Lagged and non-linear effects of Aedes surveillance indices on dengue incidence in Malaysia: evidence from sentinel-site data

BACKGROUND: Dengue remains a major public health challenge in tropical and subtropical regions, with Malaysia experiencing sustained endemic transmission despite long-standing vector control efforts. Routine Aedes entomological surveillance, particularly the ovitrap index (OI) and larval index (LI) is widely implemented to guide control activities, yet their mechanistic relationship with dengue transmission and their comparative value as early-warning signals remain insufficiently quantified. METHODS: We conducted a retrospective analysis using weekly dengue case counts and entomological surveillance data from 11 Aedes sentinel districts in Peninsular Malaysia in 2024. Dengue incidence was modelled using generalised additive models with a quasi-Poisson distribution and population offsets. Distributed lag non-linear models (DLNMs) were applied to characterise non-linear and lagged associations between dengue incidence and OI or LI over a 12-week lag window. Early-warning performance was evaluated using a rolling-origin cross-validation framework with a 4-week forecast horizon. Three models were compared: DLNM(OI), DLNM(LI), and DLNM(OI + LI). Predictive discrimination was assessed using the area under the receiver operating characteristic curve (AUC) with 95% confidence intervals estimated using the DeLong method. Sensitivity analyses were conducted using extended lag windows and temperature-adjusted models. RESULTS: Exposure-lag-response analyses showed modest and lag-dependent associations between both entomological indices and dengue incidence, with attenuation at longer lags. The LI demonstrated slightly more pronounced short-term associations at lower exposure levels, whereas the OI showed weaker and less consistent patterns. In early-warning evaluation, overall AUCs were 0.63 (95% CI 0.55-0.71) for the OI-based model, 0.68 (95% CI 0.60-0.76) for the LI-based model, and 0.65 (95% CI 0.56-0.73) for the combined model, indicating modest predictive performance. Findings were broadly consistent across lag-window specifications and after temperature adjustment. CONCLUSIONS: Routinely collected entomological surveillance indicators exhibit non-linear and delayed associations with dengue incidence and provide modest early-warning signals. The LI showed slightly higher predictive performance than the OI, although overall discrimination remained limited. These findings highlight the importance of context-specific use of entomological indicators and support their continued integration into operational dengue early-warning systems to inform timely vector control interventions.