According to Jama Network, researchers discovered that when examined by cardiology professionals educated in preventive care, a popular online AI model delivered mainly adequate answers to fundamental queries about CVD prevention.
Interactive AI has the potential to improve healthcare procedures by addressing patients’ commonly asked questions regarding CVD prevention and enhancing patient education. Such technology may deliver pre-written draft responses or conversational solutions to frequently requested questions for physicians’ electronic communications.
It is critical to investigate if these strategies may increase readability because prior research has shown that many online patient education resources for CVD prevention have low readability. The AI model was judged to have successfully solved 21 of the final 25 questions (84%).
Only four (16%) of the sixteen (16%) were thought to be suitable in both instances. There were three sets of responses where all three options were incorrect and one group where just one option was incorrect. When questioned about exercise, for example, the AI presented two contradicting and potentially dangerous recommendations: cardiovascular activity and weightlifting.
There was a lack of data in the responses that may have aided in interpreting a 200 mg/dL LDL cholesterol level, such as the presence or absence of familial hypercholesterolemia or the effect of genetics. I’ve discovered that inclisiran is no longer accessible. It was determined that a lack of replies was untrustworthy.
There are a few caveats to this study that should be mentioned. The topic of cardiovascular disease prevention is extensive, and the few simple questions posed here can only scratch the surface. Artificial intelligence is only as accurate and trustworthy as the data used to train it.
For example, a lack of current training might have resulted in an incorrect response concerning inclisiran. For this research, the most recent version of ChatGPT was employed, and no other AI language models were compared to it.
If people want to learn about the merits and downsides of other models, researchers will need to do some future comparing and contrasting. When inappropriateness ratings that have not been fully evaluated are utilized, the job must be redone using a more objective grading and evaluation approach for the replies (e.g., accuracy, readability).
Because only one person went through all of the comments, it is unclear whether the reviewers’ opinions were consistent with one another. Researchers did not compare the levels of variance across the three AI solutions. In the end, the AI tool’s statements were not substantiated by any evidence.
