Diagnostic performance of artificial intelligence in early diagnosis of mosquito-borne viral infections: Systematic review and meta-analysis

Early diagnosis of mosquito-borne viral infections is challenging because initial symptoms are often nonspecific and confirmatory testing may be delayed or unavailable. We synthesised evidence on artificial intelligence and machine learning models for the early diagnosis of dengue, chikungunya, and Zika. We conducted a PRISMA 2020-compliant systematic review and meta-analysis (PROSPERO CRD420251129670). PubMed/MEDLINE, Embase, and Scopus were searched from inception to 11 August 2025 for model-development or validation studies reporting diagnostic performance. Area under the receiver operating characteristic curve (AUC), accuracy, positive predictive value (PPV), and negative predictive value (NPV) were pooled using random-effects models, while sensitivity and specificity were jointly synthesised using a bivariate random-effects model. Risk of bias was assessed using the Prediction model Risk Of Bias Assessment Tool (PROBAST). Database searches identified 4,531 records, of which 43 studies met the inclusion criteria. The evidence was predominantly from Asia and overwhelmingly focused on dengue; no eligible studies evaluated West Nile virus, yellow fever, or Rift Valley fever. Pooled estimates were 0.954 (95% CI 0.936-0.968) for AUC, 0.904 (95% CI 0.888-0.918) for accuracy, 0.903 (95% CI 0.876-0.925) for PPV, and 0.882 (95% CI 0.859-0.902) for NPV. The bivariate pooled sensitivity was 0.861 (95% CI 0.809-0.900) and specificity was 0.902 (95% CI 0.871-0.927). Heterogeneity was substantial for the univariate outcomes (I2 95.8%-97.7%). Most studies were at high overall risk of bias, while calibration and external validation were uncommon. Artificial intelligence models showed encouraging diagnostic performance in the published literature, but the pooled estimates should be interpreted as descriptive summaries of a highly heterogeneous evidence base rather than as definitive estimates of real-world performance. Prospective, externally validated, well-calibrated, and geographically diverse studies are needed before routine clinical implementation can be recommended.