Patient

The pandorra’s box is wide open. With ChatGPT applications the discussion has started to use it for more medical applications. As for much research having assistants to support you in routine tasks in your research is a standard procedure. Now the medical profession is also discussing the use of ChatGPT for the boring and time-consuming task to draft reports. The first study, published in the Lancet Digital Health, evaluates in a preliminary form the patient-sensitive form of communication between clinics and patients. Beyond chatbots, which organise information from calling persons, the obvious application is the use of ChatGPT to draft patient clinic letters. The example in the study is the skin cancer reporting. Lengthy reporting back to patients of lots of “hot and cold spots” might be done by AI with sufficient reliability. All depends on the correctness of the data base, the screening and samples taken. The communication between clinic and patient can then focus on other issues. ChatGPT just like neuroflash has its strength in being able to control for the “level” of the language. In addition to the choice of the output language it is possible to use, as it is required in the U.S., an average understanding level of patients. In other words, easy language rather than medical expert language is an option or even a requirement. Anecdotal evidence and the PISA for adults studies show how difficult it can be to talk the same language even if you talk the same language. There is ample scope for improvement and ChatGPT or neuroflash for German applications of AI are prime candidates to fill this gap in clinic patient communication. Considering that our mobile phones (can) do already most of the scanning of skin cancer dots and AI is used in pre-scanning the images and recommends to consult medical expertise, the next step to improve health delivery seems feasible. Whereas the statistical analysis explains 62% of “median humanness”, the score of 37% of explained variance of median correctness is a surprise as the basis of the model to explain deviation from correctness should be as low as possible. Medical data, like many other data, is not simply binary. The way forward is most likely relying on a “human-in-the-loop” approach for some time. A limited human input might reassure many patients as well.
Source: Stephen R Ali, Thomas D Dobbs, Hayley A Hutchings, Iain S Whitaker (2003). Using ChatGPT to write patient clinic letters. Lancet Digit Health 2023 https://doi.org/10.1016/S2589-7500(23)00048-1