The review was designed to answer a narrower and more operational question than previous syntheses: not whether rainfall matters, but whether the published evidence can support approximate rainfall ranges, lag periods and disease-specific patterns that could inform climate-informed disease surveillance. Previous reviews had typically treated the rainfall-disease relationship qualitatively or within single-disease and single-country frames, leaving uncertainty about whether pooled effect sizes, dose-response patterns and typical delays between rainfall and peak incidence could be quantified. The researchers registered their protocol prospectively with PROSPERO, followed the PRISMA 2020 reporting guidelines, and screened 2,739 records after duplicate removal, assessing 312 full-text articles against PECO-structured eligibility criteria with substantial inter-rater agreement of kappa equal to 0.82.

Eligible studies were required to report quantitative associations between rainfall-related exposures and human disease outcomes that were laboratory-confirmed or linked to recognised surveillance definitions, a deliberate choice intended to reduce misclassification from non-specific febrile illness, a common problem in climate-health research. Studies reporting only undifferentiated fever, entomological indices without human disease data, or temperature-only exposures were excluded. Because rainfall metrics varied widely across the literature, from percentile-based contrasts such as rainfall above the 75th or 90th percentile to flooding events and continuous millimetre measurements, the team harmonised exposures carefully and graded the quality of meteorological inputs using a six-criterion framework covering data source documentation, temporal completeness, spatial representativeness, cross-validation, calibration and uncertainty reporting.

The headline result was a pooled relative risk of 1.74 for vector-borne disease incidence during elevated rainfall or flooding, but the authors emphasise that the 95 percent prediction interval ranged from 1.12 to 2.71, meaning the true effect in a comparable future setting could be considerably smaller or larger. Between-study heterogeneity was substantial, with an I-squared value of 78 percent. Disease-specific analyses showed the strongest association for dengue, with a relative risk of 1.89, followed by Japanese encephalitis at 1.67, malaria at 1.58 and chikungunya at 1.45, although the Japanese encephalitis estimate rested on only three studies and must be treated as imprecise. Geographic contrasts were striking: Bangladesh showed the highest pooled estimate at 2.13, India an intermediate 1.68, and Sri Lanka the lowest at 1.42, with statistically significant between-country differences confirmed by meta-regression.

The dose-response analyses, conducted using restricted cubic splines with three knots positioned at the 10th, 50th and 90th percentiles of the pooled rainfall distribution, suggested that risk remains relatively stable below approximately 100 millimetres of monthly rainfall and increases above that level. For dengue, the curve indicated rising risk above roughly 100 millimetres per month with an apparent peak near 200 millimetres, where the estimated relative risk reached approximately 2.3 compared with the reference level, before plateauing or declining above 400 millimetres, though confidence intervals widened considerably at the extremes of the rainfall distribution where few studies contributed data. For malaria, the spline model suggested an apparent double-peak pattern, with elevations near 125 and 350 millimetres per month, a shape that could reflect distinct hydrological mechanisms, moderate rainfall supporting temporary Anopheles breeding pools while heavier rainfall first flushes habitats and later creates post-flood transmission opportunities, but which was also sensitive to knot placement and model specification.

The authors are careful to stress that these threshold-like patterns are exploratory signals rather than fixed biological cut-points. Sensitivity analyses using alternative reference points at 50, 75, 125 and 150 millimetres per month, and four- and five-knot spline models, produced variation in curve shape, particularly for malaria, and the team interpreted such instability as evidence of uncertainty rather than stable operational thresholds. Chikungunya showed a gradual increase in risk across the rainfall range without a clear threshold pattern, limited by the small number of contributing studies. The biologically plausible mechanisms differ by disease: Aedes mosquitoes that transmit dengue and chikungunya breed in artificial containers, water-storage vessels and peri-domestic habitats replenished by moderate rainfall, whereas Japanese encephalitis transmission is more closely tied to rice agriculture, pig-rearing ecology and Culex mosquito habitats in rural settings.

One of the most operationally useful findings concerns lag periods. Across 67 studies reporting lag analyses, the pooled delay between rainfall exposure and peak disease incidence was 2.4 months, with a 95 percent confidence interval of 2.1 to 2.7 months. Disease-specific estimates were 2.2 months for dengue, 2.8 months for malaria and 1.9 months for chikungunya, differences consistent with variation in vector development rates, pathogen extrinsic incubation periods and disease-reporting dynamics. Meta-regression suggested that lag periods lengthened with latitude, increasing by 0.08 months per degree, and geographic patterns ran opposite to the effect sizes: Bangladesh showed the shortest mean lag at 2.1 months, India an intermediate 2.5 months, and Sri Lanka the longest at 2.7 months. This means country context alters both the magnitude of the rainfall signal and the warning time it may offer public health authorities.

The review also detected a temporal trend: pooled relative risks rose from 1.52 in studies covering 1990 to 2004, to 1.67 for 2005 to 2014, and to 1.91 for 2015 to 2024, with a meta-regression coefficient of 0.018 per year corresponding to an approximate 26 percent increase in the rainfall-disease association between the earliest and latest periods. The authors interpret this cautiously. A rising effect size could reflect genuine increasing vulnerability driven by urban expansion, informal settlements, drainage pressure and climate variability, but it could equally reflect improved surveillance, better diagnostics, changing publication patterns or residual confounding. Notably, when the analysis was restricted to time-series studies with comprehensive temporal adjustment, the temporal coefficient was attenuated and no longer statistically significant, suggesting that methodological differences between older and newer studies may partly explain the apparent trend.

Methodological rigour and its limits are central to the paper’s conclusions. Restricting the analysis to studies with comprehensive confounding control, defined as adjustment for four or more confounding categories, attenuated the pooled estimate to 1.62, and further restriction to high-quality studies produced a relative risk of 1.58, indicating that confounding may inflate part of the association without fully accounting for it. Publication bias analyses using Egger’s regression test and trim-and-fill adjustment suggested modest small-study effects, lowering the pooled estimate from 1.74 to 1.67, while leave-one-out and influence diagnostics showed that heterogeneity was distributed across the evidence base rather than driven by any single study. Flooding events produced larger pooled effects than heavy rainfall alone, a relative risk of 2.31 versus 1.67, which the authors attribute to the compound nature of flooding as an environmental and social exposure involving water stagnation, disrupted sanitation, displacement and altered healthcare access.

The practical message for public health is that rainfall can support early warning systems but should never be used alone, and that universal rainfall thresholds are not justified by the current evidence. The approximate 100 millimetre monthly signal and the two-to-three-month lag window may serve as useful starting points for local model development, but they require prospective validation against independent surveillance data, with estimation of sensitivity, specificity, positive predictive value, false-alarm rates and lead time before operational deployment. In resource-constrained health systems, repeated false alarms can waste limited response capacity and erode trust in early warning. The authors recommend combining rainfall monitoring with disease surveillance, temperature and humidity data, flood extent and drainage indicators, vector indices, population vulnerability measures and health-system readiness, and they call for stronger evidence from under-represented countries such as Pakistan, Nepal, Bhutan, the Maldives and Afghanistan, more studies of chikungunya and Japanese encephalitis, and sub-national and intra-urban research capable of resolving the street-level and household-level processes that country-level pooling inevitably averages away.

Subject of Research: Rainfall thresholds and temporal trends in vector-borne disease transmission in India, Bangladesh and Sri Lanka

Article Title: Rainfall thresholds and temporal trends in vector-borne disease transmission in India, Bangladesh and Sri Lanka: a systematic review and meta-analysis

Article References: Asaga, P. M., Kadukatti, V., Arsha, L., & Kroeger, A. (2026). Rainfall thresholds and temporal trends in vector-borne disease transmission in India, Bangladesh and Sri Lanka: a systematic review and meta-analysis. Environmental Challenges, 25, Article 101655. https://doi.org/10.1016/j.envc.2026.101655

Image Credits: AI Generated

DOI: Not provided

Keywords: vector-borne disease, dengue, malaria, chikungunya, Japanese encephalitis, rainfall, flooding, South Asia, meta-analysis, early warning systems, climate sensitivity, monsoon