Sepsis-Diagnosed Patients’ In-Hospital Mortality Prediction Using Machine Learning: The Use Of Local Big Data-Driven Technique In The Emergency Department
Abstract
In most cases, emergency care has seen most of its predictive analytics focus on CDR (clinical decision rule) usage, which comes in the form of simple scoring and heuristic systems. With CDR development, several limitations exist and continue to compromise the ability of the framework to select small variable sets in advance, especially those perceived to be relevant clinically. Additionally, CDRs experience frequent problems in terms if generalizability and take long to be established, besides lacking room for regular updating in response to new information. Thus, newer machine learning and analytic methods have been developed, especially those that can harness many variables contained in electronic health records, striving to facilitate data automation and deployment, as well as predict patient outcomes better. The main aim of this study was to evaluate the performance of a machine learning, big data-driven, and local technique and compare the results with those documented for traditional analytic approaches and existing CDRs, with particular focus on mortality prediction among in-hospital patients diagnosed with sepsis. In the retrospective study, findings demonstrated that the proposed random forest model AUC outperformed other algorithms with which it was compared. Thus, it was concluded that the proposed model exhibits outcome validity and reliability. In the future, emergency care predictive analytics ought to focus on the applicability of the proposed model in other healthcare environments, health conditions, and populations.