Using Machine learning to predict chronic kidney diseases among diabetic patients in Rwanda.

BackgroundChronic Kidney Disease (CKD) is a significant complication in people with diabetes, leading to serious adverse health outcomes and increased healthcare costs globally individually and on healthcare systems. This problem become more complicated when it is in Low and middle in countries including Rwanda when access to early diagnostic services is limited. Early prediction and intervention can improve patient outcomes and reduce the burden on healthcare systems. ObjectiveThis study aimed to develop and evaluate a machine learning model for predicting CKD in diabetic patients, tailored to the Rwandan population, using Electronic Medical record Data. MethodologySecondary data were extracted from OpenClinic, an electronic medical record (EMR) system used at Kigali University Hospital, covering a period of 10 years from 2013 to 2023. The final cleaned dataset was used to train four machine-learning models: Logistic Regression (LR), Random Forest (RF), Decision Tree (DT), and Extra Gradient Boosting Machine (XGBoost). XGboost was noted as the best performer with the AUC score of 0.98 and accuracy of 95.67%. ResultsThe findings revealed that XGBoost was highly effective in predicting chronic kidney disease, achieving an accuracy of 95.76% and an AUC score of 0.98. Given that the dataset was collected from the local population, this study confirms that machine learning algorithms can assist clinicians in Rwanda in diagnosing chronic kidney disease in its early stages. ConclusionThis study demonstrates the potential of machine learning models in predicting chronic kidney disease (CKD) in diabetic patients, highlighting the importance of local datasets for optimizing model performance in specific populations. These findings suggest that machine learning can effectively assist existing medical techniques in the early diagnosis of CKD in Rwanda. Author summaryIn this study, we trained machine learning model to predict the risk of chronic kidney disease (CKD) in patients with diabetes, using a dataset collected in Rwanda. Early detection of CKD is crucial, as it allows healthcare providers to intervene sooner, improving patient outcomes, potentially reducing financial, and health burden on the patients. We processed the data, by handling different available data issues and statistically created new features such as anemia status and length of hospital stay to improve the models predictions. The final model, XGBoost provides insights that it can help health providers to identify high-risk patients and plan personalized care more effectively. This study highlights how data-driven solutions can support healthcare delivery in resource-limited settings, by enhancing early diagnosis especially at primary healthcare level. By integrating this predictive tool into routine clinical workflows of Electronic Medical Record, healthcare institutions can make better clinical decisions that improve patient care and outcomes. This project contributes to the growing field of health informatics in Africa and shows the potential of applying advanced analytics to solve local health challenges.