Detecting space-time disease clusters with arbitrary shapes and sizes using a co-clustering approach

Ability to detect potential space-time clusters in spatio-temporal data on disease occurrences is necessary for conducting surveillance and implementing disease prevention policies. Most existing techniques use geometrically shaped (circular, elliptical or square) scanning windows to discover diseas...

Full description

Saved in:
Bibliographic Details
Main Authors: Ullah, S., Daud, H., Dass, S.C., Khan, H.N., Khalil, A.
Format: Article
Published: Page Press Publications 2017
Online Access:https://www.scopus.com/inward/record.uri?eid=2-s2.0-85033239575&doi=10.4081%2fgh.2017.567&partnerID=40&md5=5948a5c8185bd6ddff13a31452df31ba
http://eprints.utp.edu.my/19820/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Ability to detect potential space-time clusters in spatio-temporal data on disease occurrences is necessary for conducting surveillance and implementing disease prevention policies. Most existing techniques use geometrically shaped (circular, elliptical or square) scanning windows to discover disease clusters. In certain situations, where the disease occurrences tend to cluster in very irregularly shaped areas, these algorithms are not feasible in practise for the detection of space-time clusters. To address this problem, a new algorithm is proposed, which uses a co-clustering strategy to detect prospective and retrospective space-time disease clusters with no restriction on shape and size. The proposed method detects space-time disease clusters by tracking the changes in space–time occurrence structure instead of an in-depth search over space. This method was utilised to detect potential clusters in the annual and monthly malaria data in Khyber Pakhtunkhwa Province, Pakistan from 2012 to 2016 visualising the results on a heat map. The results of the annual data analysis showed that the most likely hotspot emerged in three sub-regions in the years 2013-2014. The most likely hotspots in monthly data appeared in the month of July to October in each year and showed a strong periodic trend. © S. Ullah et al., 2017.