There is growing interest in a hybrid control design in which a randomized controlled trial is augmented with an external control arm from a previous trial or real world data. Existing methods for analyzing hybrid control studies include various downweighting and propensity score methods as well as methods that combine downweighting with propensity score stratification. In this article, we describe and discuss methods that make use of an outcome regression model (possibly in addition to a propensity score model). Specifically, we consider an augmentation method, a G-computation method, and a weighted regression method, and note that the three methods provide different bias-variance tradeoffs. The methods are compared with each other and with existing methods in a simulation study. Simulation results indicate that weighted regression compares favorably with the other model-based methods that seek to improve efficiency by incorporating external control data. The methods are illustrated using two examples from urology and infectious disease.

The oncology phase I clinical trials aim to identify the optimal dose as the recommended phase II dose (RP2D) rather than relying on the maximum tolerated dose (MTD). This approach is preferred because novel cancer drugs often exhibit an efficacy plateau, where lower dose can achieve similar efficacy to higher dose but maintaining a favorable safety profile. With objectives to compare the accuracy of selecting true RP2D after including backfill patients and to evaluate impacts from patient allocations, we designed simulation studies including a dose escalation phase using 3 + 3 design or BOIN design, followed by a backfill phase with different strategies and with an option to further include an expansion cohort. Results based on backfill strategies across comprehensive scenarios were discussed and recommendations were provided.

In a master protocol design, multiple substudies are employed to address various research questions, with each substudy potentially having common control arms. When analyzing a substudy of interest, borrowing a control arm from another substudy may enhance the precision of the estimates. Several Bayesian methodologies have been proposed for borrowing data from other studies, particularly in the context of historical control borrowing. These have operating characteristics according to which the closer the results are between current and historical studies, the larger the amount of data borrowed from the historical study. However, such approaches often require a diagnostic assessment of convergence in Markov chain Monte Carlo methods, and pre-specification of hyperparameters, which would not be feasible for not only the primary endpoint but also the secondary and safety endpoints. The shrinkage of the difference parameters between substudies is equivalent to partial borrowing from other substudies. In this study, we propose using ridge estimator for shrinkage, in which the tuning parameter is determined based on the observed data. The ridge estimator has a closed form given the tuning parameters, and the tuning parameters can be easily computed. A simulation study demonstrated that the proposed estimator is safe and exhibits conservative performance. Supplementary materials for this article are available online.

Seamless two-stage adaptive designs with dose selection have been recognized for maximizing efficiency, minimizing resource utilization, and accelerating the drug development process. In situations where the primary endpoint takes a long time to mature, patients may have incomplete follow-up for the primary endpoint at the interim analysis for dose selection. For ethical and enrollment considerations, it is desirable to switch patients already enrolled with partial follow-up to the selected dose after the interim. However, it has been perceived in the existing literature that dose switching for seamless two-stage adaptive designs would undermine the validity of the methods and inflate the Type I error rate. In this article, we investigate such designs under dose switching and show that general combination methods with the closed testing procedure can provide strong control of the Type I error rates under realistic assumptions. Simulation studies were conducted to confirm the validity and power gains of seamless two-stage adaptive designs when dose switching is allowed.

We derive a statistical test for the dissolution-profile equivalence of two batches using samples from multivariate normal distributions where the variables correspond to time-points. We refer to this test as the general-t equivalence test. Using Monte Carlo simulation results, we determine that, for many realistic covariance matrices and realistic small-sample-size configurations, our general-t equivalence test for two dissolution profiles is more powerful than the conditional-t equivalence test derived by Saranadasa and Krishnamoorthy and therefore yields better statistically-based dissolution-equivalence decisions. Also, unlike the T2EQ procedure proposed by Hoffelder, the general-t equivalence test maintains the nominal Type I error rate regardless of the data-dimension and sample sizes within each batch. In addition, we derive a general-t equivalence confidence interval for evaluating dissolution-profile equivalence that, for realistic small-sample-size configurations, is generally more precise than the conditional-t equivalence confidence interval derived by Saranadasa and Krishnamoorthy. We demonstrate the efficacy of the general-t equivalence test and confidence interval on two real datasets. An R implementation of the proposed method is offered in the supplementary material.

Quantile difference is a critical measure for summarizing the random fluctuations of an outcome and has gained much attention. However, existing methods for quantile difference encounter challenges with general biased samples, which are common in fields such as economics, epidemiology, and medical studies. We propose two estimation approaches for quantile difference in survival data under general biased samples: a non-smoothed estimator and a more efficient smoothed estimator. The asymptotic properties of the proposed estimators are derived. Simulation studies are conducted, and two real data examples are analyzed.

Clinical trials with multiple endpoints often use prespecified weights to allocate the overall significance level unequally, reflecting the clinical importance of each endpoint, the probability of observing a treatment effect, or other considerations. To address the subjective nature of weight selection, we propose a quantitative approach where the optimal significance level allocation comes with the minimum sample size. Moreover, this innovative approach was specifically tailored and applied to weighted Hochberg-type procedures for two hypotheses, filling the existing gap in sample size optimization methods for these procedures. In addition, we propose a new Hochberg-type procedure with weights, referred to as the improved trimmed weighted Hochberg procedure, which provides increased statistical power and relaxes the dependence assumptions for familywise error rate control compared to the original weighted Hochberg procedure. Several examples and applications are provided to illustrate the methodology.

The International Council for Harmonization E14 guidance, Clinical Evaluation of QT/QTc Interval Prolongation and Proarrhythmic Potential for Non-Antiarrhythmic Drugs, recommends a thorough QT study as part of clinical drug development to investigate the potential of therapeutic drugs to delay cardiac repolarization. Moxifloxacin is often used as the positive control to demonstrate that the assay is sensitive in detecting QT effects. We performed a meta-analysis of moxifloxacin data to characterize the exposure-response relationship and determine the range of parameters related to assay sensitivity for moxifloxacin. The data are from 67 thorough QT studies submitted to FDA. We applied meta-analysis methods to evaluate the exposure-response relationship and estimated the parameters of two concentration-QTc models. We analyzed individual and pooled studies and demonstrated the homogeneity and heterogeneity of the moxifloxacin concentration-QTc relationships among the studies. We determined the reference range of study parameters that demonstrate assay sensitivity. The estimated slopes for individual studies ranged from 2.6 to 7.6 milliseconds per μ g/mL (95% tolerance interval for 90% of slope was 2.8 to 6.8 milliseconds per μ g/mL). The estimated treatment specific intercept for individual studies ranged from –2.1 to 7.3 milliseconds (95% tolerance interval for 90% of treatment specific intercept was –0.8 to 5.0 milliseconds).