Time-series analysis of climatic drivers of pediatric rotavirus and adenovirus infections in post-pandemic China

Abstract

Objective

To investigate the epidemiological characteristics and climate-sensitive transmission patterns of rotavirus (RV), adenovirus (AdV), and RV-AdV coinfections among children with acute gastroenteritis in Wuhan, China, during the post-COVID-19 era.

Methods

We conducted a retrospective time-series study of 53,088 pediatric patients tested for RV and AdV from April 2020 to August 2024. Age-stratified positivity rates were analyzed alongside temporal trends. Daily meteorological data (temperature, relative humidity, wind speed) and air pollutants were incorporated into a generalized additive model (GAM) framework with distributed lag nonlinear models (DLNMs) to assess delayed and nonlinear associations between weather exposures and virus positivity.

Results

RV was the most frequently detected virus (7.74%), peaking in preschool-aged children (3–6 years), while AdV showed broader age distribution with highest positivity in toddlers (1–3 years). Coinfections were most common in children under 2 years. Significant seasonal and interannual fluctuations were observed, particularly a post-pandemic RV surge in 2024. Spearman analysis revealed inverse correlations between RV/AdV positivity and temperature. DLNMs showed that RV risk increased significantly under low wind (RR = 1.79, lag 0), cold (RR = 1.47, lag 21), and dry conditions (RR = 1.23, lag 15). AdV exhibited a U-shaped humidity-risk curve and increased risk with cold and moderately humid conditions. Coinfection risk was primarily driven by cold temperatures. Significant interactions were found between temperature and wind (RV) and between temperature and humidity (AdV). Season-stratified analysis indicated atypical spring and summer RV peaks.

Conclusion

This study is the first in China to apply DLNMs in a large pediatric cohort to evaluate climate-driven risk of RV and AdV infections. Findings reveal pathogen-specific, delayed meteorological sensitivities and post-COVID shifts in transmission patterns, providing a foundation for climate-informed surveillance and targeted interventions in pediatric gastroenteritis control.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-24010-6.

Introduction

Acute gastroenteritis remains one of the leading causes of morbidity and mortality among children under five worldwide, particularly in low- and middle-income countries, as highlighted in the Global Burden of Disease Study 2013 [1]. Rotaviruses (RV) and adenoviruses (AdV) are among the most common viral pathogens associated with pediatric diarrhea and are responsible for a significant proportion of hospital admissions and outpatient visits worldwide [2]. While RV has long been recognized as a major cause of severe dehydrating gastroenteritis, recent evidence from China has demonstrated that AdV, particularly types 40, 41, and 3, is also significantly associated with acute pediatric diarrhea, especially in children under 3 years of age [3]. In 2019, RV alone is estimated to cause 128,500 deaths in children under five, mainly in Asia and sub-Saharan Africa, despite the introduction of vaccines in many countries [2]. Recent studies in China show that rotaviruses are responsible for 25%-35% and adenoviruses for 3%-5% of laboratory-confirmed viral gastroenteritis cases in children, with regional and seasonal differences [4, 5].

RV and AdV exhibit distinct age- and season-dependent patterns. RV primarily affects infants and young children, with peak incidence in cooler seasons, while AdV also predominantly infects young children but shows greater seasonal variability, with cases distributed throughout the year, often peaking in summer or autumn [6–9]. In China, RV and AdV are significant contributors to pediatric diarrhea. While rotavirus vaccines, such as the Lanzhou Lamb Rotavirus (LLR) vaccine and RotaTeq (RV5), have been available since 2000 and 2018 respectively, they are not included in the national immunization program. As a result, coverage remains relatively low, ranging from approximately 25–50% in urban areas such as Guangzhou and Shanghai [10, 11]. Effective control measures remain critical to reducing the burden of rotavirus-associated disease.

The COVID-19 pandemic and associated non-pharmaceutical interventions (NPIs), including lockdowns, school closures and widespread masking, have significantly altered these patterns. Studies from China and other regions reported a temporary decline in RV and AdV transmission during the peak of COVID-19 containment, followed by an atypical rebound after restrictions were eased [12]. These changes underscore the dynamic epidemiology of enteric viruses in the post-pandemic period and emphasize the need for increased surveillance and region-specific analysis. Children under the age of five are particularly at risk due to their immature immune systems, risk of behavioral exposure and frequent close contact in daycare centers [13]. In addition, younger children are more susceptible to mixed infections with several pathogens, which can worsen the severity of the disease [14]. Understanding age-specific infection patterns is crucial for the development of targeted prevention strategies and the optimization of clinical management.

Meteorological conditions are increasingly recognized as crucial factors in the transmission of enteric viruses. Research has established an inverse relationship between RV activity and ambient temperature, as well as complex relationships with relative humidity and precipitation [15–17]. Cold, dry conditions may increase virus stability or host susceptibility, while higher temperatures and humidity may reduce viral persistence or alter contact patterns. For AdV, the environmental drivers are less well defined, but are likely to be associated with similar seasonal modulations of virus survival and host behavior. Although there is growing evidence, most studies have focused on respiratory viruses, so the environmental sensitivity of enteric viruses in pediatric populations remains understudied. Furthermore, few studies have quantitatively assessed the lagged and nonlinear effects of multiple meteorologic variables on RV and AdV infections using advanced time-series models. Distributed lag nonlinear models (DLNMs) provide a robust framework for capturing lagged and nonlinear associations between environmental exposures and health outcomes [18, 19]. While DLNMs are widely used in respiratory infections, their use in the study of pediatric enteric viruses in China remains limited, especially in long-term, multicenter studies that take into account multiple meteorological factors such as temperature, humidity, and wind speed.

To address these gaps, we conducted a large-scale, retrospective, time-series study of over 53,000 pediatric patients with acute gastroenteritis at Wuhan Children’s Hospital, a large tertiary pediatric center in central China, from April 2020 to August 2024. This study is one of the first to apply DLNM modeling to quantify the effects of multiple meteorological factors on RV and AdV infections in a large pediatric cohort in China. Our objectives were: (1) to characterize the epidemiological features of RV, AdV and RV-AdV co-infections in different age groups and over time; and (2) to quantify the independent and interactive effects of temperature, humidity and wind speed on virus positivity rates. This research provides new insights into the environmental determinants of childhood transmission of enteric viruses and provides a methodological basis for climate-sensitive public health interventions. It is important to note that ADV detection in this study was performed using a commercial antigen assay specific for enteric AdV serotypes 40 and 41 (HAdV-F40/41), which are the predominant causative agents of adenoviral gastroenteritis in children.

Materials and methods

Study population

This retrospective time series study was conducted from April 1, 2020 to August 31, 2024 at Wuhan Children’s Hospital, a large tertiary pediatric center in central China. Children aged 0–18 years who presented with acute gastroenteritis and had stool viral antigen testing for RV and AdV were included. Acute gastroenteritis was defined according to WHO criteria as ≥ 3 episodes of watery or loose stools per day, with or without vomiting, fever or dehydration. Only cases with acute onset were included, and patients with prolonged or recurrent symptoms suggestive of chronic diarrhea were excluded based on clinical judgment and medical record review. The exclusion criteria were as follows: (1) noninfectious diarrhea (e.g., food allergy, inflammatory bowel disease), (2) incomplete clinical or demographic data, and (3) duplicate entries or repeat visits within a 30-day period. Notably, rotavirus vaccination history was not recorded in the hospital’s electronic medical system and was therefore unavailable for analysis, limiting our ability to assess the influence of local vaccine coverage on RV detection trends. After applying these criteria, a total of 53,088 pediatric cases were included in the final analysis. Age was analyzed as a categorical variable using clinically meaningful pediatric age groups to reflect developmental stages and facilitate epidemiological interpretation. Specifically, participants were categorized into five age groups: young infants (0–6 months), infants (6 months-1 year), toddlers (1–3 years), preschool-aged children (3–6 years), and school-aged children (6–18 years). This classification aligns with pediatric clinical standards and supports age-stratified epidemiological comparisons.

Specimen collection and detection

Stool samples were collected in clean containers by trained personnel from both outpatients and inpatients with diarrhea or fever. These samples were immediately sent to the clinical laboratory for RV and ADV antigen testing. Group A RV and ADV antigens were detected using an immunochromatographic assay with colloidal gold according to the manufacturer’s protocol (Antigen Detection Kit, Livzon Diagnostics Inc., Zhuhai, China). The AdV test kit specifically targets enteric adenovirus serotypes 40 and 41 (HAdV-F40/41), which belong to species F and are the principal agents of adenoviral gastroenteritis. Non-enteric AdV types (e.g., species B and C), typically associated with respiratory illness, are not detected by this assay and were excluded from analysis. Results were recorded as either positive or negative for each virus. For cases tested for multiple pathogens, co-infection was also recorded.

Environmental data collection

Daily meteorological data, including average temperature (°C), relative humidity (%) and wind speed (km/h), were obtained from the China Meteorological Data Service Center (CMDC). Data on air pollutants were obtained from the China National Environmental Monitoring Center (CNEMC) and included values for particulate matter (PM_2.5, µg/m³), inhalable particulate matter (PM₁₀, µg/m³), carbon monoxide (CO, mg/m³), nitrogen dioxide (NO₂, µg/m³), sulfur dioxide (SO₂, µg/m³) and ozone (O₃, µg/m³). Seasons were defined as follows: spring (March-May), summer (June-August), autumn (September-November), and winter (December-February), consistent with conventional meteorological classification in China.

Statistical analysis

Daily counts of RV and AdV infections were summarized along with the corresponding meteorological and air quality indicators using descriptive statistics, including means, standard deviations, and five-digit summaries (minimum, 25th percentile, median, 75th percentile, and maximum). Categorical and ordinal variables, such as age groups and seasonal strata, were described with frequencies and percentages. Age distributions between RV-, AdV-, and coinfection-positive groups were summarized as medians with interquartile ranges (IQRs) and compared using the Kruskal-Wallis H test, followed by Bonferroni-corrected Dunn’s post hoc tests for pairwise significance. Groupwise comparisons of positivity rates across different age groups were performed using chi-square tests. To control for multiple comparisons in these pairwise analyses, p-values were adjusted using the Bonferroni method where appropriate. For binary comparisons (e.g., sex and inpatient vs. outpatient), Bonferroni correction was not applied as each test involved a single hypothesis between independent groups, and a p-value < 0.05 was considered statistically significant. To assess monotonic associations between pathogen positivity rates and meteorological or air pollution variables, pairwise Spearman correlation analyses were performed. Given the exploratory nature of this analysis for selecting candidate predictors in subsequent DLNM modeling, Bonferroni correction was not applied. Variables with significant correlations (p < 0.05) were retained as predictors, while major air pollutants were included as covariates.

To further investigate possible lagged and nonlinear relationships between meteorological variables and daily infection counts, we implemented a DLNM within a Generalized Additive Model (GAM). The overdispersion of the daily count data was accounted for by a negative binomial distribution with a log-link function. Cross-base functions were applied to the meteorological factor using natural cubic splines for both the exposure and lag dimensions. A maximum lag of 21 days was chosen to account for lagged effects [20]. The core specification of the model was:

In this model, Y_t denotes the number of daily positive cases (e.g. for RV or ADV) on day t. The response variable follows a log-link function under a negative binomial distribution to account for overdispersion. The term α stands for the model intercept. The main exposure of interest is the meteorological factors, which are included as a crossbasis function cb (Meteorological Factorst, l) that captures both the nonlinear and lagged effects over time. The degrees of freedom (df) for the exposure-lag structures were selected based on the Akaike information criterion (AIC) to ensure optimal fit of the model. Air pollutants, including PM_2.5, PM₁₀, CO, NO₂, O₃ and SO₂, are modeled using natural cubic splines with 3 degrees of freedom to account for their potential nonlinear confounding effects [21]. A natural spline function of calendar time with 7 degrees of freedom per year is included to account for long-term trends. Additional categorical variables include the day of the week (DOW_t), the season (Season_t) and the holidays or school vacations (Holiday_t), which are coded as binary indicators. The log-transformed total number of samples tested daily log (Total_t) is included as an offset to control for variations in testing effort. Tensor product interaction terms (ti) were included in the GAMs to flexibly capture possible non-linear synergistic effects between meteorological variables. To quantify the effects of meteorological factors on RV or AdV positivity, we estimated relative risks (RRs) and 95% confidence intervals (CIs) derived from the adjusted DLNM. Both the one-day lag effects and the lag effects were calculated over a lag window of 0 to 21 days. For example, lag 0 corresponds to the day of sampling, while lag 1 reflects the exposure on the previous day. The median value of the meteorological factors during the study period was chosen as the reference for all RR estimates in order to obtain a stable and interpretable baseline. Model diagnostics were performed using residual plots and partial autocorrelation function (PACF) analysis to ensure that model assumptions were met and autocorrelation of residuals was minimized.

Results

Descriptive statistics of pathogen positivity and meteorological conditions

A total of 53,088 pediatric patients with acute gastroenteritis were tested for RV, AdV and RV-AdV coinfection between April 2020 and August 2024 (Supplementary Table S1). Of these, 4,110 (7.74%) were RV-positive, 2,656 (5.00%) were AdV-positive and 795 (1.50%) were co-infected with both viruses. The study population consisted of 61.3% men (n = 32,540) and 38.7% women (n = 20,548). Most patients were infants aged 1–3 years (33.8%), followed by infants aged 6 months-1 year (20.8%), toddlers < 6 months (20.1%), preschool children aged 3–6 years (15.6%) and school-age children aged 6–18 years (9.7%). In addition, 70.1% of patients were treated as outpatients (n = 37,190), while 29.9% of patients were hospitalized (n = 15,898). The average daily wind speed during the study period was 7.9 km/h (SD = 3.03), the average temperature was 18.77 °C (SD = 9.28) and the average relative humidity was 72.68% (SD = 14.26%).The air quality indicators showed average daily concentrations of PM_2.5 at 36.04 µg/m³, PM₁₀ at 56.72 µg/m³, NO₂ at 35.37 µg/m³, SO₂ at 8.05 µg/m³, CO at 0.81 mg/m³ and O₃ at 100.13 µg/m³. These pollutants were included as covariates or controlled confounders in the subsequent analyzes to assess their impact on the dynamics of virus transmission.

Positive distribution of pathogen across different age groups, genders, and patient types

Positivity rates of RV, AdV and coinfection varied considerably between age groups, with pairwise comparisons showing marked differences between groups (Table 1; Fig. 1A-C). Overall χ² tests confirmed significant differences for all three viruses (all p < 0.001), and pairwise p-values were Bonferroni-adjusted to account for multiple testing. For RV, the highest positivity rate was observed in the preschool group (3–6 years), which was 12.58%, significantly higher than in other age groups (adjusted p < 0.001). AdV positivity was significantly higher in toddlers aged 1–3 years (7.24%) and preschool children aged 3–6 years (7.02%) than in other age groups (adjusted p < 0.001). Although co-infection positivity was not significantly different in most age groups, the lowest rate was significantly lower in school-age children (6–18 years) (0.62%) than in the younger groups (adjusted p < 0.001). Detailed pairwise comparisons are presented in Supplementary Table S2.

Table 1.

Comparison of demographic characteristics and positive rates of RV, ADV and Co infection from 2020 to 2024

Fig. 1.

Distribution and temporal trends of RV, AdV, and coinfections among pediatric gastroenteritis cases. (A) RV positivity rates stratified by age group, sex, and patient type (left), and annual age-specific RV positivity rates from 2020 to 2024 (right). (B) AdV positivity rates stratified by age group, sex, and patient type (left), and annual age-specific AdV positivity rates from 2020 to 2024 (right). (C) Coinfection positivity rates stratified by age group, sex, and patient type (left), and annual age-specific coinfection positivity rates from 2020 to 2024 (right)

In addition to these cross-sectional results, Fig. 1D illustrates the annual trends in positivity rates. Notably, RV positivity in preschool children (3–6 years) jumped after 2021 and peaked in 2024 (28.18%), while AdV rates in groups 1–3 and 3–6 years remained relatively high. Co-infections remained uncommon but showed subtle year-to-year fluctuations, especially in younger children. These trends were also confirmed by analyzing the age distribution (Supplementary Figure S1), where the median ages of RV-, AdV- and coinfection-positive children were 2.00 years (IQR: 1.00-3.08), 1.67 years (IQR: 1.00–3.00) and 1.00 years (IQR: 0.58-2.00), respectively. Statistical comparisons confirmed significant differences between the groups (Kruskal-Wallis p < 0.001; Bonferroni-corrected Dunn tests, all p < 0.001), underlining the greater susceptibility of the younger age groups to mixed infections.

Sex-specific comparisons revealed no statistically significant differences in positivity rates between men and women for RV (7.73% vs. 7.76%, χ² = 0.01, p = 0.902), AdV (4.97% vs. 5.06%, χ² = 0.18, p = 0.668) or coinfection (1.56% vs. 1.40%, χ² = 2.20, p = 0.138), indicating a broadly comparable distribution between the sexes. In contrast, statistically significant differences were observed between inpatients and outpatients. RV positivity was significantly higher in inpatients (8.27%) than in outpatients (7.52%) (χ² = 6.75, p = 0.009), while AdV positivity was significantly lower in inpatients (3.52%) than in outpatients (5.64%) (χ² = 110.71, p < 0.001). Similarly, the co-infection positivity rate was significantly lower in inpatients (0.23%) than in outpatients (2.04%) (χ² = 246.10, p < 0.001). Each test involved a single binary comparison between independent inpatient and outpatient populations for a specific pathogen, and thus Bonferroni correction was not applied.

Associations between temporal pathogen trends and meteorological factors

The overall positivity rate for RV, AdV and coinfections showed significant interannual variability between 2020 and 2024, ranging from 6.2 to 18.8% (χ² = 734.9, p < 0.001; Fig. 2). The positivity rate peaked in 2021 (18.8%), declined sharply in 2023 (9.3%) and partially recovered in 2024 (16.8%), likely reflecting dynamic shifts in virus transmission and pandemic-related public health interventions. When stratified by pathogen, RV showed the most pronounced fluctuation, with a peak in 2021 (10.9%, 1,779 cases), followed by a decline and rebound in 2024 (12.0%, 949 cases). AdV followed a similar pattern with a peak positivity of 6.1% (1,010 cases) in 2021, followed by a gradual decline in the following years. Co-infections remained consistently low, reaching a peak of 2.4% in 2022. Statistical comparisons confirmed significant year-to-year differences in positivity rates for RV (χ² = 1106.5), AdV (χ² = 104.1) and coinfection (χ² = 87.6) (all p < 0.001), highlighting the dynamic epidemiological landscape during the five-year period.

Fig. 2.

Annual trends of RV, AdV and coinfection: number of positive cases and positivity rates from 2020 to 2024. The bar charts show the annual number of positive cases for RV, AdV, coinfection and total virus-positive samples. The overlaid lines represent the corresponding annual positivity rates (%) for each pathogen group and the overall positivity

To further investigate the temporal patterns of pathogen circulation in relation to environmental conditions, we overlaid daily positivity rates for RV, AdV and coinfections with concurrent meteorological data, including average temperature, relative humidity and wind speed (Supplementary Figure S2). RV showed clear seasonal peaks between February and May of each year, with increased positivity occurring primarily in the cooler months. AdV showed a more complex seasonal pattern with smaller peaks between December and February, often coinciding with periods of moderate humidity. Coinfection trends were more erratic but generally showed a slight increase in the winter and early spring months.

Correlation between pathogen positivity rates and environmental factors

To preliminarily investigate the relationship between pathogen activity and environmental exposure, a Spearman rank correlation analysis was performed between the daily positivity rates of RV, AdV and co-infections and the main meteorological and air pollution variables (Supplementary Figure S3). RV positivity showed a moderate negative correlation with average temperature (ρ = -0.24, p < 0.001) and a weak but significant positive correlation with wind speed (ρ = 0.17, p < 0.001). Similarly, AdV positivity was negatively associated with temperature (ρ = 0.15, p < 0.001) and showed a weak positive correlation with relative humidity (ρ = 0.07, p = 0.003) and wind speed (ρ = 0.05, p = 0.04). Coinfection positivity showed a weak inverse correlation with relative humidity (ρ = 0.06, p = 0.023). These correlation results formed the basis for the selection of variables in the subsequent Distributed Lag Nonlinear Models (DLNMs). Only environmental variables with a statistically significant correlation (p < 0.05) were retained as candidate predictors. To control for potential confounding, major air pollutants were included as covariates in the models. Highly collinear pollutants (|ρ| >0.7) [21], such as PM₁₀ and NO₂, were excluded to avoid multicollinearity, while PM_2.5, CO, SO₂, and O₃ were retained for adjustment. Given the exploratory nature of this analysis for DLNM variable selection, Bonferroni correction was not applied. Instead, subsequent multivariable modeling accounted for confounding and collinearity.

Meteorological effects on RV, adv, and coinfection positivity

Using the DLNMs, we observed clear meteorological influences on RV, AdV and RV-AdV coinfections over a period of 21 days (Table 2; Fig. 3). For RV, low temperature (-3.3 °C), low humidity (33.0%) and low wind speed (1.2 km/h) were associated with increased relative risks (RR), with wind speed having the strongest effect (RR = 1.79, 95% CI: 1.15–2.79, lag 0), followed by temperature (RR = 1.47, 95% CI: 1.25–1.73, lag 21) and humidity (RR = 1.23, 95% CI: 1.12–1.35, lag 15) (Table 2). In extreme cold (5th percentile, 1.4 °C), RR increased significantly between days 7–12 and 16–21, peaking at day 21 (RR = 1.44, 95% CI: 1.25–1.65) (Fig. 4A). For AdV, DLNM showed increased risk at low temperatures (3.3 °C, RR = 1.10, 95% CI: 1.02–1.19, lag 9), moderate humidity (57.2%, RR = 1.07, 95% CI: 1.00-1.12, lag 17) and moderate wind speeds (7.6 km/h, RR = 1.01, 95% CI: 1.00-1.01, lag 19). The temperature effect exhibited a U-shaped curve, with increased RR in cold conditions (Fig. 3D-F). At the same time, exposure to extreme cold (1.4 °C) resulted in a delayed increase in risk from day 5–15 (Fig. 4B). In coinfection, the increased RR was primarily caused by low temperature (3.3 °C, RR = 1.85, 95% CI: 1.19–2.88, lag 20) and moderate humidity (81.6%, RR = 1.08, 95% CI: 1.02–1.14, lag 20) (Fig. 3G-I). Wind speed had a smaller effect (6.6 km/h, RR = 1.02, 95% CI: 1.00-1.03, lag 2). Of note, extremely low temperatures significantly increased the risk of coinfection between lag 17–21, with a peak at lag 21 (RR = 1.64, 95% CI: 1.14–2.35), while high temperatures (95th percentile) were also associated with a moderately delayed risk (lag 11–18, RR = 1.25, 95% CI: 1.09–1.44) (Fig. 4C). Interaction analysis revealed a statistically significant interaction between temperature and wind speed for RV (p < 0.005), as illustrated in Fig. 5A. This suggests that low temperatures combined with low wind speed may synergistically enhance RV transmission. For AdV, a significant interaction was observed between temperature and relative humidity (p < 0.001; Fig. 5E). No significant interactions were found for coinfections (Fig. 5G-I; all p > 0.1). Detailed results are provided in Supplementary Table S3.

Table 2.

Maximum relative risk (RR) for meteorological variables with reference to baseline conditions

Variable-Meteorological indicators	Reference value	Exposure value	Maximum RR (95%CI)	Lag days
RV - Temperature(°C)	19.8	-3.3	1.47(1.25, 1.73)	21
RV - Relative Humidity (%)	73.0	33.0	1.23(1.12, 1.35)	15
RV - Wind Speed (km/h)	7.4	1.2	1.79(1.15, 2.79)	0
AdV - Temperature (°C)	19.8	-3.3	1.10(1.02, 1.19)	9
AdV - Relative Humidity (%)	73.0	57.2	1.07(1.00, 1.12)	17
AdV - Wind Speed (km/h)	7.4	7.6	1.01(1.00, 1.01)	19
Coinfection - Temperature (°C)	19.8	-3.3	1.85(1.19, 2.88)	20
Coinfection - Relative Humidity (%)	73.0	81.6	1.08(1.02, 1.14)	20
Coinfection - Wind Speed (km/h)	7.4	6.6	1.02(1.00, 1.03)	2

The table lists the reference values, maximum RR values, 95% confidence intervals (CI), and lag days for each variable and exposure

Fig. 3.

Three-dimensional exposure-delay-response surfaces for meteorological effects on RV, AdV and coinfection. Each panel shows the 3D risk surface estimated with DLNMs for RV (A-C), AdV( D-F) and coinfection (G-I) based on daily values of (A, D, G) average temperature (°C), (B, E, H) relative humidity (%) and (C, F, I) wind speed (km/h). The vertical axis represents the relative risk (RR), with the reference set to the median value of each meteorological variable. The lag time ranges from 0 to 21 days

Fig. 4.

Lag-specific relative risk (RR) curves at extreme meteorological percentiles for RV, AdV and coinfection. This figure shows the lag-response relationships estimated with DLNMs for (A) RV, (B) AdV, and (C) coinfection stratified by meteorological variables at the 5th (P5), 25th (P25), 75th (P75), and 95th percentiles. The Y-axis represents the RR relative to the median reference value, and the X-axis shows the lag in days (0–21). The shaded areas indicate 95% confidence intervals. The curves show the delayed effects of extreme exposures (e.g. low temperature, high humidity, low wind) on the risk of infection over the time lags

Fig. 5.

Interaction effects between meteorological variables on RV, AdV, and coinfection risk. Panels A-C show interaction surfaces for RV: (A) temperature × wind speed, (B) temperature × relative humidity, and (C) wind speed × relative humidity. Panels D-F show corresponding interactions for AdV, and panels G-I for coinfections. The color scale represents the predicted log-relative risk derived from tensor-product smooths in the GAM models. Significant interactions were identified between temperature and wind speed for RV (p < 0.005; panel A) and between temperature and relative humidity for AdV (p < 0.001; panel E). No statistically significant interactions were observed for coinfections (all p > 0.1). Full statistical details are provided in Supplementary Table S3

Seasonal effects on pathogen-specific risk

The seasonally adjusted DLNM modeling revealed clear seasonal risk patterns (Supplementary Table S4, Fig. 6). The risk of RV positivity was significantly higher in spring (RR = 2.75, 95% CI: 2.20–3.43) and summer (RR = 1.80, 95% CI: 1.41–2.29) than in fall. For AdV, the risk only increased in summer (RR = 1.37, 95% CI: 1.07–1.75). In contrast, the risk of coinfection was significantly lower in summer (RR = 0.66, 95% CI: 0.43-1.00), with no significant differences in spring and winter. These seasonal dynamics indicate different responses of the pathogens to climatic and behavioral changes in the different seasons.

Fig. 6.

Seasonal estimates of relative risk (RR) for RV, AdV and coinfection. Forest plots show season-specific RR estimates and 95% confidence intervals for (A) RV, (B) AdV, and (C) coinfection, using fall as the reference season. Values were derived from DLNM models stratified by spring, summer, fall and winter

Model diagnostics and sensitivity analyses

Model diagnostics confirmed the appropriateness of the DLNM models. Q-Q plots showed that the residuals in all three models (RV, AdV, and coinfection) were approximately normally distributed, with only minor deviations at the extremes (Supplementary Figure S4-S6). Residuals versus predictors showed no major heteroscedasticity, and histograms showed almost symmetrical distributions, especially for RV and AdV. ACF plots showed no significant autocorrelation, indicating that temporal dependencies were well controlled. In addition to degrees of freedom selection (df = 3–6), we conducted sensitivity analyses using a reduced lag window (14 vs. 21 days) for RV, AdV, and coinfection. As shown in Supplementary Figure S7, all three models retained consistent exposure-lag-response surface patterns under the shorter lag setting. The overall direction of associations remained robust, although the 14-day models slightly attenuated the peak relative risks for RV and coinfection, where effects extended beyond lag 14. These results confirm the stability and reliability of the DLNM estimates across key model specifications.

Discussion

To our knowledge, this is the first study in China to apply DLNMs in a large pediatric cohort to evaluate the independent and interactive effects of multiple meteorological factors on RV and AdV infections. This novel approach provides new insights into the environmental determinants of pediatric enteric virus transmission. In this large retrospective cohort of pediatric gastroenteritis in Wuhan (2020–2024), we observed substantial shifts in the epidemiological and environmental dynamics of RV, AdV and RV-AdV co-infections. RV was the most frequently detected virus (7.74%), with peak positivity in preschool-aged children (3–6 years). This deviates from traditional RV epidemiology, where the highest burden is typically seen in children under 2 years of age [12, 22]. Possible explanations include delayed exposure due to pandemic-related non-pharmaceutical interventions (NPIs) such as the closure of daycare centers and lower social mixing, as well as low rotavirus vaccination rates in China [23].

In contrast, AdV had a broader age distribution, with the highest positivity in young children (1–6 years) and persistent detection in preschool children, probably due to its greater environmental resistance, diversity of serotypes and persistent shedding [14, 24]. Coinfections occurred most frequently in children under 2 years of age, with an average age of 1.00 years, indicating increased susceptibility in infancy due to immature mucosal immunity and high exposure to close contact [16, 25]. These patterns underscore the importance of early interventions, including timely immunizations, breastfeeding support, and infection control strategies in child care centers. Notably, RV activity appeared to increase during periods when the mean daily temperature fell below approximately 15 °C. AdV exhibited less consistent alignment with a single meteorological factor, although periods of moderate to high humidity (≥ 70%) often overlapped with peak activity. Coinfection peaks were sporadic but more likely to occur under cold and low-humidity conditions. These descriptive results suggest that meteorological conditions may play a role in modulating seasonal transmission dynamics of enteric viruses in children. Gender comparisons showed no significant differences in virus positivity rates, which is consistent with previous studies [14]. However, stratification by patient type showed that RV was more common in hospitalized children, while AdV and co-infections were more common in outpatients. This is consistent with the typical clinical profiles of these viruses, with RV more likely to cause severe dehydrating diarrhea and AdV associated with a prolonged but milder illness [16, 26]. These patterns align with both national and global epidemiological evidence. A large pediatric study from Suzhou, China, reported that RV accounted for 20.7% of inpatient and 39.3% of outpatient diarrheal cases, suggesting a higher clinical burden among hospitalized children [23]. This supports the global consensus that RV remains a leading cause of severe gastroenteritis requiring hospitalization in young children, especially in low- and middle-income countries [27]. In contrast, recent post-pandemic surveillance in Shanghai further confirmed that AdV was more frequently detected in outpatient children, reinforcing its milder disease profile [28]. These findings underscore the importance of accounting for clinical setting and healthcare utilization patterns when interpreting virus surveillance data and evaluating disease burden.

This study period coincides with the COVID-19 pandemic, raising valid concerns about the impact of pandemic-related non-pharmaceutical interventions (NPIs) on the observed infection patterns. Evidence from Wuhan between 2011 and 2019 reported a high baseline RV positivity of approximately 25.5% among pediatric patients, which declined markedly during 2020–2021 amid stringent lockdowns, daycare closures, and reduced interpersonal contact [29, 30]. In Wuhan, strict citywide lockdowns and mobility restrictions were in effect from late January to early April 2020. Although these measures were officially lifted in April, the reopening of schools and childcare centers proceeded gradually between May and September 2020, with enhanced hygiene protocols, mask mandates, and health screenings remaining in place throughout most of 2020–2021. Correspondingly, we observed that RV and AdV positivity rates in our cohort were lowest in 2020, followed by a notable resurgence from mid-2021 onwards, which temporally coincided with the phased return to regular social contact and in-person education. Notably, RV positivity peaked in 2024, suggesting a possible “delayed epidemic compensation” effect due to accumulated susceptibility among children with limited early-life viral exposure. These findings align with regional and national surveillance data and are consistent with reports from other countries, where post-pandemic RV activity exceeded pre-pandemic levels, supporting the hypothesis of an “immunity gap” or “immunity debt” caused by suppressed transmission during prolonged NPIs [31–33]. This temporal correspondence between the lifting of NPIs and the resurgence of enteric viruses strengthens the argument that these shifts represent transient disruptions rather than lasting epidemiological transitions. Although our study lacked pre-pandemic data from the same cohort, the consistency of our findings with historical regional data and global trends suggests that the observed changes are likely temporary. Future studies incorporating longer-term, multicenter data will be essential to determine whether these trends persist and to better understand the long-term consequences of pandemic-era interventions on viral transmission dynamics.

The Spearman correlation analysis revealed pathogen-specific environmental sensitivities: RV positivity showed a moderate inverse correlation with temperature and a weak positive association with wind speed; AdV showed weak correlations with temperature and humidity. Although these associations are limited by linear assumptions for the same day, they support the biological plausibility of meteorological influences on the transmission of enteric viruses [34, 35]. To overcome the limitations of simple correlations, we applied DLNMs that more accurately capture pathogen-specific, lagged and nonlinear climate relationships.

Our analysis is one of the first to apply DLNMs to pediatric enteric viruses, revealing complex relationships between climateand infection for RV, AdV, and their coinfections. DLNMs flexibly capture both the nonlinearity and lagged effects of weather exposures. This pathogen-specific approach has revealed different dose-response curves and lag patterns and uncovered new epidemiologic insights. For RV, we found that low wind speed on the same day was the strongest immediate driver of infection risk, while low temperature and humidity increased risk with a multi-day lag. The acute impact of calm weather may reflect increased indoor transmission: Low wind speeds are associated with stagnant air and greater crowding in homes or childcare facilities. Indeed, surveillance data from China showed that RV infection clusters occurred mainly during windless periods [36]. Cold, dry weather likely increases persistence in the environment and transmission, which is consistent with studies showing RV viability at 4–20 °C and low humidity [17, 37]. Our DLNM suggests that such conditions increase risk 1–3 weeks later, indicating critical windows of opportunity. In Kolkata, high temperatures during the monsoon suppressed RV rates, supporting our finding that RV thrives in cool, dry and windless conditions [34]. In contrast, AdV showed a U-shaped response to humidity and a preference for cold temperatures. Adenoviruses are non-enveloped and survive in a wide range of environmental conditions. Our model showed that both low and high humidity increased AdV risk, with delayed peaks around 9 °C. These results are consistent with laboratory studies indicating optimal viability at high humidity (~ 80–90%) and support epidemiologic reports of AdV spikes in winter [38]. The ability of the DLNM to identify these non-monotonic effects emphasizes its utility over simpler models. Coinfections occurred most frequently during cold, moderately humid periods when both RV and AdV were circulating. However, no synergistic interaction was detected, suggesting that the co-infections were caused by overlapping main effects. This pattern suggests that DLNM can separate common environmental triggers from possible pathogen–pathogen facilitation.

These weather-virus relationships reflect the interactions between viral stability, transmission routes and host factors. The stability of RV and AdV under cool, dry (RV) or humid (AdV) conditions prolongs infectivity on surfaces and possibly in aerosols. Low wind is likely a sign that people are indoors, a critical factor for fecal-oral transmission. The peak demographic of preschool children (aged 2 to 3 years) also coincides with a high risk environment [36, 39]. The viability of the viruses and the behavior of the hosts jointly determine the observed seasonal patterns.

Our results mirror DLNM-based studies on respiratory and gastrointestinal infections, which show peaks in the winter season under similar climatic conditions [37]. For example, DLNMs applied to norovirus and cholera have found delayed risk associated with rainfall or low temperatures [40]. Such models are increasingly being used for the early prediction of disease outbreaks. For public health, this means that weather forecasts can be integrated into surveillance and response systems. Forecasts of low wind and low temperatures could lead to early warnings, expanded testing and pre-season vaccination campaigns. Measures such as better ventilation of indoor spaces during windless periods could reduce transmission. Our DLNM study confirms that rotaviruses are most prevalent in cool, dry, stagnant air, underlining the benefits of climate-tailored measures.

Interaction analyses indicated that RV risk was greatest when low temperatures combined with wind, while AdV risk increased in cold and humid conditions, underscoring the importance of modeling synergistic exposures [41]. There were no such interactions in coinfections, suggesting additive rather than multiplicative effects. Seasonal stratified modeling revealed non-traditional patterns: RV risk was elevated in spring and summer, possibly due to shifts in susceptibility and exposure after the pandemic [12, 22, 42]. AdV showed an increase in summer that is consistent with subtropical circulation patterns [14]. Interestingly, the risk of coinfection was significantly lower in summer, suggesting that a shift in seasonal peaks could limit the co-circulation of RV and AdV.

This study offers several strengths: the use of a large pediatric data set over five years, the novel application of DLNMs to enteric viruses in China, and the inclusion of interaction terms and seasonal effects. However, there are also limitations. Since this is a single-center study, the results may not be generalizable beyond urban China. The antigen-based detection may underestimate the true positivity, especially for AdV [26]. The observational design cannot fully control for confounding factors such as health-seeking behavior and co-circulating pathogens. Meteorological data were aggregated citywide, not at the individual level, and environmental exposures may be misclassified. Another important limitation of our study is the absence of individual-level socio-demographic data, such as household income, parental education, and housing conditions. These factors are known to influence children’s nutritional status, hygiene practices, health-seeking behavior, and exposure to infectious agents. However, such variables were not available in our retrospective dataset. Future prospective studies with more granular socio-economic data are needed to disentangle the interplay between environmental exposures and social determinants of health. Future studies should aim to validate these findings in multicenter, prospective cohorts with molecular diagnostics and strain-level resolution. The integration of real-time weather data into clinical surveillance systems could enable predictive risk modeling. In addition, mechanistic studies of climate-virus-host interactions are needed to understand how environmental factors modulate infection risk and clinical severity.

In summary, our results reveal significant age- and season-specific shifts in pediatric RV and AdV epidemiology after COVID-19 and identify climate-sensitive transmission patterns using DLNMs. These findings can support climate-informed surveillance strategies and targeted public health measures for viral gastroenteritis in children.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

The study was conceived and designed by Feng Tang, Xiaomei Wang and Changzhen Li. Preliminary data analysis were conducted by Xiaomei Wang, and Jingjing Rao. Figures and tables were crafted, and the initial manuscript was drafted by Changzhen Li and Lei Xi. Manuscript revisions were performed by Feng Tang and Yun Xiang. All authors have critically reviewed and approved the final manuscript.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

We confirm that our study strictly adhered to all ethical standards set forth in the Declaration of Helsinki. The study protocol was approved by the Medical Ethics Committee of Wuhan Children’s Hospital (Approval No. 2024R050-E01). Given the retrospective design of this study, it did not interfere with standard medical care or infringe upon patient rights, nor did it pose any additional risk to participants. Patient identities were protected through anonymized coding, and all medical records were securely stored and accessed only by authorized researchers. In accordance with Chinese national regulations on biomedical research ethics, informed consent was waived by the Medical Ethics Committee of Wuhan Children’s Hospital due to the retrospective nature of the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.