Sensitivity analysis with iterative outlier detection for systematic reviews and meta-analyses.

Area of Science:

• Statistics
• Biostatistics
• Medical Research Methodology

Background:

• Meta-analysis synthesizes study results but can be compromised by outliers.
• Heterogeneity in meta-analysis necessitates careful consideration of contributing factors.
• Existing outlier detection methods in meta-analysis have limitations, particularly the influence of multiple outliers.

Purpose of the Study:

• To propose an iterative outlier detection method for meta-analysis.
• To reduce the confounding impact of other outliers on detection accuracy.
• To enhance the robustness of meta-analysis through improved outlier identification and sensitivity analysis.

Main Methods:

• Development of an iterative outlier detection technique.
• Application of bagging for valid inference in sensitivity analyses.
• Simulation studies to evaluate the proposed method's performance.

Main Results:

• The iterative method demonstrated reduced bias and heterogeneity post-outlier removal.
• Improved accuracy in detecting outlying studies compared to conventional methods.
• Successful illustration of real-world performance through two case studies.

Conclusions:

• The proposed iterative method offers a more accurate and robust approach to outlier detection in meta-analysis.
• This technique enhances the reliability of systematic reviews and meta-analytic findings.
• The bagging approach provides valid inference for sensitivity analyses involving outlier exclusion.