Immortal Time Bias in Observational Studies of Time-to-Event Outcomes: Assessing Effects of Postmastectomy Radiation Therapy Using the National Cancer Database.

Research Article Immortal Time Bias in Observational Studies of Time-to-Event Outcomes: Assessing Effects of Postmastectomy Radiation Therapy Using the National Cancer Database Cancer Control Volume 25: 1-7 ª The Author(s) 2018 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/1073274818789355 journals.sagepub.com/home/ccx Parul Agarwal, PhD1, Erin Moshier, MS1, Meng Ru, MS1, Nisha Ohri, MD2, Ronald Ennis, MD3, Kenneth Rosenzweig, MD2, and Madhu Mazumdar, PhD1 Abstract The objectives of this study are to illustrate the effects of immortal time bias (ITB) using an oncology outcomes database and quantify through simulations the magnitude and direction of ITB when different analytical techniques are used. A cohort of 11 626 women who received neoadjuvant chemotherapy and underwent mastectomy with pathologically positive lymph nodes were accrued from the National Cancer Database (2004-2008). Standard Cox regression, time-dependent (TD), and landmark models were used to compare overall survival in patients who did or did not receive postmastectomy radiation therapy (PMRT). Simulation studies showing ways to reduce the effect of ITB indicate that TD exposures should be included as variables in hazardbased analyses. Standard Cox regression models comparing overall survival in patients who did and did not receive PMRT showed a significant treatment effect (hazard ratio [HR]: 0.93, 95% confidence interval [CI]: 0.88-0.99). Time-dependent and landmark methods estimated no treatment effect with HR: 0.97, 95% CI: 0.92 to 1.03 and HR: 0.98, 95% CI, 0.92 to 1.04, respectively. In our simulation studies, the standard Cox regression model significantly overestimated treatment effects when no effect was present. Estimates of TD models were closest to the true treatment effect. Landmark model results were highly dependent on landmark timing. Appropriate statistical approaches that account for ITB are critical to minimize bias when examining relationships between receipt of PMRT and survival. Keywords survival analysis, cox regression, immortal time bias, time-dependent, landmark, breast cancer, epidemiology, NCDB, observational, radiotherapy Received January 23, 2018. Received revised June 2, 2018. Accepted for publication June 22, 2018. Introduction 1 Immortal time bias (ITB), identified in epidemiology since the 1970s,1,2 occurs when there is variation in timing of treatment initiation from cohort entry and time-to-treatment is misclassified or ignored. Immortal time bias can occur in observational studies when a cohort approach is followed during which outcomes cannot occur. In the drug effectiveness literature, various cohort designs may result in ITB.3 Importantly, analyses can be flawed if ITB is not accounted for. To simplify, we generalize “treatment” to mean exposure, treatment, or response to exposure or treatment, as ITB can arise when the Corresponding Author: Madhu Mazumdar, Department of Population Health Science and Policy, Institute for Healthcare Delivery Science, Tisch Cancer Institute (TCI), Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA. Email: Department of Population Health Science and Policy, Institute for Healthcare Delivery Science, Tisch Cancer Institute (TCI), Icahn School of Medicine at Mount Sinai, New York, NY, USA 2 Department of Radiation Oncology, Icahn School of Medicine at Mount Sinai, New York, NY, USA 3 Department of Radiation Oncology, Rutgers Cancer Institute of New Jersey, New Brunswick, NJ, USA Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-NonCommercial 4.0 License (http://www.creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/open-access-at-sage). 2 timing to initiation of any of these states varies between patient groups. Various studies have addressed ITB by applying statistical approaches such as time-dependent (TD) Cox regression, landmark analyses, prescription time-distribution matching, and the sequential Cox approach.4-6 In a systematic review, over 40% of studies with a survival end point and timevarying treatment were susceptible to ITB.7 Thus, it is important to identify and control ITB, because it may alter study conclusions by underestimating the hazard ratio (HR), a measure of an effect of an intervention on an outcome of interest over time. Consequently, researchers may falsely conclude that a treatment is effective in influencing outcomes. For example, a systematic review of studies using the Surveillance, Epidemiology, and End Results Program database comparing survival among patients who did and did not receive postoperative radiotherapy observed that ITB was not addressed or controlled for in most studies, which may have led to false conclusions.8 A systematic review and metaanalysis on the effects of b-adrenergic receptor antagonists on cancer survival drew similar conclusions.9 Although systematic reviews identified ITB as an issue in the oncology literature, optimal methods for addressing ITB arising from variations in timing of initiation of radiation therapy are lacking. Immortal time bias may be more pronounced in the oncology literature due to greater availability and use of observational data to explore various research questions. Thus, it is important for clinical researchers to be aware of ITB and statistical methods to address it. The effectiveness of postmastectomy radiation therapy (PMRT) is well established in the literature10-12; however, when results of studies are analyzed, variations in the timing of initiation of radiation therapy may lead to ITB. It is vital to address ITB in radiation oncology studies, since uncontrolled bias may affect estimates of therapeutic effectiveness of radiation on survival among patients with cancer and could generate spurious associations. Therefore, the present study used an oncology outcomes database and sought to quantify through simulations the magnitude and direction of ITB when different analytic techniques are used. Methods Study Design A retrospective cohort study design was used to compare overall survival in patients who did and did not receive PMRT. Using parameter estimates obtained from the cohort (see subsequently), we conducted simulations to provide recommendations on including TD exposure in hazard-based analyses. Because no patient, provider, or hospital identifiers are included in the analytic or simulation components of this study and no protected health information is present, institutional review board approval was not required. Informed consent was not required for the analytic sample or simulation study. Cancer Control Data Source Data were obtained for the years 2004 to 2008 from the National Cancer Database (N (...truncated)