Clinical and commercial laboratories are required to report all severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) molecular test results (positive, negative, indeterminate) for New York state residents to the New York State Electronic Clinical Laboratory Reporting System (9). For NYC residents, this reporting system transmits reports to DOHMH. Laboratory reports include specimen collection date and patient demographics, including residential address, which we geocoded by census tract. Patient symptoms and illness onset date, if any, are obtained through interviews, although not all patients are interviewed.

To detect emerging clusters, the space-time scan statistic uses a cylinder in which the circular base covers a geographic area and the height corresponds to time (10). This cylinder is moved, or scanned, over space and time to cover different areas and periods. At each position, the number of cases inside the cylinder is compared with the expected count under the null hypothesis of no clusters by using a likelihood function, and the position with the maximum likelihood is the primary candidate for a cluster. The statistical significance of this cluster is then evaluated, adjusting for the multiple testing inherent in the many cylinder positions evaluated.

To quickly detect emerging hotspots, prospective analyses are conducted daily (11). To adjust for the multiple testing stemming from daily analyses, recurrence intervals are used instead of p values (12). A recurrence interval of D days means that under the null hypothesis, if we conduct the analysis repeatedly over D days, then the expected number of clusters of the same or larger magnitude is 1.

The space-time scan statistic can be used with different probability models; we used the Poisson model (10), adjusting not for population size (which would fail to account for changing testing rates) but rather for persons tested. Because the goal was to detect newly emerging hotspots rather than areas with consistently high percent positivity, we further adjusted analyses nonparametrically for purely geographic variations that were consistent over time. Fitting a log-linear function, we also adjusted for citywide temporal trends in percent positivity because the goal was to detect local hotspots rather than general citywide trends. For each day and location, the expected count was calculated as the number of persons tested × temporal trend function × a location-specific constant to ensure that, summed over all days in the study period, the location has the same number of observed and expected cases. To prioritize quickly emerging clusters to identify epidemiologic linkages, we used a short maximum temporal window of 7 days. To detect sustained clusters to inform place-based resource allocation, starting July 15, we also ran secondary analyses with a maximum temporal window of 21 days.

We developed SAS code (SAS Institute, https://www.sas.com; https://github.com/CityOfNewYork/communicable-disease-surveillance-nycdohmh) to generate daily input and parameter files (Table 1; Appendix Table). The SAS code then invoked SaTScan in batch mode, read analysis results back into SAS for further processing, output files to secured folders (including patient line lists with demographics and map and time-trend visualizations), and sent an investigator notification email.

Figure

Figure. Cluster case counts and severe acute respiratory syndrome coronavirus 2 test percent positivity inside and outside cluster area for selected clusters detected in New York City, NY, USA, 2020. A)...

We launched the system on June 11, 2020, and 2 clusters detected by July 31 prompted public health action (Table 2). First, on June 22, in the context of waning case counts citywide, the only cluster detected was of 6 patients residing in a 0.6-km radius, all with specimens collected on June 17 (Figure, panel A). Consequently, DOHMH staff interviewed patients to collect and compare potential common exposures, such as attending the same event or visiting the same location. On June 23, a DOHMH surveillance investigator (D.B.) determined that 2 patients had attended the same gathering, where recommended social distancing practices had not been observed. In response, DOHMH launched an effort to limit further transmission, including testing, contact tracing, community engagement, and health education emphasizing the importance of isolation and quarantine. No other epidemiologic linkages were identified after attempts to investigate ≈65 additional clusters detected through July 2020. Second, detection on July 15 of a sustained cluster (lasting >1 week) with a high percent positivity (Figure, panel B) contributed to geographically targeted testing, outreach, and education, as part of NYC’s hyper-local plan to prevent COVID-19 transmission (14).

COVID-19 community clusters detected by SaTScan prompted localized public education, testing, and community engagement (15). In addition, prioritizing interviews of patients in clusters can identify epidemiologic linkages and opportunities for interrupting further transmission, as is done for other reportable diseases (6–8). Identification of only 1 linkage in this study could be attributable to changing cluster investigation protocols, low patient response rates, or transmission occurring diffusely in small gatherings. Because all patients are referred for contact tracing, DOHMH discontinued reactively interviewing cluster patients for linkages and instead used clusters to proactively target resources.

The first limitation in this study was timeliness. Analyses were based on specimen collection date; however, given delays in testing availability and care seeking, these dates did not necessarily represent recent infections. Timeliness was further limited by delays from specimen collection to laboratory testing and reporting. Clusters dominated by asymptomatic patients or patients with illness onset >14 days before diagnosis may not require intervention because positive PCR results indicate presence of viral RNA but not necessarily viable virus. The second limitation involved the need to geocode for spatial precision. Of unique NYC residents for whom a specimen was collected for SARS-CoV-2 RNA PCR testing during June–July 2020, residential address was not geocodable for ≈3% of residents, so they were excluded. Third, although recurrence interval thresholds can be used to prioritize responding to clusters (6), COVID-19 cluster interpretation can be more complex. Other characteristics for prioritizing COVID-19 clusters, besides statistical significance, include percent positivity, relative risk, case count, epidemic curve trajectory, radius, demographics, and persistence. Prioritization can differ by response activity (e.g., establishing new testing sites, conducting outreach) and how quickly resources can be reallocated. Deciding when and where to initiate interventions in response to COVID-19 clusters cannot be fully automated and requires epidemiologic interpretation.

In summary, our COVID-19 early detection system highlighted areas warranting a rapid response. Targeted, place-based approaches for education and outreach efforts and for localized high transmission warnings could better protect persons at high risk for severe illness and death.