Health Services Research
SS 44 - HSR 2 - Health Services Research
326 - Inclusion of Subjects With Internally Inconsistent or Anomalous Data Within the National Cancer Database Significantly Alters Survival Results
Wednesday, October 24
3:55 PM - 4:05 PM
Location: Room 008
Inclusion of Subjects With Internally Inconsistent or Anomalous Data Within the National Cancer Database Significantly Alters Survival Results
C. D. Jacobs, D. J. Carpenter, J. C. Hong, and J. P. Chino; Duke University Medical Center, Durham, NC
A prior analysis quantified anomalous data within the National Cancer Database (NCDB) that was internally inconsistent or outside accepted norms of adjuvant radiation therapy (RT) using node positive uterine cancer as a test clinical scenario. This project seeks to determine the effect of anomalous data on overall survival (OS).
All NCDB subjects with FIGO stage IIIC1-IIIC2 uterine cancer treated with hysterectomy and adjuvant RT between 1998-2012 were identified. Variables reviewed to identify anomalous data included RT site, modality, dose, fractions, timing, duration and stage. Log rank test was used to calculate OS for subjects with and without anomalous data. Univariate and multivariate (MVA) cox regression analyses of anomalous data with respect to OS were performed with and without insufficient, anomalous, or unknown RT dose, with both models controlling for the following factors: facility type, facility location, race, median income, educational status, Charlson-Deyo score, and diagnosis year.
2,288 (16.0%) of the 14,298 analyzed subjects had ≥1 anomalous datum; 538 (3.8%) subjects had anomalies limited to insufficient dose or fractions which could be from an incomplete RT course. 5 year OS in those with ≥1 anomalous datum was 51.32% (95% CI 48.89-53.71%) versus 57.97% (95% CI 57.89-58.07%) for those without anomalous data (p<0.0001). MVA showed that facility location (HR 1.212, 95% CI 1.062-1.383; p=0.0008), age at diagnosis (HR 1.043, 95% CI 1.040-1.047; p<0.0001), race (HR 1.263, 95% CI 1.110-1.438; p<0.0001), Charlson-Deyo score (HR 1.149, 95% CI 1.053-1.253; p=0.0001), diagnosis year (HR 0.972, 95% CI 0.958-0.987; p=0.0002), RT dose per Gy, (HR 0.998, 95% CI 0.997-0.999; p=0.0007), and any anomalous data (HR 1.133, 95% CI 1.086-1.183; p<0.0001) were significantly associated with OS. After excluding subjects with an insufficient, anomalous, or unknown total RT dose (unknown, n=5022; <40 Gy, n=671; >90 Gy, n=172), the 814 (9.7%) of 8,433 subjects with ≥1 anomalous datum had overlapping 5 year OS (57.19%, 95% CI 53.27-61.11%) versus those without anomalous data (61.46%, 95% CI 60.15-62.77%). MVA of this subset showed OS correlated significantly with facility location (HR 1.187, 95% CI 1.012-1.392; p=0.0014), age at diagnosis (HR 1.044, 95% CI 1.040-1.048; p<0.0001), race (HR 1.248, 95% CI 1.068-1.458; p<0.0001), Charlson-Deyo score (HR 1.141, 95% CI 1.027-1.267; p=0.0132), diagnosis year (HR 0.973, 95% CI 0.955-0.991; p=0.032), and RT dose (HR 1.012, 95% CI 1.005-1.019; p=0.0008); however, there was no longer an association with anomalous data (HR 1.000, 95% CI 0.929-1.077; p=0.996).
Unaccounted-for anomalous data within the NCDB in this test clinical scenario, particularly insufficient RT dose, was significantly associated with worse OS. Careful accounting for inconsistently recorded entries and anomalous data is critical for future investigations using the NCDB, particularly when radiation details are examined.
Author Disclosure: C.D. Jacobs: None. D.J. Carpenter: None. J.C. Hong: None. J.P. Chino: Partner; Duke University Cancer Center. Stock; NanoScint. Co-Founder/Owner; NanoScint.