A Bayesian Hierarchical Modelling Approach for Indirect Comparison of Response Outcomes in Histology-Independent Therapies


Mackay E1, Springford A1, Nagamuthu C1, Dron L2
1Cytel, Toronto, ON, Canada, 2Cytel, Vancouver, BC, Canada

Presentation Documents

OBJECTIVES: Development of new cancer therapies targeting rare mutations is resulting in increased use of basket trial designs for studying effectiveness of histology-independent therapies (HIT). This trend presents challenges for health technology assessments (HTA) as basket trial sample sizes tend to be limited within each histology, and response rates may vary by histology. Murphy et al. (2021) have proposed the use of a Bayesian hierarchical modelling (BHM) approach to account for heterogeneity in response rates across histologies. We extend this BHM model to allow for comparison of response outcomes between two basket trials, relying on aggregate-level data.

METHODS: We develop a BHM model for indirectly comparing treatment effects of HITs for a response endpoint. This method requires response data by histology for two basket trials evaluating HITs with a common target mutation, as well as sufficient overlap in histologies included. The model relies on several key assumptions: (i) the relative treatment effect is constant across histologies, (ii) histology-specific response outcomes satisfy a conditional exchangeability requirement, and (iii) patient characteristics within each histology do not systematically differ between the two trials. We demonstrate the method using simulated data with 100 patients per trial and differing distributions across 12 prognostically important histologies in each trial.

RESULTS: Among 500 simulated datasets, the 95% credible interval captured the true treatment effect 91.6% of the time versus 59.4% for the pooled estimate. However, MCMC convergence was not achieved for 1.2% of cases and, despite reduction in bias, relative treatment effects still tended to be underestimated.

CONCLUSIONS: In the presence of response heterogeneity across histologies, the proposed BHM model was able to reduce bias in indirect comparisons for HITs relative to pooled comparisons. When paired with careful selection of priors and sensitivity analysis, the method may facilitate more balanced comparisons of HITs between basket trials.

Conference/Value in Health Info

2022-11, ISPOR Europe 2022, Vienna, Austria

Value in Health, Volume 25, Issue 12S (December 2022)




Clinical Outcomes, Methodological & Statistical Research, Study Approaches

Topic Subcategory

Comparative Effectiveness or Efficacy, Confounding, Selection Bias Correction, Causal Inference, Meta-Analysis & Indirect Comparisons


STA: Personalized & Precision Medicine

Your browser is out-of-date

ISPOR recommends that you update your browser for more security, speed and the best experience on ispor.org. Update my browser now