The debate of these first months of 2023 has been stimulated by the performance of the ChatGPT artificial intelligence (AI) program (Generative Pretrained Transformer). The program automatically creates texts based on written instructions and is freely accessible via a web portal created by the developer, OpenAI.
ChatGPT uses a technique called Reinforcement Learning from Human Feedback to train the language model, making it very conversational.In response to a written request, ChatGPT can compose emails, write computer code, and even create screenplays for movies. Some simulations in the scientific medical field amaze, demonstrating his competence in passing medical license exams (1) or as .
Reactions range from enthusiasm for potential to ethical concerns that could, and perhaps should, limit adoption of the tool. Given the premises, the applications of ChatGPT in the field of healthcare could be countless and among the many is considered, in a article published on The Lancet Digital Health, the use to generate patient discharge letters (2).
ChatGPT and health documentation: new conceivable scenarios
With the development of electronic health records, health data storage systems have moved towards progressive digitization, almost entirely replacing the traditional method of collecting paper medical record data (3).
Accurate discharge summary documentation plays a crucial role in the continuity of healthcare for patients discharged from hospital.This information is of paramount importance to enable the general practitioner (GP) to meet the patient’s health needs at the time of discharge in the community (4).
The possibility of automation of this process seems a logical and almost obvious application of ChatGPT that would allow on the one hand doctors to lighten a workload focused on the provision of services, with more time to devote to clinical activity, research and training; on the other hand, it could improve the quality of resignation letters.
How to implement ChatGPT: open issues
When asked to get a simple request for a summary for a patient’s discharge, ChatGPT responds like this (2): «Although this summary is likely to be accurate for most patients, additional information not included in the prompt was automatically generated by ChatGPT, highlighting the need for manual control by clinical staff before finalization».
A premise that makes us understand how the implementation of ChatGPT in clinical practice will not be immediate, but a process that will require the overcoming of several barriers.First, the program depends on the data it receives and therefore requires manual entry of relevant information. Future interactions may allow automatic extraction of data from the patient’s electronic medical record, but the problem is also open with respect to data storage and access.
Electronic records are still critical and are often considered incomplete. Until a few years ago, less than 5% of anesthesia departments in Italy used a specific folder for the specialty (5) and a recent trial in the field of primary care analyzed the quality of the exchange of aggregated data by MMG teams detecting many errors that required human intervention and manual repair, precluding full automation of processes (6).
In addition, concern that automation could lead to depersonalization of care could lead to resistance to this technology. Data from pilot trials will be needed to gather the views of stakeholders (doctors and patients) and demonstrate outcomes on process efficiency and quality.All this must also be done taking into account the consequences of a possible failure of the technology. However, the power of this new generation of tools applied to health care, and in particular in the areas of documentation, is a challenge that will have to be played: it will be necessary to understand how.
Access to the site is restricted and reserved for healthcare professionals
You have reached the maximum number of visits
Source — https://www.univadis.it/viewarticle/lettere-di-dimissioni-e-chat-gpt-una-sfida-possibile-2023a10005dn