Table 3.

Summary of recommendations for assessing and fine-tuning a system to detect reportable communicable disease outbreaks using SaTScan.

Section and subtopic			Recommendations
System assessment
	Proof of concept	1. Identify at least 1 known historical acute, localized, community-acquired outbreak and mimic prospective surveillance around that period to assess the speed and precision with which the outbreak could have been detected.
	System performance	2. If the system consistently produces clusters that are too large or too delayed to be useful, misses signals, or produces too many false or weak signals, first examine data quality by inspecting input files and line list output files for data errors before fine-tuning the system.
System fine-tuning when outbreak detection is delayed or missed
	Difficulty detecting geographically small outbreaks or outbreaks at the boundaries of geographic units	3. Disaggregate data into smaller geographic units.
	Difficulty detecting outbreaks affecting a particular age group	4. Conduct additional age-restricted subgroup analyses, such as for children aged <5 years.
	Difficulty detecting outbreaks in which people are exposed far from their home	5. Collect and analyze multiple addresses per patient, such as both home and work.
	Difficulty detecting outbreaks because of missing data	6. Conduct quality assurance to detect and resolve drop-offs in laboratory reporting by laboratory and disease. 7. To avoid missing outbreaks affecting unsheltered patients experiencing homelessness and without a geocodable address, assign them to an artificial census tract.
	Difficulty detecting new outbreaks in areas with previous outbreaks	8. From the baseline period in the case file, exclude days during major previous outbreaks. 9. For previous outbreaks limited to a specific building or institution, exclude those events from the baseline period in the case file.
	Difficulty detecting outbreaks affecting people who reside in areas not connected in the network file	10. Add more connections to the network file.
	Difficulty detecting weak clusters that are potentially actionable outbreaks	11. If using an RIa threshold for signaling of ≥1 year, decrease the threshold.
System fine-tuning when consistently finding large, uninteresting clusters
	Clusters span hard-to-cross boundaries, such as lakes, rivers, and mountain ranges	12. Use the network input file.
	Clusters with a relative risk near 1 are of limited public health interest	13. Impose a minimum relative risk restriction.
System fine-tuning when there are too many signals
	Clusters are driven by duplicate events or incorrect patient addresses	14. Proactively check for and correct data errors.
	Clusters are driven by changing nature of input data, such as recent adoption of culture-independent diagnostic testing	15. Shorten the baseline study period.
	Clusters are driven by reports that do not represent true illnesses, such as reports of patients with only negative laboratory test results	16. Use stricter disease definition inclusion criteria and impose a time lag to allow for investigating and ruling out cases.
	Clusters are driven by within-household transmission, so they do not represent community transmission	17. Retain only 1 event per household.
	Clusters represent weak outbreaks that are not actionable given available resources	18. If using an RI threshold for signaling of <1 year, increase the threshold.

^aRI: recurrence interval.