Predictive Analytics in Health Information Management: The Impact of Electronic Health Record (EHR) Data on Patient Outcomes in Nigerian Tertiary Health Facilities

Background – The trend towards the implementation of Electronic Health Records (EHR) into healthcare systems has offered a treasure trove of data which can be used in predictive analytics. Patient care and clinical decision-making can be enhanced by learning the factors that are important in determining patient outcomes and predictive modeling in tertiary health facilities in Nigeria.  Objective – The purpose of the study was to determine the primary elements in EHR data, that have a significant impact on patient outcomes and to construct and test predictive models to predict these outcomes, and any such outcome can lead to improved clinical decision-making in Nigerian hospitals.  Methods – An analysis of 500 patient records of two tertiary hospitals of Nigeria was carried out in the retrospective manner. The information contained demographics, medical history, diagnosis, laboratory data, and treatment information. Correlation analysis was done to come up with factors that significantly affect patient outcomes. Logistic Regression, Random Forest and Support Vector Machine (SVM) were some of the predictive models that were developed and tested based on accuracy, precision, recall, and F1 score.  Results – The most important variables that were regarded as significant predictors of mortality, readmission, and recovery were age, high blood pressure, diabetes, and blood pressure. Random Forest model did better and reported the accuracy and precision of 90 and 89 respectively, recall of 91 and F1 score of 90. It was found that hypertension and diabetes were most strongly correlated with adverse outcomes.  Conclusion – The research shows that EHR data may be successfully employed to create predictive models that may be utilized to improve patient outcomes and clinical decision-making. Random Forest model proved to be the most efficient in terms of patient outcome prediction and this hints on its possible further application in the healthcare environment. The study ought to be extended in terms of data volume and incorporating new variables to maximize predictive models to apply in a larger context.