Design of cancer trials based on progression‐free survival with intermittent assessment

Composite endpoints are very common in clinical research, such as recurrence‐free survival in oncology research, defined as the earliest of either death or disease recurrence. Because of the way data are collected in such studies, component‐wise censoring is common, where, for example, recurrence is an interval‐censored event and death is a right‐censored event. However, a common way to analyze such component‐wise censored composite endpoints is to treat them as right‐censored, with the date at which the non‐fatal event was detected serving as the date the event occurred. This approach is known to introduce upward bias when the Kaplan‐Meier estimator is applied, but has more complex impact on semi‐parametric regression approaches. In this article we compare the performance of the Cox model estimators for right‐censored data and the Cox model estimators for interval‐censored data in the context of component‐wise censored data where the visit process differs across levels of a covariate of interest, a common scenario in observational data. We additionally examine estimators of the cause‐specific hazard when applied to the individual components of such component‐wise censored composite endpoints. We found that when visit schedules differed according to levels of a covariate of interest, the Cox model estimators for right‐censored data and the estimators for cause‐specific hazards were increasingly biased as the frequency of visits decreased. The Cox model estimator for interval‐censored data with censoring at the last disease‐free date is recommended for use in the presence of differential visit schedules.

Section: Discussionmentioning

confidence: 99%

Analysis of composite endpoints with component‐wise censoring in the presence of differential visit schedules

Eaton

Zabor

2022

“…Often, in trials assessing PFS , assessments happen at regular intervals. How to account for that type of measurements in the likelihood is discussed by Zeng et al…”

Section: Statistical Inferencementioning

confidence: 99%

Joint modeling of progression‐free and overall survival and computation of correlation measures

Meller

Beyersmann²,

Rufibach

2019

In this paper, we derive the joint distribution of progression-free and overall survival as a function of transition probabilities in a multistate model. No assumptions on copulae or latent event times are needed and the model is allowed to be non-Markov. From the joint distribution, statistics of interest can then be computed. As an example, we provide closed formulas and statistical inference for Pearson's correlation coefficient between progression-free and overall survival in a parametric framework. The example is inspired by recent approaches to quantify the dependence between progression-free survival, a common primary outcome in Phase 3 trials in oncology and overall survival. We complement these approaches by providing methods of statistical inference while at the same time working within a much more parsimonious modeling framework. Our approach is completely general and can be applied to other measures of dependence. We also discuss extensions to nonparametric inference.Our analytical results are illustrated using a large randomized clinical trial in breast cancer. KEYWORDS event history analysis, illness-death model, multistate model, randomized clinical trials INTRODUCTIONAn established way to evaluate performance of therapies in clinical trials is time-to-event endpoints. While overall survival (OS) remains the gold standard for demonstrating clinical benefit, especially in oncology, alternative endpoints such as progression-free survival (PFS), are also accepted by Health Authorities. 1,2 PFS is not only considered a surrogate for OS but may provide clinical benefit by itself, eg, by delaying symptoms or subsequent therapies. 3 Further advantages of the use of PFS are shorter trial durations and the fact that, as opposed to OS, PFS is not confounded by the use of later line, ie, post-progression therapies, and PFS is less vulnerable to competing causes of death than OS. 4 Methodology has been developed to assess surrogacy of one endpoint for another 5 and these methodologies have been applied to a wide range of indications. 6 One important aspect in the assessment of surrogacy is to quantify the correlation between the surrogate and the real endpoint. 7 This aspect has received quite some attention in the literature lately. 8,9 These two papers consider Pearson's correlation coefficient and rely on an illness-death model without recovery (IDM) to model the association between PFS and OS. To specify the underlying statistical model for the likelihood function for parametric estimation, they use a latent failure time approach. One characteristic of the latent failure time approach is that it allows for PFS events after OS, see Section 2.3. Within this model formulation, Fleischer et al 8 then derive closed formulas for the survival functions S PFS and S OS for PFS and OS as well as the correlation coefficient Corr (PFS, OS ) by assuming an exponential distribution for all transition intensities in the underlying multistate model. Li and Zhang 9 generalized these results to Weibull transition intensitie...

“…Panel data of this type, for which only snapshots of the underlying disease process are obtained, are interval censored, in that changes in health status can occur at some time point between assessments, and only part of the (latent) disease process is observed 12 . In cancer trials that use progression assessed via imaging or other tests, ignoring this interval censoring has been shown to result in sample size estimates that are up to 7.2% lower than required for the stated power 18 . Therefore, sample size estimation should consider interval censoring at the design stage, and MSM together with parametric transition models provide an appropriate method for accommodating it 10‐12 …”

Section: Introductionmentioning

confidence: 99%

“…Fifth, MSM can be used to assess the impact of frequency of assessments. A simulation study by Zeng et al explored efficiency gains due to increasing the frequency of patient assessments in an illness death model, concluding that, in their context, the gain in power was small in comparison to increasing the sample size 18 . Of note, their model constrained the well‐to‐disease progression and well‐to‐death transition intensities to be equal, with power estimates based on the effect of treatment on time to any transition out of the well state.…”

Section: Introductionmentioning

confidence: 99%

“…12 In cancer trials that use progression assessed via imaging or other tests, ignoring this interval censoring has been shown to result in sample size estimates that are up to 7.2% lower than required for the stated power. 18 Therefore, sample size estimation should consider interval censoring at the design stage, and MSM together with parametric transition models provide an appropriate method for accommodating it. [10][11][12] Third, in practice, due to administrative and patient-related events, assessments are not necessarily conducted at the same time point for all participants, and time intervals between assessments may vary.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Power and sample size for multistate model analysis of longitudinal discrete outcomes in disease prevention trials

Smith

Nixon

Sharples

2021

For clinical trials where participants pass through a number of discrete health states resulting in longitudinal measures over time, there are several potential primary estimands for the treatment effect. Incidence or time to a particular health state are commonly used outcomes but the choice of health state may not be obvious and these estimands do not make full use of the longitudinal assessments. Multistate models have been developed for some diseases and conditions with the purpose of understanding their natural history and have been used for secondary analysis to understand mechanisms of action of treatments. There is little published on the use of multistate models as the primary analysis method and potential implications on design features, such as assessment schedules. We illustrate methods via analysis of data from a motivating example; a Phase III clinical trial of pressure ulcer prevention strategies. We clarify some of the possible estimands that might be considered and we show, via a simulation study, that under some circumstances the sample size could be reduced by half using a multistate model based analysis, without adversely affecting the power of the trial.