A systematic review identifies a lack of standardization in methods for handling missing variance data

Area of Science:

• Biostatistics
• Evidence Synthesis
• Medical Research Methodology

Background:

• Missing variance data is a common challenge in meta-analysis (MA).
• Numerous methods exist to address this issue, leading to potential inconsistencies in systematic reviews.
• The current landscape lacks standardized approaches for handling missing variance data.

Purpose of the Study:

• To systematically review and critically appraise existing methods for handling missing variance data in meta-analysis.
• To categorize and assess the theoretical basis and practical application of these methods.
• To identify areas for improvement in methodology and reporting.

Main Methods:

• A systematic literature search was conducted across multiple databases (MEDLINE, EMBASE, Web of Science, etc.) and relevant texts.
• Included were any texts describing methods for handling missing variance data in meta-analysis.
• Extracted information included method descriptions, theoretical underpinnings, and input/output variables.

Main Results:

• Eight distinct classes of methods were identified for handling missing variance data.
• These methods include algebraic recalculations, imputations of study-level standard deviations (SDs) and correlations, and MA-level adjustments.
• Methods range from true imputations to those obviating the need for an SD or recalculating it.

Conclusions:

• The wide array of methods indicates a lack of consensus within the systematic review community.
• The appropriate application of these methods is often questionable, underscoring the need for expert statistical guidance.
• Further research is needed to optimize method selection, minimize bias, enhance accuracy, and improve reporting standards.