Abstract
Early detection of dengue virus infection (DVI) in healthcare settings is frequently limited by the availability and turnaround time of tests such as nonstructural protein 1 (NS1) antigen detection and Reverse TranscriptionPolymerase Chain Reaction (RT-PCR). Routine hematological evaluations are widely accessible and could be employed for screening at no additional cost. This study aims to assess the efficacy of a hematology analyzer’s integrated dengue scoring system utilizing a Random Forest algorithm, referencing RT-PCR and NS1 antigen detection for diagnostic performance. A cross-sectional analysis of 119 patient samples was conducted using the Yumizen H500 hematology analyzer and a Dengue score algorithm based on Random Forest machine learning, comparing dengue scores with RTPCR and NS1 antigen alongside immunoglobulin M (IgM) and immunoglobulin G (IgG) antibody quantification via fluorescence immunoassay technology. Out of 119 patients (70 with confirmed DVI and 49 without), a dengue score cutoff of ≥50 yielded sensitivities of 90.0%, specificity of 57.1%, positive predictive value of 75.0%, negative predictive value of 79.4% and overall accuracy of 77.3%, with an area under the receiver operating characteristic curve of 0.77. The stratification of patients into risk categories based on dengue score ranges may aid in enhancing DVI diagnostic processes. The Yumizen H500 Dengue Score demonstrates substantial validity as a diagnostic screening tool for DVI, offering efficient implementation for rapid management.
Keywords: Dengue Virus Infection, Diagnostic Accuracy, Hematology Analyzer, Machine Learning, Random Forest Algorithm.