The burden of childhood hand-foot-mouth disease morbidity attributable to relative humidity: a multicity study in the Sichuan Basin, China

Hand, foot and mouth disease (HFMD) is a growing threat to children's health, causing a serious public health burden in China. The relationships between associated meteorological factors and HFMD have been widely studied. However, the HFMD burden due to relative humidity from the perspective of attributable risk has been neglected. This study investigated the humidity-HFMD relationship in three comprehensive perspectives, humidity-HFMD relationship curves, effect modification and attributable risks in the Sichuan Basin between 2011 and 2017. We used multistage analyses composed of distributed lag nonlinear models (DLNMs), a multivariate meta-regression model and the calculations of attributable risk to quantify the humidity-HFMD association. We observed a J-shaped pattern for the pooled cumulative humidity-HFMD relationship, which presented significant heterogeneity relating to the geographical region and number of primary school students. Overall, 27.77% (95% CI 25.24–30.02%) of HFMD infections were attributed to humidity. High relative humidity resulted in the greatest burden of HFMD infections. The proportion of high humidity-related HFMD in the southern basin was higher than that in the northern basin. The findings provide evidence from multiple perspectives for public health policy formulation and health resource allocation to develop priorities and targeted policies to ease the HFMD burden associated with humidity.


Figures
. The stratified analyses for regional differences in humidity effects in the first peak. Figure S5. The stratified analyses for regional differences in humidity effects in the second peak. Tables   Table S1 Description Figure S4. The stratified analyses for regional differences in humidity effects in the first peak. Text S1. Sensitivity Analysis 1. The choice of degrees of freedom to control long-term trends and seasonality.
To control long-term trends and seasonality, we performed a sensitivity analysis to change the dfs from 1 to 15 per year. The results are presented in Figure S6. When the dfs of natural cubic splines exceeded 8, both the average estimates of the relative humidity effect and goodness of fit remained stable. Therefore, we set the dfs per year to 8 to control long-term trends and seasonality. 2. The choice of the reference value of relative humidity.
We performed an alternative analysis with references values of 75%, 80%, and 40%. As shown in Figure S7 and Table S3, all the shapes of the humidity-HFMD curves and the main results of modification effects in the second-stage analysis were robust. We used 70% humidity as the reference value mainly based on the following reasons. Firstly, most studies have used the mean or median value of exposure as a reference to achieve higher statistical power and narrower confidence intervals. In our study, the range of the mean relative humidity was 70%-80% in the 17 cities (Main text: Table 1). According to the results of the sensitivity analysis, setting references in this range yielded the same shapes of exposure-response curves and narrower confidence intervals than other references such as 40%. Secondly, the exposure-response curve was initially relatively flat. However, once the relative humidity exceeded approximately 70%, the curve began to rise, indicating that the risk of HFMD increased with increasing relative humidity (Main text: Figure 1). Therefore, we used 70% humidity as the reference value. Figure S7. Overall humidity-HFMD relationship curves with 70%, 75%, 80%, and 40% relative humidity as the reference values. For each confounder, we compared 9 different models by varying the association pattern (i.e., linear or nonlinear) and lag structure (i.e., how many lags should be included). Table S4 shows the detailed settings of the confounder models. C0 is the null model with no confounder.
C1-C4 are based on the assumption of a linear relationship of the association pattern. C1 includes only one lag, which is set to the median of the incubation period. C2 to C4 include multiple lags but use different methods to summarize the multiple lags through simple and exponential moving weighted averages and distributed lag models. A natural cubic spline with 4 dfs is applied to constrain the distributed lag model. Similarly, C5-C8 are based on the assumption of a nonlinear relationship of the association pattern. A natural cubic spline with 3 dfs is applied to describe the nonlinear relationship. The average estimates of humidity effects and model fits are presented in Figures S8-S11. As shown in Figure S8, for the mean temperature results, the confounder model (C6) based on the assumption of a nonlinear relationship with multiple lags and simple moving weighted averages has the best model fit.
Similarly, for the other three confounders, we simultaneously considered the parsimony and the consistency of controlling for the time-varying confounders. Finally, we controlled these factors (atmospheric pressure, sunshine duration, and wind velocity) by calculating exponential moving weighted averages in the same lag range as relative humidity.   4. The choice of the lag structure of relative humidity effects on HFMD.
To explore the whole lag structure of humidity effects, we used a series of lag settings up to 30 days to construct an unconstrained distributed lag model for the 17 cities. According to the results shown in Figure S12, we chose a max lag of 17 days for two main reasons. Firstly, as shown in Figure S12, the average estimates of humidity effects after the lag of 17 days for the 17 cities gradually decreased to zero and then regularly fluctuated around zero, which may be negligible. Secondly, considering the incubation period (3-5 days) and the approximately two-week infection period for HFMD, 17 days was chosen to reflect the lag structure of humidity effects on HFMD based on the natural history of HFMD infection. Splines with 3-5 dfs were commonly used to capture the lag distribution when modeling the relationship between weather and infectious diseases. As our main focus is shaping the exposure-response relationship instead of the lag-response relationship, we chose natural cubic 4-dfs splines, which seem to be sufficient to express the complexity of the lag distribution according to the lag structure figure ( Figure S12). Figure S12. Average estimates of relative humidity effects at different lags (up to 30 days).
5. The choice of the dfs of splines that define the relative humidity-HFMD relationship.
We performed a systematic sensitivity analysis to determine the dfs of natural cubic splines for the effects of relative humidity by varying the dfs from 3 to 8. The results are presented in Figure S13. Natural cubic splines for the effects of relative humidity with 3 dfs show the best goodness of fit, and different dfs yielded robust shapes of the humidity-HFMD curve. Therefore, we used natural cubic splines for the effects of relative humidity with 3 dfs in the final model. Figure S13. Overall model fits and humidity-HFMD relationship curves for different dfs settings of the exposure-response association.