LAUSR

Establishing reliable early warning models for severe dengue cases is a high priority to facilitate triage in dengue-endemic areas and optimal use of limited resources. However, few studies have identified the complex interactive relationship between potential risk factors and severe dengue. This research aimed to assess the potential risk factors and detect their high-order combinative effects on severe dengue. A structured questionnaire was used to collect detailed dengue outbreak data from eight representative hospitals in Dhaka, Bangladesh, in 2019. Logistic regression and machine learning models were used to examine the complex effects of demographic characteristics, clinical symptoms, and biochemical markers on severe dengue. A total of 1,090 dengue cases (158 severe and 932 non-severe) were included in this study. Dyspnoea (Odds Ratio [OR] = 2.87, 95% Confidence Interval [CI]: 1.72 to 4.77), plasma leakage (OR = 3.61, 95% CI: 2.12 to 6.15), and hemorrhage (OR = 2.33, 95% CI: 1.46 to 3.73) were positively and significantly associated with the occurrence of severe dengue. Classification and regression tree models showed that the probability of occurrence of severe dengue cases ranged from 7% (age >12.5 years without plasma leakage) to 92.9% (age ≤12.5 years with dyspnoea and plasma leakage). The random forest model indicated that age was the most important factor in predicting severe dengue, followed by education, plasma leakage, platelet, and dyspnoea. The research provides new evidence to identify key risk factors contributing to severe dengue cases, which could be beneficial to clinical doctors to identify and predict the severity of dengue early.