Effects of temperature, humidity, and diurnal temperature range on influenza incidence in a temperate region

Abstract Background The effect of temperature and humidity on the incidence of influenza may differ by climate region. In addition, the effect of diurnal temperature range on influenza incidence is unclear, according to previous study findings. Objectives The aim of this study was to analyze the effects of temperature, humidity, and diurnal temperature range on the incidence of influenza in Seoul, Republic of Korea, which is located in a temperate region. Methods We used Korean National Health insurance data to assess the weekly influenza incidence between 2010 and 2016, and used meteorological data from Seoul. To investigate the effect of temperature, relative humidity, and diurnal temperature range levels on influenza incidence, we used a distributed lag non‐linear model. Results The risk of influenza incidence was significantly increased with low daily temperatures of 0‐5°C and low (30%–40%) or high (70%) relative humidity. We found a positive significant association between diurnal temperature range and influenza incidence in this study. Conclusions Influenza incidence increased with low temperature and low/high humidity in a temperate region. Influenza incidence also increased with high diurnal temperature range, after considering temperature and humidity.


| INTRODUC TI ON
Influenza places a great burden on public health and the economy. 1 Various factors affect influenza transmission, including household structure, 2 individual behavior, 3 and climate variables. 4 Previous studies have reported that environmental factors such as temperature, humidity, and precipitation can affect influenza transmission. 5 Humidity and temperature conditions contribute to the seasonality of influenza outbreaks 6 ; two types of environmental conditions are associated with influenza outbreaks: "cold-dry" and "humid-rainy." 7 Multiple studies have measured the association between low temperatures and influenza infection. 8,9 Several studies have reported the significant effect of temperature on influenza. 10,11 For example, Jaakkola et al 8 showed that a 1°C decrease in temperature increased the estimated risk of influenza occurrence by 11% among military conscripts in Finland. However, other studies have found no significant association between temperature and influenza incidence. 9 The relationship between humidity and influenza activity is unclear. Shaman et al 12 showed that low absolute humidity drives seasonal variations in influenza transmission. The same researchers also reported that influenza activity modulated by seasonal changes in absolute humidity was consistent with the timing of pandemic influenza outbreaks. 13,14 A study based in the United States investigating the effect of humidity and temperature on influenza mortality reported that low absolute humidity was associated with increases in influenza mortality. 15 However, one Japanese study reported that the number of patients with influenza increased significantly during times of high relative humidity. 16 In addition, several studies did not find a significant association between humidity and influenza activity. 10,17 The association of temperature or humidity with influenza may differ by region. For example, Soebiyanto et al 18 reported that seasonal influenza showed a positive association with humidity in El Salvador and Panama but a negative association in Guatemala, which is a similarly temperate region. Another study investigating the relationship between weather factors and influenza epidemics reported different exposure-response relationships by region. 19 Additional studies assessing the effect of weather on influenza in various regions are needed.
Previous works have reported that the diurnal temperature range (DTR) is a risk factor for mortality, 20 including respiratoryrelated mortality, 21 cardiovascular mortality, 22,23 and asthma. 24 Although several studies have reported the effect of DTR on influenza hospitalization 25 or seasonal influenza, 26 research assessing the effect of DTR on influenza is scarce and the findings controversial.
In this study, we aimed to assess the effect of temperature, humidity, and DTR on the incidence of influenza in Seoul, Republic of Korea, which is located in a temperate region.

| Data collection
We extracted the data of patients diagnosed with influenza between 2010 and 2016 from the database of the Korea National Health Insurance Service (NHIS-2018-1-075). This database includes information of the obligatory national health insurance system, and beneficiaries include nearly the entire Korean population. 27 Patients are diagnosed when they visit primary clinics or are admitted to hospitals. We extracted data from patients with disease codes J09 (influenza derived from a specific influenza virus strain), J10 (influenza derived from the influenza virus, strain not identified), and J11 (influenza, virus not identified), according to the Korea Informative Classification of Disease (KOICD). 28 In this study, we used influenza incidence in Seoul, the capital of the Republic of Korea. Nearly 10 out of 50 million Koreans reside in Seoul. 29 As daily incidence data are not provided in the Korea National Health Insurance Service to protect the personal information of patients, we used weekly influenza incidence data in this study.
Previous studies have shown a relationship between influenza and air pollution. In a Taiwan study, influenza among adults and older people was affected by ambient fine particles. 30 A study in China showed that increased air pollutant concentrations were associated with an increase in cases of influenza-like illness. 31 A study by Nakao et al 32 using Korean data reported that air pollution causes respiratory symptoms. Therefore, we included air pollution data as a covariate in this study.
Temperature, humidity, and air pollution data were obtained from the Korea Meteorological Administration Web site. 13 As Korean meteorological stations only provide data of relative humidity without absolute humidity, relative humidity was used for the present analysis. Seoul meteorological stations provide data of daily temperature and relative humidity. Air pollution data were obtained from Air Quality Information of the Seoul Metropolitan Government. 33 Air pollutants included in the analysis were ozone (O 3 ), particulate matter <10 µm in diameter (PM 10 ), nitrogen dioxide (NO 2 ), and sulfur dioxide (SO 2 ). Levels of these pollutants were obtained from the same source. For weekly meteorological data, mean data values in each week were used.

| Statistical analysis
Associations between temperature, humidity, and air pollutants and the incidence of influenza were assessed using Spearman's correlation analysis. To investigate the non-linear and delayed effects of meteorological factors on influenza incidence, a distributed lag nonlinear model was used. The effects of daily temperature and humidity were assessed using 5 df natural cubic spline for mean temperature and 1 df natural cubic spline for lag up to 2 weeks. The effect of DTR was assumed to be linear, based on previous studies. 20 We used lags up to 2 weeks for temperature/humidity 10,17,34 and DTR. 20,35 PM 10 , O 3 , and NO 2 were controlled using 3 df natural cubic spline. 35 where Y t denotes the influenza incidence count on day t, μ t is the expected influenza incidence on day t, and ns (temperature), ns (humidity), and ns (DTR) indicate a natural cubic spline of ambient temperature, relative humidity, and DTR on day t, respectively. COVs are potential confounders including ozone, NO 2 , SO 2 , and PM 10 . After comparison of Poisson, quasi-Poisson, and negative binomial models, the quasi-Poisson model was selected based on the residual and assumption tests. The quasi-Poisson likelihood model was used to consider overdispersion. First, we investigated the effect of temperature and humidity. Subsequently, the effect of DTR was assessed, controlling for temperature and humidity. We reported the relative risk (RR) with 95% confidence intervals (CIs) of temperature, humidity, and change in DTR on influenza incidence with 10°C, 60%, and 8°C as reference values, respectively, which are close to the mean values.
Residuals were evaluated to assess adequacy of the model.
For all statistical tests, two-tailed P < .05 was considered statistically significant. The dlnm package in R software version 3.4.4 (The R Project for Statistical Computing, Vienna, Austria) was used to fit all the models. All analyses were conducted in 2019. For sensitivity analysis, we changed the degrees of freedom for time to control for season, and maximum lag for temperature, humidity, and DTR.  (Table 1). Figure 1 shows the trends of daily incidence for influenza, temperature, and humidity. Whereas one peak in influenza outbreaks can be seen in 2011, 2013, and 2014, two peaks are visible in other years. The incidence of influenza was higher during winter months (December-February) than summer months. Temperature and humidity were highest during June to August, which is the summer season in Korea (Figure 1).

| Correlation between influenza incidence and meteorological factors
Influenza incidence was negatively associated with temperature  Figure 2 shows the three-dimensional relationship between weekly meteorological data and influenza cases across 2 lag weeks. Overall, the estimated effects of temperature and humidity on influenza incidence were non-linear. A visual inspection of the figure suggests that there was increased RR of influenza at low temperatures and a large DTR. The effect of a large DTR decreased over the lag period ( Figure 2). With respect to humidity, since data of 6 weeks with over 90% relative humidity showed a high risk of influenza occurrence, the relationship between influenza and humidity lower than 90% was also shown in Figure 2C.

| Effects of temperature, humidity, and DTR on influenza incidence
The RR of influenza for specific lag periods (0, 1, and 2 weeks) and specific values of temperature, humidity, and DTR are presented in Table 3. Values of temperature, humidity, and DTR were selected, considering the distribution of each factor. Temperatures lower than the reference value (10°C) were associated with increased influenza incidence across all lag days, and the overall effect was significant only at 0°C (RR: 4.04, 95% CI: 1.25-13.02) and 5°C (RR: 1.96, 95% CI: 1.09-3.51), shown in Table 3.
Low humidity (30%) showed increased risk for the current (RR: We changed the lag periods for meteorological data, and the effect of these factors and trends were not substantially changed, and the effect of DTR was significant in all models. Changes in the degrees of freedom for seasonal control showed similar results ( Figure   S1). A residual plot was also created, to check the assumption of analysis (Table S2). study in Japan showed an increased influenza incidence in high humidity. 16 Further studies of the association between humidity and influenza incidence in temperate regions are needed.

| D ISCUSS I ON
Even though previous studies have reported the association between DTR and mortality 20,22 and respiratory disease, 21,24 studies investigating the effect of DTR on influenza are rare. One Chinese study reported that DTR was significantly related to influenza TA B L E 2 Spearman's correlation between meteorological factors and air pollutant levels and influenza incidence Abbreviations: DTR, diurnal temperature range; NO 2 , nitrogen dioxide; PM 10 , particulate matter <10 µm in diameter; SO 2 , sulfur dioxide. *P < .05. ***P < .001.
F I G U R E 2 Three-dimensional plot of relative risk (RR) for temperature (A), humidity (B), and diurnal temperature range (DTR) (C) and lags for influenza incidence seasonality in dry periods but not in humid periods. 25 In another study conducted in France, the effect of daily temperature variation on seasonal influenza was not significant. 26 The results of that study showed that high DTR was positively associated with influenza incidence even if the temperature is same.
We found that temperature and humidity had a short lagged effect, which was limited in one week, on influenza diagno- There is clearly a need to investigate the rates of influenza incidence according to both climate and virus strain.
One limitation of this study is that weekly data rather than daily data were used; daily influenza incidence can reflect the change in daily weather better than weekly influenza incidence.
However, this study was focused on the trend of influenza incidence and weather over a long period (ie, 6 years) rather than minor changes over a short duration. Thus, the present results provide information about the association between weather and influenza incidence.
Apart from temperature and humidity, previous studies have reported that seasonal changes in individual behavior and family and social structures 5 might be associated with the incidence of influenza. In addition, a recent study showed an association between a decline in vaccination coverage and an increase in influenza incidence. 40 We could not control for those factors in this study because of limited data. Social and behavioral factors should be considered in future studies. (No. CRC-16-01-KRICT) and Korea Institute of Oriental Medicine (G17273).

CO N FLI C T O F I NTE R E S T
The author declares that they have no competing interests.

R E S E A RCH I N VO LV I N G H U M A N PA RTI CI PA NTS A N D/ O R A N I M A L S
The study did not include human participants or animals.