Spatio-Temporal Cluster Detection for Legionellosis using Multiple Patient Addresses


  • Eric R. Peterson Bureau of Communicable Disease, New York City Department of Health and Mental Hygiene, Queens, NY, USA
  • Sharon K. Greene Bureau of Communicable Disease, New York City Department of Health and Mental Hygiene, Queens, NY, USA



ObjectiveTo improve timeliness and sensitivity of legionellosis clusterdetection in New York City (NYC) by using all addresses availablefor each patient in one analysis.IntroductionThe Bureau of Communicable Disease (BCD) at the NYCDepartment of Health and Mental Hygiene performs daily automatedanalyses using SaTScan to detect spatio-temporal clusters for37 reportable diseases.1Initially, we analyzed one address per patient,prioritizing home address if available. On September 25, 2015, aBCD investigator noticed two legionellosis cases with similar workaddresses. A third case was identified in a nearby residential facility,and an investigation was initiated to identify a common exposuresource. Four days later, after additional cases living nearby werereported, the SaTScan analysis detected a corresponding cluster.In response to this signaling delay, we implemented a multiple address(MA) analysis to improve upon single address (SA) analyses by usingall location data available on possible exposure sites.2MethodsPositiveLegionellatest results for NYC residents are reported toBCD with patient demographic and address data. BCD interviews allcases to elicit additional locations of potential exposure and enters theaddresses into a disease surveillance database (Maven). Addressesare assigned X/Y coordinates in near real-time via integration with ageocoding webservice.We used the prospective space-time permutation scan statistic inSaTScan,3enabling the advanced input feature on the spatial neighborstab to “include location ID in the scanning window if at least one setof coordinates is included.” This option considered a case as includedin a given cluster ifanyof the case’s addresses were within the cluster.The case file included: unique case ID (as the location ID), number ofcases, onset date, and day of week. The coordinate file included: caseID and X/Y coordinates for each address per case, resulting in one ormore rows per case. We searched for alive clusters with a temporalrange of 2 to 30 days and a maximum spatial size of 50% of observedcases. The study period was 1 year. Monte Carlo simulations (N=999)were used to determine statistical significance.We mimicked prospective surveillance to determine when theSeptember 2015 cluster would have been detected had this analysisbeen in place, by performing daily SA and MA analyses fromSeptember 21 (when the first outbreak-linked case was reported)to September 29 (when the initial SaTScan analysis signaled). Anycluster with a recurrence interval (RI)≥100 days was summarized ina map and linelist. Prospective, automated analyses were launchedin April 2016 and run daily using Microsoft Task Scheduler, SAS9.4, and SaTScan 9.4.1. Signals through July 2016 were summarized.ResultsIn mimicked prospective analysis, the SA and MA SaTScananalyses identified clusters of 13 and 11 cases, respectively, startingSeptember 27, 2015. The MA cluster was more spatially focused(2.11 km vs. 5.42 km) and more unlikely to occur by chance alone(RI of 16,256 days vs. 8,758 days). In prospective analyses, a MAcluster of 6 cases was identified on July 5, 2016 with a radius of1.69 km (RI=100 days). On July 6, the MA cluster case countincreased to 7 and maintained the same radius (RI=685 days), whilea cluster of the same 7 cases was identified by the SA analysis witha larger radius (1.97 km) and lower RI (292 days). The RI for bothclusters peaked on July 7 (MA: 2348 days, SA: 713 days).ConclusionsIn preliminary evaluation, the MA analysis facilitated clusterdetection using non-residential possible exposure sites, such asworkplaces. Timeliness was slightly improved, but the larger practicalbenefit was identifying more spatially focused clusters. Smallerclusters are useful for more precisely targeting legionellosis infectionsource identification and remediation activities, especially in urbanenvironments with high population and building densities.




How to Cite

Peterson, E. R., & Greene, S. K. (2017). Spatio-Temporal Cluster Detection for Legionellosis using Multiple Patient Addresses. Online Journal of Public Health Informatics, 9(1).



Communicable Disease Surveillance Use Cases for Human, Animal, and Zoonotic Diseases