Talk to the Veterans Crisis Line now
An official website of the United States government
DeLisle S (VA Maryland Health Care System; University of Maryland, Baltimore, MD), South BR
(University of Utah, Salt Lake City, UT), Smith G
(School of Veterinary Medicine, University of Pennsylvania, PA), Gaff H
(University of Maryland, Baltimore, MD), Samore M
(University of Utah, Salt Lake City, UT), Perl TM
(Johns Hopkins Medical Institutions, Baltimore, MD)
Objectives:
This works compares the delay and workload required to detect a community influenza epidemic using a CPRS-based automated surveillance system that targets either (1) all cases of acute respiratory infections (ARI), or (2) only those ARI cases that are febrile (F-ARI).
Methods:
Using an explicit definition of ARI and F-ARI, we reviewed 15,377 outpatient encounters at two VA systems. Found ARI and ILI cases served as a reference to develop case-detection algorithms (CDAs) that utilized combinations of structured EMR data and text analyses of clinical notes. We recreated historical background casecount time series by applying the most successful CDAs to historical CPRS data. We used mathematically modeled influenza epidemic to inject factitious influenza cases to CDA-specific backgrounds and recreated a prospective surveillance using a modified CUSUM statistic to detect the outbreak. The injection/50-day-prospective-surveillance cycle was repeated each week of the study year. To distinguish between true- and background-positive alarms, the daily statistics were performed on paired background+injection vs. background-only time series. We computed two benchmarks: 1) the average “Detection Delay,” from the time of each injection to the first true-positive alarm; and 2) the “Workload,” defined as the yearly number of cases included in all the background-positive alarms. We compared these benchmarks for systems optimized to detect either ARI or F-ARI.
Results:
CDAs that minimized both Detection Delay and Workload were those that maximized specificity and positive predictive value. Compared to “respiratory” ICD-9 codesets alone, the best ARI CDA decreased Detection Delay from 38 to 30 days, and Workload from 2397 to 483 cases/year. The best F-ARI-targeted CDA further reduced Delay to 22 days and Workload to 121 cases/year.
Implications:
Case detection methods that take advantage of information from the full EMR and that focus only on those ILI cases that are febrile can lower both the delay and the workload required to detect an influenza epidemic in the community.
Impacts:
This work illustrates how CPRS can be used to enhance nationwide influenza surveillance.
