Effect of meteorological factors on influenza-like illness from 2012 to 2015 in Huludao, a northeastern city in China

Study setting, ILI surveillance data collection and ethical considerations

This work was based on the ILI dataset supplied by Huludao Municipal Center for Disease Control and Prevention, which served a total of 2.78 million inhabitants. Huludao is a coastal city (40°N and 120°E) of Liaoning province in the northeastern part of China (Fig. 1) (Guan et al., 2009). The dataset included all ILI cases reported by experienced clinicians in Huludao Municipal Center Hospital on a weekly basis between January 1st, 2012 and December 31st, 2015. A total of 208 weeks’ ILI data was collected, and the start time of the first week is January 2nd, 2012.

Huludao Municipal Center Hospital is one of the China national ILI sentinel surveillance sites (Fig. 1), the number of ILI cases together with their age information and the total visits to outpatient and/or emergency departments of internal medicine and pediatrics were recorded daily and reported weekly through an Internet-based platform maintained by the Chinese Center for Disease Control and Prevention. The surveillance was conducted according to the existing guidelines provided in the 2010 Edition of China National Influenza Surveillance Program (General Office of the Ministry of Health, China, 2010). The standard case definition of ILI was fever (body temperature higher or equal to 38 °C) plus cough or sore throat, in the absence of other alternative diagnoses (Xu et al., 2013; Yang et al., 2009). The data were aggregated by week (Monday to Sunday) and age category.

Research institutional review boards of China Medical University and Huludao Municipal Center for Disease Control and Prevention both determined that the collection of data from ILI cases was part of continuing public health surveillance of a legally mandated notifiable infectious disease and was exempt from institutional review board assessment. All the data were supplied and analyzed in the anonymous format, without any access to the personal identifying information.

Meteorological data

Daily meteorological data from Xingcheng Weather Station (the closest national meteorological station nested in Huludao city, within 20 kilometres of the population-weighted centre of the city) were retrieved from the China Meteorological Data Service Center and collected over the same time period of time as the ILI data (Fig. 1). The meteorological variables were average, maximum, minimum air temperature (°C), vapor pressure (0.1 hPa) and relative humidity (%).

Statistical analysis

Surveillance data of ILI cases and meteorological variables were exported to Microsoft Excel 2003, continuous variables are expressed as the mean ± standard deviation or median (P₂₅, P₇₅), while categorical variables are described by absolute and relative frequencies.

The Spearman correlation analyses was performed to explore the correlations between the meteorological factors. Generalized additive model was used to seek the relationship between the number of ILI cases and the meteorological factors. GAM proposed by Hastie & Tibshirani (1990) is a generalized linear model in which part of the linear predictor is specified in terms of sum of smooth functions of predictor variables.

The basic form of GAM applied in the present study is expressed using the following Eq. (1). (1)${\displaystyle g\left[E\left(Y_i\right)\right]={\beta }_0+\sum _{i=1}^m{s}_i\left(x_{i-l}\right)+{\epsilon }_i.}$

In the present study, Y_i is the count of weekly number of ILI, β₀ denotes the intercept, i indicates the calendar week, multiple smoothing parameter estimation was made based on the basis of Poisson distribution g, s_i(x_i−l) denotes a smooth function of meteorological variables x_i−l, ε_i is error. Lag time l was determined by the minimum AICs. The lag effect of weather for ILI cases was explored for one and two weeks according to previous literature (Davis, Rossier & Enfield, 2012; Zhao et al., 2018).

The Spearman correlation analyses were conducted by using the software IBM SPSS Statistics 21.0 for windows (IBM, Asian Analytics Shanghai) and the mgcv package in R software was used to construct the GAM models (Code S1).

Description of ILI surveillance dataset

A total of 29 622 weekly reported ILI cases from 2012 to 2015 in Huludao city were collected (Fig. 2, Table 1, Dataset S1). During the study period, the total percentage of children with ILI was approximately 86.77%.

Spearman correlation coefficients between meteorological variables

The correlation analysis between the meteorological factors indicates that average air temperature, maximum air temperature, minimum air temperatures and vapor pressure are strongly correlated with each other, and correlated with relative humidity moderately (Table 2).

Lag effect selection for individual meteorological variable

The lag time for average air temperature, maximum air temperature, minimum air temperature, vapor pressure and relative humidity are 2, 2, 1, 1 and 0 weeks, respectively (Table 3).

Summary for individual meteorological variable and two explanatory variables for GAM models

In the single variable models, the adjusted R² for average air temperature, maximum air temperature, minimum air temperature, vapor pressure and relative humidity is 11.5%, 4.7%, 13.2%, 12.4%, 3.2%, respectively, and the models can each explain 16.5%, 9.53%, 18.0%, 15.9%, 7.7% of the deviance of the number of ILI cases (Table 4). The effective degree of freedom for each variable is greater than 8, indicating the complex spline relationship between meteorological variables and the occurrence of ILI cases. Minimum air temperature and relative humidity were included in the two variable model, because minimum air temperature is the best explanatory temperature index among the three temperature indexes, relative humidity is the least relevant explanatory variable with the other four meteorological variable. The two variable model could explain 23.2% of the deviance.

ILI activity and Relative risk of the meteorological variables

In winter, relative risk (RR) of ILI decreases from 1.14 (exp(0.13)) to 0.90 (exp(−0.11)), when the average air temperature changes from −4 °C to 1 °C (Fig. 3). When the maximum air temperature changes from 2 degrees to 8 degrees, RR decreases from 1.08 (exp(0.08)) to 0.88 (exp(−0.13)) (Fig. 4). When the minimum air temperature changes from −13 °C to −6 °C, RR decreases from 1.11 (exp(0.10)) to 0.90 (exp(−0.10)) (Fig. 5). RR decreases from 1.11 (exp(0.10)) to 0.92 (exp(−0.08)) as the vapor pressure changes from 10 (0.1 hPa) to 45 (0.1 hPa) (Fig. 6). When relative humidity changes from 43% to 64%, RR decreases from 1.15 (exp(0.14)) to 0.95 (exp(−0.05)) (Fig. 7).

In summer, RR increases from 0.88 (exp(−0.13)) to 1.16 (exp(0.15)), when the average air temperature changes from 18 °C to 24 °C (Fig. 3). When the maximum air temperature changes from 24 °C to 29 °C, RR increases from 0.95 (exp(−0.05)) to 1.14 (exp(0.13)) (Fig. 4). RR increases from 0.88 (exp(−0.13)) to 1.21 (exp(0.19)), when the minimum temperature changes from 13 °C to 18 °C (Fig. 5). When the vapor pressure changes from 195 (0.1 hPa) to 255 (0.1 hPa), the RR value increases from 0.94 (exp(−0.06)) to 1.32 (exp(0.28)) (Fig. 6). RR increases from 1.00 (exp(0.00)) to 1.35(exp(0.30)), when relative humidity changes from 76% to 98% (Fig. 7).