Bayesian Integrated Learning of Longitudinal Dose-Response Relationships via Decentralized Clinical Trials.

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Pharmacometrics

Background:

• Decentralized clinical trials (DCTs) improve access and convenience but introduce data variability and bias concerns.
• Chronic conditions like diabetes and obesity require long-term studies, making DCTs particularly relevant.
• Analyzing dose-response relationships in DCTs with mixed data sources presents statistical challenges.

Purpose of the Study:

• To develop a novel Bayesian integrated learning procedure for analyzing dose-response relationships in DCTs.
• To address biases and uncertainties associated with decentralized data collection.
• To enable data-adaptive integration of information from both centralized and decentralized sources.

Main Methods:

• Generalization of a parametric exponential decay model to accommodate mixed data sources (centralized and decentralized).
• Application of Bayesian spike-and-slab priors to mitigate biases and uncertainties from decentralized measurements.
• Utilizing longitudinal data from a phase II DCT combining centralized and decentralized data collection.

Main Results:

• The proposed Bayesian approach demonstrated favorable performance across various simulation scenarios.
• The method achieved efficiency comparable to traditional trials when decentralized data had no additional error.
• Even with data variability, the approach showed lower bias and higher efficiency than naive pooling methods.

Conclusions:

• The novel Bayesian integrated learning procedure effectively analyzes dose-response relationships in DCTs with mixed data.
• This method provides a robust framework for handling data from both centralized and decentralized sources.
• The approach offers a statistically sound and efficient alternative for analyzing DCT data, particularly for chronic disease studies.