Use of graph theory to identify patterns of deprivation and high morbidity and mortality
Use of Graph Theory to Identify Patterns of Deprivation and High Morbidity and Mortality in Public Health Data Sets. JAMIA. 2005; 12(6): 630-641
Introduction The purpose of this paper was to see if graph algorithms could be adapted to finding regions of high morbidity and deprivation that are adjacent and run along significant geological features. Prior work has identified clusters of such regions that were adjacent sites with high health risks; e.g. landfills and nuclear powerstations. Other features such as roads and rivers represent more complex patterns which could also be a source of pollutants or contamination. Once identified, these regions could then be allocated more health resources to improve the quality of life more efficiently.
Methods A graph was constructed using the Enumeration Districts (ED) from the Trent region in England. ED refers to an area assigned to one census taker and was chosen because it represented the lowest level of census data. A search was then performed over this graph to search for a line of fives EDs exhibiting high levels of deprivation. Deprivation, was defined as a collection of variables representing unemployment, overcrowding, lack of owner-occupied accomodation, lack of car ownership, etc. The first search produced 3,484 patterns which included 195 EDs. To trim this down, patterns in an alternative set that represented five EDs in a cycle were removed from the solution. The MCS algorith, used in the RASCAL program was used to match up patterns between the search with the original pattern and the searches with the alternative patterns. Two methods were used to remove EDs, one to remove all patterns identified in the alternative set, and another to remove all EDs in the alternative set. The initial search was reduced to 1,704 patterns with 161 EDs using the former option and was further culled down to 12 EDs using the latter option.
Discussion The application was able to identify a small set of regions which could be scrutinized further by a human researcher. There were some complications with the method however, when sets of areas with high deprivation overlapped. This makes it difficult to distinguish between chains and clusters of areas with poor public health. Nevertheless, the algorithm does provide a method of searching for complex patterns of geographic features that are detrimental to public health. Once identified, these high-risk regions could be allocated more resources and education to improve the quality of life for the general public.
Comments This article piqued my interest somewhat because it tries to identify health risks at a high level. Such information could be useful for both small clinics and large hospitals alike if they were able to stock up on the more relevant vaccines and medicines. Also, such information could be used to raise the awareness of potential health hazards at the community level. If an interoperable health database were developed at the nation, or even the state level, this data would become a byproduct of the information system and could then be used in such a fashion