Interrupted time series segmented regression analysis for detecting waterborne disease outbreaks by syndromic surveillance

Aidan Yuen, Davoud Pourmarzi, Suzie Sarkis, Carmela Luisetto, Kamal Khatri, Angie Bone, Jim Black

    Research output: Contribution to journalArticlepeer-review

    Abstract

    Introduction: Pathogens can enter the drinking water supply and cause gastroenteritis outbreaks. Such events can affect many people in a short time, making them a high risk for public health. In Australia, the Victoria State Government Department of Health is deploying a syndromic surveillance system for drinking water contamination events. We assessed the utility of segmented regression models for detecting such events and determined the number of excess presentations needed for such methods to signal a detection. Methods: The study involved an interrupted time series study of a past lapse in water treatment. The baseline period comprised the four weeks before the minimum incubation period of suspected pathogens, set at two days post-event. The surveillance period comprised the week after. We used segmented linear regression to compare the count of gastroenteritis presentations to public hospital emergency departments (EDs) between the surveillance and baseline periods. We then simulated events resulting in varying excess presentations. These were superimposed onto the ED data over fifty different dates across 2020. Using the same regression, we calculated the detection probability at p < 0.05 for each outbreak size. Results: In the retrospective analysis, there was strong evidence for an increase in presentations shortly after the event. In the simulations, with no excess presentations (i.e., with the ED data as is) the models signalled 8% probability of detection. The models returned 50% probability of detection with 28 excess presentations and 100% probability of detection with 78 excess presentations. Conclusions: The transient increase in presentations after the event may be attributed to microbiological hazards or increased health-seeking behaviour following the issuing of boil water advisories. The simulations demonstrated the ability for segmented regressions to signal a detection, even without a large excess in presentations. The approach also demonstrated high specificity and should be considered for informing Victoria's syndromic surveillance system.

    Original languageEnglish
    JournalCommunicable diseases intelligence (2018)
    Volume47
    DOIs
    Publication statusPublished - 28 Feb 2023

    Fingerprint

    Dive into the research topics of 'Interrupted time series segmented regression analysis for detecting waterborne disease outbreaks by syndromic surveillance'. Together they form a unique fingerprint.

    Cite this