Journal of Biopharmaceutical Statistics最新文献

This paper introduces a methodology for processing continuous monitoring device data, such as data from a wearable digital device or continuous telemetered data, to estimate outcomes like systolic blood pressure or treatment effects. One of the challenges of analyzing this type of data is to find a suitable binning or scaling to compress the information for improving outcome predictions. Another challenge is to select and weight the features to be included in the computational model. The new methodology consists of a combination of feature selection and feature weighting incorporated into the LASSO and the elastic net methods, which addresses both issues simultaneously. The compression of continuous data into weighted discretized data is a prominent issue in the development of AI methodology that is applied to wearable DHT devices. The new methodology was applied to a Fitbit data set from a Hong Kong elderly center study.

In oncology dose-finding trials, small cohorts of patients are often assigned to increasing dose levels, with the aim of determining the maximum tolerated dose. In the era of targeted agents, this practice has come under intense scrutiny as treating patients at doses beyond a certain level often results in increased off-target toxicity without significant gains in antitumor activity. Dose optimization for targeted agents becomes more challenging in proof-of-concept trials when the experimental treatment is tested in multiple indications of low prevalence and there is the need to characterize the dose-response relationship in each indication. To provide an alternative to the conventional "more is better" paradigm in oncology dose finding, we propose a Bayesian model averaging approach based on robust mixture priors (rBMA) for identifying the recommended phase III dose in randomized dose optimization studies conducted simultaneously in multiple indications. Compared to the dose optimization strategy which evaluates the dose-response relationship in each indication independently, we demonstrate the proposed approach can improve the accuracy of dose recommendation by learning across indications. The performance of the proposed approach in making the correct dose recommendation is examined based on systematic simulation studies.

In recent years, the role of the p-value in applied research has been heavily scrutinized. Several new proposals have been put forward from a Bayesian viewpoint, including the analysis of credibility. By using the reverse Bayes theorem, and reasoning in terms of subverting the significance or the non-significance denoted by the p-value, this analysis provides the credibility, in a Bayesian sense, of an experimental result. We discuss a normalized indicator of credibility, namely $C$, a variant of the index $\tilde{C}$ (Quatto et al. J. Biopharm. Stat. 32, 308-329, 2022). This can be used to assess the degree of credibility of experimental results and can also be compared with a fixed threshold. The index is extended to the case of one-sided hypotheses. A simulation study is conducted to empirically assess the behavior of the index $C$. Two illustrative examples in the contexts of pharmacotherapy for COVID-19 and heart failure are presented. We then propose adopting the credibility index for meta-analyses, in which it can provide a suitable diagnostic value for modeling fixed and random effects.

Today's challenges around global healthcare emphasize the need for large-scale collaborations between the clinical and sciesntific communities. However, regulatory constraints around data sharing and patient privacy might hinder access to data genuinely representing clinically relevant patient populations. We have developed an open-source federated learning framework, NVIDIA FLARE, to work around such restrictions while maintaining patient privacy using modern cryptographic and information-theoretic methods such as homomorphic encryption and differential privacy. In this work, we show how NVIDIA FLARE addresses clinical questions, such as predicting clinical outcomes in patients with COVID-19 and other real-world applications, including federated statistics and parameter-efficient adaptation of large language models under a collaborative setting, while allowing participants to retain governance over their data.

With the continuous advancement of medical treatments, there is an increasing demand for clinical trial designs and analyses using cure rate models to accommodate a plateau in the survival curve. This is especially pertinent in oncology, where high proportions of patients, such as those with melanoma, lung cancer, and endometrial cancer, exhibit usual life spans post-cancer detection. A Bayesian clinical trial design methodology for multivariate time-to-event outcomes with cured fractions is developed. This approach employs a copula to jointly model the multivariate time-to-event outcomes. We propose a model that uses a Gaussian copula on the population survival function, irrespective of cure status. The minimum sample size required to achieve high statistical power while maintaining reasonable control over the type I error rate from a Bayesian perspective is identified using point-mass sampling priors. The methodology is demonstrated in simulation studies inspired by an endometrial cancer trial.

In the field of medicine, evaluating the diagnostic performance of new diagnostic methods can be challenging, especially in the absence of a gold standard. This study proposes a methodology for assessing the performance of diagnostic tests by estimating the posterior distribution of the $F_1$ score using latent class analysis, without relying on a gold standard. The proposed method utilizes Markov Chain Monte Carlo sampling to estimate the posterior distribution of the $F_1$ score, enabling a comprehensive evaluation of diagnostic test methods. By applying this method to internet addiction, we demonstrate how latent class analysis can be effectively used to assess diagnostic performance, offering a practical solution for situations where no gold standard is available. The effectiveness of the proposed approach was evaluated through simulation studies by examining the coverage probability of the 95% highest density interval of the estimated posterior distributions.

During randomized controlled trials, it is critical to remain vigilant in safety monitoring. A common approach is to present information over time, such as frequency tables and graphs, when analyzing adverse events. Nevertheless, there is still a need for developing statistical methods for analyzing safety data of a dynamic nature. The process is typically challenging due to small sample sizes, a lack of observational data sources, difficulties in false-positive control, and the necessity for early detection of serious adverse events. In this article, we propose a simple and effective framework called Bayesian Efficient sAfety Monitoring (BEAM) to analyze evidence aggregation of potentially serious adverse events that may arise during the trial, as well as a timeline for when concrete evidence for safety concerns of unlikely outcomes becomes available. BEAM can be easily tabulated and visualized before the trial starts, making evaluations transparent and easy to use in practice, while maintaining flexibility when the underlying adverse event rate varies. Simulation studies have shown that BEAM supports continuous monitoring, can incorporate external information, and demonstrates good operating characteristics across various scenarios. In most practical situations, it has a reasonable likelihood of detecting elevated risks and identifying safety signals early on when safety concerns arise regarding the investigational drug.

Biomarkers are measured repeatedly in clinical studies until a pre-defined endpoint, such as death from certain causes, is reached. Such repeated measurements may present a dynamic process for understanding when to expect the study's endpoint. Joint modelling is often employed to handle such a model. Typically, shared random effects are assumed to be common to both the longitudinal component and the study's endpoint. These shared random effects usually assume homogeneous and follow a normal distribution. However, identifying homogeneous subgroups is important when the underlying population is heterogeneous. This issue has received little attention in the literature, particularly for multi-phase longitudinal responses. In this paper, we propose a joint modelling approach for longitudinal and survival models using a bent-cable mixed model for longitudinal measurements and a Weibull distribution for the survival component. We also incorporate a finite mixture of normal distribution assumptions to account for the unobserved heterogeneity in the shared random effects model. A Bayesian MCMC is developed for parameter estimation and inferences. The proposed method is evaluated using simulation studies and the Tehran Lipid and Glucose Study dataset.

Randomized clinical trials (RCTs) can benefit from using Real-World Data (RWD) as a supplementary data source to enhance their analysis. An Augmented RCT combines randomized treatment and control groups with synthetic controls derived from RWD. This way, the trial can achieve less prospective enrollment, higher statistical power, and lower costs. However, to ensure scientific validity, the synthetic controls must satisfy the same eligibility criteria as the trial participants. A major challenge is that RWD often have missing data that hinder the eligibility assessment. This problem has been overlooked in the literature and this paper offers statistical solutions to address it. We use multiple imputations to handle missing data in the variables involved in the eligibility criteria. We also propose a generalized propensity score weighting procedure to adjust for the life expectancy requirement, a common eligibility criterion in oncology clinical trials but usually unavailable in RWD. Since the life expectancy is an unmeasured confounder, we discuss the statistical assumptions required to correct its bias. We validate the proposed solutions through simulation studies and the analysis of an Augmented RCT in oncology.