An evaluation of SaTScan for the prospective detection of space-time Campylobacter clusters in the North East of England

Spatiotemporal Cluster Detection for COVID-19 Outbreak Surveillance: Descriptive Analysis Study

BACKGROUND

During the peak of the winter 2020-2021 surge, the number of weekly reported COVID-19 outbreaks in Washington State was 231; the majority occurred in high-priority settings such as workplaces, community settings, and schools. The Washington State Department of Health used automated address matching to identify clusters at health care facilities. No other systematic, statewide outbreak detection methods were in place. This was a gap given the high volume of cases, which delayed investigations and decreased data completeness, potentially leading to undetected outbreaks. We initiated statewide cluster detection using SaTScan, implementing a space-time permutation model to identify COVID-19 clusters for investigation.

OBJECTIVE

To improve outbreak detection, the Washington State Department of Health initiated a systematic cluster detection model to identify timely and actionable COVID-19 clusters for local health jurisdiction (LHJ) investigation and resource prioritization. This report details the model's implementation and the assessment of the tool's effectiveness.

METHODS

In total, 6 LHJs participated in a pilot to test model parameters including analysis type, geographic aggregation, cluster radius, and data lag. Parameters were determined through heuristic criteria to detect clusters early when they are smaller, making interventions more feasible. This study reviews all clusters detected after statewide implementation from July 17 to December 17, 2021. The clusters were analyzed by LHJ population and disease incidence. Clusters were compared with reported outbreaks.

RESULTS

A weekly, LHJ-specific retrospective space-time permutation model identified 2874 new clusters during this period. While the weekly analysis included case data from the prior 3 weeks, 58.25% (n=1674) of all clusters identified were timely-having occurred within 1 week of the analysis and early enough for intervention to prevent further transmission. There were 2874 reported outbreaks during this same period. Of those, 363 (12.63%) matched to at least one SaTScan cluster. The most frequent settings among reported and matched outbreaks were schools and youth programs (n=825, 28.71% and n=108, 29.8%), workplaces (n=617, 21.46% and n=56, 15%), and long-term care facilities (n=541, 18.82% and n=99, 27.3%). Settings with the highest percentage of clusters that matched outbreaks were community settings (16/72, 22%) and congregate housing (44/212, 20.8%). The model identified approximately one-third (119/363, 32.8%) of matched outbreaks before cases were associated with the outbreak event in our surveillance system.

CONCLUSIONS

Our goal was to routinely and systematically identify timely and actionable COVID-19 clusters statewide. Regardless of population or incidence, the model identified reasonably sized, timely clusters statewide, meeting the objective. Among some high-priority settings subject to public health interventions throughout the pandemic, such as schools and community settings, the model identified clusters that were matched to reported outbreaks. In workplaces, another high-priority setting, results suggest the model might be able to identify outbreaks sooner than existing outbreak detection methods.

A machine learning approach for rapid early detection of Campylobacter spp. using absorbance spectra collected from enrichment cultures

Enumeration of Campylobacter from environmental waters can be difficult due to its low concentrations, which can still pose a significant health risk. Spectrophotometry is an approach commonly used for fast detection of water-borne pollutants in water samples, but it has not been used for pathogen detection, which is commonly done through a laborious and time-consuming culture or qPCR Most Probable Number enumeration methods (i.e., MPN-PCR approaches). In this study, we proposed a new method, MPN-Spectro-ML, that can provide rapid evidence of Campylobacter detection and, hence, water concentrations. After an initial incubation, the samples were analysed using a spectrophotometer, and the spectrum data were used to train three machine learning (ML) models (i.e., supported vector machine - SVM, logistic regression-LR, and random forest-RF). The trained models were used to predict the presence of Campylobacter in the enriched water samples and estimate the most probable number (MPN). Over 100 stormwater, river, and creek samples (including both fresh and brackish water) from rural and urban catchments were collected to test the accuracy of the MPN-Spectro-ML method under various scenarios and compared to a previously standardised MPN-PCR method. Differences in the spectrum were found between positive and negative control samples, with two distinctive absorbance peaks between 540-542nm and 575-576nm for positive samples. Further, the three ML models had similar performance irrespective of the scenario tested with average prediction accuracy (ACC) and false negative rates at 0.763 and 13.8%, respectively. However, the predicted MPN of Campylobacter from the new method varied from the traditional MPN-PCR method, with a maximum Nash-Sutcliffe coefficient of 0.44 for the urban catchment dataset. Nevertheless, the MPN values based on these two methods were still comparable, considering the confidence intervals and large uncertainties associated with MPN estimation. The study reveals the potential of this novel approach for providing interim evidence of the presence and levels of Campylobacter within environmental water bodies. This, in turn, decreases the time from risk detection to management for the benefit of public health.

Prospective Spatiotemporal Cluster Detection Using SaTScan: Tutorial for Designing and Fine-Tuning a System to Detect Reportable Communicable Disease Outbreaks

Staff at public health departments have few training materials to learn how to design and fine-tune systems to quickly detect acute, localized, community-acquired outbreaks of infectious diseases. Since 2014, the Bureau of Communicable Disease at the New York City Department of Health and Mental Hygiene has analyzed reportable communicable diseases daily using SaTScan. SaTScan is a free software that analyzes data using scan statistics, which can detect increasing disease activity without a priori specification of temporal period, geographic location, or size. The Bureau of Communicable Disease's systems have quickly detected outbreaks of salmonellosis, legionellosis, shigellosis, and COVID-19. This tutorial details system design considerations, including geographic and temporal data aggregation, study period length, inclusion criteria, whether to account for population size, network location file setup to account for natural boundaries, probability model (eg, space-time permutation), day-of-week effects, minimum and maximum spatial and temporal cluster sizes, secondary cluster reporting criteria, signaling criteria, and distinguishing new clusters versus ongoing clusters with additional events. We illustrate how to support health equity by minimizing analytic exclusions of patients with reportable diseases (eg, persons experiencing homelessness who are unsheltered) and accounting for purely spatial patterns, such as adjusting nonparametrically for areas with lower access to care and testing for reportable diseases. We describe how to fine-tune the system when the detected clusters are too large to be of interest or when signals of clusters are delayed, missed, too numerous, or false. We demonstrate low-code techniques for automating analyses and interpreting results through built-in features on the user interface (eg, patient line lists, temporal graphs, and dynamic maps), which became newly available with the July 2022 release of SaTScan version 10.1. This tutorial is the first comprehensive resource for health department staff to design and maintain a reportable communicable disease outbreak detection system using SaTScan to catalyze field investigations as well as develop intuition for interpreting results and fine-tuning the system. While our practical experience is limited to monitoring certain reportable diseases in a dense, urban area, we believe that most recommendations are generalizable to other jurisdictions in the United States and internationally. Additional analytic technical support for detecting outbreaks would benefit state, tribal, local, and territorial public health departments and the populations they serve.

Detecting high-risk neighborhoods and socioeconomic determinants for common oral diseases in Germany

BACKGROUND

Ideally, health services and interventions to improve dental health should be tailored to local target populations. But this is not the standard. Little is known about risk clusters in dental health care and their evaluation based on small-scale, spatial data, particularly among under-represented groups in health surveys. Our study aims to investigate the incidence rates of major oral diseases among privately insured and self-paying individuals in Germany, explore the spatial clustering of these diseases, and evaluate the influence of social determinants on oral disease risk clusters using advanced data analysis techniques, i.e. machine learning.

METHODS

A retrospective cohort study was performed to calculate the age- and sex-standardized incidence rate of oral diseases in a study population of privately insured and self-pay patients in Germany who received dental treatment between 2016 and 2021. This was based on anonymized claims data from BFS health finance, Bertelsmann, Dortmund, Germany. The disease history of individuals was recorded and aggregated at the ZIP code 5 level (n = 8871).

RESULTS

Statistically significant, spatially compact clusters and relative risks (RR) of incidence rates were identified. By linking disease and socioeconomic databases on the ZIP-5 level, local risk models for each disease were estimated based on spatial-neighborhood variables using different machine learning models. We found that dental diseases were spatially clustered among privately insured and self-payer patients in Germany. Incidence rates within clusters were significantly elevated compared to incidence rates outside clusters. The relative risks (RR) for a new dental disease in primary risk clusters were min = 1.3 (irreversible pulpitis; 95%-CI = 1.3-1.3) and max = 2.7 (periodontitis; 95%-CI = 2.6-2.8), depending on the disease. Despite some similarity in the importance of variables from machine learning models across different clusters, each cluster is unique and must be treated as such when addressing oral public health threats.

CONCLUSIONS

Our study analyzed the incidence of major oral diseases in Germany and employed spatial methods to identify and characterize high-risk clusters for targeted interventions. We found that private claims data, combined with a network-based, data-driven approach, can effectively pinpoint areas and factors relevant to oral healthcare, including socioeconomic determinants like income and occupational status. The methodology presented here enables the identification of disease clusters of greatest demand, which would allow implementing more targeted approaches and improve access to quality care where they can have the most impact.

Epidemiological Characteristics and Spatio-Temporal Distribution of Hepatitis A in Spain in the Context of the 2016/2017 European Outbreak

The aim of our study was to describe the results of the epidemiological surveillance of hepatitis A infections in Spain in the context of the 2016/2017 European outbreak, particularly of hepatitis A outbreaks reported in the MSM population, incorporating the results of a spatio-temporal analysis of cases. Hepatitis A cases and outbreaks reported in 2016-2017 to the National Epidemiological Surveillance Network were reviewed: outbreaks in which some of the cases belonged to the MSM group were described, and clusters of hepatitis A cases in men and women were analysed using a space-time scan statistic. Twenty-six outbreaks were identified, with a median size of two cases per outbreak, with most of the outbreak-related cases belonging to the 15-44 years-old group. Nearly 85% occurred in a household setting, and in all outbreaks, the mode of transmission was direct person-to-person contact. Regarding space-time analysis, twenty statistically significant clusters were identified in the male population and eight in the female population; clusters in men presented a higher number of observed cases and affected municipalities, as well as a higher percentage of municipalities classified as large urban areas. The elevated number of cases detected in clusters of men indicates that the number of MSM-related outbreaks may be higher than reported, showing that spatio-temporal analysis is a complementary, useful tool which may improve the detection of outbreaks in settings where epidemiological investigation may be more challenging.

Spatiotemporal clusters and the socioeconomic determinants of COVID-19 in Toronto neighbourhoods, Canada

The aim of this study is to identify spatiotemporal clusters and the socioeconomic drivers of COVID-19 in Toronto. Geographical, epidemiological, and socioeconomic data from the 140 neighbourhoods in Toronto were used in this study. We used local and global Moran's I, and space-time scan statistic to identify spatial and spatiotemporal clusters of COVID-19. We also used global (spatial regression models), and local geographically weighted regression (GWR) and Multiscale Geographically weighted regression (MGWR) models to identify the globally and locally varying socioeconomic drivers of COVID-19. The global regression model identified a lower percentage of educated people and a higher percentage of immigrants in the neighbourhoods as significant predictors of COVID-19. MGWR shows the best fit model to explain the variables affecting COVID-19. The findings imply that a single intervention package for the entire area would not be an effective strategy for controlling COVID-19; a locally adaptable intervention package would be beneficial.

Detection of COVID-19 case clusters in Québec, May-October 2020

OBJECTIVES

The Quebec Public Health Institute (INSPQ) was mandated to develop an automated tool for detecting space-time COVID-19 case clusters to assist regional public health authorities in identifying situations that require public health interventions. This article aims to describe the methodology used and to document the main outcomes achieved.

METHODS

New COVID-19 cases are supplied by the "Trajectoire de santé publique" information system, geolocated to civic addresses and then aggregated by day and dissemination area. To target community-level clusters, cases identified as residents of congregate living settings are excluded from the cluster detection analysis. Detection is performed using the space-time scan statistic and Poisson statistical model, and implemented in the SaTScan software. Information on detected clusters is disseminated daily via an online interactive mapping interface.

RESULTS

The number of clusters detected tracked with the number of new cases. Slightly more than 4900 statistically significant (p ≤ 0.01) space-time clusters were detected over 14 health regions from May to October 2020. The Montréal region was the most affected.

CONCLUSION

Considering the objective of timely cluster detection, the use of near-real-time health surveillance data of varying quality over time and by region constitutes an acceptable compromise between timeliness and data quality. This tool serves to supplement the epidemiologic investigations carried out by regional public health authorities for purposes of COVID-19 management and prevention.

Simulation of Legionnaires' disease prospective spatiotemporal cluster detection, Allegheny County, Pennsylvania, USA

Legionnaires' disease (LD) incidence in the USA has quadrupled since 2000. Health departments must detect LD outbreaks quickly to identify and remediate sources. We tested the performance of a system to prospectively detect simulated LD outbreaks in Allegheny County, Pennsylvania, USA. We generated three simulated LD outbreaks based on published outbreaks. After verifying no significant clusters existed in surveillance data during 2014-2016, we embedded simulated outbreak-associated cases into 2016, assigning simulated residences and report dates. We mimicked daily analyses in 2016 using the prospective space-time permutation scan statistic to detect clusters of ⩽30 and ⩽180 days using 365-day and 730-day baseline periods, respectively. We used recurrence interval (RI) thresholds of ⩾20, ⩾100 and ⩾365 days to define significant signals. We calculated sensitivity, specificity and positive and negative predictive values for daily analyses, separately for each embedded outbreak. Two large, simulated cooling tower-associated outbreaks were detected. As the RI threshold was increased, sensitivity and negative predictive value decreased, while positive predictive value and specificity increased. A small, simulated potable water-associated outbreak was not detected. Use of a RI threshold of ⩾100 days minimised time-to-detection while maximizing positive predictive value. Health departments should consider using this system to detect community-acquired LD outbreaks.

Daily Reportable Disease Spatiotemporal Cluster Detection, New York City, New York, USA, 2014-2015

Each day, the New York City Department of Health and Mental Hygiene uses the free SaTScan software to apply prospective space–time permutation scan statistics to strengthen early outbreak detection for 35 reportable diseases. This method prompted early detection of outbreaks of community-acquired legionellosis and shigellosis.

Online platform for applying space-time scan statistics for prospectively detecting emerging hot spots of dengue fever

Background

Cases of dengue fever have increased in areas of Southeast Asia in recent years. Taiwan hit a record-high 42,856 cases in 2015, with the majority in southern Tainan and Kaohsiung Cities. Leveraging spatial statistics and geo-visualization techniques, we aim to design an online analytical tool for local public health workers to prospectively identify ongoing hot spots of dengue fever weekly at the village level.

Methods

A total of 57,516 confirmed cases of dengue fever in 2014 and 2015 were obtained from the Taiwan Centers for Disease Control (TCDC). Incorporating demographic information as covariates with cumulative cases (365 days) in a discrete Poisson model, we iteratively applied space–time scan statistics by SaTScan software to detect the currently active cluster of dengue fever (reported as relative risk) in each village of Tainan and Kaohsiung every week. A village with a relative risk >1 and p value <0.05 was identified as a dengue-epidemic area. Assuming an ongoing transmission might continuously spread for two consecutive weeks, we estimated the sensitivity and specificity for detecting outbreaks by comparing the scan-based classification (dengue-epidemic vs. dengue-free village) with the true cumulative case numbers from the TCDC’s surveillance statistics.

Results

Among the 1648 villages in Tainan and Kaohsiung, the overall sensitivity for detecting outbreaks increases as case numbers grow in a total of 92 weekly simulations. The specificity for detecting outbreaks behaves inversely, compared to the sensitivity. On average, the mean sensitivity and specificity of 2-week hot spot detection were 0.615 and 0.891 respectively (p value <0.001) for the covariate adjustment model, as the maximum spatial and temporal windows were specified as 50% of the total population at risk and 28 days. Dengue-epidemic villages were visualized and explored in an interactive map.

Conclusions

We designed an online analytical tool for front-line public health workers to prospectively detect ongoing dengue fever transmission on a weekly basis at the village level by using the routine surveillance data.

Presence and survival of culturable Campylobacter spp. and Escherichia coli in a temperate urban estuary

Urban estuaries throughout the world typically contain elevated levels of faecal contamination, the extent of which is generally assessed using faecal indicator organisms (FIO) such as Escherichia coli. This study assesses whether the bacterial FIO, E. coli is a suitable surrogate for Campylobacter spp., in estuaries. The presence and survival dynamics of culturable E. coli and Campylobacter spp. are compared in the water column, bank sediments and bed sediments of the Yarra River estuary (located in Melbourne, Australia). The presence of E. coli did not necessarily indicate detectable levels of Campylobacter spp. in the water column, bed and bank sediments, but the inactivation rates of the two bacteria were similar in the water column. A key finding of the study is that E. coli and Campylobacter spp. can survive for up to 14days in the water column and up to 21days in the bed and bank sediments of the estuary. Preliminary data presented in this study also suggests that the inactivation rates of the two bacteria may be similar in bed and bank sediments. This undermines previous hypotheses that Campylobacter spp. cannot survive outside of its host and indicates that public health risks can persist in aquatic systems for up to three weeks after the initial contamination event.

Environmental monitoring of waterborne Campylobacter: evaluation of the Australian standard and a hybrid extraction-free MPN-PCR method

Campylobacter is the leading agent of diarrheal disease worldwide. This study evaluates a novel culture-PCR hybrid (MPN-PCR) assay for the rapid enumeration of Campylobacter spp. from estuarine and wastewater systems. To first evaluate the current, culture-based, Australian standard, an inter-laboratory study was conducted on 69 subsampled water samples. The proposed Most-Probable Number (MPN)-PCR method was then evaluated, by analysing 147 estuarine samples collected over a 2 year period. Data for 14 different biological, hydrological and climatic parameters were also collated to identify pathogen-environment relationships and assess the potential for method specific bias. The results demonstrated that the intra-laboratory performance of the MPN-PCR was superior to that of AS/NZS (σ = 0.7912, P < 0.001; κ = 0.701, P < 0.001) with an overall diagnostic accuracy of ~94%. Furthermore, the analysis of both MPN-PCR and AS/NZS identified the potential for the introduction of method specific bias during assessment of the effects of environmental parameters on Campylobacter spp. numbers.

A critical look at prospective surveillance using a scan statistic

The scan statistic is a very popular surveillance technique for purely spatial, purely temporal, and spatial-temporal disease data. It was extended to the prospective surveillance case, and it has been applied quite extensively in this situation. When the usual signal rules, as those implemented in SaTScan(TM) (Boston, MA, USA) software, are used, we show that the scan statistic method is not appropriate for the prospective case. The reason is that it does not adjust properly for the sequential and repeated tests carried out during the surveillance. We demonstrate that the nominal significance level α is not meaningful and there is no relationship between α and the recurrence interval or the average run length (ARL). In some cases, the ARL may be equal to ∞, which makes the method ineffective. This lack of control of the type-I error probability and of the ARL leads us to strongly oppose the use of the scan statistic with the usual signal rules in the prospective context.