Design and analysis of partially randomized preference trials with propensity score stratification

Area of Science:

• Clinical Trial Design
• Biostatistics
• Health Services Research

Background:

• Standard two-stage randomized designs may face challenges in clinical practice due to patient non-compliance with randomization.
• Partially randomized preference designs (PRPD) offer a pragmatic alternative but can introduce bias in effect estimation.
• Patient treatment preferences can significantly impact study outcomes, necessitating methods to account for this.

Purpose of the Study:

• To apply propensity score stratification (PSS) within a PRPD framework to mitigate bias.
• To enable unbiased estimation of overall treatment, selection, and preference effects.
• To derive and validate sample size formulas for the PSS-PRPD.

Main Methods:

• Utilized propensity score stratification (PSS) to create a conditional first-stage randomization within the PRPD.
• Derived closed-form sample size formulas for estimating treatment, selection, and preference effects.
• Conducted simulation studies to assess bias reduction and validate sample size formulas.

Main Results:

• PSS effectively reduces bias in effect estimates within the PRPD.
• The proposed sample size formulas accurately predict required sample sizes.
• 5 to 10 propensity score strata are generally sufficient for bias correction, with optimal numbers depending on effect heterogeneity.

Conclusions:

• PSS-PRPD provides a robust method for unbiasedly estimating treatment, selection, and preference effects in pragmatic clinical trials.
• The derived sample size formulas are valuable tools for planning future studies using this design.
• The methodology was successfully demonstrated using data from the Harapan Study.