Spatio-temporal forecasting of dengue in the Americas through hybrid mechanistic and data-driven models: Systematic...

This systematic review and meta-analysis (PROSPERO: CRD420251130769) synthesises 30 dengue modelling studies conducted in the Americas between 2016 and 2025, evaluating the integration of mechanistic and data-driven approaches. We quantified the reliability of diverse modelling frameworks by applying a Standardised Predictive Fidelity Index (SPFI). Our synthesis reveals a robust positive association between temperature and dengue risk across all methodologies (pooled relative risk (RR) = 1.26 [95 % confidence interval (CI): 1.18-1.35]). However, a critical performance dichotomy remains: while mechanistic models exhibit high variance dependent on calibration quality, temporal regression analysis confirms that machine learning architectures have achieved statistically significant convergence towards high predictive fidelity (median SPFI: 0.89) since 2023. Despite their precision, data-driven models remain disconnected from the causal logic necessary for intervention simulation. To address this methodological fragmentation, we have developed a functional "glass-box" hybrid architecture, which is defined by three evidence-based pathways: the dynamic parameterisation of mechanistic cores via machine learning; the enforcement of biological constraints on predictive algorithms; and the continuous assimilation of data. We conclude that transitioning from descriptive science to this operational, data-assimilating hybrid framework is essential for enabling precise, location-specific public health responses to the escalating dengue crisis in the Americas.