Developing a patient flow visualization and prediction model using aggregated data for a healthcare network cluster in Southwest Ethiopia

A health information system has been created to gather, aggregate, analyze, interpret, and utilize data collected from diverse sources. In Ethiopia, the most popular digital tools are the Electronic Community Health Information System and the District Health Information System. However, these systems lack capabilities like real-time interactive visualization and a data-driven engine for evidence-based insights. As a result, it was challenging to observe and continuously monitor the flow of patients. To address the gap, this study used aggregated data to visualize and predict patient flow in a South Western Ethiopia healthcare network cluster. The South-Western Ethiopian healthcare network cluster was where the patient flow datasets were collected. The collected dataset encompasses a span of 41 months, from 2019 to 2022, and has been obtained from 21 hospitals and health centers. Python Sankey diagrams were used to develop and build patient flow visualizations. Then, using the random forest and K-Nearest Neighbors (KNN) algorithms, we achieved an accuracy of 0.85 and 0.83 for the outpatient flow modeling and prediction, respectively. The imbalance in the data was further addressed using the NearMiss Algorithm, Synthetic Minority Oversampling Technique (SMOTE), and SMOTE-Tomek methods. In conclusion, we developed a patient flow visualization and prediction model as a first step toward an end-to-end effective real-time patient flow data-driven and analytical dashboard in Ethiopia, as well as a plugin for the already-existing digital health information system. Moreover, the need for and amount of data created by these digital tools will grow along with their use, demanding effective data-driven visualization and prediction to support evidence-based decision-making.