Methods for Detecting Reportable Disease Clusters in Space and Time

Key Objectives:

• Describe methods currently used by public health departments for reportable disease cluster detection.
• Identify strengths and limitations of implementing different approaches.
• Compile best practices for accounting for common data and analytic challenges.

Brief Summary:

Detecting disease clusters in near real-time can guide investigations and lead to timely public health interventions. However, health departments face common practical challenges to cluster detection. At this roundtable discussion, presenting authors from two local health departments will provide an overview of the methods they use to detect clusters of reportable communicable diseases in space and time. These methods include a version of the Historical Limits Method modified to account for biases¹ and the prospective space-time permutation scan statistic in SaTScan.² Real-life examples of occasions when these methods successfully detected true clusters and informed public health investigations will be highlighted. 

Attendees will have the opportunity to discuss how they may have approached decisions in implementing cluster detection analyses, including: Which reportable diseases are considered to be worthwhile to analyze for near real-time cluster detection? How do we account for reported cases that are still pending investigation or confirmation? How do we account for lags in data accrual? What is the time period of interest for recent clusters? What is the baseline time period for comparison? How do we account for secular trends or past clusters in the baseline data? How do we account for discontinuities in the baseline, e.g., changes in surveillance case definitions? How should data be aggregated geographically for analysis? How frequently should we conduct analyses? What statistical test(s) should be used to determine if case counts are greater than would be expected by chance alone? How do we succinctly present cluster summary information to our colleagues to guide investigations? To what degree can analyses be automated rather than run manually? 

Through discussion, participants will learn about the strengths and limitations of various cluster detection methods currently in use. Attendees might identify colleagues with sharable resources (e.g., SAS code, parameter settings) that could be useful for enhancing the cluster detection processes in their own jurisdiction.

References:

1. Levin-Rector A, Wilson EL, Fine AD, Greene SK. Refining historical limits method to improve disease cluster detection, New York City, New York, USA. Emerging Infectious Diseases, 2015; 21:265-272. 

2. Kulldorff M, Heffernan R, Hartman J, Assunção RM, Mostashari F. A space-time permutation scan statistic for the early detection of disease outbreaks. PLoS Medicine, 2005; 2:216-224.