Meta-analysis on continuous outcomes in minimal important difference units: an application with appropriate variance...

Area of Science:

• Biostatistics
• Clinical Research Methodology
• Health Economics

Background:

• Meta-analyses are crucial for synthesizing evidence but rely on assumptions about key parameters.
• The minimal important difference (MID) is a critical threshold in interpreting treatment effects.
• Treating MID as a constant may introduce biases in meta-analysis results.

Purpose of the Study:

• To compare meta-analysis outcomes when the minimal important difference (MID) is modeled as a random variable versus a constant.
• To evaluate the impact of different assumptions on treatment effect estimates and heterogeneity P-values.

Main Methods:

• Conducted meta-analyses on published data, including osteoarthritis studies.
• Calculated the variance of MID (MD_MID) using the delta method, treating MID as both a random variable and a constant.
• Assessed performance under varying assumptions for rho and coefficient of variation of MID (CoV_MID).

Main Results:

• Estimates of treatment effect and P-values differed significantly depending on whether MID was treated as a constant or a random variable.
• Under specific conditions (rho=0.5, CoV_MID=0.8), treating MID as a constant overestimated treatment effects by 33-110% and reduced heterogeneity P-values substantially.
• Similar effect magnitudes were observed when rho=0.8 and CoV_MID=0.5.

Conclusions:

• Modeling MID as a random variable is preferable to treating it as a constant.
• This approach avoids unrealistic assumptions inherent in the constant MID model.
• Utilizing MID as a random variable leads to more appropriate and reliable treatment effect estimations in meta-analyses.