Targeting What Matters in Pediatric Trials: Bayesian Basket and Umbrella Trials for Heterogeneous Airway Disease

Abstract

Introduction
Children with chronic inflammatory airway disease often present with similar signs and symptoms despite markedly different underlying biology. Nasal cytokine profiling routinely reveals five endotypes: eosinophilic, IL-5–high, IL-13–high, mixed, and pauci-inflammatory, each associated with differential response to targeted therapies with biologics. In trials with small sample sizes and challenging long-term spirometric follow-up, early change in small-airway resistance (ΔR5–R20 at 14 days) is a common surrogate outcome. Traditional parallel-group RCTs struggle to capture this heterogeneity. To address this, we developed a Bayesian basket-and-umbrella trial framework tailored to the pediatric context and evaluated its performance through an extensive simulation study.

Objective

To develop and evaluate Bayesian basket and umbrella clinical trial designs for pediatric chronic inflammatory airway disease, assessing their ability to improve the efficiency, precision, and reliability of treatment-effect estimation across cytokine-defined endotypes under realistic pediatric trial conditions

Methods
Two designs were constructed around the same motivating example. In the basket configuration, one biologic agent was tested across the five cytokine-defined endotypes. In the umbrella configuration, three mechanism-specific agents, anti-IL-5, anti-IL-13, and dual-pathway modulation, were compared within a shared clinical population, with treatment assignment informed by cytokine signatures at enrollment.
Hierarchical Bayesian models were used to estimate subgroup-specific treatment effects, allowing structured information borrowing across biologically related endotypes. Hyperparameters were elicited from pediatric pulmonology experts to reflect realistic variability, and weakly informative priors ensured interpretability. Posterior probabilities guided interim rules for early stopping (superiority/futility), adaptive randomization, and dynamic modification of subgroup enrollment.
A simulation study of 10,000 trials per scenario incorporated constraints typical of pediatric research: unequal subgroup prevalence, rare strata (<5%), biomarker misclassification, early dropout due to symptom flares, and heterogeneous effect sizes. Operating characteristics included posterior power, false-positive/false-negative rates, precision of subgroup effect estimation, probability of correct endotype identification, expected sample size, and time to decision.

Results
Across most clinically plausible scenarios, Bayesian basket and umbrella designs outperformed classical RCTs. Precision in treatment-effect estimation improved by 20–50% in rare endotypes, and adaptive randomization reduced expected sample size by 15–30% while maintaining strict control of false discoveries. Posterior decision rules enabled earlier identification of beneficial therapies without compromising robustness. When substantial biomarker misclassification was introduced, hierarchical models automatically reduced inappropriate borrowing across endotypes, preventing misleading shrinkage and preserving inferential reliability. Clinically, the Bayesian designs provided sharper, endotype-specific therapeutic signals than achievable with parallel-group trials of comparable size.

Conclusion
Starting with a realistic pediatric airway disease example, this simulation study shows that Bayesian basket and umbrella trials offer a flexible, efficient, and biologically aligned approach for evaluating targeted therapies in children. By integrating mechanistic knowledge, hierarchical modeling, and adaptive decision rules, these designs address key limitations of conventional RCTs, small samples, heterogeneity, and operational constraints, while delivering clinically interpretable subgroup insights. Their use may substantially accelerate treatment development in pediatric settings where traditional designs are underpowered or insufficiently granular.