Design and Analysis of Clinical Trials by Integrating Real-World Data for Better Decision Making in Drug and Device Development

It is common to have a treatment-predictive continuous biomarker with unknown cutoff points so that the biomarker-positive subgroup that benefits most from the new treatment cannot be determined at the time of designing the biomarker-guided clinical trial. Biomarker-guided multi-stage design is often preferred as it allows adaptively identifying biomarker threshold that defines biomarker-positive subgroup and flexible patient enrichment in the late stage of the ongoing trial. In this talk, we will focus on designing biomarker-guided multi-stage adaptive trials with threshold detection, patient enrichment, and possible information borrowing from historical controls. We will discuss algorithms that adaptively identify optimal biomarker threshold to define the patient subgroup that benefits most from the new treatment, decision-making for patient enrichment for better efficiency, and information borrowing for threshold detection and treatment effect inference from historical controls that are subject to measured or unmeasured baseline confounders. The operating characteristics of the proposed design with competing designs are evaluated by extensive simulation.  

In comparative effectiveness research (CER) for rare types of cancer, it is appealing to combine primary cohort data containing detailed tumor profiles together with aggregate information derived from cancer registry databases. Such integration of data may improve statistical efficiency in CER. A major challenge in combining information from different resources, however, is that the aggregate information from the cancer registry databases could be incomparable with the primary cohort data, which are often collected from a single cancer center or a clinical trial. We develop an adaptive estimation procedure, which uses the combined information to determine the degree of information borrowing from the aggregate data of the external resource. The proposed method yields a substantial gain in statistical efficiency over the conventional method using the primary cohort only, and avoids undesirable biases when the given external information is incomparable to the primary cohort. We apply the proposed method to evaluate the long-term effect of trimodality treatment inflammatory breast cancer by tumor subtypes, while combining the IBC patient cohort at MD Anderson and external information. 

This presentation provides an overview of different types of external controls and their unique features when used in master protocols. Some key considerations in master protocols with external controls are discussed including construction of estimands, assessment of fit-for-use real-world data, and considerations for different types of master protocols. Similarities and differences between regular randomized controlled trials and master protocols when using external controls are discussed. A targeted learning-based causal roadmap is presented which constitutes three key steps: (1) define a target statistical estimand that aligns with the causal estimand for the study objective, (2) use an efficient estimator to estimate the target statistical estimand and its uncertainty, and (3) evaluate the impact of causal assumptions on the study conclusion by performing sensitivity analyses. Two illustrative examples for master protocols using external controls are discussed for their merits and possible improvement in causal effect estimation. 

Randomized controlled trials augmented by external controls involve two distinct sampling processes. First, the clinical trial or intervention arm is sampled from a population defined by the clinical and operational characteristics of the study design, such as the enrollment period, choice of enrolling sites, site enrollment rate, trial inclusion criteria, and patient consent. Second, an external control data set arises from a distinct sampling process. When samples from these two populations are combined for analysis, the covariate distributions are mixed. The mixing proportion may be proportional to the contributing sample sizes or dependent on the choice of balancing weights, thus giving rise to different target populations. We define relevant estimands for studies that augment clinical trials with real-world data using balancing weights. We highlight the advantages and disadvantages of different estimands with respect to interpretation. Finally, we compare alternative estimators through simulation and in application to an augmented clinical trial of idiopathic pulmonary fibrosis.