Dose-response meta-analysis of differences in means.

Area of Science:

• Biostatistics
• Pharmacometrics
• Clinical Trial Analysis

Background:

• Existing meta-analytical methods are limited for combining quantitative outcomes expressed as mean differences.
• There is a need for robust statistical approaches to synthesize dose-response data from multiple studies when outcomes are continuous.

Purpose of the Study:

• To develop and validate a novel two-stage meta-analytical method for combining dose-response findings presented as mean differences.
• To establish a flexible approach for modeling non-linear dose-response relationships in quantitative outcomes across studies.

Main Methods:

• A two-stage approach was proposed: flexible dose-response modeling within each study, followed by a multivariate random-effects model to combine study-specific parameters.
• The method utilizes spline models, which do not require pre-specification of the dose-response curve shape and can accommodate various data characteristics.
• Covariance of data points (mean differences, standardized mean differences) is accounted for in the modeling process.

Main Results:

• The method was illustrated using clinical trial data for aripiprazole in shizoaffective patients, assessing symptom improvement via the Positive and Negative Syndrome Scale (PANSS).
• A non-linear dose-response association was identified, with a maximum mean PANSS score change of 10.40 (95% CI: 7.48, 13.30) at 19.32 mg/day.
• A dose of 10.43 mg/day was estimated to yield 80% of the maximum predicted response, with no significant improvement beyond this dose.

Conclusions:

• The proposed two-stage approach is suitable for combining correlated mean differences from multiple studies to establish dose-response relationships.
• Sensitivity analyses are recommended to ensure the robustness of the derived dose-response curves.
• A user-friendly R package is available to facilitate the practical application of this methodology by researchers and practitioners.