The Impact of Deidentification on Statistical Information in Limited Access Datasets

• Activity Number: 477
• Type: Contributed
• Date/Time: Wednesday, August 1, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): April Slee*+ and Paula Diehr and Kenneth Rice
• Companies: Axio Research and University of Washington and University of Washington
• Keywords: HIPAA ; Limited Access Data Set

Abstract

HIPAA guidelines focus on removal of Private Health Information (PHI), or "deidentification," when creating limited access data sets (LADS). LADS are used to estimate prevalence, calculate sample size, compare event rates and gather other information. Removing PHI impacts the amount of information available in LADS. We explored methods of quantifying the loss of information due to deidentification. We selected 14 published electronic data sets, recalculated the primary analysis, removed all PHI, and repeated the analysis on the deidentified data. We compared the ratio of confidence intervals for the predictor of interest from the original and deidentified data. The mean ratio of confidence intervals was 0.80 (p-value 0.039). Information loss was significantly greater for observational studies than randomized clinical trials (mean ratio 0.65 versus 1.0, p= 0.046).