Data-driven Predictive Classification Of ICU Patients Readmission

The intensive care unit is one of the key departments in the hospital.It is the unit with the highest cost of care in all hospitals.The number of critically ill patients in the intensive care unit has increased significantly in the past two decades,far exceeding the available medical resources in the intensive care unit,resulting in the intensive care unit often operating near full capacity.Readmission into the intensive care unit for critically ill patients who are out of danger can lead to a significant increase in patient mortality and cost of care and treatment,but in most cases this result can be avoided.Therefore,the medical management community has made tremendous efforts to predict the classification of readmission of critically ill patients out of danger in order to reduce the mortality of patients and the cost of treatment and care.Based on the research based on 4697 electronic medical records from March 2017 to December 2018 in the intensive care unit of the First Affiliated Hospital of Anhui Medical University,the thesis studied the problem of predictive classification of ICU patients' readmission.The paper describes the original data set and feature engineering based on describing the ICU reentry problem.Feature engineering includes the processing method of the original data.It introduces the combined methods of missing value analysis,coefficient of variation analysis,and likelihood ratio test.For the selection of prediction classification indicators,the paper establishes a random forest model with weight attenuation for prediction classification.In order to measure the test effect,the paper introduces the evaluation criteria for prediction classification and the baseline model for prediction classification.The model enlightenment and conclusion of the predictive classification of ICU patients' readmission are obtained.The main conclusion of the thesis is that the most relevant factors that affect the readmission of critically ill patients include the length of time the patient stays in the intensive care unit,the patient's age,heart rate,respiratory rate and body temperature.The conclusion drawn in the trial was that the missing indicator that had the greatest impact on the predicted classification of patient readmission was the lack of oxygen concentration fraction in the inhaled gas.