Using Data Mining to Differentiate Dengue with Warning Signs from Severe Dengue: A Predictive Model from Oaxaca, Mexico.

Dengue with warning signs (DWS) and severe dengue are significant public health concerns in tropical and subtropical regions globally. Accurate and timely differentiation between these clinical forms of dengue, although crucial, is often complex. In this study, data mining techniques were applied to enhance diagnostic accuracy for clinical dengue (DWS and severe dengue) and to optimize clinical decision-making and patient management, particularly in resource-limited settings. A dataset of 1,341 patients from Oaxaca City, México was analyzed. The database was used to train and test different machine learning algorithms. Precision rates were 62.4% and 82.9% in differentiating between DWS and severe dengue, respectively, using the efficient linear support vector machine model. In addition, other performance metrics were used in the models, such as recall/sensitivity (90.8%), F1 score (73.8%), area under the curve (71.3%), and Cohen kappa (0.3593). The predictor weights were also analyzed with their Shapley values; the three most significant were hematemesis, ascites, and edema. In other words, these signs emerged as features that contribute most to the classification between DWS and severe dengue. These findings highlight the potential of data mining and machine learning techniques to enhance the differential diagnosis of dengue, particularly in resource-limited settings where timely intervention is crucial. Identifying critical risk factors with high specificity makes these models valuable tools for improving clinical decision-making and optimizing patient management. Future applications could enable health systems to prioritize resources more effectively during dengue outbreaks, ultimately contributing to better health outcomes in endemic regions.