Application of Machine Learning Models to Predict Clinical Outcomes in Low-Resource Settings

Abstract

Healthcare systems in low-resource environments are usually challenged by limited medical staff, poor diagnostic tools, and delayed clinical decision processes. These restrictions add to the risk of preventable morbidity and mortality. Recent developments in machine learning (ML) provide promising prospects to assist healthcare workers with the predictive analysis of the outcomes of clinical cases based on routine patient data. This study investigates the use of ML models (logistic regression, random forests, support vector machines and artificial neural networks) to predict key factors such as disease course, risk of death, race, response to treatment and risk of readmission to the hospital. Based on evidence in low and middle-income settings (LMIC), the main message of the paper is the potential of ML to help improve the accuracy of diagnostics, optimize triage and resource allocations. However, challenges such as the lack of a large quantity of data, poor quality of data, limited computational infrastructure and lack of technical expertise have impeded large-scale implementation. The study concludes that integration of ML models into clinical workflows could play a major role in improving patient outcomes that they say, in low resource setting futures, as long as ethical considerations, stakeholder training and context-specific model customisation are prioritized.