On the Use of Sampling Weights for Retrospective Chart Reviews

• Activity Number: 185
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Teaching of Statistics in the Health Sciences
• Author(s): Ernest Shen* and Adam Sharp and Corrine Munoz-Plaza and Tania Tang and Erin Hahn and Michael Gould
• Companies: Kaiser Permanente and Kaiser Permanente and Kaiser Permanente and Kaiser Permanente and Kaiser Permanente and Kaiser Permanente
• Keywords: Survey sampling

Abstract

Retrospective chart review (RCR) is often used in medical research to answer the "why" questions that statistical modeling cannot. In addition to its utility as an explanatory tool, it can also be used to generate hypotheses using available retrospective data, and so is a convenient guide for developing future prospective studies. A recent review of papers that used RCR discussed ten "best practices," including sampling strategies for cases where one can only conduct RCR for a sample of the target population. While that review acknowledged the importance of carefully selecting a sampling strategy, and outlined three commonly used methods (convenience, simple random, and systematic), they do not discuss the use of sampling information at the data analysis stage. This paper demonstrates why the use of sample weights ought to be another "best practice" to add to the list, by reviewing well-known theoretical details and some data analysis examples. For one such example, the raw proportion of length of symptoms less than 7 days was about 30% higher than when weighted by the sampling fractions of the three care settings from which charts were sampled.