Exploration of Disease Occurrence Prediction Model Based on Clinical Pathology Results Using Machine Learning

Abstract

Background/Objectives: The purpose of this study is to perform predictive modeling for disease onset by automating the analysis of diagnostic test results using machine learning and big data analysis for small- and medium-sized hospitals. Methods/Statistical analysis: Methods/Statistical analysis: Excel and CatBoost algorithms were used for preprocessing and analysis of the collected data. 21,140 pieces of valid data consisting of 27 attributes were obtained, and they were classified into a training set and a test set necessary for the development and use of machine learning-based predictive models. The training set was used as input data to develop the predictive models, and the test set was used as input data to evaluate the performance of the developed models. A decision tree analysis algorithm was applied, and performance analysis was performed by using accuracy, precision, and the receiver operating characteristic (ROC) area as indicators. Findings: First, the predictive models developed for three diseases, i.e., diabetes, hypertension, and hyperlipidemia, were found to have a prediction accuracy exceeding 80 to 90% and a large AUC. Second, the collected data was useful in predictive modeling research, based on a cluster analysis of the data of patients who tested positive for metabolic syndromes (diabetes, hypertension, hyperlipidemia) and the data of patients who tested negative. Third, from the above study results, diabetes, hyperlipidemia, and hypertension were found to be adult diseases, with a high level of association with blood circulation, and it was determined that it would be possible to predict cases in which a single person had multiple diseases at the same time. Fourth, it was deemed that it would be possible to obtain useful results by changing the data structure into a form of test result data for time series analysis and data with a display of disease diagnosis time point. Improvements/Applications: Through the development of predictive models for diseases using machine learning technology, predictive modeling is expected to evolve to enable prediction of diverse diseases, thereby improving the clinical environment and enhancing the reliability and competitiveness of medical care by preventing potential diagnostic errors.