Online Program Home
  My Program

Abstract Details

Activity Number: 68 - Government Health Statistics
Type: Contributed
Date/Time: Sunday, July 30, 2017 : 4:00 PM to 5:50 PM
Sponsor: Government Statistics Section
Abstract #323039
Title: Comparing Algorithmic Alerts Using Provisional Case Counts to Alerts Using Finalized Data--National Notifiable Diseases Surveillance System
Author(s): Hong Zhou* and Howard Burkom and Ruth Jajosky and Tara Strine
Companies: CDC and Johns Hopkins Applied Physics Laboratory and Centers for Disease Control and Prevention and Centers for Disease Control and Prevention
Keywords: aberration detection ; , Historical Limits method ; disease surveillance
Abstract:

In the US National Notifiable Diseases Surveillance System (NNDSS), provisional weekly counts of notifiable disease case notifications are analyzed for timely aberration detection. Provisional data were assessed because finalized data are generally delayed by about 1.5 years. We applied the traditional Historical Limits method (HLM), using a 4-week data unit, and a modification (MHLM), with a 1-week data unit, to detect anomalous increases in weekly time series. Data for 7 diseases from 9-26 selected states were evaluated using 2009-2013 as the baseline and 2014 for testing. At a threshold of 2 standard deviations above predicted values, sensitivity for weekly alerts in state-level provisional data, using alerts in finalized data as ground truth, ranged from 27-85% for HLM and 34-92% for MHLM. The false alarm rate of weeks alerted in provisional data relative to alerting in finalized data ranged from 5%-14% for HLM and 3%-7% for MHLM. Given that aberration detection sensitivity and false alarm rate in the provisional data vary by disease and state, reporting timeliness may affect aberration detection performance for some diseases in NNDSS.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2017 program

 
 
Copyright © American Statistical Association