Analysis of Surveillance, Epidemiology, and End Results Program (SEER) PSA coding error using primary data.

Abstract

21 Background: SEER did not report PSA values for the 2014 submission because of inaccurate coding. This inaccuracy was a result of a confusing data entry guideline where PSA is coded in a 3-digit field with an implied decimal between digits 2 and 3. Our study uses original registry data to assess the magnitude and implications of these coding errors. Methods: The National Oncology Data Alliance is a database of more than 150 Commission-on-Cancer compliant tumor registries that use proprietary cancer registry software sold by Elekta AB (Stockholm, Sweden) and contains the same data sent to state tumor registries and SEER. De-identified data from all newly diagnosed prostate cancer cases from 2005-2013 were extracted (n = 89,379). PSA data were acquired from both the Collaborative Staging (CS) Site Specific Factor 1 (SSF1) for prostate cancer and from de-identified text fields. SSF1 contained the error prone datum that was transferred to SEER. The PSA data contained in the text fields, which are used to record lab results, etc. verbatim, were taken to be definitive. PSA values from both SSF1 and text fields were complete in 63,051 patients. AJCC Stage was determined using the clinical T stage field and the SSF fields that code for Gleason score. We calculated the error rates and their directions for PSA stratification into three levels (0 to < 10, 10 to < 20, and ≥ 20) and for the AJCC stage. Results: The measured error rate in PSA values in SSF1 caused by erroneous decimal placement was 9.0%. The resulting error rate in PSA stratification was 7.5%, with the SSF1 PSA value giving a higher PSA category in 4.8% of cases and lower in 2.7%. The consequent error rate in AJCC staging was 2.8% overall, with the PSA error resulting in a higher stage group in 1.9% and lower in 0.9%. Conclusions: Many factors can contribute to the overall error rate in PSA values in SEER data, but we are addressing only the errors associated with the implied decimal. The consequences of this error are modest despite a 9% error rate. However, SSF1 for prostate cancer is not the only confusing SSF introduced by the CS system. The current cancer registry database design is derived from outdated technology and needs restructuring to be a more effective surveillance and research tool.