Epidemiological trends of laboratory-confirmed influenza cases driven by meteorological factors in Anhui Province, China: a multi-city time-series analysis

Abstract

Background

Influenza poses a significant threat to public health, potentially influenced by environmental factors. However, the role of meteorological factors (MFs) on influenza risks in China remains underexplored. This study explored the effect of MFs on laboratory-confirmed influenza (LCI) cases in Anhui, China.

Methods

We analysed daily meteorological and influenza data between January 2015 and March 2023, to determine the relationship between temperature, relative humidity, wind speed and LCI cases, using two-stage time series analysis. First, we used distributed lag nonlinear models (DLNMs) to construct cross-basis functions capturing the non-linear and lagged effects of MFs, which were then incorporated into a generalized additive quasi-Poisson regression model for each city. Second, we conducted a random-effects meta-analysis to combine city-specific estimates. We further performed sub-group analysis by age and gender and explored effect modifications by population density, median MFs levels, longitude, and latitude through meta-regression.

Results

A total of 43,872 LCI cases were recorded in Anhui. A slight, non-significant negative association between temperature and influenza cases was observed at a single-day lag (RR = 0.9778; 95% CI: 0.9468–1.0098), but a positive association was found over cumulative lags (RR = 1.0263; 95% CI: 0.9721–1.0836). Relative humidity showed a positive association with influenza on single-day lag (RR = 1.0056; 95% CI: 0.9899–1.0216), but a slight negative association over cumulative lags (RR = 0.9974; 95% CI: 0.9927–1.0022). Wind speed displayed a slight, non-significant positive association at both single-day (RR = 1.0105; 95% CI: 0.9965–1.0246) and over cumulative lags (RR = 1.0083; 95% CI: 0.9498–1.0704). Temperature negatively associated with LCI cases across all genders and ages, at p = 0.0001, marginally moderated by population density (p = 0.0506).

Conclusions

In conclusion, while MFs showed non-significant associations with influenza in general population, sub-group analysis showed statistically significant temperature-LCI cases association. Population density marginally modified this association. Our findings enhance evidence-based knowledge for developing targeted interventions like early warning systems to reduce influenza risks.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-24182-1.

Background

Influenza remains a significant public health concern worldwide, contributing to high morbidity and mortality rates besides socio-economic disruption [1, 2]. Each year, it affects about 1 billion people worldwide, with 3 to 5 million severe cases and 290,000 to 650,000 related respiratory deaths [3, 4]. The World Health Organization previously linked between 250,000 and 500,000 annual deaths from all causes to seasonal influenza [4, 5]. In 2019, the Global Burden of Disease (GBD) study linked about 99,000 to 200,000 annual deaths resulting from lower respiratory tract infections to influenza [5, 6]. Loss of lives due to the affliction of influenza-related illness [3] has affected socio-economic growth through reduced workforce productivity and increased economic burden on healthcare amenities [2]. Considerable budgetary allocation to public health sector and epidemic areas for influenza control and research has also been observed [7]. Presently used intervention measures, such as vaccination, surveillance, public awareness, pharmacological and non-pharmacological strategies and ongoing research have not yet provided an optimal solution to the influenza burden in China. Consequently, identifying risk factors with a potential to elevate influenza risks across various populations in China is crucial.

Meteorological factors (MFs) have been recognized as significant health risk factors, with considerable impact on morbidity and mortality in populations globally [8, 9]. The associations between MFs and respiratory diseases-related mortality are anticipated to upsurge with the rising frequency and intensity of extreme weather driven by climate change [10, 11]. Already, the significant role of adverse temperature, relative humidity, and precipitation in shaping influenza seasonality has been documented [12–14]. Wind patterns influence the geographic spread of influenza, by facilitating the dispersion of respiratory droplets containing influenza virus, contributing to the rapid transmission of the virus across regions [15, 16].

Although some consistent results emerge from previous epidemiological studies about the association between MFs and influenza, there are notable disparities in the reported estimates. For instance, while some studies found no significant association between relative humidity and influenza incidence, others observed a significant positive association with long-term lag effect [13, 17, 18]. Variations in data types, geographical scale, analysis methods, and region-specific characteristics, may account for the differing study outcomes [19]. For instance, while stratified analysis is crucial in public health for profiling susceptible population subgroups [20], some previous studies have focused on the risks in overall population or single city. Analyzing overall risks might hide diverging patterns between different population subgroups and thwart the identification of vulnerable ones, complicating mitigation and adaptation actions. Location-specific factors, such as population density and levels of MFs are expected to trigger variation in vulnerability to MFs-mediated influenza risks [21]. Numerous studies have focused on influenza-like illness (ILI). Such studies may fail to capture clear and accurate insights into the impact of MFs, following the diagnostic uncertainty associated with ILI.

Despite the increasing use of distributed lag non-linear models (DLNMs) in climate-health research [22], there is still a limited application of two-stage time-series approaches incorporating DLNMs in sub-national regions like Anhui Province [23]. Some studies apply models without adequately accounting for both city-level heterogeneity and long-term lag structures that capture delayed and cumulative meteorological effects. This may lead to misinterpretations of exposure-response relationships and inaccurate risk assessments [24, 25]. Addressing these gaps is crucial for generating more regionally tailored evidence to guide public health interventions. In view of these, we employed a multi-city two-stage time-series analysis to explore the influence of MFs on LCI cases in Anhui province, China during 2015–2023.

Materials and methods

Study area

Figure 1 is a map of China showing Anhui province and the study locations. Anhui is located in East China along the Yangtze and Huai Rivers valleys. It spans from 114°54’E to 119°37’E and 29°41’N to 34°38’N, with a total population of 61.13 million people by 2021. The region is characterized by diverse geographical features and experiences a humid subtropical monsoon climate.

Fig. 1.

A map of China showing Anhui province and the study locations

Data collection and processing

Data on daily LCI cases in 16 prefecture-level cities of Anhui province was obtained from the China Influenza Surveillance Information System (CISIS) from January, 2015 to March, 2023. The system compiles data on patients presenting with ILI and all the LCI cases were originally observed from the 16 municipal Centers for Disease Control (CDC) and 24 sentinel hospitals. In this study, all patients presenting with ILI at the 16 municipal CDCs and 24 sentinel hospitals were tested for influenza viruses. Thus, our data included individual-level test outcomes (positive or negative) for each ILI case. The variables extracted from the database include onset date, collection date, diagnosed date, case classifications, spatial attributes (latitude and longitude), virus type, gender, reporting method, and age. According to the National Health Commission of China (NHCC), influenza is a class C notifiable disease requiring immediate reporting of cases to the CDC upon diagnosis. The NHCC further guides on “Diagnostic Criteria for Influenza (WS285-2008),” and it states: a LCI case involves influenza-like symptoms (fever and cough or sore throat) confirmed by laboratory tests i.e., reverse transcriptase polymerase chain reaction, viral culture, or serological tests for specific influenza A or B virus or a ≥ 4-fold increase in specific antibody to seasonal influenza A or B virus between acute and convalescent serum samples.

Daily ambient mean temperature (^oC), relative humidity (%), and wind speed (m/s) data were obtained from the National Meteorological Information Center (http://data.cma.cn/site/index.html). Small number of missing meteorological data was repaired using computed mean value of adjacent points, while days with no number of LCI cases were recorded as zero. Meteorological and influenza datasets were aligned and merged using crosswalking technique, ensuring consistency of the monitoring date. The city-specific population was obtained from the National Population Census of China (http://www.stats.gov.cn/).

Statistical analyses

We first conducted a descriptive analysis of the MFs and influenza data in the 16 cities of Anhui. Spearman correlation analysis was conducted to give a preliminary perspective of the MFs and LCI cases relationship. To visually assess the delayed and non-linear effects of MFs on influenza risk, we plotted three-dimensional (3D) exposure–lag–response curve. The impact of MFs on the LCI cases was determined using DLNMs [26], based on two-stage time series analysis to enhance accuracy and statistical power [27]. The DLNM was used to simultaneously take into account the non-linear associations and delayed effects [22, 26]. Each meteorological variable was modeled using the crossbasis() function, in the dlnm R package, with two dimensions: (1) the exposure–response dimension and (2) the lag–response dimension. For the exposure–response dimension, we specified a natural cubic spline (ns) with 3 degrees of freedom (df) (argvar = list(fun = “ns”, df = 3)), allowing flexibility to capture potential non-linear exposure-response relationships. For the lag–response dimension, we specified a ns with 3 df (arglag = list(fun = “ns”, df = 3)), enabling modeling of potential non-linear and delayed effects across the 14-day lag period. This approach is suitable given the non-linear and lagged influence of weather on respiratory outcomes [28, 29]. It aligns with literature [22] and is consistent with common practice in environmental epidemiology studies using DLNM [26].

First stage analysis

In the first stage analysis, we estimated MFs impact on LCI cases in the 16 cities of Anhui, using a quasi-Poisson regression to account for over-dispersion and autocorrelation (via a smooth function of time) in city-specific time series data. In each city, the model was fitted on overall data, age and gender specific data. The basic model for MFs-LCI cases focused on meteorological variables and their lagged effects on LCI cases. The general structure of the single MFs-LCI cases model was as follows:

1

Where, Yt represents the daily number of LCI cases reported on the day t; α_o denotes the intercept; cb represented the cross-basis matrix of MFs, x_i is one of the MFs variables; x_j represented the climate variables other than x_i; ns and df denote the smooth natural cubic spline function and the degree of freedom, respectively; DOW is an indicator variable meaning “day of the week”. The time trend was smoothed using ns with 7 df for each year to account for the unmeasured long-term trends and seasonality of influenza cases. This choice was justified by the strong seasonal patterns of influenza. The 7 df provided sufficient flexibility to capture these periodic fluctuations without over-fitting, aligning with standard practices in environmental epidemiology studies, where 6–8 df per year is commonly recommended. Additionally, potential confounders were adjusted using ns with 3 df [30], providing adequate flexibility to capture non-linear associations without introducing unnecessary complexity. Including meteorological variables along with time and weekday controls helps isolate the effects of weather on LCI cases, enhancing the validity of the estimated exposure–response relationships. To address collinearity, variables with high correlation (Spearman’s correlation coefficient ≥ 0.7) were not included in the same model.

Given that the impact of MFs on influenza can last for more than a week [31], both single-day lag models (ranging from the current day [lag 0] to 14 days prior [lag 14]) and cumulative-day lag models (e.g., lag 0–1 to lag 0–14) were employed to investigate the MFs-LCI cases associations. Lag 0 represents exposure on the index day, while lag 1–14 correspond to the fourteen prior days, accounting for both single-and-cumulative day effects [32]. Lag days are used to reflect the delay between exposure (e.g., to MFs) and the onset or reporting of infection. A maximum lag of 14 days is appropriate to capture the delayed effects of MFs on influenza cases. Using a distributed lag structure helps elucidate the temporal dynamics of MFs in influenza epidemiology.

We conducted sub-group analysis, at single-day lag, to assess whether effects of MFs varied by age and gender. The city-specific estimated effects were presented as risk ratio (RR) and 95% confidence intervals (CIs) in daily influenza cases.

Second stage analysis

In the second-stage analysis, city-specific associations between MFs and LCI cases from the first stage analysis were pooled using the random effects meta-analysis to get an overall estimate for the entire province. Following the diverse topographies and populations across cities, significant heterogeneity was anticipated between individual studies, justifying the use of random-effects model to account for variability in effect sizes. The lag with the most acute effect on the incidence of influenza was selected as the representative effect size for each city. Acute lag helps pinpoint the critical exposure window and thus the most sensitive time frame for preventive actions [33]. The meta-analysis was evaluated using restricted maximum likelihood method [34], which accounts for both within-and-between study variability.

A multivariate meta-regression model was fitted on single-day lag data to explore effect modifications of population density, multi-year median (50th percentile) levels, longitude, and latitude [35]. Cochran’s Q test and the I² (%) statistic were used to assess the heterogeneity of associations. Heterogeneity was considered low when I² was less than 25%, moderate when I² ranged from 25% to less than 75%, and high when I² was 75% or greater, at a significance level of p ≤ 0.05. The significance (p ≤ 0.05) of the pooled effect estimates was determined using the Z test. We performed sensitivity analyses to confirm the robustness of our findings by adjusting the df for the time trend (2–4 df per year), and varying df for the exposure–lag response functions of MFs (2–4 df). We also extended maximum lag to 21 days and conducted leave-one-city-out analyses to assess if results were driven by any single city. To assess potential confounding due to COVID-19, we re-ran our analyses (on single-day lag data) excluding data from 2020 onwards. The analyses were conducted in R (Version 4.3.1) using the “readxl”, “dlnm”, “splines”, “metafor”, “mgcv”, “plot3D”, “plotly”, and “ggplot2” packages, with a significance threshold of p < 0.05.

Results

Descriptive statistics for MFs and LCI cases in Anhui

Table 1 shows the provincial-level descriptive statistics of daily MFs and daily LCI cases from January, 2015 to March, 2023. Average means ± SD for temperature was 16.34 ± 9.34 °C, relative humidity was 75.58 ± 14.09%, and wind speed was 2.24 ± 1.14 m/s. The mean prevalence of influenza cases differed by age, gender and virus type. Similar observation was made for the city-specific annual differences and mean ± SD of daily MFs and the LCI cases as shown in Additional file 1 Tables 1, 2 and 3 and Additional file 1 Figs. 1, 2, 3, 4, 5 and 6.

Table 1.

Provincial-level descriptive statistics of daily MFs and daily LCI cases in Anhui (January, 2015-March, 2023)

	variable	mean	SD	max	min	P25	P50	P75	sum
Meteorological factors	Tmean	16.34 °C	9.34	36.3	−15.9	8.2	16.7	24.4
	Rhumidity	75.58%	14.09	100.0	4.0	67	77	86
	Wspeed	2.24ms	1.14	13.9	0.0	1.5	2	2.8
LCI cases	Overall	4.37	7.03	217	1	0	0	0	43,872
	Gender
	Male	3.22	4.56	115	1	1	2	4	23,928
	Female	2.91	3.85	102	1	1	2	3	19,944
	Age (years)
	0–4 years	2.0	1.91	21	1	1	1	2	7520
	5–14 years	3.61	6.60	189	1	1	2	3	19,200
	15–24 years	2.02	2.40	38	1	1	1	2	5470
	25–59 years	2.20	2.31	41	1	1	1	2	9707
	≥ 60 years	1.30	0.68	6	1	1	1	1	1975
	Virus type
	H3N2	3.07	3.89	50	1	1	2	3	10,561
	H1N1	4.86	8.79	217	1	1	2	5	14,997
	B/yamagata	2.72	3.31	43	1	1	2	3	4246
	B/victoria	3.71	5.48	58	1	1	2	4	13,983
	H7N9	1.08	0.89	4	1	1	1	1	85
	A un-typed	1.0	0	1	1	1	1	1	10

Abbreviations: Tmean temperature, Wspeed wind speed, Rhumidity relative humidity, SD standardized deviation, Min minimum, Max maximum, P25 25^th percentile, P50 50^th percentile, P75 75^th percentile

The time series plot (Fig. 2) reveals a discernible seasonal pattern in daily influenza cases and MFs in Anhui. The plot shows clear seasonal peaks in influenza cases, predominantly occurring during the winter months (December to February). This seasonal pattern is consistent across the years, highlighting the cyclical nature of influenza outbreaks. There is a notable variation in the magnitude of these peaks across different years, with some influenza seasons being more severe than others.

Fig. 2.

Time series plot for daily LCI cases and MFs in Anhui (January, 2015-March, 2023). N/B- Tmean: temperature mean, Wspeed: wind speed, Rhumidity: relative humidity

DLNM analysis

Figure 3 is a 3D plot illustrating the effects of wind speed, temperature, relative humidity and lag (days) on the risks of LCI cases. The association between temperature and influenza risk displayed a complex, non-linear pattern across lag days. The RR values are highest (RR > 1) around moderate temperatures (10 °C to 20 °C) and decline (RR < 1) at both lower (< 0 °C) and higher temperature (around ~ 25 °C), before increasing at very high temperatures (~ 30 °C). The lag effect reveals an immediate but diminishing influence on influenza risk as the lag increases. The exposure–lag–response surface for wind speed reveals a nonlinear and lag-dependent association. Notably, low wind speeds (< 1.5 m/s) are associated with elevated RR at short lags (0–5 days), indicating an immediate risk amplification under stagnant air conditions. As wind speed increases to moderate levels (around 2–3 m/s), the RR remains elevated but gradually declines, with a secondary peak in RR emerging at longer lags (10–15 days) before dropping sharply beyond 10 m/s. The lag day shows both immediate and delayed effects. The RR values peak at low humidity (10–30%) and shorter lags (0–3 days), and decrease at longer lags. There is a wave-like or oscillating pattern in RR across both lag days and relative humidity levels, suggesting a non-linear relationship.

Fig. 3.

Three-dimensional plot illustrating the exposure–lag–response relationships between meteorological factors (MFs) and the risk ratio (RR) of LCI cases across various lag days. These plots were generated using a distributed lag non-linear model (DLNM), visualized via the crosspred() function in R

Figure 4 shows RRs (95% CI) for the pooled effect estimates of the association between MFs and LCI cases under single-and-cumulative day lags. An increase in temperature on a single day was associated with a slight, non-significant decrease in the risk of influenza cases: RR = 0.9778 (95% CI: 0.9468–1.0098). There was a non-significant increase in the risk of influenza cases for cumulative effect of temperature: RR = 1.0263 (95% CI: 0.9721–1.0836). There is a slight, non-significant increase in influenza risk with higher relative humidity on a single day: RR = 1.0056 (95% CI: 0.9899–1.0216). There is a slight, non-significant decrease in influenza risk with the cumulative effect of relative humidity: RR = 0.9974 (95% CI: 0.9927–1.0022). There is a slight, non-significant increase in influenza risk with higher wind speed on a single day: RR = 1.0105 (95% CI: 0.9965–1.0246). There is a non-significant increase in influenza risk with the cumulative effect of wind speed: RR = 1.0083 (95% CI: 0.9498–1.0704). To assess the COVID-19 impact, we re-ran our analyses (single day lag) excluding data from 2020 onwards. We found that the associations between MFs and LCI cases remained directionally consistent, with temperature showing a slightly stronger negative association (RR = 0.9306; 95% CI: 0.9239–0.9373) and relative humidity and wind speed showing similar positive but non-significant associations (Additional file 1 Fig. 10).

Fig. 4.

Forest plots comparing the associations between MFs and daily LCI cases. The associations are presented as the pooled RRs for 43,872 cases reported from January 2015 to March 2023 in Anhui, China. Blue diamond = single-day lag; Red diamond = cumulative-day lag. Risk ratio (RR) represent the relative risk of LCI at different exposure levels of MFs, with respect to a reference value (e.g., median), as estimated from the DLNM exposure–response surfaces using the crosspred () function in R

Figure 5 shows the results of the population stratification analysis by gender. There was a statistically significant negative association between temperature and LCI cases with slightly higher effect estimates in males; RR = 0.9942 (95% CI: 0.9936–0.9948) than females: RR = 0.9941 (95% CI: 0.9934–0.9948). There was no strong evidence that relative humidity: RR = 1.0000 (95% CI: 0.9992–1.0008) is associated with influenza risk in both males (p = 0.9259; I² = 90.6%) and females (p = 0.9880; I² = 92.9%). The effect sizes: RR = 1.0109 (95% CI: 0.9939; 1.0282) and RR = 1.0033 (95% CI: 0.9904; 1.0163) suggest a positive, but not statistically significant association between wind speed and LCI cases in males (p = 0.2103) and females (p = 0.6192), respectively.

Fig. 5.

Forest plots comparing the associations between MFs and daily LCI cases defined by gender. The associations are presented as the pooled RRs for 43,872 cases reported from January 2015 to March 2023 in Anhui, China. Blue diamond = Male; Red diamond = Female. Risk ratio (RR) represent the relative risk of LCI at different exposure levels of MFs, with respect to a reference value (e.g., median), as estimated from the DLNM exposure–response surfaces using the crosspred () function in R

Table 2 and Additional file 1 Fig. 7 show the results of the sub-group analysis by age. Temperature negatively associated with LCI cases in all ages, at p = 0.0001. There was a slight statistically significant decrease in influenza risk with higher relative humidity for age ≥ 60 years (p = 0.0024). Contrarily, there was no statistically significant effect for relative humidity on influenza risk for all other age groups. Similarly, wind speed showed no significant association with LCI cases across all age groups.

Table 2.

Pooled effect estimates for the associations between MFs and LCI cases defined by age groups

Age group		Temperature	Relative humidity	Wind speed
0–4 years	RR (95%CI)	0.9953 (0.9947–0.9960)	1.0000 (0.9990–1.0010)	1.0141 (0.9958–1.0327)
	I2%	92	79	87
	p	0.0001	0.9813	0.1329
5–14 years	RR (95%CI)	0.9927 (0.9922–0.9932)	0.9997 (0.9986–1.0008)	1.0053 (0.9848–1.0263)
	I2%	87	94	97
	p	0.0001	0.5624	0.6125
15–24 years	RR (95%CI)	0.9942 (0.9934–0.9949)	1.0000(0.9989–1.0011)	1.0119 (0.9969–1.0271)
	I2%	85	86	70
	p	0.0001	0.9984	0.1201
25–59 years	RR (95%CI)	0.9938 (0.9931–0.9945)	0.9996 (0.9987–1.0006)	1.0061 (0.99881–1.0244)
	I2%	91	85	88
	p	0.0001	0.4625	0.5109
≥ 60 years	RR (95%CI)	0.9956 (0.9946–0.9966)	0.9990 (0.9984–0.9996)	1.0234 (0.9944–1.0531)
	I2%	83	58	47
	p	0.0001	0.0024	0.1144

CI confidence interval, RR risk ratio, I2%= Heterogeneity, p statistical significance (p ≤ 0.05) of the pooled effect estimates (RR) as determine using the Z test. The associations are presented as the pooled RR for 43, 872 cases reported from January 2015-March 2023 in Anhui, China

Additional file 1 Fig. 7: Forest plots for the associations between MFs and LCI cases defined by age groups. The associations are presented as the pooled RRs for 43,872 cases reported from January 2015 to March 2023 in 16 cities in Anhui province.

Table 3 shows the results of the meta-regression modeling. Population density marginally modified the relationship between temperature and LCI cases (Estimate = 1.3667, SE = 0.7151, p = 0.0506). Other moderators, including median pollution levels, latitude, and longitude, do not show significant effects. Thus these factors may not play a substantial role in modifying the temperature-influenza relationship in Anhui. Population density, median pollution levels, latitude, and longitude do not moderate relationship between relative humidity, wind speed, and LCI cases. Sensitivity analysis by varying the df for the time trend and MFs showed that RR estimates remained largely stable (Additional file 1 Fig. 8). On extending the maximum lag to 21 days, the results showed slight modifications in estimates compared to the 14-day lag analysis. Nonetheless, the direction and magnitude of the associations remained consistent (Additional file 1 Fig. 9). These findings were further supported by leave-one-city-out analysis, which showed that no single city significantly influenced the observed relationships.

Table 3.

Moderating effects of population density, median MFs levels, latitude and longitude on the MFs-LCI associations

Variable	Population density			Median level of MFs value			Latitude			Longitude
	Coefficient (ρ)	SE	P Value	coefficient (ρ)	SE	P Value	coefficient (ρ)	SE	P Value	coefficient (ρ)	SE	P Value
Temperature	1.3667	0.7151	0.0506	0.0729	0.4581	0.8736	−0.3033	0.2857	0.2884	0.0000	0.0000	0.7494
Relative humidity	−1.3698	2.2395	0.5407	−1.3251	1.0755	0.2179	−1.2574	0.9728	0.1962	−0.0000	0.0000	0.6407
Wind speed	0.2148	0.1637	0.1894	−0.1710	0.1544	0.2681	−0.0818	0.0627	0.1920	−0.0000	0.0000	0.9803

SE standard error, p = statistical significance (p ≤ 0.05), MFs meteorological factors

Discussion

Although influenza epidemiology has been extensively researched in China, numerous concerns remain unanswered. Here in, we explored the impact of MFs on LCI cases in Anhui during 2015–2023, using a multi-city time-series analysis. The descriptive statistics revealed significant disparities in MFs levels and influenza cases across cities, age, and gender. These population dynamics underlines the necessity for targeted interventions to decrease influenza risks in vulnerable populations. The results of DLNM show that while the MFs showed general population associations with influenza risk, none of the relationships were statistically significant. These findings align with previous studies which acknowledge that whereas some climatic factors increase the risk of influenza, others such as wind speed, atmospheric pressure, and sunlight duration show a generally weak or not statistically significant correlation [36, 37].

The slight, non-significant decrease in influenza risk associated with single-day lag temperature increases and the positive association for cumulative-day lag effects align with other studies showing the non-linear influence of temperature on respiratory infections [38, 39]. Wang et al. [37] observed that in northern China, low temperatures increased the risk of influenza, while both high and low temperatures increased the risk of ILI in central and southern China. Also, the GBD study 2019, reported that both high and low ambient temperatures are leading environmental risk factors [40]. Low temperatures increase influenza spread by enhancing virus stability, altering its envelope, and weakening respiratory defenses like mucociliary clearance and leukocyte activity [41, 42]. Inhaling cold air cools the nasal epithelium, impairing immune responses and aiding virus transmission [43, 44]. Extreme temperatures also encourage indoor activity, increasing close human contact and further promoting virus spread [45].

A slight increase in relative humidity was associated with a very small increase in the risk of influenza for single-day lag model. Conversely, for the cumulative-day lag, there was slight decrease in influenza risk as humidity rises. Although the results were not statistically significant, they suggest a time-dependent effect of relative humidity on influenza risk. These findings support previous studies suggesting that both high and low humidity levels influence influenza transmission [13, 17, 46]. Low humidity promotes influenza virus transmission by enabling droplet nuclei formation and enhancing viral stability [47–49]. It also affects mucus properties, thickening it and slowing ciliary movement, which reduces mucociliary clearance efficiency [47, 50]. Breathing cold air worsens this by drying the respiratory tract, weakening protective barriers and increasing vulnerability to viral infections. High humidity, conversely, may inactivate viral particles, reducing transmission risk [50]. Low wind speeds were associated with elevated RR at short lags, possibly due to limited air dispersion of infectious droplets in stagnant conditions [51]. The lack of significant association between wind speed and influenza risk in this study may indicate that, in urban settings, other factors such as population density and indoor crowding may play a more dominant role in influencing influenza risks.

The non-significant associations between MFs and influenza in general population contradicts some previous studies that have found significant relationships. These discrepancies may be attributed to a range of contextual and methodological factors. While MFs influence influenza spread, their impact may vary by region, following environmental differences [52]. For instance, Anhui region characterized by a humid subtropical climate with relatively moderate seasonal fluctuations may not experience the same intensity of meteorologically driven influenza transmission observed in regions with more extreme weather conditions [53]. Differences in healthcare systems and public health strategies could contribute to the inconsistencies. Anhui has made notable improvements in healthcare accessibility and surveillance systems in recent years [54], which may have enhanced case detection and control, potentially mitigating the influence of environmental triggers. Furthermore, population immunity levels, influenced by past exposure, vaccination rates, and circulating viral strains, can significantly modify the relationship between MFs and influenza spread [55, 56]. These variables are rarely standardized across studies.

In terms of methodological differences, variability in measurement and data quality, such as differences in how meteorological variables and outcome are measured, resolution of exposure data, and influenza case definitions may impact the associations [57]. Unlike previous studies that have largely relied on ILI data, this study used LCI cases, offering greater diagnostic specificity and reducing potential misclassification bias. Additionally, age distribution and comorbidity patterns of the studied populations may influence susceptibility and response to environmental stressors. Studies focused on elderly populations or children may detect stronger associations than studies analyzing general populations, as these subgroups are more vulnerable to environmental fluctuations [58]. While our findings diverge from some prior studies, this divergence underscores the complexity of MF-influenza interactions and the importance of contextualizing environmental epidemiological results within local ecological, socio-demographic, and methodological frameworks.

Effect of MFs on the risk of influenza by age and gender

Sub-group analysis showed a statistically significant negative association between temperature and LCI, with greater vulnerability observed in age ≥ 60 and 0–4 years. The substantial susceptibility in young children and elderly persons may be due to their weaker immune systems, making them more vulnerable to weather changes under varying climatic conditions [59]. The stronger negative association between temperature and influenza in males than females suggests potential gender differences in behavior or immune response to environmental changes [60]. Our results imply that different genders and ages may be affected deferentially and directly by changes in temperature.

The slight but statistically significant decrease in risk for individuals aged ≥ 60 at higher relative humidity reflects a non-linear association, potentially linked to the reduced stability and transmissibility of influenza viruses in more humid environments The elderly may experience more pronounced effects from dry air, particularly during the influenza season, making them more vulnerable to infection under low humidity conditions [45, 47]. Wind speed showed positive but non-significant association with LCI across the general population and subgroups. This could be due to the indirect nature of wind’s influence on influenza transmission, as wind is more likely to affect the dispersal of airborne pathogens outdoors rather than influencing person-to-person transmission indoors, where most influenza cases occur [61].

The differences in statistically significant and non-significant associations between sub-groups and the general population, particularly for temperature and relative humidity (ages ≥ 60), can be attributed to varying levels of vulnerability and exposure. Some sub-groups may be more susceptible to influenza, amplifying the effects of MFs on influenza risk in the populations. In contrast, stronger immune responses in some age groups may mitigate the effect of MFs in the general population leading to a non-significant association. Also, the general population may be better equipped to handle fluctuations in MFs, either through behavioral adaptation or prior exposure. This could attenuate any direct association between MFs and influenza risk in the general population. The current findings highlight the need for influenza mitigation strategies that consider both climate data and population characteristics.

Moderating effects of population density, median MFs levels, latitude, and longitude

Population density marginally modified the relationship between temperature and LCI cases, with higher density being associated with an increase in influenza cases. In line with our findings, previous studies have reported on the role of population density in modifying influenza seasonality [58, 62]. In areas with higher population density, the chances of close contact are greater, particularly when people tend to stay indoors more, which can result in higher rates of influenza transmission [32]. Median MFs level, latitude, and longitude did not show significant modifying effects. Our findings align with those of Tamerius et al. [58] who submitted that relative humidity is a strong predictor of influenza peaks in both high and low latitudes, but not in middle-latitude regions (Anhui Province, our study area, is located in a middle latitude region). The author further reported that in high-latitude regions, humidity enhance virus survival and aerosol transmission [58], while in low latitudes, direct or fomite transmission may dominate. Middle latitudes might represent a transitional zone where either low humidity or high precipitation influences influenza seasons, depending on local climate [63, 64].

The current study’s strengths are: First, focusing on LCI cases removes diagnostic uncertainty linked to ILI, providing clearer associations and aiding targeted public health policies. Second, a two-stage modeling strategy for time series data allows for the explicit estimation of the level of autocorrelation and over-dispersion (via a smooth function of time) in specific cities. This strategy also enables inference on the appropriately lagged acute effects of MFs, aggregated across all cities or stratified by age and gender. Third, both single-day and cumulative-day lag analyses are applied, capturing the temporal dynamics of MFs on influenza in Anhui, China. Finally, this study fills a research gap by prioritizing LCI over ILI, strengthening evidence for interventions against MFs and public health threats.

Despite the strengths, this study has some limitations. First, its ecological design prevents causal inference on the association between MFs and influenza. Second, confounding variables, such as population mobility, vaccination rates, NPIs, healthcare-seeking behavior, socio-economic status, and changing diagnostic practices, were not included. Third, underreporting of LCI illness likely occurred, as patients with mild symptoms may have avoided sentinel hospitals; however, this potential bias is likely random and non-differential. Fourth, the use of city-level aggregate meteorological data may not accurately reflect individual-level exposures due to spatial and temporal variability within cities. This could lead to exposure misclassification and bias in effect estimates. Fifth, although, the use of a DLNM allowed us to capture long-term exposure-lag-response relationships across the entire 2015–2023 period, the COVID-19 impact slightly modified the magnitude of the associations. However, this did not fundamentally alter the main conclusion of our study. Finally, the study’s focus on Anhui Province, China, limits generalizability to regions with different environmental and demographic profiles.

Conclusion

The relationship between MFs and influenza is varied and complex. While MFs showed non-significant associations with LCI cases in general population, sub-group analysis showed a statistically significant association between temperature and LCI, with population density marginally modifying the association. This study finding enhances evidence-based knowledge which may guide developing targeted interventions like early warning systems to reduce influenza risks.

Abbreviations

DLNMs

Distributed lag nonlinear models

GBD

Global Burden of Disease

LCI

Laboratory-confirmed influenza

MFs

Meteorological factors

NHCC

National Health Commission of China

Authors’ contributions

HAM: Conceptualization, Methodology, Software, Formal analysis, Investigation, Data curation, Writing-original draft, Visualization. YSH: Conceptualization, Methodology, Software, Validation, Writing-review & editing, Visualization. LG: Investigation, Validation, Writing-review & editing. SH: Investigation, Validation, Writing-review & editing. JY: Investigation, Validation, Writing-review & editing. WC: Investigation, Validation, Writing-review & editing. PW: Conceptualization, Validation, Writing – review & editing, Supervision. JH: Investigation, Validation, Resources, Writing-review & editing, Project administration. HFP: Conceptualization, Validation, Resources, Writing – review & editing, Supervision, Funding acquisition, Project administration.

Funding

This study was funded by grants from the Research and Innovation team, School of Public Health, Anhui Medical University (kctd200402), National Key Research and Development Project of the Ministry of Science and Technology of China. (No. 2022YFE0110100), Opening foundation of the State Key Laboratory for Diagnosis and Treatment of Infectious Diseases, The First Affiliated Hospital, Zhejiang University School of Medicine (SKLID2021KF04) and Peak Discipline Project from School of Public Health, Anhui Medical University (Collaborative Education Innovation Project) (2024GWXTYRZ005).

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.