ABSTRACT In medical contexts, especially in the diagnosis process of patients into different ordered stages of a disease (e.g., not diseased, early disease, and advanced disease) based on a continuous biomarker, the concept of the receiver operating characteristic (ROC) curve has been extended to that of the ROC surface and, consequently, the corresponding diagnostic performance measures as well. Specifically, the maximum absolute determinant (MADET) has been recently proposed, extending the classic concept of the Youden index. However, no method has been proposed for constructing confidence intervals (CIs) for the MADET index and its respective optimal cutoff points. In this work, we introduce an approach to construct CIs for these parameters based on the generalized pivotal quantity (GPQ) methodology assuming a trinormal model. A simulation study was conducted to evaluate the performance of the new CIs for the MADET index and its corresponding cut‐off points. The results indicate a good performance in terms of coverage probability and average width for small sample sizes, unbalanced designs and different values across the range of the MADET index. This methodology was also applied to a real dataset on Alzheimer disease, which provides several biomarkers within the normality assumption, requiring or not a Box‐Cox power transformation.
ABSTRACT When observation of recurrent events ends before the planned follow‐up time, conclusions may depend on untestable assumptions about outcomes during the unobserved period. This is especially challenging for recurrent outcomes such as hospitalizations or infections, for which each event creates a temporary period during which a new event cannot occur and at‐risk time stops accruing. In such settings, multiple imputation must jointly determine future event counts and future at‐risk exposure. This paper proposes an off‐risk‐aware extension of controlled multiple imputation that simulates post‐dropout trajectories in calendar time while allowing exposure to accrue only during on‐risk intervals. Standard assumptions, including missing at random and jump‐to‐reference, are implemented by modifying only the post‐dropout rate assignment while leaving the exposure‐generation mechanism unchanged. In simulation, missing‐at‐random imputation closely matched direct likelihood, whereas jump‐to‐reference progressively attenuated the treatment effect toward the null as missingness increased. In the rhDNase illustration, missing‐at‐random imputation again yielded results very close to direct likelihood, while jump‐to‐reference produced only modest attenuation. The method provides a practical sensitivity‐analysis framework for trials in which recurrent events temporarily remove participants from risk, helping align missing‐data assumptions, outcome generation, and analysis on the at‐risk time scale.
ABSTRACT When the number of patients is limited in clinical trials for rare diseases and pediatric developments, enrolling sufficient participants for confirmatory randomized clinical trials is difficult. Alternative trial designs using external controls have been proposed: augmented randomized clinical trials (ARCTs) and supplemented single‐arm trials (SSATs). In order to assess the favorable situations in which to implement each trial design, we propose to specify the sensitivity analysis for confounding at the planning phase, and call it “pre‐sensitivity analysis.” In addition, we assessed the probability that the pre‐sensitivity analysis would show a treatment effect, which we define as “power with confounders.” A simulation was carried out across multiple cases, which considered the number of participants from external control data sources and the randomization ratio in an ARCT. The simulation indicated that allocating relatively more participants to the control arm in an ARCT could be considered when the available historical data are equal to or less than the number of concurrent trial participants. Conversely, when the available historical data are sufficient, allocating more participants to the test arm in an SSAT or ARCT could be considered.
Sample size re-estimation based on interim data has become an increasingly important component of adaptive clinical trial designs. Many novel anti-cancer therapies under development demonstrate enhanced efficacy in biomarker-defined subpopulations rather than the overall population. To address the inherent uncertainty in the predictive value of biomarkers, biomarker-based adaptive designs have gained considerable interest. Given the complexity of these designs, it is critical to evaluate whether existing sample size re-estimation methods maintain adequate control of the overall Type I error rate. In this paper, we focus on the promising zone method applied within biomarker-based adaptive designs. We show that the promising zone method can produce Type I error inflation when embedded in biomarker-based adaptive designs, with the magnitude of inflation depending on the underlying true treatment effect in the biomarker-positive subgroup and specific design parameters such as the biomarker subgroup prevalence, the interim analysis timing, and the maximum inflation factor. We derive exact analytical results for normally distributed endpoints and extend these findings to time-to-event (TTE) endpoints through the asymptotic normal approximation of the log-rank test statistic, with the conclusions supported by simulation studies. We also discuss regulatory considerations for the implementation of such complex adaptive designs in confirmatory settings.
Stability assessment of vaccine drug products typically requires years of data collection. Accelerated stability studies offer a practical alternative by leveraging high-temperature data collected over shorter time periods. We model degradation using arbitrary nth order kinetics and derive a closed-form solution that enables extrapolation across time and temperature. We compare interval estimation methods including the delta method, bootstrap resampling, and Bayesian inference and propose a frequentist-Bayesian hybrid (FBH) posterior approximation framework based on a multivariate Student's t-distribution. FBH is closely related to objective Bayesian inference under noninformative priors in simple Gaussian settings and fully propagates uncertainty in both model parameters and residual variance. Simulation results show that FBH outperforms standard approaches by consistently achieving nominal coverage, particularly in small-sample settings where competing methods exhibit undercoverage. The approach provides well-calibrated confidence and prediction intervals while maintaining computational efficiency comparable to analytical methods and avoiding the cost of full posterior sampling. The methodology is illustrated using two vaccine development case studies and implemented in the AccelStab R package.
