Flexibly modeling the baseline risk in meta-analysis

Area of Science:

• Biostatistics
• Clinical Trials
• Statistical Modeling

Background:

• Meta-analysis in clinical trials often faces challenges with accurately measuring treatment effects, especially when baseline risk varies.
• Existing methods may struggle with potential misspecification of the baseline risk distribution and measurement errors.

Purpose of the Study:

• To develop and evaluate a robust likelihood-based approach for meta-analysis that accounts for errors in treatment effect and baseline risk measures.
• To model baseline risk using a flexible skew-normal distribution to overcome limitations of traditional normality assumptions.

Main Methods:

• A novel likelihood-based framework was developed for meta-analysis incorporating baseline risk as an explanatory variable.
• The skew-normal distribution was employed to flexibly model baseline risk, addressing potential misspecification.
• Simulation studies were conducted to assess the performance of the proposed method.

Main Results:

• The proposed skew-normal based likelihood approach demonstrated improved performance compared to standard methods.
• Comparison with routine likelihood methods assuming normality and weighted least-squares regression highlighted the advantages of the new approach.
• Application to two published datasets validated the method's practical utility.

Conclusions:

• The flexible likelihood-based approach using skew-normal distributions offers a more robust and accurate method for meta-analysis in clinical trials.
• This method effectively handles errors in baseline risk and treatment effect measures, leading to more reliable results.