Analysis of heterogeneous dengue transmission in Guangdong in 2014 with multivariate time series model

Qing Cheng, Xin Lü, Joseph T. Wu, Zhong Liu, Jincai HuangScientific 2016 · DOI

Abstract

Guangdong Province experienced a record-breaking dengue epidemic in 2014, accounting for over 90% of cases in the preceding decade. This study utilizes a multivariate time series model to decompose transmission into endemic, autoregressive, and spatiotemporal components. By incorporating random effects and a power-law approach for spatial interaction, the research identifies significant heterogeneity in transmission drivers. High-incidence urban centers in the Pearl River Delta were primarily driven by endemic background risk, while lower-incidence peripheral areas relied heavily on spatiotemporal spread from neighbors. These findings provide a framework for targeted public health interventions based on local transmission profiles.

Introduction

Dengue fever remains a significant public health challenge in mainland China, with Guangdong Province serving as the primary epicenter. The 2014 outbreak was unprecedented, involving over 45,000 cases—more than the total of the previous ten years combined. Understanding whether local outbreaks are driven by internal persistence, environmental suitability, or importation from adjacent areas is crucial for effective resource allocation. Standard epidemiological models often assume uniform transmission, but real-world data is highly heterogeneous. This paper seeks to quantify that heterogeneity to inform targeted control strategies.

Methodology

The authors employed a multivariate time series model to analyze weekly case counts across 21 districts. The core approach is an additive decomposition of dengue risk into three parts:

  1. Endemic Component: Background risk driven by external factors like climate and population density, independent of the epidemic's history.
  2. Autoregressive Component: Local transmission within a district based on its own past case counts.
  3. Spatiotemporal Component: Transmission originating from other districts.

To account for unobserved differences between districts, random effects were introduced. Crucially, the model implemented a Power Law (PL) approach to spatial interaction, assuming that the influence of one district on another decays as the distance increases. Finally, clustering analysis was used to group districts based on their specific transmission fingerprints.

Results

The 2014 epidemic was characterized by a sharp, synchronous peak across the province, occurring around week 40.

Weekly and cumulative dengue cases in Guangdong during the 2014 epidemic, showing the peak at week 40.

Spatial analysis revealed significant disparities in incidence and infection counts. The Pearl River Delta area, particularly Guangzhou and Foshan, exhibited the highest incidence rates, while peripheral districts remained relatively low.

Spatial distribution of dengue incidence and total infected cases across Guangdong districts in 2014.

Clustering analysis identified three distinct transmission patterns:

  • Pattern A (The Engines): High-incidence areas like Guangzhou where endemic factors provide a massive background risk, contributing over 70% of cases in the early stages.
  • Pattern B (The Transitionals): Districts influenced by a mix of endemic risk and spatial spread.
  • Pattern C (The Satellites): Low-incidence areas where the outbreak is almost entirely "imported" from neighbors via spatiotemporal spread.

Discussion and Conclusion

The study demonstrates that dengue transmission in Guangdong is highly heterogeneous. While the autoregressive component (local persistence) showed little variation across the province, the endemic and spatiotemporal components varied significantly. High-incidence regions require interventions focused on reducing environmental suitability, whereas low-incidence regions should prioritize monitoring and controlling importation from neighboring hotspots.

Limitations include the use of the Power Law as a proxy for mobility; future models could incorporate actual human movement data from mobile phone logs. Additionally, explicitly adding environmental covariates like mosquito density could refine the model's predictive power. This multivariate framework provides a robust tool for understanding and managing complex infectious disease outbreaks in urbanized regions.

Summary updated

Abstract

Guangdong experienced the largest dengue epidemic in recent history. In 2014, the number of dengue cases was the highest in the previous 10 years and comprised more than 90% of all cases. In order to analyze heterogeneous transmission of dengue, a multivariate time series model decomposing dengue risk additively into endemic, autoregressive and spatiotemporal components was used to model dengue transmission. Moreover, random effects were introduced in the model to deal with heterogeneous dengue transmission and incidence levels and power law approach was embedded into the model to account for spatial interaction. There was little spatial variation in the autoregressive component. In contrast, for the endemic component, there was a pronounced heterogeneity between the Pearl River Delta area and the remaining districts. For the spatiotemporal component, there was considerable heterogeneity across districts with highest values in some western and eastern department. The results showed that the patterns driving dengue transmission were found by using clustering analysis. And endemic component contribution seems to be important in the Pearl River Delta area, where the incidence is high (95 per 100,000), while areas with relatively low incidence (4 per 100,000) are highly dependent on spatiotemporal spread and local autoregression.
BibTeX
@inproceedings{cheng2016analysis,
  author = {Qing Cheng and Xin Lü and Joseph T. Wu and Zhong Liu and Jincai Huang},
  title = {{Analysis of heterogeneous dengue transmission in Guangdong in 2014 with multivariate time series model}},
  booktitle = {Scientific},
  year = {2016},
  doi = {10.1038/srep33755},
}
Figure