INTRODUCTION: We sought to characterize the effects of PSA registry errors on clinical research by comparing cohorts based on cancer registry PSA values with those based directly on results in the electronic health record.METHODS: We defined example cohorts of men with prostate cancer using data from the Veterans Health Administration: those with a PSA values less than 4.0 ng/mL, 4.0 to 10.0 ng/mL, 10.0 to 20.0 ng/mL, and 20.0 to 98.0 ng/mL. We compared the composition of each cohort and overall patient survival when using PSA values from either the VA Central Cancer Registry versus the gold standard electronic health record laboratory file results.RESULTS: There was limited agreement between cohorts defined using either the cancer registry PSA values versus the laboratory file of the electronic health record. The least agreement was seen in patients with PSA values < 4.0 ng/mL (58%) and greatest among patients with PSA values between 4.0 and 10.0 ng/mL (89%). In each cohort, patients assigned to a cohort based only on the cancer registry PSA value had significantly different overall survival when compared with patients assigned based on both the registry and laboratory file PSA values.CONCLUSIONS: Cohorts based exclusively on cancer registry PSA values may have high rates of misclassification that can introduce concerning differences in key characteristics and result in measurable differences in clinical outcomes.
View details for PubMedID 29630980