Researchers explored whether ChatGPT could help identify which hospital patients need hemodialysis, a treatment that filters waste from the blood when kidneys fail. They analyzed anonymized notes from 22 patients at a university hospital. Three kidney specialists reviewed each case and agreed on whether dialysis was needed. Their consensus served as the gold standard.
ChatGPT, using a simple prompt, gave its own recommendations. The tool matched the experts' decisions in most cases. Specifically, the experts determined that 8 out of 22 patients (about 41%) needed dialysis, while 13 did not. ChatGPT's recommendations aligned with these expert judgments, though the exact number of matches was not reported.
The study also checked how consistently the human experts agreed with each other. Their agreement was excellent, with a statistical score of 0.814, which adds confidence to the reference standard.
This was an exploratory study with a very small sample, so the results are preliminary. No safety issues were reported, but the tool is not ready for real-world clinical use. The researchers suggest ChatGPT might serve as an educational aid for training future nephrologists, not as a replacement for clinical judgment.
For now, patients and doctors should view this as an early step. More research with larger groups is needed before drawing firm conclusions.
