A study to reduce the number of preventable emergency visits at community level

Abstract

Introduction: Emergency Department overcrowding is a worldwide issue. The impact of increased unnecessary and preventable emergency visits in hospitals leads to reduced quality of care, inability to care for critically ill patients, increased errors, and mortality rates. The study aims at identifying ways to predict avoidable ER visits through implementing machine learning algorithms and intervening them at a community level. Materials and Method: The data was collected on a community level Electronic Health Record (CCMO) database and was provided for further investigation by the Randolph Caring Community Partnership (RCCCP). The exploratory data analysis was conducted in the RStudio and later the machine learning models were implemented in the jupyter notebook. 14 features were selected out of the 43 features through step wise logistic regression method to predict the ER visits. Result: The data of 595 patients was collected during the period of 2018 to 2021. The AUC score of the Decision Tree and the logistic regression model was 0.54 and 0.61 respectively. The support vector machine had the accuracy of 0.58 in predicting ER visits. Conclusion: The study depicted that it is possible to have prediction models at the community level to reduce the burden at emergency department. A more focused data collection specific for the prediction of unnecessary ER visit will be able to produce a more feasible model.