An app detecting dengue fever in children: Using sequencing symptom patterns for a web-based assessment

Abstract

Background: Dengue fever (DF) is one of the most common arthropod-borne viral diseases worldwide, particularly in South East Asia, Africa, the Western Pacific, and the Americas. However, DF symptoms are usually assessed using a dichotomous (ie, absent vs present) evaluation. There has been no published study that has reported using the specific sequence of symptoms to detect DF. An app is required to help patients or their family members or clinicians to identify DF at an earlier stage. Objective: The aim of this study was to develop an app examining symptoms to effectively predict DF. Methods: We extracted statistically significant features from 17 DF-related clinical symptoms in 177 pediatric patients (69 diagnosed with DF) using (1) the unweighted summation score and (2) the nonparametric HT person fit statistic, which can jointly combine (3) the weighted score (yielded by logistic regression) to predict DF risk. Results: A total of 6 symptoms (family history, fever ≥39°C, skin rash, petechiae, abdominal pain, and weakness) significantly predicted DF. When a cutoff point of >–0.68 (P=.34) suggested combining the weighted score and the HT coefficient, the sensitivity was 0.87, and the specificity was 0.84. The area under the receiver operating characteristic curve was 0.91, which was a better predictor: specificity was 10.2% higher than it was for the traditional logistic regression. Conclusions: A total of 6 simple symptoms analyzed using logistic regression were useful and valid for early detection of DF risk in children. A better predictive specificity increased after combining the nonparametric HT coefficient with the weighted regression score. A self-assessment using patient mobile phones is available to discriminate DF, and it may eliminate the need for a costly and time-consuming dengue laboratory test.