Interobserver variability in collecting data from medical records

As part of a quality assurance program for monitoring medical efficacy of laboratory tests, three registered nurses collected data from the records of over 3000 patients who had plasma parathyroid hormone measurements. At the close of this investigation, medical records were reabstracted to assess the consistency of data collection. Overall agreement between data entries varied from 99.5% to 93.4% to 82% for 1672 numerical entries, 940 interpretative entries, and 50 classification codes, respectively. Statistics for 68 variables showed 3% poor agreement, 10% slight to fair agreement, 15% moderate agreement, 24% substantial agreement, and 48% almost perfect agreement. Since interobserver variability is a potential source of bias, statistical analysis of reabstracted data is a useful way to evaluate data consistency.