Faculty of Veterinary Science, The University of Sydney, 425 Werombi Road, Camden, NSW 2570, Australia.
Prev Vet Med. 2011 Dec 1;102(3):185-95. doi: 10.1016/j.prevetmed.2011.07.006. Epub 2011 Aug 17.
This paper explores methods for describing the dynamics of early epidemic spread and the clustering of infected cases in space and time when an underlying contact network structure is influencing disease spread. A novel method of describing an epidemic is presented that applies social network analysis to characterise the importance of both spatial location and contact network position. This method enables the development of a model of how these clusters formed, incorporating spatial clustering and contact network topology. Based on data from the first 30 days of the 2007 equine influenza outbreak in Australia, clusters of infected premises (IPs) were identified and delineated using social network analysis to integrate contact-tracing and spatial relationships. Clusters identified by this approach were compared to those detected using the space-time scan statistic to analyse the same data. To further investigate the importance of network and spatial location in epidemic spread, kriging by date of onset of clinical signs was used to model the spatio-temporal spread without reference to underlying contact network structure. Leave-one-out cross-validation was used to derive a summary estimate of the accuracy of the kriged surface. Observations poorly explained by the kriged surface were identified, and their position within the contact network was explored to determine if they could be better explained through analysis of the underlying contact network. The contact network was found to lie at the core of a combined network model of spread, with infected horse movements connecting spatial clusters of infected premises. Kriging was imprecise in describing the pattern of spread during this early phase of the outbreak (explaining only 13% of the variation in date of onset of IPs), because early dissemination was dominated by network-based spread. Combined analysis of spatial and contact network data demonstrated that over the first 30 days of this outbreak local spread emanated outwards from the small number of infected premises in the contact network, up to a distance of around 15km. Consequently, when a contact network structure underlies epidemic spread, we recommend applying a method of spatial analysis that incorporates the network component of disease spread. Linking the spatial and network analysis of epidemics facilitates inference of the methods of disease transmission, providing a description of the sequence of spatial cluster formation.
本文探讨了在底层接触网络结构影响疾病传播时，描述早期疫情传播动态和时空感染病例聚集的方法。提出了一种新的疫情描述方法，该方法应用社交网络分析来描述空间位置和接触网络位置的重要性。该方法能够开发一种模型，说明这些集群是如何形成的，包括空间聚类和接触网络拓扑。利用澳大利亚 2007 年马流感爆发的前 30 天的数据，通过社会网络分析将接触追踪和空间关系相结合，识别和划定感染场所（IP）集群。使用这种方法识别的集群与使用时空扫描统计分析相同数据时检测到的集群进行了比较。为了进一步研究网络和空间位置在疫情传播中的重要性，通过发病日期进行克里金插值，在不考虑底层接触网络结构的情况下，对时空传播进行建模。使用逐一剔除交叉验证方法，从克里金插值表面得出疾病传播的准确摘要估计。识别出克里金插值表面无法解释的观察结果，并探索它们在接触网络中的位置，以确定通过分析底层接触网络是否可以更好地解释这些观察结果。接触网络是传播综合网络模型的核心，感染马匹的移动将感染场所的空间集群连接起来。在疫情爆发的早期阶段，克里金插值在描述传播模式方面不够精确（仅解释了感染场所发病日期变化的 13%），因为早期传播主要是基于网络的传播。对空间和接触网络数据的综合分析表明，在疫情爆发的前 30 天，局部传播从接触网络中少数受感染的场所向外传播，传播距离可达 15 公里左右。因此，当接触网络结构是疫情传播的基础时，我们建议应用一种空间分析方法，该方法包含疾病传播的网络成分。将疫情的空间和网络分析联系起来，可以推断出疾病传播的方法，提供了对空间集群形成顺序的描述。
