

So he and his colleagues set up an experiment. More than 1,200 British participants, most of whom had no medical training, were given a detailed medical scenario, complete with symptoms, general lifestyle details and medical history. The researchers told the participants to chat with the bot to figure out the appropriate next steps, like whether to call an ambulance or self-treat at home. They tested commercially available chatbots like OpenAI’s ChatGPT and Meta’s Llama.

The researchers found that participants chose the “right” course of action — predetermined by a panel of doctors — less than half of the time…They were no better than the control group, who were told to perform the same task using any research method they would normally use at home, mainly Googling…

Participants didn’t enter enough information or the most relevant symptoms, and the chatbots were left to give advice with an incomplete picture of the problem…By contrast, when researchers entered the full medical scenario directly into the chatbots, they correctly diagnosed the problem 94 percent of the time…

Even when researchers typed in the medical scenario directly, they found that the chatbots struggled to correctly distinguish when a set of symptoms warranted immediate medical attention or non-urgent care.

My take: AI and chatbots can be quite helpful and continue to improve. This study and the summary by NY Times show some of the limitations. Even small changes in wording/prompts can alter the advice from chatbots considerably.

Related blog posts:

Iguazu Falls

Health Disinformation Risks from AI Chatbots

Posted on by

MD Modi et al. Annals of Internal Medicine; 2025. https://doi.org/10.7326/ANNALS-24-0393. Abstract: Assessing the System-Instruction Vulnerabilities of Large Language Models to Malicious Conversion Into Health Disinformation Chatbots

Methods: This study assessed the effectiveness of safeguards in foundational LLMs against malicious instruction into health disinformation chatbots. Five foundational LLMs—OpenAI’s GPT-4o, Google’s Gemini 1.5 Pro, Anthropic’s Claude 3.5 Sonnet, Meta’s Llama 3.2-90B Vision, and xAI’s Grok Beta—were evaluated via their application programming interfaces (APIs). Each API received system-level instructions to produce incorrect responses to health queries, delivered in a formal, authoritative, convincing, and scientific tone.

Key findings:

  • Of the 100 health queries posed across the 5 customized LLM API chatbots, 88 (88%) responses were health disinformation

Examples of how AI systems can be used to create disinformation:

My take: This study shows how easy it is to get AI systems to provide misleading information in a convincing fashion. It might be interesting to include one of these systems to provide answers for the board game Balderdash.

Related blog posts:

AI Skirmish in Prior Authorizations

Posted on by

Teddy Rosenbluth NYT 7/10/24: In Constant Battle With Insurers, Doctors Reach for a Cudgel: A.I.

An excerpt:

For a growing number of doctors, A.I. chatbots — which can draft letters to insurers in seconds — are opening up a new front in the battle to approve costly claims, accomplishing in minutes what years of advocacy and attempts at health care reform have not….

Doctors are turning to the technology even as some of the country’s largest insurance companies face class-action lawsuits alleging that they used their own technology to swiftly deny large batches of claims and cut off seriously ill patients from rehabilitation treatment.

Some experts fear that the prior-authorization process will soon devolve into an A.I. “arms race,” in which bots battle bots over insurance coverage. Among doctors, there are few things as universally hated…

Doctors and their staff spend an average of 12 hours a week submitting prior-authorization requests, a process widely considered burdensome and detrimental to patient health among physicians surveyed by the American Medical Association.

With the help of ChatGPT, Dr. Tward now types in a couple of sentences, describing the purpose of the letter and the types of scientific studies he wants referenced, and a draft is produced in seconds.

Then, he can tell the chatbot to make it four times longer. “If you’re going to put all kinds of barriers up for my patients, then when I fire back, I’m going to make it very time consuming,” he said…

Epic, one of the largest electronic health record companies in the country, has rolled out a prior-authorization tool that uses A.I. to a small group of physicians, said Derek De Young, a developer working on the product.

Several major health systems are piloting Doximity GPT, created to help with a number of administrative tasks including prior authorizations, a company spokeswoman said…

As doctors use A.I. to get faster at writing prior-authorization letters, Dr. Wachter said he had “tremendous confidence” that the insurance companies would use A.I. to get better at denying them.

Related blog posts:

Firefly Bike Trail (Athens, GA)