The effects of meteorological factors on influenza among children in Guangzhou, China

Background Influenza seriously affects the health of children, yet little evidence is available on the association between meteorological factors and the occurrence of influenza among children in subtropical regions. The current study aimed to explore the effects of meteorological factors on influenza among children in Guangzhou, a subtropical city in China. Methods The distributed lag nonlinear model (DLNM) was used to assess the effects of meteorological factors on children influenza occurrence in Guangzhou, China. Daily number of influenza cases among children aged 0‐17 years from 2013 to 2017 were obtained from the National Information System for Disease Control and Prevention. Results Mean temperature, relative humidity, and atmospheric pressure were associated with influenza cases. The relative risks (RRs) increased as temperature fell below 20°C. The relationship between relative humidity and influenza cases could be described with a U‐shaped curve, and the RRs increased if relative humidity was lower than 50% or higher than 80%. The risk of influenza increased with rising atmospheric pressure with 1005 hPa as the break point. The cold effect, humid effect, dry effect, high‐pressure effect, and low‐pressure effect showed statistical significance both in female and male. The cold effect increased with age. The humid‐effect affects all age ranges of children, but dry effect mainly affected 4‐14 years old. High‐pressure effect mainly affected the 0‐3 years old, whereas low‐pressure effect protected preschool children aged 0‐6 years old. Conclusion Mean temperature, relative humidity, and atmospheric pressure might be important predictors of the influenza occurrence among children in Guangzhou.


| INTRODUC TI ON
Influenza is a global public health concern; annual epidemics attack 3-5 million people of all ages and contributes to 500 000 deaths worldwide. 1 Children aged 6-59 months face the highest risk of complications. In temperate regions, seasonal epidemics occur mainly in winter. While in tropical and subtropical areas, irregular epidemics may break out throughout the year. 1,2 In southern China, influenza is prevalent throughout the year; it has a clear peak in summer and a less pronounced peak in winter. 3 Hong Kong, close to Guangzhou city, experiences semi-annual cycles with epidemics usually peaking in spring and summer. 4,5 The seasonality of influenza suggests that meteorological factors might influence the spread of the disease.
Several studies have suggested an association between climate change and influenza spread, but the findings are inconsistent. [6][7][8] Experimental and modeling studies have suggested that cold temperature and low relative/absolute humidity increase the risk of influenza transmission and survival in temperate settings, [3][4][5]9 while rainfall fluctuations may drive influenza activity in low latitudes, but this has not been well documented. 5,10 In tropical and subtropical regions, where the annual average temperature and humidity are much higher, there is insufficient evidence to prove a relationship between influenza risk and temperature or humidity. Furthermore, since temperature, precipitation, humidity, and other parameters may vary significantly within a climatic zone, it needs to explore the local information on influenza seasonality.
In Guangzhou, a subtropical city in China which often experiences extreme weather, the effects of temperature with low humidity and other meteorological factors on influenza transmission in children are not clearly understood. In this study, we established a distributed lag nonlinear model (DLNM) with the daily time series data during 2013-2017 to assess the effects of meteorological factors on the occurrence of influenza among children in Guangzhou. We also examined whether the association was regulated by age and sex. The findings of this study may provide evidence for health authorities to develop an early warning system for influenza epidemic, so as to limit the health impacts on children.

| Setting
Guangzhou, the capital of Guangdong Province in southern China, is located at 22°26′N to 23°56′N and 112°57′E to 114°3′E. 11,12 F I G U R E 1 Geographical location of the study area in China Covering an area of 7434 square kilometers and 11 districts, Guangzhou has a population of 14.0 million in 2017, including 1.71 million (12.21%) aged 0-14 years, 12.33 million (87.79%) aged over 15 years. The monsoon in subtropical Asia gives Guangzhou a long, hot, and humid summer, and a short, mild, and dry winter. January and February have the lowest mean temperatures (6.5-12.1°C).
Extremely wet days, locally known as "Hui-nan-tian" in February and March, appear when the dew point temperature of the outdoor air exceeds the surface temperature of indoor objects, along with high air humidity up to 100%. 13,14 The period between May and September is hot and humid, with the mean temperature (33.0-34.9°C) peaking between July and August. The average annual rainfall ranges from 1370 to 2353 mm and exhibits a seasonality. 15 Figure 1 shows the geographical location of Guangzhou.

| Data collection
Influenza is a notifiable disease in China and categorized as a Class "C" Demographic information of all cases included gender, age (between the date of influenza onset and the date of birth), and the form of child care. According to our data, more than 83.08% influenza cases were children aged 0-17 years. Therefore, we focused on the cases among children aged 0-17 years in this study.
The form of child care might impact influenza infection in children.
According to our data, children aged 0-3 years are usually cared for at home, 4-6 years attend kindergarten, and 7-14 years go to primary and middle schools, 15-17 years go to senior school. To investigate which subpopulation was most susceptible to meteorological changes, we conducted the analyses for different age subgroups (0-3, 4-6, 7-14, and 15-17 years) and sex (male and female).
The daily meteorological data including maximum, mean, and minimum temperature, and relative humidity, daily precipitation, wind velocity, and atmospheric pressure were obtained from Guangdong Meteorological Administration.

| Data analysis
We used Pearson correlation analysis to examine the correlation between daily meteorological factors and influenza cases. Scatter plots were drawn to observe the patterns of meteorological factors and influenza cases over time. Since the daily number of influenza cases follows a Poisson distribution, and the meteorological factors such as temperature have lag effects, we used a distributed lag nonlinear model (DLNM) to assess the association between meteorological factors and the daily number of influenza cases. 12,18 DLNM is a flexible model that simultaneously estimates the nonlinear and delayed exposure-response relationship, especially the effects of meteorological factors on health. 18 We calculated Pearson's correlation coefficients matrix within the meteorological factors for a preliminary analysis. Because of the strong correlation coefficient between temperature and atmospheric pressure (Table 2), we developed two DLNMs to independently analyze the two variables. A Poisson regression with quasi-Poisson function was established to test the over-dispersion in the daily number of influenza cases. All models were adjusted with indicator variables, like time trend, day of week (DOW), and public holidays. The calculation was as follows: In this equation, Y t is the daily number of influenza cases on day t; α is intercept; NS is a natural cubic spline; M is one of the two strongly correlative variables to estimate including temperature (or atmospheric pressure); df is the degree of freedom; X i is the other explanatory variables of meteorological factors including relative humidity and atmospheric pressure; DOW t is day t of the week, and γ is vector of coefficients; Holiday t is a binary variable the value of which is "1" if day t is a public holiday; Time refers to duration of seasonality in the calendar year and long-term trend, and we found that using a NS with 6 df per year for time produced the best model fitting based on the Akaike Information Criterion for quasi-Poisson (Q-AIC), which was supported by other references. 11,18 We used the Q-AIC to choose the df for the meteorological factors and the maximum lag. The effects of daily mean temperature, relative humidity, and atmospheric pressure were controlled using NS of 2 df, 2 df and 3 df, and the maximum lag of all 27 days, respectively. The reference level was defined as the median value of mean temperature, relative humidity, and atmospheric pressure to calculate the relative risks (RRs) and 95% confidence intervals (CI). All statistical tests were two-sided, and P-value < 0.05 was considered statistically significant. All analyses were performed using the R software (version 3.0.2; R Development Core Team 2013), and DLNM was carried out using the "dlnm" package.   However, in the further stratified analyses by age and sex, temperature, humidity, and air pressure were significantly correlated with influenza cases. Thus, we used the statistically significant variables above (mean temperature, relative humidity, and atmospheric pressure) for the subsequent analyses. Meanwhile, there was a high correlation between temperature and atmospheric pressure (correlation coefficient 0.86, P < 0.05). Thus, temperature and atmospheric pressure were not included in the same model due to the potential collinearity.

| RE SULTS
For a better interpretation, the relative risks (RRs) and 95% confidence intervals (CI) of daily influenza occurrence (relative to the risk at the median value of the covariate) were plotted against covariate values for mean temperature, relative humidity, and atmospheric pressure over the corresponding lag days in Figure 3.
In general, mean temperature, relative humidity, and atmospheric pressure were associated with influenza cases. The relative risks (RRs) increased as temperature fell below 20°C. The relationship between relative humidity and influenza cases could be described with a U-shaped curve. The RRs increased if relative humidity was lower than 50% or higher than 80%. The risk of influenza increased with rising atmospheric pressure with 1005 hPa as the break point.
To identify the extreme effects, the estimated effects of mean temperature, relative humidity, and atmospheric pressure comparing the 95 th percentile to the median value and the 5 th percentiles to the median value were plotted in Figure 4. The hot effect was not significant, whereas cold effect was a significant risk factor occurring within 0-27 lag days. A significant humid effect was observed along 0-27 lag days, but the dry effect was not significant.

F I G U R E 3
The estimated overall effects of mean temperature, relative humidity, and atmospheric pressure on children influenza cases over the corresponding lag days. Confounding factors included time trend, DOW, and public holidays. The red lines represent mean relative risks, and gray regions represent 95% confidence intervals. The median value of each meteorological factor (mean temperature: 23.8°C, humidity: 80%, pressure:1005 hPa) is as a reference level The high-pressure effect peaked at the current day and appeared within 0-15 lag days. The low-pressure effect appeared within 0-20 lag days.
We also calculated the cumulative extreme effect of meteorological factors on influenza (

| D ISCUSS I ON
To the best of our knowledge, this is the first in China to apply a DLNM model and use population-wide monitoring data to evaluate the effects of meteorological factors on influenza occurrence among children in a subtropical city. In addition to temperature and humidity which were often mentioned in previous studies, 6,7,19 the role of other potential meteorological factors including daily precipitation, wind velocity, and atmospheric pressure were also explored in the present study. Our study showed that the correlation between relative humidity and influenza cases was illustrated with a U-shaped F I G U R E 4 The estimated extreme effects at the 95th and the 5th percentile of mean temperature, relative humidity, and atmospheric pressure at different lag days on children influenza cases. The median value of each meteorological factor (mean temperature: 23.8°C, humidity: 80%, pressure: 1005 hPa) is as a reference level. The red lines are mean relative risks, and gray regions are 95% confidence intervals curve, and influenza cases increased with colder temperature and higher pressure.
Our study indicated that influenza cases increased significantly with a low temperature below 20°C, which is consistent with many previous studies. 5,6,10,[19][20][21] For example, a study using a guinea pig model found the most efficient transmission of influenza virus was at 5°C and the most inefficient transmission at 30°C. 22 Two studies from tropical and subtropical areas showed that cool climate (14-19°C  Previous studies found dry weather was associated with the seasonal influenza epidemic in winter in temperate areas, 4-6,9 but this association was not significant in tropical and subtropical areas such as Hong Kong. 5,7,10,20,23 Our investigation verified a U-shaped relationship between the relative humidity and influenza cases, which is similar to the results of some previous experimental and epidemiological studies. 5,24,25 The influenza virus is active when the relative humidity is below 50%, especially between 20% and 35%. 6 Influenza virus can attack the innate defense of host nasal epithelia, and it is more productive and transmittable (via respiratory aerosols and droplets) in dry weather conditions. 6,9,25 Our findings indicate that it will be useful to humidify the indoor air to curb the spread of influenza virus in dry winter.
Influenza cases normally peak in winter in temperate regions, 3,5 but the influenza peak appears between January and June in Guangzhou, Hong Kong and some other tropical and subtropical cities. 4,5,23 The spring and summer of Guangzhou are warm, humid and rainy, with occasional extreme humid wet days. High relative humidity or precipitation might swing the seasonality of influenza epidemics. 5,7,10,23,26  The effects of an extreme high value (hot effect, humid effect and high-pressure effect) are presented by the relative risk (RR) and 95% confidence interval (CI) of the meteorological factors on children influenza cases by comparing the 95th to the 90th percentile of their own effect (for all age children, the reference level of hot effect was 29.2°C, humid effect 92%, high-pressure effect 1014.10 hPa). The maximum lag day of 10 d was selected for hot effect, of 27 d for humid effect, of 15 d for high-pressure effect.
The effects of an extreme low value (cold effect, dry effect and low-pressure effect) are presented by the relative risk (RR) and 95% confidence interval (CI) of the meteorological factors on children influenza cases by comparing the 5th to the 10th percentile of their own (for all age children, the reference level of cold effect was 12.96°C, dry effect 63%, low-pressure effect 996.60 hPa). The maximum lag day of 27 d was selected for cold effect, of 20 d for dry effect and of 20 d for low-pressure effect.
Guangdong is a coastal province frequented by tropical cyclones, mostly typhoons, which bring low pressure and rich rainfall. 18,28,29 Low pressure can also reduce children's outdoor activities and interpersonal body contacts.
Several limitations exist in this study. First, the meteorological data were for outdoor measurements, but children spend most of their daytime in air-conditioned rooms where the humidity might be different from that of the outside. Second, other potentially influential factors (such as school closure, vaccination, economy) were not analyzed in our investigation. 8,30 Third, this is a single city study and more cities in different areas should also be covered.

| CON CLUS IONS
Mean temperature, relative humidity, and atmospheric pressure might be important predictors of the influenza occurrence among children in Guangzhou. Accordingly, the warning system for influenza should be improved to protect this population.

ACK N OWLED G EM ENTS
We thank the physicians from sentinel surveillance hospitals, community medical services centers, and Guangzhou Center for Disease Control and Prevention (GZCDC) for providing the data on children influenza disease cases and the Guangdong Provincial Institute of Public Health for providing the data on daily meteorological data.
We also appreciate all study participants for sharing their thoughts and experiences.

CO M PE TI N G I NTER E S TS
The authors declare that they have no competing interests.

DATA S H A R I N G
No additional data is available.