Evaluating the Capacity of New York City's (NYC) Syndromic Surveillance System to Track Injuries in Real-Time

Seil, Kacie

BACKGROUND: Surveillance for injuries uses administrative data systems with an inherent prolonged lag-time in data delivery. Syndromic surveillance systems (SSS) provide timely data to identify geographic and temporal clusters to inform rapid response. Although existing SSS are used for communicable disease detection, there is limited experience with monitoring of non-communicable disease. NYC’s SSS archives ED visit data from 49 of 52 qualifying NYC hospitals daily. We aim to evaluate SSS’s capacity to track injuries in near real-time.

METHODS: Drawing on conventional injury outcome and mechanism language as well as common misspellings, first we developed eight injury syndromes reflecting NYC’s injury prevention priorities. Syndromes were based on injury outcomes [laceration, fracture, and traumatic brain injury (TBI)] and mechanisms (drowning, traffic, electrocution, gunshot, and stabbing). Using the SAS index function, we then scanned free text of the SSS chief complaint field for the language constituting each injury syndrome. We computed annual and monthly injury totals from 2008–2010 for each syndrome. For the same years, we computed annual and monthly totals from comparable injury indicators in NYC’s administrative ED data from the Statewide Planning and Research Cooperative System (SPARCS). Injury visits in ED SPARCS were identified using International Classification of Diseases, 9^th revision, Clinical Modification coding system. Then, we graphed monthly volume from both systems to examine concordance in increases and decreases over time. Finally, we assessed the validity of the developed injury syndromes by reviewing full text of SSS records for all cases with any mention of syndrome terms. Syndrome specificity was calculated for each syndrome; sensitivity analysis is underway.

RESULTS: Overall, the SSS identified 19% more TBI visits than ED SPARCS. For all other injuries, SSS identified fewer injury visits than ED SPARCS (8% fewer gunshot injuries; 14% fewer traffic injuries; 15% fewer electrocutions; 20% fewer lacerations; 45% fewer stabbing injuries; 70% fewer drowning injuries; 87% fewer fractures). Despite differences in volume generated by the two systems, monthly patterns tracked in parallel. Greatest concordance was seen with lacerations, electrocution, gunshot, stabbing, and traffic injuries. Specificity values exceeded 75%: gunshot (97%), stab (97%), TBI (97%), fracture (91%), traffic (89%), electrocution (86%), drowning (84%), and laceration (77%).

CONCLUSIONS: While SSS may not accurately assess volume, patterns in injury-related ED visits reliably track ED SPARCS data. This evaluation demonstrates the potential utility of SSS for tracking injury in near real-time, ameliorating the lag in data delivery from usual surveillance systems.