Reliability of Cancer Recurrence Data: Institution Cancer Registry vs. Trained Chart Abstraction

Abstract

<h3>Purpose/Objective(s)</h3> While large national cancer registries have significant advantages, data regarding cancer outcomes (e.g., recurrence) are often lacking. Although cancer recurrence data are collected by institutional cancer registries, they are often not published without manual review due to large quantities of missing data and inaccuracy. This study aims to compare recurrence data from trained manual chart abstraction to an institutional cancer registry for patients with head and neck (HN) cancer treated with radiation therapy (RT). <h3>Materials/Methods</h3> Cancer recurrence status was collected by a trained research student under the guidance of a radiation oncologist. Recurrence status and type, date, method of detection / confirmation, location of follow-up, treatment location, and survival status were collected and compared with the cancer registry for continuous patients treated with RT for HN cancer at a tertiary cancer center. The sensitivity and specificity of cancer registry data was calculated using manual review as the gold standard. False negatives were compared to true positive recurrences to assess for differences in patient characteristics and compared via chi-square analysis. <h3>Results</h3> A total of 1,337 patient charts were reviewed who were treated from 2010 to 2017. A total of 375 (28%) confirmed cancer recurrences were identified, 45 (3%) had concern for recurrence without radiologic or pathologic confirmation, and 30 had persistent disease (2%). Of those who had confirmed recurrence, most were distant (36%) or distant plus locoregional (27%) while fewer were local (13%), regional (10%), or locoregional (15%). The cancer registry was found to have an accuracy of 82% with a sensitivity of 52% and specificity of 96%. There were 191 false negatives, 208 true positives, 832 true negatives, and 36 false positives. When comparing recurrences correctly identified (true positives) to those missed (false negatives), there was less time from RT to recurrence (283 days vs 748 days), differences between years of diagnosis (p=0.03), and less pathologic confirmation (67% vs 88%, p<0.01). A total of 92% of cancer recurrences identified by the registry occurred within 2 years of RT, versus 66% on manual review (p<0.01). There was no difference in age (63 vs 62 years), percent men (76% vs 79%), or treatment location. <h3>Conclusion</h3> Currently, cancer registries have high specificity and poor sensitivity. It is likely that with longer follow-up of cancer status, sensitivity will improve. Automated data abstraction tools may improve detection of cancer recurrences and minimize manual chart review.