Lead Investigator: Rifaquat Rahman, Dana Farber Cancer Institute
Title of Proposal Research: Evaluation of Differences in Trial and Non-Trial Patients and Leveraging of External Data for More Efficient Clinical Trial Designs in Newly Diagnosed Glioblastoma
Vivli Data Request: 6739
Funding Source: None
Potential Conflicts of Interest: None
Summary of the Proposed Research:
Glioblastoma (GBM) is an aggressive primary central nervous system tumor, and it is the most common primary malignant brain tumor in adults. There is a critical need for better therapies, as GBM continues to be associated with a dismal prognosis despite maximal therapy with surgery, radiation therapy and chemotherapy. While there have been some improvements in understanding the molecular pathways underlying GBM, there have been very few advances in the treatment of GBM in the last two decades.
In glioblastoma, the current clinical trial landscape is felt to be suboptimal for the development of new therapies, and there is a significant need for both better therapies and for a better system for testing and developing such therapies. Given these issues, glioblastoma has a dismal success rate in resource-intensive, time-consuming phase III trials with only one successful trial in the last several decades. Many failures in drug development in glioblastoma have been attributed to the overestimation of the treatment effect in early phase studies.
I. Trial Effects
While most oncologists believe that patients with cancer who enroll in clinical trials have better outcomes, scant evidence exists on whether there is an effect of trial participation itself on outcome (i.e. a trial effect). Of note, the largest limitation in the reviewed studies was inconsistent and insufficient strategies to control for potential confounders. There are broad efforts to improve clinical trial participation in GBM, but better understanding of patient characteristics, tumor-related factors and clinical outcomes among clinical trial and non-trial patients is important to provide insight into steps to increase this.
II. Incorporating External Data into Trial Designs
While early phase randomized controlled trials have been suggested as a possible solution, they require larger sample sizes, longer enrollment periods, and discourage patients who do not want to be randomized to a control arm There is increasing interest in using data from outside of clinical trials to replace or support controls within trials. The methodological development underpinning these “external control arms” is unclear, however. The rationale for this project is to develop an external control arm for clinical trials of newly diagnosed GBM and test the viability of such an approach given GBM-specific details such as endpoint variability in progression-free survival (PFS), overall survival (OS), and relevant prognostic factors.
This research can have a tremendous impact in the field of neuro-oncology. It will allow us to explore a type of trial design that could significantly accelerate the progress in glioblastoma to find more effective therapies faster and more efficiently. It will also help clinicians better understand how to interpret and contextualize clinical trials by comparing clinical trial patients with non-trial patients.
Statistical Analysis Plan:
General approach to design and evaluation of the externally controlled trial (ECT)
Our approach for designing an ECT, estimating sample size of targeted power and evaluating relevant operating characteristic is as follows:. First, we’ll define the patient population for the ECT for each trial setting in glioblastoma (post-chemoradiation and recurrence). Next, we will identify a set of prognostic factors associated with the outcome of interested and estimated the prevalence in the study population. Finally, we will specify the control therapy and identify available datasets (trials and RWD) using that therapy and extract relevant outcomes and patients’ characteristics.
As described below, to validate the ECT design, the control arm of each study is compared (using adjustment methods) to an external control that is defined by the remaining available data for patients that received the same control treatment. For the newly diagnosed GBM setting, we will use radiation therapy and temozolomide as the standard of care therapy. In these above comparisons the treatment effect will be zero by construction, which facilitates interpretability of the validation analyses and produce bias and variability summaries for ECT’s treatment effect estimates, and type I error rates estimates. If the ECT design maintains (approximately) the targeted type I error rate, we will determine the sample size required for ECT, single arm trial and RCT for a targeted probability of treatment discovery at a pre-defined treatment effect.
Terminology and Notation
The binary variable A indicates the assignment of a patient to the experimental treatment, A=1, or to the control arm, A=0, and Y denotes the outcome. We will focus on binary endpoints, such as survival at 12 months from enrollment (OS12) and expand the discussion to time-to-event outcomes. The vector X indicates a set of pre-treatment patient characteristics including known or likely prognostic factors. We will evaluate whether characteristics X are sufficient to obtain unbiased treatment effect estimates or not, and use Pr(Y=1│A=a,X) to indicate the probability of a positive response to treatment a given the pre-treatment characteristics X.
Externally Controlled Trial design
The ECT uses the study data and external data to conduct inference on treatment effects. More specifically, we estimate -for a hypothetical randomized study- the average treatment effect by averaging the conditional outcome probabilities with respect to a distribution Pr_X (x). Possible definitions for Pr_X (x) used by existing adjustment methods are the distribution of patient characteristics X in the single arm study, Pr_SAT (x), or the distribution of X in the external (historical) control, Pr_HC (x). We will initially consider four adjustment methods, all based on the usual hypothesis of no unmeasured confounders, to estimate average treatment effects: direct standardization, matching, inverse probability weighting and marginal structural methods. These methods use different reweighting schemes to obtain estimates of TE_Ave.
Model-free evaluation of the ECT design
We will also evaluate the ECT design by mimicking the comparison of an ineffective experimental arm to an external control. Hypothetical ECT experimental arms are generated from data of the control arm of one the included datasets. For each study, we will iterate the following three steps to generate ECTs:
We randomly select n patients (without replacement) from the control arm of the study and use the clinical profiles X and outcomes Y of these patients as experimental arms of the ECT. Here n is the number of enrolled patients. The control arms of the remaining studies are then used as the external control. We then estimate the treatment effect comparing the “experimental” arm and the external control using one of the candidate adjustment methods and test the null hypothesis of no-benefit at target type I error rate of 10%. We then repeat the steps several times with different sets of n randomly selected patients. For each study, we also conduct steps with n equal to the sample size of the control arm. A similar model-free evaluation allows one to evaluate the operating characteristics of ECTs in presence of positive treatment effects, by reclassifying in step (a) – randomly and with fixed probability – negative individual outcomes into positive.
Model-based evaluation of ECT
We also will a model-based approach to evaluate the ECT. Based the included studies, we will estimate a logistic model for the response to the control treatment arm given patient characteristics. A positive treatment effect (regression parameter) is added to specify a probability model Pr(Y|A=1,X) for an effective experimental treatment. We then will generate for each study a hypothetical ECT with effective experimental arm and fixed sample size n:
(a) Select n patient profiles X (with replacement) from the study and generate the corresponding outcomes Y from the probability Pr(Y|A=1,X) for the experimental arm.
(b) Randomly select N patient profiles X from the remaining studies (where N is the sample size of external control data) and generate outcomes Y from the probability Pr(Y|A=0,X) for control treatment.
(c) Conduct a covariate-adjusted ECT test using “experimental” arm data (step a) and the external control (step b).
We repeat steps (a-c) 10,000 times and compute for each study the proportion of ECT tests that reject the null hypothesis at α=0.1. By repeating this calculation over a grid of sample sizes we determine the smallest size that achieve an 80% power.
Comparison, ECT versus single-arm and RCT designs
We will compare the ECT against a single-arm trial and RCT designs in the setting of new diagnosis and post-chemoradiation glioblastoma to evaluate if the ECT design can provide unbiased treatment effects estimates. We used the following criteria to compare designs:
(a) bias and variability of treatment effect estimates,
(b) deviations of Type I error rates from targeted control of false positive results, and
(c) the sample size to achieve a targeted power.
Requested Studies:
A Randomized, Double-Blind, Placebo-Controlled, Multicenter Phase III Trial of Bevacizumab, Temozolomide and Radiotherapy, Followed by Bevacizumab and Temozolomide Versus Placebo, Temozolomide and Radiotherapy Followed by Placebo and Temozolomide in Patients With Newly Diagnosed Glioblastoma
Data Contributor: Roche
Study ID: NCT00943826
Sponsor ID: BO21990
Cilengitide in Subjects With Newly Diagnosed Glioblastoma and Unmethylated MGMT Gene Promoter – a Multicenter, Open-label Phase II Study, Investigating Two Cilengitide Regimens in Combination With Standard Treatment (Temozolomide With Concomitant Radiation Therapy, Followed by Temozolomide Maintenance Therapy). [The CORE Study]
Data Contributor: Project Data Sphere
Study ID: NCT00813943
Sponsor ID: EMD121974-012
Cilengitide for Subjects With Newly Diagnosed Glioblastoma and Methylated MGMT Gene Promoter – A Multicenter, Open-label, Controlled Phase III Study, Testing Cilengitide in Combination With Standard Treatment (Temozolomide With Concomitant Radiation Therapy, Followed by Temozolomide Maintenance Therapy) Versus Standard Treatment Alone (CENTRIC)
Data Contributor: Project Data Sphere
Study ID: NCT00689221
Sponsor ID: EMD 121974-011
Public Disclosure:
Rahman, R., Ventz, S., Redd, R., Fell, G., Tan, Y., Orio, P., Tanner, K., Wen, P.Y. and Trippa, L., 2025. Identifying appropriate external control datasets in support of future glioblastoma clinical trials leveraging external data. Neuro-Oncology, p.noaf031. Doi: 10.1093/neuonc/noaf031