Contributed Poster Presentations: International Chinese Statistical Association

Randomized control trials (RCTs) are considered as gold standard, encounter challenges due to insufficient sample size. Single-arm trials provide an approach in such circumstance, but its outcomes might be biased. The increasing application of RWD has drawn attention to the study of hybrid RCT, which features a control arm that integrates concurrent control and external control. Our article introduces an adaptive two-stage design, consisting of a single-arm trial stage and a hybrid RCT stage. Researchers allow to assess treatment efficacy and futility or decide continuation to hybrid RCT at interim analysis. Prior to conducting hybrid RCT, two steps are needed: 1) Sample size re-estimation adjust the number of subjects in treatment arm; 2) Propensity score matching are applied to identify well-matched subjects between external control and treatment arm. Concurrent control subjects are incorporated to supplement control arm when external control does not match requirement. Our approach offers a promising alternative to RCTs and single-arm trials by adjusting design after interim analysis, which enhances the efficiency and flexibility, especially rare diseases and pediatric trials. 

Molecular heterogeneity is a hallmark of cancer. Network-based analysis can be more informative than that based on simpler statistics. Downstream molecular measurements, such as protein expression, are highly regulated by gene expression and also show heterogeneous patterns across populations in regulation. Incorporating gene expression and gene regulation can better delineate the "source" of molecular heterogeneity. Gene expression networks typically exhibit a block structure, where correlations within blocks are stronger than those between blocks. This block-wise organization extends to regulatory patterns as well. In this work, we propose a novel heterogeneity analysis framework based on gene expression network and gene regulation accounting for block structures among genes. This approach can simultaneously identify sample subgroups, gene block structures, and subgroup-specific networks and regulatory mechanisms. An effective computational algorithm and theoretical properties are provided. In the analysis of cancer datasets, the proposed approach identifies heterogeneity and molecular characteristics different from the alternatives and with sound biological implication. 

The joint model offers a strategy to simultaneously incorporate latent associations between different types of outcomes. Recent Bayesian approaches have enhanced the flexibility and application of these models. However, there is a lack of Bayesian joint models for multilevel hierarchical data that include both longitudinal and binary outcomes. We aim to propose such a prognostic joint model for timely intervention before pulmonary exacerbation during CF progression. Additionally, we will demonstrate the biases that can arise if center effects are ignored in multicenter data.

Reference: Zhou GC, Song S, Szczesniak RD. Multilevel joint model of longitudinal continuous and binary outcomes for hierarchically structured data. Statistics in Medicine. 2023; 42(17): 2914–2927. doi: 10.1002/sim.9758