Forecasting Levels and Standard Errors to Enable Prospective

• Activity Number: 490
• Type: Topic Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Howard Burkom*+ and Yevgeniy Elbert and Steven Babin
• Companies: The Johns Hopkins University Applied Physics Laboratory and The Johns Hopkins University Applied Physics Laboratory and The Johns Hopkins University Applied Physics Laboratory
• Keywords: disease surveillance ; absenteeism ; control chart ; monitoring ratios

Abstract

Disease surveillance systems must periodically monitor key ratios such as those of positive laboratory tests, of patients from outside a facility catchment area, of patients in a certain age group, or of absentees. P-charts are widely used for this purpose. A key feature of p-charts is that normalization of deviations from expected ratios uses the binomial approximation to the normal distribution. However, series of ratios from authentic datasets often violate underlying assumptions. A ratio-monitoring alternative is to replace the standard errors by empirical standard deviation estimates. Strategies to produce robust monitoring include a) stratification by geographic region or patient class, b) outlier removal (dates/locations of quality problems, true aberrations), and c) regression modeling if sufficient historical data exist and if independent variables may be assumed available for future estimates. This presentation will use a 7-year absenteeism dataset from over 200 public schools to compare approaches for estimating expected ratios and standard errors relative to these strategies and will recommend charts appropriate to specific data types and monitoring objectives.