Pneumonia: VA Informatics and Computing Infrastructure (VINCI)

Assessing confirmation bias effects on classifying supporting, refuting or uncertain evidence for suspected pneumonia case review
A typical retrospective clinical research study involves manual review of a superset of patients’ charts to identify experimental and control groups. Even when patient records are available electronically, chart review can involve many minutes or even hours reading through records for a single patient. Ultimately, research studies using clinical data are time-consuming and cumbersome due to the massive effort required for manual chart review. In addition to the time, effort, and expense involved in chart review, manual analysis of a chart is susceptible to cognitive biases that affect the accuracy of case classification. The overarching goal of this study is to evaluate potential improvements in the efficiency and accuracy of human chart review with the goal of reducing confirmation bias at the level of annotation of mentions for supporting, refuting or uncertain evidence. We hypothesize that integrating a linguistic model of confirmation and uncertainty with visualization techniques will decrease confirmation bias in manual chart review. Our long term study objective is to understand confirmation bias for pneumonia case assessment and develop models of uncertainty.
Selected Publications, Posters, and Presentations

Project ID: HIR 08-204
PI: Wendy Chapman

Publications: 

 South, BR, Kramer HS, Jones, B, Tharp, W, Chapman, W. Classifying Supporting, Refuting, or Uncertain Evidence for Pneumonia Case Review. Online Journal of Public Health Informatics 7 (1)