Virtual Library

Start Your Search

Y. Won



Author of

  • +

    P3.06 - Poster Session with Presenters Present (ID 492)

    • Event: WCLC 2016
    • Type: Poster Presenters Present
    • Track: Trial Design/Statistics
    • Presentations: 1
    • +

      P3.06-007 - The Consequence of Incomplete Follow-up in Hospital-based Survival Study as Compared with National Vital Status-based Results (ID 5887)

      14:30 - 15:45  |  Author(s): Y. Won

      • Abstract

      Background:
      Loss to follow-up (FU) is an important issue in survival analysis using the data based on hospital records. To better address the magnitude of this issue in a real clinical setting, we compared survival outcomes from hospital database with those from national cancer registry data which incorporated national vital status record.

      Methods:
      From the hospital database of National Cancer Center Hospital, Korea, we identified 970 small cell lung cancer (SCLC) patients who were treated between 04/2001 and 04/2013. Most of them were male (n = 854) and smokers (n = 906). Median age was 63 years (range, 32–80 years). We made two survival datasets, hospital-based dataset (HD) and cancer registry-based dataset (CD).

      Results:
      Of 352 LD-SCLC patients, there were 144 deaths in the HD and 107 additional deaths were identified in the CD (Total= 251). There was no difference in median progression free survival (PFS) between the HD and CD (12.7 months [95% CI, 10.9-14.6] vs. 12.3 months [95% CI, 10.8-14.2]). But, median OS in the HD was significantly longer than in the CD (55.7 months [95% CI, 35.8-115.6] vs. 26.3 months [95% CI, 22.8-30.8]). The 5-year survival rate of LD-SCLC was 48.7% vs. 29.6% in the HD and CD, respectively. For 618 ED-SCLC patients, there were 234 deaths in the HD while 341 additional deaths were confirmed in the CD (Total= 575). Median PFS from the HD was similar to that from the CD (6.5 months [95% CI, 6.2-6.9] vs. 6.4 months [95% CI, 6.1-6.8]). Median OS of HD was 14.5 months [95% CI, 13.5-16.9], significantly longer than that of CD (11.9 months [95% CI, 11.2-12.9]). The 5-year survival rate of ED-SCLC in the HD and CD was 11.5% and 3.5%, respectively. In the simulation analysis, the estimated median OS increased as the proportion of patients who were actually dead but censored in the HD increased. When this proportion was 25%, 50% and 75%, the estimated median OS was 27.8 months, 33.8 months, and 37.2 months for LD-SCLC, respectively, and 12.5 months, 13.1 months, and 13.7 months for ED-SCLC. Obviously, this discrepancy reflects the limitation of HD-based survival analysis since medical records do not trace all patients until death, especially for those who did not return for subsequent follow-up care.

      Conclusion:
      Incomplete follow-up, by increasing the number of censoring events, could result in spurious prolongation of overall survival, which warrants caution in interpreting the HD-based survival analysis.