Adjusting for Misclassification in Observational Studies: Getting from Cleaner to Messier Scenarios

• Activity Number: 273 - Challenges and Strategies in Analysis of Complex Data
• Type: Invited
• Date/Time: Tuesday, August 1, 2017 : 8:30 AM to 10:20 AM
• Sponsor: SSC
• Author(s): Tanja Högg* and Paul Gustafson and Yinshan Zhao and John Petkau and Helen Tremlett
• Companies: University of British Columbia and University of British Columbia and BC Center for Improved Cardiovasular Health and University of British Columbia and University of British Columbia
• Keywords: outcome misclassification ; administrative health data ; Bayesian methods ; matched case-control study

Abstract

In some situations, there is quite a clean story on how measurement error or misclassification, if ignored, can distort inferences about associations of interest. And commensurately the approach to adjusting for measurement error is straightforward and intuitive, provided amenable information about the measurement error is available (say in the form of a validation study). Other problems are more complex, with a less predictable story arc. We give an example of such an arc, in a problem involving administrative health data. Here the ascertainment of disease status (control or case) is error-prone, yet this ascertainment is used as a matching factor in the creation of a for-research dataset. Using Bayesian analysis and multiple sources of information it is possible to adjust for the imperfect ascertainment of disease status.