TL09-14 Roundtable Discussions

Covariate-adjusted response-adaptive (CARA) designs use the available responses to skew the treatment allocation towards the treatment found to be best at an interim stage of a clinical trial, for a given patient's covariate profile. There has recently been extensive research on CARA designs with parametric distributional assumption on the patient responses. However, the range of application for such designs become limited in real clinical trials. Sverdlov,Rosenberger and Ryzenik (2013) has pointed out that irrespective of a specific parametric form of the survival outcomes, their proposed CARA designs based on the exponential model provide valid statistical inference, provided the final analysis is performed using the appropriate accelerated failure time (AFT) model. In real survival trials, however, the planned primary analysis is rarely conducted using an AFT model. The proposed CARA designs are developed obviating any distributional assumptions about the survival responses, relying only on the proportional hazards assumption between the two treatment arms. To meet the multiple experimental objectives of a clinical trial, the proposed designs are developed based on optimal allocation approach. The covariate-adjusted doubly-adaptive biased coin design and the covariate-adjusted efficient randomised adaptive design are used to randomise the patients to achieve the derived targets on expectation. These expected targets are functions of the Cox regression coefficients that are estimated sequentially with the arrival of every new patient into the trial. The merits of the proposed designs are validated using extensive simulation studies assessing their operating characteristics and has also been implemented to re-design a real-life confirmatory clinical trial.

Key Questions for Discussion :

1) Which Phases are such designs applicable in real clinical trials ?
2) How can this be extended for multi-arm platform trials for a phase 2 clinical trial ?
3) How are the derived optimal design methods are comparable to the Bayesian Optimal methods that are being used in designing adaptive phase 2 trials.
4) How well does the proposed methodology fits into the FDA's Complex Innovative Design Programme ?
5) How does the proposed designs help align with the primary analysis methods in survival trials and how it fits to the ICH E9 guidelines ?
6) Can the Estimand framework be implemented given that such a design is for confirmatory trials ?
7) Is an optimal design necessary for a real-life pivotal trial ? 

Roundtable Session Proposal

Title: "Derivations of real-world data (RWD) oncology endpoints: challenges and opportunities."

Roundtable Leader:
Ilana Trumble, PhD
303-817-4892
[email protected]

Proposal:

The role of real-world data (RWD) in the design and execution of clinical trials is ever increasing. Of note is the recent surge in the use of RWD to build external control arms, especially for rare disease or biomarker driven oncology trials where a perceived effective standard of care does not exist (Izem et al., 2022; Majumdar et al., 2022; Mishra-Kalyani et al., 2022). However, RWD can be useful more broadly to a clinical development program, beyond external controls. For example, RWD can be used to validate associations between short-term response-based endpoints and long-term survival endpoints. RWD can be used to assess endpoint differences between similar classes of drugs, for example, to differentiate between multiple PD(L)-1 agents. Such use cases require careful derivations of endpoints, often involving making clinically reasonable assumptions, a detailed understanding of the RWD patient experience, and knowing the limitations of translating the RWD experience to clinical trials. In this roundtable session, we will discuss challenges and opportunities specific to the use of RWD data, particular as related to endpoint derivations to inform clinical development in the Phase I to III oncology setting.

The roundtable discussion will benefit from the participation of statisticians from industry, regulatory agencies, and academia. Topics may include: challenges, tradeoffs, and solutions when defining clinical, pathological, and survival endpoints for RWD; challenges, tradeoffs, and solutions when defining I/E criteria for RWD; the use of RWD to justify novel oncology endpoints; regulatory agencies' considerations when assessing the use of RWD in oncology trial design and analysis; the importance of involving multidisciplinary teams in decision making and strategies for effectively collaborating with those teams; participants own experiences and advice pertaining to the use of RWD to inform oncology trials; and how RWD and these discussions have the potential to create positive impacts on global health.

About the Roundtable Leader:

Dr. Ilana Trumble is a Principal Statistician in the Oncology Biostatistics division of GSK. As a member of the Disease Area Strategy team, Dr. Trumble uses RWD to make trials more efficient and less risky for patients. Dr. Trumble obtained her degree in Biostatistics from the Colorado School of Public Health.

References

Izem, R., Buenconsejo, J., Davi, R. et al. Real-World Data as External Controls: Practical Experience from Notable Marketing Applications of New Therapies. Ther Innov Regul Sci 56, 704–716 (2022). https://doi.org/10.1007/s43441-022-00413-0

Majumdar, A., Davi, R., Bexon, M. et al. Building an External Control Arm for Development of a New Molecular Entity: An Application in a Recurrent Glioblastoma Trial for MDNA55. Stat Biosci 14, 285–303 (2022). https://doi.org/10.1007/s12561-022-09337-7

Mishra-Kalyani PS, Amiri Kordestani L, Rivera DR, Singh H, Ibrahim A, DeClaro RA, Shen Y, Tang S, Sridhara R, Kluetz PG, Concato J, Pazdur R, Beaver JA. External control arms in oncology: current use and future directions. Ann Oncol. 2022 Apr;33(4):376-383. https://doi.org/10.1016/j.annonc.2021.12.015. Epub 2022 Jan 10. PMID: 35026413. 

Abstract: Different drugs are eliminated from the body in various mechanisms which includes renal excretion. It has been shown that impaired renal functions can alter the drug's PK to an extent that a change in dosing regimen is required in many cases. So PK studies in renally impaired population has become an important research topic and integral part of dosing recommendations in labeling. FDA published a draft guidance on how to conduct renal impairment studies (September 2020).
The round table discussion will include:
Question 1: For a full PK study design, FDA recommends inclusion of subjects with kidney failure (eGFR <15 or dialysis patients on non-dialysis days). How will dialysis impact the study design?
Question 2: For the sample size calculation, FDA's example targets a 95% CI within 60% and 140% of the GM (geometric mean) of the PK parameters with at least 80% power. In practice, is there any flexibility for the range of the CI, and 90 vs 95% CI? Is there any practical criteria of different options that is allowed?
Question 3: For the data analysis part, FDA recommends a regression approach where estimated renal function and the PK parameters are treated as continuous variables. For those drug's PK impacted by hemodialysis (HD), should different slopes of eGFR be applied for HD vs other groups in the model where common slope doesn't fit well? 

The estimand framework enables trialists define causal effects in clinical trials in the presence of estimation challenges, including protocol deviation, noncompliance, and missing data. The benefits of the estimand framework have been emphasized in enabling trialists in defining a clear, precise causal question and identifying a strategy for answering the question in the presence of these challenges. However, estimand framework has been underemphasized in terms of its benefits in facilitating biometrics quality assurance and quality control. Additionally, some aspects of good biometrics leadership entail developing people and supporting them as they scale their activities and avoid learned helplessness, but estimand framework has been underemphasized in terms of it providing an opportunity for biometrics leadership. Therefore, this session will address the following questions:
1) What role does the estimand framework play in quality assurance and quality control?
2) How does the estimand framework affect response to questions during regulatory inspection?
3) Which biometrics standard operating procedure will likely retire as sponsors adopt the estimand framework?
4) How does the estimand framework serve as an opportunity for biometric leadership?
The session will be particularly beneficial for participants interested in learning about compliant, robust, lean, fit-for-purpose effective biometrics strategies. 

The ICH E9 (R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials was finalized in Nov 2019 and has been subsequently implemented by many regulatory agencies, including the FDA. While the ICH E9 (R1) suggests aligning trial design to the estimands that reflect trial objectives, sample size calculation, as a critical component in trial design, has not been fully updated to reflect the chosen estimands and the associated intercurrent event (ICE) strategies for many neuroscience clinical trials.

Our proposed roundtable discussion focuses on estimand-oriented sample size calculation for neuroscience clinical trials with ICEs. Neuroscience clinical trials share numerous common characteristics including high cost, high failure rate, and many operational challenges such as difficulty in collecting post-ICE efficacy follow up data and difficulty in estimating the pattern and extent of data missingness resulted from ICEs. While a few methods exist to adjust for the impact of ICEs in sample size calculation, these operational challenges often call for a more simulation intensive approach which provides the study sponsor and key stakeholders a better understanding of how different ICE strategies and missing data pattern/extent will affect the sample size and study design prior to study execution, thereby enhancing the quality and feasibility of the trial. A key point for discussion may therefore be the consideration of various missing data handling approaches as well as choice of estimators and corresponding analysis methods in the parametrization of such simulation-based methodology to better facilitate estimand-oriented sample size calculation.

This roundtable discussion is intended for audiences across regulatory agencies, pharmaceutical companies, and academia who are interested in further developing estimand-oriented sample size calculation methods in neuroscience clinical trials. 

Overall survival (OS) is both a safety and an efficacy endpoint. When other endpoints have been used for approval, the FDA has still required submission of OS data at the time of approval and/or as a post-marketing requirement after approval. When OS is not the primary endpoint, it has often been analyzed in a descriptive manner without formal statistical power calculations or Type I error control. In this setting, particularly in diseases with long natural histories, the number of events may be small at the time of analysis of the primary endpoint and there can be significant uncertainty regarding the estimates.

This round table discussion will focus on the following questions:
• What amount of OS information is feasible and useful for benefit-risk evaluation in diseases with a long natural history?
• When pre-specifying analyses for OS, when it is not the primary endpoint and there may or may not be planned formal statistical testing, what analyses would be useful to rule out harm?
• What should be prespecified regarding maturity of the OS data to adequately inform a benefit-risk assessment for regulatory evaluation?
• What safety and efficacy analyses of OS should be conducted post-hoc, especially if there were no pre-specified analyses to assess for harm?
• What would be considered "immature" data such that the observed HR could be considered an unreliable estimate?