Generation of Prediction Intervals to Assess Data Quality in the Distribute System Using Quantile Regression

• Activity Number: 336
• Type: Topic Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): Ian Painter*+ and Julie Eaton and Debra Revere and Bill Lober and Donald Olson
• Companies: University of Washington and University of Puget Sound and University of Washington and University of Washington and International Society of Disease Surveillance
• Keywords: Syndromic surveillance ; Data Quality ; Quantile regression

Abstract

Distribute is a national influenza-like-illness (ILI) surveillance project that integrates data from multiple jurisdictions. Distribute works solely with summarized (aggregated) data. Timeliness of the data varies considerably between sites; for many sites data for each encounter date arrives piecemeal, spread over several days. This spread adds additional noise into the data received by the Distribute system. Systematic differences in the timeliness between sources of data can introduce bias into the indicator of interest, the ILI ratio.

Quantile regression using the observed relationship between incomplete and complete data is used to calculate prediction intervals for complete data. Some sites have very narrow prediction intervals that indicate the ILI-ratio calculated from incomplete data approximates the complete data ratio very accurately. Other sites show considerable asymmetry