Sensitivity to Design Choices When Deriving Variables from Electronic Health Records

• Activity Number: 374
• Type: Invited
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Risk Analysis
• Author(s): Robert Alan Greevy*+ and Mary Margaret Huizinga and Christianne L Roumie and Carlos G Grijalva and Harvey Murff and Xulei Liu and Marie R Griffin
• Companies: Vanderbilt University and The Johns Hopkins University School of Medicine and Vanderbilt University and Vanderbilt University and Vanderbilt University and Vanderbilt University and Vanderbilt University
• Keywords: comparative effectiveness research ; diabetes ; electronic health records

Abstract

Two important dilemmas in the design of studies using prescription data from electronic health records are what minimum level of adherence to require for an episode of continuous drug use and how to handle stockpiling of medications. Generally, the sensitivity of the study's conclusions to these design choices is not analyzed. Covariate adjusted Cox models compared persistence and durability on three common oral anti-diabetic therapies in a cohort of 12,697 incident users. Allowing 50% stockpiling, sulfonyurea therapy showed a significantly lower risk of nonpersistence (changing or stopping therapy) compared to metformin when 0 gap days were allowed, 95% confidence interval for the adjusted hazard ratio (0.91, 1.00), no significant difference when 14 gaps days were allowed (0.94, 1.03), and significantly greater risk of nonpersistence when 30 or 90 gaps days were allowed (1.00, 1.10) and (1.07, 1.20) respectively. All drug comparisons showed statistically significant effects in both directions, showing decreased or increased risk of nonpersistence depending on the design parameters. (www.nature.com/clpt/journal/vaop/ncurrent/full/clpt2011228a.html)