J. O. Jansen1, J. J. Morrison2, T. Cornulier3 1University Of Alabama At Birmingham,Division Of Acute Care Surgery,Birmingham, AL, USA 2University Of Maryland,R Adams Cowley Shock Trauma Center,Baltimore, MD, USA 3University Of Aberdeen,School Of Biological Sciences,Aberdeen, SCOTLAND, United Kingdom
Introduction:
The Centers for Disease Control and the World Health Organisation have promoted the concept of “injury surveillance”, to inform the provision of services. Such analyses tend to rely either on the evaluation of temporal trends or of geographical variations in case volume, both having important implications for trauma system configuration. However, spatial variation in these temporal trends (or changes in these distributions) are more difficult to estimate particularly in sparsely populated areas, and have received relatively little attention as a consequence. The aim of this study was to propose a model to facilitate the spatio-temporal surveillance of injuries, using Scotland as a case study.
Methods:
This is a retrospective analysis of five years’ of trauma incident location data, as collected routinely by the Scottish Ambulance Service, for incidents attended from 2009 to 2013. The source data was geocoded by postcode district (PCD), a medium-sized spatial unit. There are 444 PCDs in Scotland. We analysed the study population as a whole, as well as a number of predefined subgroups, such as those with abnormal physiological signs. Our analysis aimed at characterising the geographical distribution of expected incident numbers and identifying spatial variation in their temporal trends. In order to leverage sufficient statistical power to detect temporal trends in rare events over short time periods and small spatial units, we used a geographically weighted regression model, which assumed a Poisson distribution for the counts of incidents per PCD and per year, and used a Markov random field to condition estimates for each PCD on those from adjacent PCDs. The results are displayed as choropleth maps, showing percentage change per year, with hatched areas indicating statistically significant changes over 5 years.
Results:
There were 509,725 incidents. Overall, there were increases in case volume in the Glasgow area, the central Southern part of the country, the Northern parts of the Highlands, the North-East, and the Orkney and Shetland Islands. Statistically significant increases were largely restricted to major cities, with the notable exception of Edinburgh. Significant decreases in the number of incidents were seen in Western Scotland, Fife and Lothian, and the Borders. Subgroup analyses showed markedly different spatio-temporal patterns.
Conclusion:
This project has demonstrated the feasibility of population-based spatio-temporal injury surveillance. Even over a relatively short period, the geographical distribution of where injuries occur may change, and different injuries present different spatio-temporal patterns. These findings have potential implications for health policy and service delivery.
