CAJ | 학술논문

Objective: To develop and evaluate predictive models by quantifying warning signs prior to the development of severe dengue. <br> Methods: A retrospective cohort study was conducted in which the total number of warning signs each day was compared between dengue with warning signs and severe dengue. Multivariate logistic regression with forward likelihood ratio method was employed to achieve the best fit models for the prediction of severe dengue. The models were also being explored by adding diarrhoea and removing lethargy. Receiver operating characteristics were then used in these best fit models to identify suitable cut-off probability values derived from the equation of the models. <br> Results: Median age of patients was 26 years old (interquartile range was 15 years) and 65.3%(1110) were males. Age with total number of warning signs at day one of illness (model T1) and age with total number of warning signs at day two of illness (model T2) were identified as the best fit models. The best probability cut-offs for model T1 was 0.0506 with 10.1% positive predictive value, 96.4% negative predictive value, 99.4% sensitivity, 1.8% specificity; for model T2 was 0.0503 with 10.2% positive predictive value, 96.4% negative predictive value, 99.4% sensitivity, 1.8% specificity. <br> Conclusions: The models developed in this study might not reduce the burden effectively. Clinicians may use the models but the models must be re-validated in their clinical settings as the effect size might vary. Furthermore, the risk and benefit in selecting the cut-off values should be evaluated before implementing such models.