Large language modeling and classical AI methods for the future of healthcare

Abstract

Large Language Modeling (LLM) is ubiquitous in the healthcare industry guiding clinical decisions. With the increase in demand, we must proceed with caution in the AI industry. In this study, we evaluated the accuracy of the Random Forest model in comparison to other similar models. From the 2005 to 2010 National Health and Nutrition Examination Survey (NHANES) dataset, we assessed if there was a relationship between depression and hypertension and if depression predicted hypertension. Depression was determined using the Patient Health Questionnaire (PHQ)−9 ≥ 10. Hypertension was determined by taking the average of three systolic pressure readings that were elevated. Current smoking was determined by self-reported data. We tested several Random Forest models, compared with logistic regression, naïve Bayes, decision tree model and assessed these for accuracy. The percentage of the population with diabetes was 7.7%. We found that in comparison to logistic regression (87.8%), naïve Bayes (84.6%), and decision tree model (89.3%), the Random Forest model (98.4%) was considered most accurate. We also found that out of all the variables, according to the Gini impurity index, employment (150) received the highest score in relative importance. The next highest score was depression (140). This system demonstrates the importance of using traditional AI systems such as Random Forest modeling in conjunction with LLM. ChatGPT and LLM’s must be further understood to integrate with classical machine learning techniques to make further advances in healthcare. LLM’s have been mobilized to write history and physical assessment, extracting drug names from medical notes, and condensing radiology reports. Abstraction of medical records and other applications in healthcare can further be enhanced by using the full potential for AI systems such LLM.