• No results found

Chapter 3 Spatial Distribution and Cluster Analysis

3.4 Discussion

The SARS outbreak in Hong Kong was characterised by large proportion of hospital infections, with half of the hospital cases occurring in two hospitals (PWH and AHN). Half of those infected in hospital were healthcare workers. The other large outbreak at Amoy Gardens accounted for one- quarter of all cases and was caused by environmental factors in the affected building. These three large outbreaks certainly have implications for the spatial distribution of cases in the epidemic as a whole. My work has explored the impact of these outbreaks on the spatial disease clusters that are identified, and on the variability in incidence spatially.

I found that the distribution of cases in the SARS outbreak in Hong Kong was spatiotemporally clustered, in agreement with a previous study (Lai, Wong et al. 2004). A considerable proportion of administrative districts remained uninfected, and the incidence rates of the infected areas varied substantially. Half of the total cases were contained in the spatial clusters I identified, and the majority of the clusters were located around major outbreak hospitals including AHN and PWH, and the Amoy Gardens. Before the PWH outbreak, the epidemic was sporadic and the cases were scattered over the territory. The PWH outbreak generated a significant increase in both incidence and spatial spread of SARS, and subsequently the formation of spatiotemporal clusters, the majority of which could be identified within a week of the start of the PWH outbreak.

Hospital and community transmission were inter-related in the SARS outbreak - people infected in hospitals generated secondary cases in the community, both in households and other community settings, and all the infected subjects were hospitalised (sub-clinical infections being negligible in SARS (Leung, Lim et al. 2006)). This made hospitals treating SARS cases places with high transmission risk. In this study I have examined two aspects of the influence of hospitals on the epidemic - disease cluster formation and the size of community transmission.

In the spatial clusters located near hospitals, 27% - 41% of cases were infected in the community. On comparing spatiotemporal clusters identified examining solely hospital or solely community cases the same clusters were identified around AHN and PWH, indicating the importance of the location of major outbreak hospitals on both hospital and community disease clusters. Temporally these common clusters first started as hospital-only clusters and then became simultaneous hospital- transmission and community-transmission clusters, suggesting hospital outbreaks triggered community transmission.

The size of the community outbreak in an area increased with the size of the outbreak in the nearest hospital treating SARS, and an area was more likely to have no community-infected cases if it was far

85

from hospitals treating SARS, or had less hospital-infected cases within the area. Therefore the distance to hospitals treating SARS patients can be considered as intrinsic factor of spatial heterogeneity of exposure of an area in the transmission of SARS. Population distribution of the elderly was related to the spatial heterogeneity of susceptibility of an area, with a higher proportion of individuals aged 75+ in the area increasing the size of the community outbreak, while a higher proportion of the age group 50-74 increasing the likelihood that an area to had no community- infected cases. This could be related to the over-representation of the aged 75+ and under- representation of the 50-74 age group in the SARS cases compared to the general population (Leung, Hedley et al. 2004).

Given the driving role of hospital transmission in the overall spatial spread of the epidemic, I examined the spatial distribution of the three main types of hospital infections – healthcare workers, inpatients and visitors. Considerable variation was observed in the spatial distribution of the three main outbreak hospitals cases, and in the types of people infected in individual hospitals (i.e. healthcare workers, in-patients, or visitors). The latter is likely to have been related to the implementation of infection control measures in the different hospitals and the number of SARS patients treated. For example, the TPH outbreak was sourced from the transfer of 14 patients who were considered as not infected with SARS from AHN for convalescence (Severe Acute Respiratory Syndrome Expert Committee 2003), and thus seeded a high proportion of inpatients.

The control policy implications from the study are that areas near hospitals that treat SARS patients are of higher risk of forming a disease cluster and inducing further community transmission, and the final size of the community outbreak in the area is related to the size of the nearest hospital outbreak. With the observation that key hospital spatial clusters lead the start of community clusters in a short period of time, close monitoring of the emergence of hospital clusters may provide opportunities to prevent further community transmission and formation of community clusters. Therefore, timely and localised infection control measures in these areas could help to maximize the benefit and cost-effectiveness of the control measures.

Timely implementation of localized control measures partly relies on real-time detection of the emerging spatial disease clusters during the epidemic. My work shows that real-time application of the Kulldorff’s prospective scan statistic method would have been effective in identifying some real- time clusters accurately, and would have detected the Amoy Garden cluster earlier than in reality by the health officials during the outbreak, and before peak transmission was reached, though the result was masked by a few early non-Amoy cases.

However, because the method is conditional on the total number of cases in the data, at the initial period of the epidemic when the number of total cases was small, it tended to detect clusters as a group of TPUs where the cases where concentrated. As more cases were records later and with the spatial distribution changed over time, these clusters were no longer significant. Because the total number of cases in the complete dataset was much greater, the retrospective scan did not pick up these TPUs as clusters for that time period. In the public health perspective, when these clusters are detected during the epidemic, it is worth putting them under observation, because it is uncertain if these clusters would grow further. Given that the prospective scan can be run regularly upon new data arrival, any existing cluster which does not grow further will be indicated as insignificant cluster in the later prospective scans.

There were a few TPUs which did not have any SARS cases but detected as part of the retrospective clusters. This justified the much smaller maximum circle size settings applied in the prospective and retrospective scans than the default settings. If the default settings are applied, it is believed more false TPUs would have been included as part of the cluster. The retrospective and prospective scan results verified that using smaller maximum circle in space and time can still detect clusters of large size and long duration as more than one significant cluster. On the contrary, a study on the spatial patterns of Hong Kong H1N1 pandemic that used the default settings of 50% spatial and temporal window as the maximum resulted in some relatively big clusters (Lee and Wong 2010), though it is not directly comparable to my study.

So far my work has focused on cluster formation and the interactions between hospital and community transmission. Between-areas spread and its dynamics are best studied using contact tracing data or methods of infection tree inference. Complete contact tracing data is scarcely available for any large-scale emerging outbreak, and this is the case of the SARS outbreak in Hong Kong in 2003. In chapter 6, I use an infection tree inference method derived by extending the Wallinga and Teunis method spatially to derive the between areas and within area spread of SARS in Hong Kong.

87

Related documents