Artificial intelligence models designed to diagnose mosquito-borne viral infections frequently suffer from a lack of external validation, according to a systematic review and meta-analysis published on October 27, 2026, in PLOS Digital Health. The study, conducted by Pennisi F and colleagues, examined how machine learning algorithms perform when spotting early cases of diseases like dengue, Zika, and chikungunya before confirmatory lab tests become available.
Search Parameters and Methodology
To capture the state of AI diagnostics, the research team searched PubMed/MEDLINE, Embase, and Scopus up to August 11, 2025. They screened records using Rayyan, following PRISMA 2020 guidelines and the PRISMA-AI framework. The protocol was registered prospectively in PROSPERO under the identifier CRD420251129670.
The review evaluated models that processed clinical features, demographics, routine laboratory data, biomarkers, and imaging. The authors sorted individual algorithms into four methodological families:
- Classical machine learning, such as support vector machines and logistic regression.
- Tree-ensemble methods, including random forests and gradient boosting.
- Deep learning architectures, such as artificial neural networks and convolutional neural networks.
- Heuristic approaches, including belief rule-based expert systems.
Performance Metrics and Validation Gaps
Early diagnosis remains vital for clinical triage and outbreak response, but early symptoms often mimic other endemic febrile illnesses. While laboratory reference standards like RT-PCR or NS1 antigen tests provide certainty, they are often resource-intensive or unavailable at the point of care. AI models attempt to fill this gap by analyzing readily available patient data.
However, the study found that published reports vary widely in their modeling pipelines and validation strategies. Many studies omit external validation and calibration measures, a shortfall that can inflate apparent accuracy and limit generalizability across different populations and clinical settings.
Public Data Availability
The dataset generated through data extraction for this meta-analysis has been made publicly available on Figshare via https://doi.org/10.6084/m9.figshare.33354231. The authors declared no specific funding for the work and reported no competing interests.
Diagnostic Model Performance Details
| Model Family | Common Algorithms | Primary Inputs |
|---|---|---|
| Classical ML | Logistic Regression, SVM, Naive Bayes | Clinical features, routine lab variables |
| Tree-Ensemble | Random Forest, XGBoost, LightGBM | Structured clinical and demographic data |
| Deep Learning | ANN, CNN, Multilayer Perceptron | High-dimensional data, biomarkers, imaging |
| Heuristic | Belief Rule-Based Expert Systems | Expert rules and clinical parameters |
Frequently Asked Questions
What specific mosquito-borne viruses did the review cover?
The systematic review evaluated diagnostic models for dengue, Zika, chikungunya, West Nile virus, yellow fever, and Rift Valley fever.
When was the systematic review formally accepted for publication?
The manuscript was accepted on October 6, 2026, and subsequently published on October 27, 2026.
Where can researchers access the study’s data extraction files?
The complete extraction dataset is hosted on Figshare and can be accessed directly at https://doi.org/10.6084/m9.figshare.33354231.
Worth a look