Using Geographically Weighted Regression to Study the Seasonal Influence of Potential Risk Factors on the Incidence of HFMD on the Chinese Mainland

Jingtao Sun, Sensen Wu, Zhen Yan, Yadong Li, Cheng Yan, Feng Zhang, Renyi Liu, Zhenhong DuIsprs Int. J. geo inf 2021 · DOI

TL;DR

This study uses Geographically Weighted Regression (GWR) to map how environmental and socioeconomic factors influence Hand, Foot, and Mouth Disease (HFMD) incidence across China, revealing significant seasonal and spatial variations in these relationships. The findings highlight the need for localized, season-specific prevention strategies to combat this widespread infectious disease.

What Problem Does This Paper Address?

Hand, Foot, and Mouth Disease (HFMD) is a common, yet potentially severe, infectious disease primarily affecting young children. While its spread is known to be influenced by various factors, understanding the complex interplay between these factors and HFMD incidence, especially across a vast and diverse region like China, has been challenging. Previous studies often focused on single cities or provinces, or a limited time frame, failing to capture the full spatiotemporal heterogeneity of HFMD. This paper aims to bridge this gap by investigating the seasonal and spatial variations of potential risk factors on HFMD incidence across the entire Chinese mainland over a decade.

Methods

The core of this research lies in the application of Geographically Weighted Regression (GWR). Unlike traditional Ordinary Least Squares (OLS) regression, which assumes a single set of coefficients for the entire study area, GWR allows regression coefficients to vary spatially. This means the influence of each risk factor can be different in different locations.

The study employed two types of GWR models:

  • GWR-AFG (Adaptive Bi-square kernel): This model uses an adaptive bandwidth, meaning the size of the neighborhood considered for regression varies depending on the data density.
  • GWR-AAB (Fixed Gaussian kernel): This model uses a fixed bandwidth, considering a consistent neighborhood size across the study area.

Both models were optimized using the corrected Akaike information criterion (AICc) and compared against each other and OLS regression for each season. The researchers also performed correlation and multicollinearity tests (Pearson correlation and VIF) to ensure the validity of their models. Spatial autocorrelation was assessed using Moran's I. Model performance was evaluated using R², RMSE, MAE, MAPE, and AICc.

Datasets

The study utilized the following datasets:

  • HFMD Incidence Data: Monthly incidence data for each of the 31 provinces of mainland China from 2007 to 2017, collected by the Chinese Center for Disease Control, Prevention and Health.
  • Environmental Factors:

    • Temperature (TEMP)
    • Precipitation (PREC)
    • Specific Humidity (HU)
    • Air Pressure (AP)
    • Wind Speed (WS)
    • Altitude (ALT) These were derived from meteorological observation data, reanalysis data, and remote sensing data, with spatial resolutions down to 1 km.
  • Socioeconomic Factors:

    • Child Population Density (CHD)
    • Per Capita GDP (mGDP) These were extracted from national statistical yearbooks at the provincial administrative unit level.

Key Insights and Ideas

  • Spatiotemporal Non-Stationarity is Key: The most significant takeaway is that the relationship between risk factors and HFMD incidence is not uniform. It varies dramatically across different regions of China and throughout the year, underscoring the limitations of global models.
  • Seasonal Shifts in Influence: Factors like wind speed (WS) and precipitation (PREC) show stark seasonal shifts in their influence, being beneficial in some seasons and detrimental in others. This highlights the need for dynamic, not static, prevention strategies.
  • Socioeconomic Factors Matter, But Differently: While both child population density (CHD) and per capita GDP (mGDP) are important, their roles fluctuate. For instance, mGDP consistently shows a positive correlation with HFMD, suggesting economic activity and associated social contact drive transmission, while CHD's influence is more complex and season-dependent.
  • Altitude's Indirect Role: Altitude's inhibitory effect in high-altitude provinces suggests that geographical isolation might play a role in limiting disease spread, a nuanced finding beyond simple environmental exposure.
  • GWR Outperforms OLS: The consistent improvement in model performance metrics (R², RMSE, MAE, etc.) when using GWR over OLS strongly validates the necessity of accounting for spatial heterogeneity in HFMD research.

Key Results

  • Model Performance: GWR models (both adaptive and fixed kernel) significantly outperformed Ordinary Least Squares (OLS) regression across all seasons, with R² values increasing substantially (e.g., from 60.01% with OLR to 75.47% with GWR-AAB in spring).
  • Seasonal Variations:
    • Spring: mGDP had the greatest influence in the southwest and southeast. AP showed inhibitory effects in the northeast, southwest, and northwest, but positive effects in the southeast and north.
  • Summer: WS positively influenced HFMD in northeastern and eastern China. mGDP had a positive effect in most provinces.
  • Autumn: CHD had a negative effect in high-risk northern, central, and southern regions. AP positively affected HFMD in northern and eastern China. WS had a positive effect in 80.6% of provinces.
  • Winter: mGDP had a positive effect nationwide. AP positively influenced HFMD in 75% of provinces. PREC had a negative effect across all provinces, potentially due to low precipitation inhibiting virus reproduction.
  • Statistical Significance: The "pass rate" (proportion of statistically significant provinces) for risk factors varied by season, with mGDP consistently showing high significance. Winter exhibited the highest overall statistical significance for risk factors.

Limitations and Future Work

The authors acknowledge two main limitations:

  • Medical Resources: The study did not incorporate data on medical resources (e.g., number of doctors, hospital beds per capita), which could significantly impact HFMD risk and outcomes. Future work should include these factors.
  • Combined Environmental Effects: The study analyzed environmental factors individually. However, the combined effects of different climatic factors (e.g., temperature and humidity interactions) might create unique influences on HFMD transmission that were not explored. Future research should investigate these synergistic effects.

Despite these limitations, the study's findings provide a valuable foundation for developing more targeted and effective HFMD prevention strategies by considering the unique spatiotemporal dynamics of risk factors across China.

Summary updated

BibTeX
@article{sun2021using,
  author = {Jingtao Sun and Sensen Wu and Zhen Yan and Yadong Li and Cheng Yan and Feng Zhang and Renyi Liu and Zhenhong Du},
  title = {{Using Geographically Weighted Regression to Study the Seasonal Influence of Potential Risk Factors on the Incidence of HFMD on the Chinese Mainland}},
  journal = {Isprs Int. J. geo inf},
  year = {2021},
  doi = {10.3390/ijgi10070448},
}
Figure