An evaluation of an automated hospital outbreak detection system (WHONET-SaTScan) versus standard outbreak detection approach

Background:

A recent survey on current hospital outbreak detection systems (ODS) show hospitals rely on a manual review of empirical rules of no more than 9 pre-identified organisms to identify potential outbreaks.¹ We use a novel software package, WHONET-SaTScan (WS), to facilitate an automated outbreak detection system via space/time to improve standard outbreak detection approach (SODA).

Methods:

We established WS along with a database to detect, store, link electronic health record (EHR) data and record results from investigations/interventions of clusters to create an automated ODS.  We conducted surveillance for 490 bacterial/fungal organisms along with antibiotic resistance phenotypes for 11 organisms.  For patients identified as part of a cluster, the following information was extracted from the EHR: possible environmental organism; shared unit, room, medical providers and antibiotic susceptibility patterns.  These data were used to stratify clusters into 2 risk groups (high vs low). Using the CDC’s guidelines for evaluating a surveillance system, we assessed system attributes (i.e. simplicity, flexibility, sensitivity, timeliness and usefulness).² 

Results:

During an 8 month time period (Sep 1, 2015 – Apr 30, 2016), 146 clusters were detected, 39 of which were considered high risk clusters.  Of these 39 clusters, 16 organisms were identified, with S. aureus (21%), S. maltophilia (15%) and K. peumoniae (13%) representing the majority (Fig 1).  Three transmission events were determined to have likely occurred based on demographic factors and review of disinfection practices.  Only 1 of these transmission events was identified through an astute clinician.  WS did not miss any clusters identified by SODA. The median number of days for WS to detect a cluster was 16.

Conclusion:

WS is a relatively simple ODS that offers flexibility in creating the parameters (e.g. outbreak window, significance threshold, baseline data) and can identify clusters in space/time.  The output is simple to navigate; however, the process to implement software and link with EHR data required additional IT skills. The system has high sensitivity, albeit low specificity. The median time to detection was 16 days; however, many clusters were detected within 3 days.  WS is a useful addition to a robust infection control program.