Date of Award

January 2012

Document Type


Degree Name

Medical Doctor (MD)



First Advisor

James B. Yu

Subject Area(s)

Oncology, Epidemiology



Dozens of high-impact observational studies evaluating adjuvant therapy currently inform national physician practice. However, an often-neglected source of bias known as "immortal time bias" may artificially skew the results of these analyses in favor of the cohort receiving adjuvant therapy. In order to account for this, some investigators have used "landmark analysis" to measure post-diagnosis survival exclusively among patients who have survived a pre-selected length of time. As clinicians increasingly rely on observational studies to judge the efficacy of postoperative radiotherapy (PORT), thorough evaluation of these issues are essential.


First, we reviewed previous studies of the National Cancer Institute's (NCI) Surveillance, Epidemiology, and End Results (SEER) database to determine how frequently this bias was considered. Second, we used SEER to select three tumor types (glioblastoma multiforme, stage IA-IVM0 gastric adenocarcinoma, and stage II-III rectal carcinoma) for which prospective trials demonstrated an improvement in survival associated with PORT, as well as two tumor types (breast ductal carcinoma in situ and cT1-2/pT3N0M0 prostate carcinoma) for which prospective trials demonstrated no survival difference associated with PORT. For each tumor type, we calculated conditional survivals and adjusted hazard ratios of PORT vs. postoperative observation cohorts while restricting the sample at sequential monthly landmarks.


In the first part of the study, we found that sixty-two percent of previous SEER publications evaluating PORT failed to use a landmark analysis. In the second part of the study, a total of 78,043 patients were included. For GBM, PORT was associated with improved 1-year overall survival if we included all patients in the analysis (HR 0.41, 95% CI 0.37-0.45), consistent with clinical trial data, though the magnitude of the survival benefit diminished significantly as the landmark increased. Similar trends were seen in gastric and rectal cancer. Inconsistent with clinical trial data are the associations of PORT with improved 5-year cancer-specific survival for breast DCIS and with worsened 5-year cancer-specific survival for cT1-2/pT3N0M0 prostate cancer, regardless of landmark selection.


Though the majority of previous SEER papers do not correct for it, immortal time bias leads to altered estimates of PORT effectiveness, which are very sensitive to landmark selection. Conclusions from observational studies estimating the survival benefit of PORT are limited by three competing selection biases: 1) immortal time bias favoring PORT in low-survival tumors like GBM (patients dying soon after surgery may be less likely to receive PORT); 2) "low-risk patient" bias favoring PORT in high-survival tumors like breast DCIS (healthier patients with longer life expectancy may be more likely to receive PORT); and 3) "high-risk cancer" bias may favor OBS in high-survival tumors like cT1-2/pT3N0M0 prostate cancer (patients with higher-risk tumor characteristics not captured by SEER may be more likely to receive PORT). Sequential landmark analysis is necessary to account for the impact of the immortal time bias in any observational study evaluating the efficacy of PORT.


This thesis is restricted to Yale network users only. This thesis is permanently embargoed from public release.