A First Approach to the Generation of Linguistic Summaries from Glucose Sensors Using GPT-4

Abstract

The use of activity monitoring sensors on users with some type of disease or dependence is very useful for health technicians, for family members or for the user himself. The knowledge of these values in real time allows alerting of a possible crisis or starting correcting actions to prevent a serious health problem. For this reason, many proposals have been made to summarize in words the huge amount of measures taken by these sensors in order to highlight only what is really important for the end user, family or medical staff. The emergence of new text generation tools based on Artificial Intelligence (AI), such as the latest GPT-4, is having a major impact in the healthcare field. In this article we analyze how the latest version of ChatGPT, allows the generation of linguistic summaries in natural language from glucose sensor measurements. We also learn how to ask the right questions to obtain the type of output adapted to the user, whether or not it is necessary to perform some kind of preprocessing on the data to be analyzed and what are the strengths and drawbacks of this technology.