Podcast 61: Evaluating Analytic Models for IRGT Trials

This episode of the NIH Pragmatic Trials Collaboratory podcast explores the often-overlooked yet highly prevalent design of individually randomized group treatment trials (IRGTs), where individuals are randomized to treatment arms but receive interventions in group formats led by shared agents such as therapists, coaches, or healthcare providers. Co-hosts Adrian Hernandez and Patrick Haggerty are joined by NIH biostatisticians Jonathan Moyer and David Murray, who discuss their recent publication on analytic models for IRGTs with complex clustering. The hosts emphasize that IRGTs are at least as common as traditional RCTs—potentially even more so—but are rarely recognized as such in published research. The episode highlights critical methodological challenges: nested designs (especially partially nested) can inflate type I error rates due to unaccounted correlations, while cross-designs—where agents deliver interventions in both arms—can be highly advantageous if workloads are balanced, effectively allowing analysis like a standard RCT. The discussion underscores the importance of documenting agent involvement, ensuring sufficient numbers of independent agents for power, and using appropriate statistical models that account for clustering and multiple membership. The team also recommends the NIH's Research Methods Resources (RMR) website as a key tool for trial planning and sample size calculations. Key takeaways include: (1) Recognize IRGTs when they occur—many behavioral, psychological, and implementation trials fall into this category; (2) Prioritize balanced cross-designs when feasible to simplify analysis and avoid model complexity; (3) Always account for shared agents in both design and analysis to prevent inflated type I error; (4) Use the RMR website’s sample size calculator to explore 'what-if' scenarios; (5) Avoid single-agent designs due to poor power and generalizability; (6) In multiple membership settings, account for all relevant agents if exposure is evenly distributed; (7) Report agent numbers and roles transparently in publications; (8) Consider the fraction of time participants spend with multiple agents when choosing analytical approaches. The overall tone is constructive, urgent, and empowering, urging researchers to embrace the complexity of real-world trial designs with better tools and awareness.