How Tired Docs Write: A Tech Take

Doctors jotting down notes after seeing patients might not just be recording medical details. These notes could also hint at something else: how tired the doctor is. A recent study dug into this idea using a whopping 129, 228 emergency room visits. The goal? To train a computer model to spot notes written by doctors who are likely running on empty. Here's how they did it: they looked at doctors who had worked at least five out of the last seven shifts in the emergency department. The model learned to recognize patterns in these notes. It didn't stop there. The model also picked up on signs of fatigue in other high-pressure situations, like overnight shifts or when there were too many patients to handle. Now, here's where it gets interesting. When the model flagged a note as potentially written by a tired doctor, the decisions made for that patient weren't as good. For example, the chances of correctly identifying a heart attack dropped by 19% for every step up in predicted fatigue. One surprising finding? Notes from tired doctors had a certain predictability. The next word was easier to guess based on what came before. This is similar to how large language models (LLMs) work. In fact, notes generated by LLMs were 74% more likely to show signs of fatigue compared to those written by actual doctors. This raises a big question: Are LLMs introducing biases or errors we don't even know about yet?