Cure Models in Cancer Clinical Trials

• Published in: iTextbook of Clinical Trials in Oncology pp. 465 - 492
• Main Authors: Legrand, Catherine, Bertrand, Aurélie
• Format: Book Chapter
• Language: English
• Published: CRC Press 2019
• Edition: 1
• Subjects: Semi-parametric Cox PH Model; Cumulative Hazard Function; PH Model; Cox PH; Aft Model; Parametric PH Model; Survival Function; Semiparametric Cox PH Model; Van Keilegom; Cox PH Model; Childhood Acute Lymphocytic Leukemia; Conditional Survival; IFN Group; Population Survival Function; Standard Cox PH Model; Estimated Survival Curves; Conditional Survival Function; Sample Size Calculation; PH Assumption; Binary Covariate; MCM; Promotion Time Cure Model; Exponential Link Function; Uncured Patients; Cure Fraction
• ISBN: 1138083771; 9781138083776
• DOI: 10.1201/9781315112084-22

Due to the advances in medical research in the past decades, cancer is not necessarily a fatal disease anymore. For specific cancer types, one can now reasonably expect a fraction of long-term survivors to show-up in cancer clinical trials. The presence of short-and long-term survivors may lead to a violation of the proportional hazards assumption and therefore jeopardize the use of of the popular Cox model. Furthermore, in such a situation, the proportion of "cured" patients becomes a crucial component of the assessment of patient benefit, and being able to distinguish a curative from a life-prolonging effect conveys important additional information in the evaluation of a new treatment. To address these issues, specific "cure models" have been proposed in the statistical literature. In this chapter we introduce the two main families of such models: mixture cure models and promotion time cure models. We elaborate on how and when to use these models and discuss that in practice, it is not only a matter of whether or not there are cured patients in the data, but that as in classical survival analysis, the appropriate model to be used should be carefully chosen, based on the main features of the data and with a strong emphasis on the proportional hazards assumption. It has become more and more common in cancer clinical trials to observe patients experiencing long-term relapse-free survival, and cure has become a reality for both patients and clinicians. Cure has become a reality for both the patients and the clinicians in some types of cancer; however, despite the fact that cure models can therefore be an interesting way to characterize and study patient survival, they are still an underused statistical tool in the context of oncology trials. A general and informal rule that holds for all cure models requires the follow-up of the study to be sufficiently long: the estimated survival function should exhibit a long plateau containing many censored observations. More formally, the maximum possible event time should be smaller than the maximum possible censoring time. The estimation of mixture cure models is classically based on the maximization of the likelihood function.