1269. Evaluation of the Utility of CMS Claim Data for Early Detection of Increasing Influenza Activity
Session: Poster Abstract Session: Clinical Infectious Diseases: Respiratory Infections
Friday, October 28, 2016
Room: Poster Hall
Background: To evaluate the utility of CMS Medicare claim data for early detection of increasing influenza (flu) activity using weekly volume of flu diagnoses (V_dx) or flu rapid tests (V_test).

Methods: We compared CMS Medicare outpatient claims (OTP) with positive test volumes using rapid flu tests from the CDC National Respiratory and Enteric Virus Surveillance System (NREVSS), referred to as the “gold standard” for flu activity. We compared both V_dx and V_test to the gold standard for each of 10 selected states (one per HHS region). Spline models were developed and applied to fit the time series of weekly volumes of V_dx, V_test, and NREVSS test positives (dependent variables) from 2007 to 2012, with the week number as the independent variable. Using the fitted models, we calculated predicted weekly volumes and standard deviations (SD) for V_dx and V_test. Criteria for an initial alert were defined as: 1. an increase of the predicted volume from the previous week by two SD (2 SD criterion), or 2. an increase of the predicted volume from previous week by 0.45 SD for two consecutive weeks (0.45 SD criterion). A flu “episode” was defined as the interval between an initial alert and a drop in volume meeting the same criterion.


We detected 67 flu episodes from the 10 states in NREVSS (2007-12) using Spline models. With the 0.45 SD and 2 SD criteria respectively, the OTP predictors using V_dx data yielded 60 and 41 out of the 67 episodes, and 64 and 49 episodes using V_test data. With the 0.45 SD criterion, 83% of episodes were detected for V_dx and 85% using V_test in NREVSS predictions. Using the 2 SD criterion, 93% (98%) of episodes were detected with V_dx and 98% using V_test in NREVSS predictions. With the 0.45 SD criterion, V_dx (V_test) detected flu activity an average of 3.1 (5.7) weeks earlier than NREVSS predictions. With the 2 SD criterion, V_dx (V_test) detected flu activity an average of 3.6 (2.6) weeks later than NREVSS predictions.


In this retrospective analysis, Spline models performed well for early detection of an increase in flu activity. The 0.45 SD criterion demonstrated greater sensitivity over the 2 SD criterion yielding earlier detection by an average of over three weeks compared to predictions using the gold standard data. The consecutive week exceedance requirement using 0.45 SD also produced practical positive predictive values.

Lin Fan, PhD, Centers for Disease Control and Prevention, Atlanta, GA, Rex Astles, PhD, Centers for Disease Control and Prevention, atlanta, GA, Howard Burkom, PhD, Applied Physics Laboratory, johns hopkins University, baltimore, MD and Zheyu Hu, Ph.D, M.D, Emory University, atlanta, GA


L. Fan, None

R. Astles, None

H. Burkom, None

Z. Hu, None

Findings in the abstracts are embargoed until 12:01 a.m. CDT, Wednesday Oct. 26th with the exception of research findings presented at the IDWeek press conferences.