Precision Clinical Medicine Through Machine Learning: Using High and Low Quantile Ranges of Vital Signs for Risk Stratification of ICU Patients

Remote patient monitoring in the intensive care unit (ICU) is a critical observation and assessment duty required for precision medicine. We recently developed a cloud-based intelligent remote patient monitoring (IRPM) framework in which we adhere to the state-of-the-art in risk stratification via machine learning-based prediction, but with minimal features that rely on vital signs, the most commonly used physiological variables obtained inside and outside hospitals. In this paper, we create three machine learning models for readmission, abnormality, and next-day vital sign readings to considerably improve the efficacy of the basic IRPM framework. We give a formal representation of a feature engineering technique and describe the creation and testing of three replicable machine learning prediction models. CU patient readmission, irregularity, and vital sign measures the next day We provided two solutions for data with unbalanced classes for the readmission model and applied five binary classification techniques to each approach. We used the same five algorithms to predict whether a patient will have abnormal health conditions in the abnormality model. Our findings show that by focusing on low and high quantile ranges of vital signs, we may still get an acceptable performance with these machine learning models. We discovered that using the most current vital sign values results in the lowest prediction error. Given the medical industry's large investment in patient monitoring devices, the developed models will be incorporated into an Intelligent ICU Patient Monitoring (IICUPM) module.

KEYWORDS: Machine Learning, Random Forest, SVM, KNN, LDA, Logistic Regression.