Variability in Causal Effects and Noncompliance in a Multisite Trial: A Bivariate Hierarchical Generalized Random...

Area of Science:

• Biostatistics
• Health Services Research
• Public Health

Background:

• Colorectal cancer screening (CRCS) rates can be improved with targeted interventions.
• Online tools show promise in increasing patient engagement with preventative health measures.
• Understanding causal effects of interventions and their variation across healthcare providers is crucial.

Purpose of the Study:

• To estimate the average causal effect and complier average causal effect of an e-assist tool on CRCS.
• To analyze the variation of these causal effects across 170 physicians.
• To develop a statistical methodology for estimating physician-specific screening probabilities with partially observed compliance.

Main Methods:

• Randomized trial comparing an e-assist tool to a control group for CRCS.
• Maximum likelihood estimation incorporating random effects to model physician-specific probabilities.
• Adaptive Gauss-Hermite quadrature for likelihood marginalization.
• Factor analytic structure for shared random effects to reduce dimensionality.

Main Results:

• The study developed a novel statistical approach to estimate causal effects in the presence of partial compliance and physician-level variation.
• The methodology successfully addressed challenges in jointly estimating physician-specific screening probabilities.
• Simulation studies informed recommendations for optimal sample sizes for accurate and precise estimation.

Conclusions:

• The developed statistical framework provides a robust method for analyzing interventions with partial compliance and hierarchical data structures.
• The findings contribute to understanding the effectiveness of online tools in promoting colorectal cancer screening.
• The research offers guidance on study design, including sample size recommendations, for similar future interventions.