As shown in Figure 1, Qingdao is a coastal city of Shandong province, which is situated in the eastern of China between longitude 119°30′-121°00′ E and latitude 35°35′-37°09′ N. The city has a mid-temperate continental monsoon climate with an annual average of 12.7°C and annual cumulative precipitation of 662.1 mm. Additionally, as a harbor city, Qingdao is the economic center of Shandong province with a population density of 801 persons per km2 (in 2014: population=9,046,200; land size=11282 km2).
Data collection and management
Data on disease
Daily data on scarlet fever from 2014 to 2018 in Qingdao were obtained from the Notifiable Disease Surveillance System (NDSS). According to the Chinese Infectious Diseases Law, clinicians must report to the NDSS when they identify any probable, clinical, or laboratory-confirmed case of scarlet fever within 24 hours of diagnosis, and it ensures that the morbidity of scarlet fever is a relative real figure of the city. Additionally, all cases of scarlet fever including probable, clinical, and laboratory-confirmed infections were diagnosed according to the diagnostic criteria for scarlet fever issued by the Ministry of Health of the People’s Republic of China in 2008 . This study was approved by the Ethics Commission of Municipal Centre of Disease Control and Prevention of Qingdao and written informed consent was not applicable.
Air pollution data
Air pollution data during 2014-2018 in Qingdao were obtained from China National Environmental Monitoring Center, including data of daily air quality index and air pollutant concentrations, such as PM2.5, PM10, sulfur dioxide (SO2), carbon monoxide (CO), nitrogen dioxide (NO2) and ozone (O3). According to Ambient Air Quality Standards issued by Ministry of Ecology and Environment of the People’s Republic of China in December 2012, the standard limits of PM2.5, PM10, SO2, CO and NO2 concentrations, equivalently to the 24-hour means, are 75μg/m3, 150μg/m3, 150μg/m3, 4mg/m3 and 80μg/m3, respectively, followed by the O3 concentration limit with 200μg/m3 on eight hours average.
Air pollution is defined as the phenomenon or event that the content of any substance in atmospheric are varied harmfully for ecological stability and the condition of human survival, causing hazards for human, animals, vegetation or material. Severity of air pollution is indicated by different air quality index (AQI) value ranges. AQI is a number used by government agencies to communicate to the public how polluted the air is currently, which is summarized by considering several main air pollutants and calculated by Individual Air Quality Index (IAQI) of each pollutant. IAQI represents the state of individual contaminant. The IAQI was calculated as follows according to the Technical Regulation on Ambient Air Quality Index (on trial):
See formula 1 in the supplementary files.
IAQIP represents the Individual Air Quality Index of P contaminant. Cp represents the mass concentration of P contaminant. BPHi and BPLo represent the highest and lowest value of concentration limit like CP, respectively. IAQIHi and IAQILo represent the Individual Air Quality Index of BPHi and BPLo, respectively.
The AQI was calculated as followed :
See formula 2 in the supplementary files.
IAQI represents the Individual Air Quality Index of contaminants. n represents the specific contaminant.
AQI values are divided into four ranges, and each range is assigned a descriptor for air pollution level. According to the Technical Regulation on Ambient Air Quality Index (on trial), air pollution is divided into 4 levels on the basis of AQI value, including mild pollution (AQI:101-150), moderate pollution (AQI:151-200), severe pollution (AQI:201-300) and most severe pollution (AQI:＞300). In our study, we divided the air pollutions into 4 levels according to the AQI values mentioned above.
Meteorological data from 2014 to 2018 were collected from the China Meteorological Data Sharing Service System (http://cdc.cma.gov.cn/), which includes daily data such as cumulative precipitation, average temperature and average air pressure.
First, the distribution of scarlet fever morbidity and air pollution variables were described during the study period. Second, a generalized additive Mixed Model (GAMM) combined with a distributed lag non-linear model (DLNM) was applied to quantify the distributed lag effects of air pollutions on scarlet fever, with daily morbidity of scarlet fever as the dependent variable and air pollutions as the independent variable adjusted for potential confounders. A quasi-Poisson regression was used to deal with the over dispersion of Poisson distribution. In order to control the potential confounds, such as meteorological factors, long-term and seasonal trend, day of the week (DOW) and public holidays were introduced into the model simultaneously. The model is as follows:
See formula 3 in the supplementary files.
Where t referred to the day of the observation. Yt denoted the daily morbidity of scarlet fever on day t. α was the intercept. Pollutiont,l and Pollutantt,l, were matrixes obtained by applying the DLNM to air pollution and air pollutants over a lag of 0 to l days. γ and δ were the vectors of corresponding air pollution and pollutant variables. Pollutiont,l was a ordinal categorical variable, using 1, 2, 3, 4 to present mild, moderate, severe and most severe air pollution, respectively. Pollutantt,l was a metric variable presenting the concentration of air pollutant. NS() represents the natural spline function. DF was the degree of freedom of the nonparametric smoothing spline function. Prect, Tempt and Pressuret refered to cumulative precipitation, mean temperature and mean air pressure on day t, respectively. Time, as the number of days (1, 2, 3…), was used to control for long-term trend and seasonality confounding. DOWt was the day of the week on day t, which was a categorical variable. Holiday was a binary variable that the value was “1” if day t was a public holiday.
Air pollutants usually have a highly interaction effect, which may result in collinearity in the model. In order to avoid the collinearity, the pairwise correlation was applied by spearman correlation in all air pollutants. Among the six air pollutants, there were three pollutants such as two PM pollutants (PM2.5 and PM10) and O3 with no significant correlation (P>0.05). However, considering the previous findings of researches, we focused on PM2.5 and O3 as the pollutant variables included in the model (Table S1). All degrees of freedom of variables were selected according to the empirical researches. In order to completely capture the effects of air pollution and air pollutant concentrations on daily morbidity of scarlet fever, the DLNM was applied for air pollution and air pollutants in our study with both 3 degrees of freedom (DF) [20-22]. Using a natural cubic spline, we chose DF as 7 per year for Time to remove long term trends and seasonality . Additionally, we used smooth function of natural cubic splines with 3 DF in the model for cumulative precipitation, mean temperature and air pressure [23,24]..
Previous studies have shown that the lagged effect of air pollutants on respiratory diseases were usually short [25,26]. The incubation period of scarlet fever is usually between 1 and 3 days . However, considering the delayed environmental transport of pathogens and delayed onset of clinical symptoms, morbidity of scarlet fever was expected to peak several days after the exposure of air pollution. Therefore, a lag effect at a maximum of 7 days were applied in the DLNM.
Sensitive analysis was performed by altering DF (6-9 per year) for Time, and DF (2-5) for cumulative precipitation, mean temperature and air pressure. R software (version 3.2.2, R Development Core Team 2015) was used to perform all statistical analyses. The “dlnm” package was used to create the DLNM model. All statistical tests were two-sided, and p values with less than 0.05 were considered statistically significant.