Study area
This study was conducted in Hefei, which is the capital and largest city of Anhui province in Eastern China with a population of 8.09 million inhabitants (from 2018 census data). Hefei has a humid subtropical climate with mean temperature was 16.8°C.
Arthritis data
Daily counts of outpatient admission for OA/RA during 2014-2017 were obtained from The First Affiliated Hospital of University of Science and Technology of China (Anhui Provincial Hospital). The patient data included date of outpatient admission, age, gender, residential address. Diagnosis of OA (ICD-10: M13.9) and RA (ICD-10: M06.9) was coded according to the International Classification of Disease, 10th Revision (ICD-10). Ethical approval was obtained from the Ethics Committee of Anhui Provincial Hospital prior to data collection.
Weather and air pollutants data
Meteorological data on daily mean temperature, relative humidity, rainfall, barometric pressure and wind velocity during the same period were obtained from Hefei Bureau of Meteorology. Air pollution data including average daily level of sulfur dioxide (SO2), nitrogen dioxide (NO2), carbon monoxide (CO), ozone (O3), particulate matter of less than 10µm and 2.5µm (PM10 and PM2.5) were collected from the Environmental Protection Bureau in Hefei. Daily 24-h mean concentrations of SO2, NO2, CO, PM10 and PM2.5, and daily maximum 8-hour mean concentrations of O3 were calculated. Consistent with previous study [15], we chose the 50th percentile of temperature (P50, 17.8°C) as the reference in analyses.
Statistical analysis
We first examined the correlations among weather indicators and pollutants with Spearman’s correlation test. Then, we applied a Poisson generalized linear regression combined with distributed lag non-linear model (DLNM) to examine the non-linear and lagged effects of ambient temperature on outpatient admission for OA/RA, after controlling for long-term trend and seasonality, day of week (DOW), public holidays (Holiday), humidity, wind velocity, PM2.5, SO2, NO2 and O3. The core model is expressed as follows:
Yt~Poisson(µt)
Log(µt)=α+βTemperaturet,l+ns(Humidityt,l,3)+ns(Windt,l, 3)+ns(PM2.5t,l, 3)+ns(SO2t,l, 3)+ns(NO2t,l, 3)+ns(O3t,l, 3)+ns(Timet, 8)+ŋDOWt+ γHolidayt
Where Yt is the number of OA/RA admission on day t; α represents the intercept; Temperaturet,l, Humidityt,l, and Windt,l are the cross-basis matrix produced by DLNM, β is vector of coefficients for Temperaturet,l, and l is the lag days; ns() denotes a natural cubic spline; three degrees of freedom (dfs) with lags 0-7 was used to adjust for humidity, wind velocity, PM2.5, SO2, NO2 and O3; 8 dfs per year for time was used to control for long-term trend and seasonality; DOW and Holiday were controlled for in the model as a categorical variable, respectively.
On the basis of the lowest Akaike Information Criterion (AIC), we selected the maximum lag of four days to capture any single and cumulative effects of temperature. Because the plot of overall exposure-response did not find the relationship between temperature and OA admission (Fig. 1), we only quantified the relative risks (RRs) of temperature change on RA admission by single day lags at low temperature (25th percentile, P25) compared to the reference temperature (50th percentile, P50). Furthermore, we examined the specific cumulative effects of temperature decrease on RA admission by gender (male and female) and age (0-17 years, 18-40 years, 41-65 years and ≥66 years).
To test the robustness of our results, sensitivity analyses were performed by varying df for time (7-9 dfs/year), humidity (3-5 dfs) and wind velocity (3-5 dfs), respectively. Data manipulation and analyses were conducted using R software (version 3.1.1), with the “dlnm” package to fit the DLNM [14].