A recent analysis published in Academic Radiology suggests that large language models (LLMs) could help alleviate the administrative workload of writing appeal letters for denied imaging exams by insurance companies. Researchers evaluated four LLMs—Claude 3.5, Nova Pro, Llama-3.1–70B, and ChatGPT-4o—by assigning them to generate appeal letters for simulated clinical scenarios. The generated letters were then assessed by interventional radiologists for content accuracy, grammar, structure, and usability. Overall, the LLMs achieved mean scores of 3.9 for content and 4.3 for grammar on a five-point scale, though the assessment revealed variability among raters. While the letters were largely accurate, instances of hallucinations and fabricated references were noted, particularly with ChatGPT-4o. Despite these issues, 73% of the radiologists found the LLM-generated letters to be useful templates, indicating potential for reducing administrative burdens. The study emphasizes that while LLMs can be beneficial, oversight is still necessary before finalizing the appeal letters for submission.
Thu, 07 May 2026 17:27:55 GMT | Radiology Business