Increasing intensity of enterovirus outbreaks projected with climate change

The study investigates how environmental (climate) and demographic (schooling) factors shape the seasonality and intensity of enterovirus transmission and projects how climate change may alter future outbreaks. Directly transmitted infections often display climate sensitivity, with many wintertime pathogens negatively correlated with absolute humidity, while enteroviruses typically peak in summer and are correlated with warmer conditions. The authors aim to quantify the effect of temperature and other climate variables on enterovirus transmission across multiple serotypes and settings (polio in the pre-vaccine US; EVA71 and CVA16 associated with HFMD in China and Japan) and to assess how late 21st-century warming could change epidemic peak sizes across Chinese provinces. Understanding these relationships is important for public health planning, as warming and changes in seasonal temperature ranges may intensify outbreaks and alter periodicity.

Prior work shows latitudinal gradients in the timing of enterovirus epidemics, with earlier outbreaks at lower latitudes. Dew point temperature has been associated with non-polio enterovirus transmission in the US, and humidity has influenced poliovirus viability in laboratory studies. Many respiratory diseases show negative correlations with absolute humidity in winter, with possible nonlinear responses in the tropics. HFMD has high burdens in Asia (notably China and Japan), and several studies have characterized its epidemiology and serotype-specific dynamics. Evidence from Japan and elsewhere indicates multi-annual cycles (biennial, triennial) and interactions among enterovirus serotypes. These findings motivate testing temperature and humidity metrics as transmission drivers and considering schooling as a key demographic aggregator for childhood infections.

Data: Weekly serotype-specific HFMD cases for EVA71 and CVA16 in Chinese provinces (2009–2013) constructed from province-level HFMD syndromic reports and serotype prevalence; national weekly EVA71 and CVA16 cases for Japan (1982–2015) from prior publications; weekly polio cases for US states (1940–1960) from Project Tycho. Climate data: ERA5 temperature, relative humidity, specific humidity for China/Japan; CHIRPS precipitation. For US polio, Berkeley Earth historical temperature (other variables unavailable). All gridded climate data were spatially and temporally averaged to match disease data. Demographic data: births and populations from national statistics (China, Japan) and CDC (USA). School timing: semester dates estimated for China (2022 Beijing data) and Japan; weekly dummy for polio (historic data on schooling unavailable).

Transmission estimation: The time-varying empirical transmission rate was estimated using a discrete-time TSIR framework at weekly resolution (generation time ~1 week). Empirical transmission Emβ = α I_t /(S_t N_t), with α fixed at 0.975. Susceptibles S_t were reconstructed using births and reported cases via linear regression of cumulative births on cumulative cases to estimate reporting rate p and deviations Z_t; p was ~1% for polio and ~2% for EVA71/CVA16. Seasonal transmission β_t was estimated biweekly via log-linearization. Assumption of long-lasting, largely immunizing infections was supported by serology; sensitivity analyses with low waning (SIRS) did not alter main results.

Regression analysis: Panel regressions of log empirical transmission on climate drivers (temperature, specific humidity, relative humidity, precipitation) with location and year fixed effects and clustered SEs at the location level. EVA71 and CVA16 (China+Japan) analyzed together with location dummies; polio analyzed separately with temperature only and weekly fixed effects. Dew point temperature was also tested in alternate models.

Seasonality metrics: Mean timing of cases computed as center of gravity on circular week-of-year axis. Epidemic intensity defined as inverse Shannon entropy of average weekly case distribution and normalized per pathogen. Periodicity assessed from simulated and observed series.

Simulation experiments: Present-day seasonal simulations used estimated temperature and schooling coefficients to generate weekly transmission: a sinusoidal temperature forcing across a range of seasonal temperature ranges (10–50 °C) and varying baseline transmission (log scale), with fixed school semesters (China), to examine regime transitions (semi-annual to annual, biennial/triennial, higher-order, chaotic dynamics) and the relative size of schooling vs temperature-driven peaks.

Climate change projections: Deterministic TSIR simulations for Chinese provinces used CMIP6 SSP585 weekly temperature climatologies (2080–2100 vs 1990–2010) bias-corrected via delta change added to ERA5 observations. Simulations incorporated 30 years of observed interannual variability to evaluate effects on average and maximum annual epidemic peak sizes over 30-year windows. Multiple CMIP6 models were run to capture model uncertainty; outputs compared across provinces.

Uncertainty analysis: For Beijing, an ANOVA partitioned variance in simulated peak sizes into contributions from climate model choice, uncertainty in the estimated temperature coefficient (sampled from its normal estimate), and interannual temperature variability, using 12,600 annual outbreaks generated from combinations of factors.

• Spatial patterns: All three pathogens (polio in US; EVA71 and CVA16 in China, EVA71 in Japan) showed latitudinal gradients with earlier outbreaks in warmer, lower-latitude locations. Mean temperature was negatively associated with mean timing of cases (earlier timing with higher temperatures): OLS p << 0.001 for polio and EVA71; p = 0.024 for CVA16. Southern Chinese provinces exhibited two annual peaks for EVA71/CVA16, with the secondary (often fall) peak growing in magnitude toward the tropics (e.g., Hainan). Epidemic intensity increased with larger seasonal temperature ranges (p << 0.001 across pathogens).
• Climate–transmission relationship: Panel regressions on log empirical transmission found temperature significantly and positively associated with transmission across polio, EVA71, and CVA16, robust to location and year fixed effects. Specific humidity, relative humidity, and precipitation were not significant predictors. Dew point temperature was significant but slightly worse than temperature as a predictor, suggesting it largely captures temperature effects.
• Schooling: School semester timing had a significant positive effect on transmission for EVA71 and CVA16 when included with temperature, and aligning schooling improved cross-pathogen consistency of temperature estimates.
• Mechanistic simulations: Increasing seasonal temperature range shifted modeled dynamics from semi-annual (temperature-driven spring peak plus schooling-driven fall peak) toward a single annual peak where temperature forcing dominates; higher ranges and baseline transmission produced biennial and higher-order cycles (consistent with observed bienniality in Liaoning). The ratio of schooling-to-temperature peak was >1 in tropical Hainan but declined and the schooling peak vanished as temperature range increased.
• Climate change projections (SSP585, 2080–2100): Across Chinese provinces, average and maximum epidemic peak sizes generally increased for EVA71 and especially CVA16. Worst-case scenarios projected up to 40% increases in peak size, depending on province and climate model. Model-averaged changes in maximum annual peak size across provinces ranged from approximately -5% to +24% for CVA16 and -4% to +8% for EVA71. Some models (e.g., INM-CM4-8) and northern provinces (e.g., Heilongjiang) showed decreases, linked to greater projected winter than summer warming reducing seasonal temperature range and thus outbreak intensity.
• Nonlinearity and heterogeneity: Nonlinear temperature–transmission responses can cause increases in seasonal transmission ranges even when seasonal temperature ranges decline, depending on local baselines and model projections, contributing to inter-model heterogeneity.
• Variability and uncertainty: Interannual temperature variability was the dominant contributor to projected outbreak size in Beijing (ANOVA): EVA71—variability 51%, temperature coefficient 48%, climate model 1%; CVA16—variability 79%, temperature coefficient 20%, climate model 1%. Large outbreaks tended to occur when a cooler year (accumulating susceptibles) was followed by a warmer year.

Findings demonstrate that temperature positively influences enterovirus transmission and, together with schooling, explains observed seasonality, including semi-annual peaks in warmer regions. Seasonal temperature range is a key determinant of epidemic intensity via its control of susceptible replenishment: warmer winters facilitate more continuous transmission, reducing summer peaks; increased seasonal contrast amplifies concentrated summer outbreaks and can lead to multi-annual cycles. Projections indicate that climate change will generally increase epidemic peak sizes in China, with magnitude depending on regional warming patterns, especially the balance of winter vs summer warming. The dominance of interannual temperature variability in driving extreme future outbreaks underscores the need to consider variability, not just mean changes, when assessing climate impacts on infectious diseases. These insights support integrating near-term climate forecasts into outbreak early warning systems and highlight potential benefits of targeted nonpharmaceutical interventions and vaccines during periods of predicted high transmission. The work addresses the research question by quantifying climate and schooling effects across multiple enteroviruses and demonstrating their implications for future outbreak intensity under warming scenarios.

This study integrates mechanistic transmission modeling, empirical climate–transmission inference, and climate projections to show that temperature and schooling jointly shape enterovirus seasonality and that climate change is likely to intensify outbreaks in many Chinese provinces, with worst-case increases up to 40%. Seasonal temperature range, rather than mean temperature alone, is pivotal for outbreak intensity and periodicity. Interannual temperature variability and uncertainty in the temperature coefficient drive much of the projection uncertainty, suggesting priorities for improved parameter estimation and operational use of seasonal-to-interannual climate forecasts in public health planning. Future work should expand analyses to other regions, incorporate longer and finer-resolution serotype-specific datasets, evaluate additional drivers such as mobility and holidays, refine demographic projections (e.g., declining birth rates), and assess vaccination strategies, including potential multivalent vaccines, to mitigate intensifying outbreaks.

• Predictors: Temperature and schooling alone do not fully capture the shape of seasonal transmission; unmodeled factors (e.g., seasonal mobility, national holidays) may contribute.
• Data constraints: Chinese serotype-specific series span only 5 years, limiting detection of longer-term periodicities; Japanese data are national-level, potentially masking prefecture-level heterogeneity. Polio analysis is limited to pre-vaccination era data and temperature only (other climate variables unavailable).
• Scope: Projections restricted to China; external generalizability to other regions not assessed here. Polio not projected due to current limited circulation and vaccine-era dynamics.
• Demography and vaccination: Main projections hold births and population fixed at 2013 levels; demographic change (e.g., declining birth rates) could alter susceptible supply, though not the direction of climate effects in sensitivity analyses. EVA71 vaccination (introduced 2016) and potential multivalent vaccines are not fully integrated in main results; sensitivity with 60% coverage did not substantially change conclusions.
• Model assumptions: TSIR assumes largely immunizing infections; small recurrence rates exist. SIRS sensitivity with low waning did not change main findings but adds uncertainty to immunity duration. Reporting rate estimates are approximate (~1–2%). Climate model biases addressed via delta method but residual biases may persist.