LEAP: the latent exchangeability prior for borrowing information from historical data

Area of Science:

• Biostatistics
• Statistical Modeling
• Clinical Trial Design

Background:

• Eliciting informative priors from historical data is increasingly popular in statistical analysis.
• Existing methods like power prior, commensurate prior, and robust meta-analytic predictive prior offer blanket discounting, which may be inappropriate when only a subset of historical data is exchangeable with current data.
• Propensity score approaches address covariate distribution but not outcome-based exchangeability.

Purpose of the Study:

• To introduce the Latent Exchangeability Prior (LEAP) for more appropriate use of historical data.
• To address the limitations of existing priors when historical data is not fully exchangeable with current data.
• To improve the augmentation of control arms in clinical trials, especially those with unbalanced randomization.

Main Methods:

• The Latent Exchangeability Prior (LEAP) classifies historical data observations into exchangeable and non-exchangeable groups.
• LEAP discounts historical data by identifying the most relevant subjects.
• The approach was compared against alternatives via simulations and applied to a phase 3 clinical trial in plaque psoriasis.

Main Results:

• The proposed LEAP approach offers a more refined method for utilizing historical data compared to blanket discounting.
• LEAP effectively identifies and discounts non-exchangeable data, improving prior elicitation.
• The case study demonstrated LEAP's utility in augmenting a control arm with an unbalanced randomization scheme.

Conclusions:

• The Latent Exchangeability Prior (LEAP) provides a flexible and effective method for incorporating historical data when exchangeability is partial.
• LEAP enhances statistical modeling by selectively leveraging relevant historical information.
• This method has significant implications for clinical trial design and analysis, particularly in augmenting control arms.