Dynamic borrowing from a single prior data source using the conditional power prior.

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Bayesian Statistics

Background:

• The conditional power prior method allows borrowing information from a single prior data source.
• The power parameter, controlling information borrowing, is typically fixed before a new study, which can be problematic if study outcomes differ significantly from prior data.
• Treating the power parameter as a random variable may reduce control over borrowing, especially under regulatory guidance.

Purpose of the Study:

• To introduce modifications to existing conditional power prior methods for determining the power parameter.
• To limit potential operational bias by determining the power parameter using interim study data.
• To introduce a clinically relevant measure of similarity for assessing appropriate data borrowing.

Main Methods:

• Determining the power parameter based on the similarity between a percentage of current interim study outcome data and prior outcome data.
• Introducing a new similarity measure that incorporates a pre-specified clinical margin, defining a 'clinical similarity region'.
• Evaluating the approach through simulations to assess bias, power, and type I error rates.

Main Results:

• The proposed approach demonstrates low bias and good statistical power.
• The method effectively reduces the type I error rate in areas outside the defined 'similarity region'.
• The introduced clinical similarity region is understandable for clinicians assessing borrowing appropriateness.

Conclusions:

• The modified conditional power prior method offers a more flexible and clinically relevant approach to borrowing information from prior data sources.
• This method addresses limitations of fixed power parameters and uncontrolled random variable approaches.
• The approach has potential applications in practice, illustrated by a medical device study example.