Introduction

Errors in prostate specific antigen (PSA) values included in prostate cancer registries have called into question clinical research studies that rely on this information. We sought to characterize the potential effects of PSA registry errors on clinical research by comparing cohorts based on registry PSA values with those based on laboratory values extracted from an integrated national health care system._x000D_

Methods

We defined three example cohorts of men with prostate cancer using data from the VA integrated health care system: those with &[Prime]very low&[Prime] (<4.0 ng/mL), &[Prime]low&[Prime] (<10.0 ng/mL), and &[Prime]high&[Prime] (20-100 ng/mL) PSA values. We compared the composition of each cohort when using the cancer registry versus the electronic health record PSA values. We compared overall survival for each cohort as an example clinical outcome. We fit multivariable proportional hazards models to determine the importance of the PSA source in each cohort._x000D_

Results

There was significant discordance when using cancer registry versus electronic health record PSA values to identify a cohort of patients with &[Prime]very low PSA&[Prime] values. While 7,286 were included in both cohorts, one third (n=3,515) of the cohort defined using cancer registry PSA values was misclassified and 1,800 additional patients were identified when using electronic health record data. The concordance was highest for patients with &[Prime]low&[Prime] PSA values, with 21,860 (98%) of patients identified in both the cancer registry and electronic health record based cohorts. Cancer registry PSA values misclassified 41% (604) of the &[Prime]high&[Prime] PSA cohort, and 133 additional patients were identified using electronic health record data. Comparisons of overall survival in the examples cohorts identified a difference in overall survival in the &[Prime]very low&[Prime] (log rank P=0.03), but not the &[Prime]low&[Prime] or &[Prime]high&[Prime] PSA cohorts.

Conclusions

Patient cohorts based on cancer registry PSA values may have high rate of misclassification, particularly among patients with &[Prime]very low&[Prime] or &[Prime]high&[Prime] PSA values. In some cases, differences in cohorts resulted in measurable differences in overall survival. Attempts should be made to validate cancer registry PSA data to ensure accurate and reproducible results._x000D_

Funding

None