Study area
Suzhou is located in the northern Anhui Province, in the Yangtze River Delta and is known as the northern gate of Anhui Province. It lies between 116°09′-118°10′ east longitude and 33°18′-34°38′ north latitude, with a total area of 9939 km2. In 2020, Suzhou has a permanent population of 5,324,476 people. It’s a warm temperate semi-humid monsoon climate zone, and the main characteristics of Suzhou are mild climate, four distinct seasons, sufficient sunshine and moderate rainfall. Figure 1 presented the geographical location information of Suzhou.
Data collection
In this study, daily depression cases from January 1, 2017, to December 31, 2019, were obtained from Suzhou Second People’s Hospital (Suzhou Mental Health Center), whose diagnosis and treatment of depression have a good credibility. The diagnosis of depression was based on the International Classification of Diseases, 10th edition (ICD-10 code: F32-F33). Case information includes gender, age, outpatient visits date, residential address, and visit types. Patients whose residential addresses were not in Suzhou were excluded.
Meteorological data, including daily mean temperature, rainfall, relative humidity, as well as sunshine duration, were obtained from China Meteorological Data Sharing Service System (http://data.cma.cn/). Daily air pollution data including particulate matter with aerodynamic diameter less than 2.5 μm (PM2.5), nitrogen dioxide (NO2) and sulfur dioxide (SO2) were retrieved from China National Environmental Monitoring Centre (http://www.cnemc.cn/).
So far, there is no unified description of the concept of extreme precipitation. In view of the regional and seasonal differences of precipitation distribution, extreme precipitation was defined by using the percentile method, which was also the method applied by many scholars [18, 19]. By using the 95th percentile as the cutoff points, we divided precipitation into three categorical variables, namely no precipitation (equal to 0 mm), moderate precipitation (> 0 mm and < 95th percentile) and extreme precipitation (≥95th percentile) [20].
Statistical analysis
Previous studies have shown that the DLNM can better evaluate the nonlinear and delayed effects of environmental exposure on health outcomes [21]. Therefore, we performed a quasi-Poisson generalized linear regression model combined with DLNM to quantitatively access the impact of extreme precipitation on outpatient visits for depression. Potential confounding factors including long-term trends and seasonality, weekdays (DOW), public holiday (Holiday), daily mean temperature (MT), relative humidity (RH), and sunshine duration (SD) were included in the model. The model was shown as follows:
$${\displaystyle \begin{array}{c} Yt\sim quasi- Poisson\ \left({\mu}_t\right)\\ {} Log\ \left({\mu}_t\right)=\propto +{\beta EP}_{t,l},4+ ns\left( MT,3\right)+ ns\left( RH,3\right)\\ {}+ ns\left( SD,3\right)+ ns\left( Time,7\right)+\eta {DOW}_t+\gamma {Holiday}_t\end{array}}$$
In the formula, t represented the observation time (day); μt was the expected number of depression outpatient visits on day t; ∝ meant the intercept of the model; β was the cross-basis matrix coefficient produced by DLNM; EPt, l referred to the extreme precipitation on day t; l was the number of lag days. In our study, extreme precipitation and lagged effects were comprised using a “natural cubic spline-natural cubic spline” method [22], and the degree of freedom (df) for exposure and lag dimensions are set to 1 and 4, respectively [23]. The ns() represented the natural cubic splines. Ns with 7 df per year was used to control long-term trend and seasonality. And ns with 3 df were used to accommodate the delayed effects of MT (lag 0–14), RH (lag 0–14) and SD (lag 0–14) [20]. Holiday and DOW were also controlled in the model as binary and categorical variables, respectively. The effect estimates were calculated as extreme precipitation relative to no precipitation.
According to the minimum Akaike Information Criterion (AIC), we selected 14 days as the maximum lag days to capture the effect of extreme precipitation (Table S1). Furthermore, subgroup analysis was performed to identify the susceptible population of depression caused by extreme precipitation based on gender (male, female), age (≤18 years, 19–39 years, 40–64 years and ≥ 65 years) and visit types (first visit, multiple visits). Those with two or more outpatient visits were considered as multiple visits. The statistical significance of the differences between subgroups was identified by calculating the 95% confidence interval (CI) of the formula [23]: \(\left(\hat{{\mathcal{Q}}_1}-\hat{{\mathcal{Q}}_2}\right)\pm 1.96\sqrt{{\left(\hat{SE_1}\right)}^2+{\left(\hat{SE_2}\right)}^2}\), Where \(\hat{{\mathcal{Q}}_1}\) and \(\hat{{\mathcal{Q}}_2}\) were the estimates for the two groups, and \(\hat{SE_1}\) and \(\hat{SE_2}\) were their respective standard errors [24].
Attributable risk can better reveal disease burden of depression caused by exposure to extreme precipitation. In our study, we used the following formulae to calculate AF and AN, which can assess the burden of depression caused by extreme precipitation [23].
$${\displaystyle \begin{array}{c}A{F}_t=R{R}_t-1/R{R}_t\\ {}A{N}_t=A{F}_t\ast {N}_t\end{array}}$$
In the formulae, Nt meant the number of outpatient visits for depression on day t. AF represented the ratio of the number of depression cases attributed to extreme precipitation to the number of depression outpatient visits.
The “splines” and “dlnm” packages were used in R software (version 4.1.2) to perform all statistical analysis. Two-sided P values less than 0.05 were considered statistically significant.
Sensitivity analysis
In this study, four sensitivity analyses were performed to test the robustness of the results: (1) changing the df for MT (3–6), RH (3–6) and SD (3–6); (2) varying the df (5–8) for time to adjust for long-term trend and seasonality; (3) replacing the P95 cut off value with different percentiles (P90, P92.5, P97.5 and P99) to check the stability of the model. (4) Air pollutants such as PM2.5, NO2, and SO2 have been shown to be associated with the risk of outpatient visits for depression [25], we compared the results before and after adding air pollutants to the model to test its robustness.