It took only four seconds for ChatGPT-4 to generate a detailed letter appealing the denial of an inpatient admission for sepsis, making it a timesaver for hospitals faced with a growing volume of payer denials, according to a physician advisor. But the letter had a material error even though the patient’s information was fed into the artificial intelligence (AI) tool, underscoring its limits in fighting claim denials.

“This is assistive intelligence, not artificial intelligence,” said Kendall Smith, M.D., chief medical officer at PayerWatch/AppealMasters in Towson, Maryland. “Do I need to double check this? The answer is absolutely yes.” The technology has value, however, in terms of allowing people to sift through data very quickly and find “what’s pertinent in a split second,” added Vivek Radhakrishna, president and chief technology officer. That makes a difference when writing a well-written rebuttal to a denial involving a 3,000-page chart.

Smith walked through the sepsis appeal at a March 27 webinar sponsored by PayerWatch/AppealMasters. He used a HIPAA-compliant AI tool powered by ChatGPT. Here's what he asked it to do: “Write a letter to a medical director for a health insurance company arguing that inpatient care, not observation care, was appropriate for their member, an 85-year-old gentleman admitted to the hospital with a fever of 103, respiratory rate of 24, a platelet count of 100,000 and a bilirubin of 2.0 and a creatinine of 2.5. Calculate the patient’s SOFA score and mortality rate. The tone of the letter should be professional and the audience is another physician. Cite Sepsis-3.”