Multivariate fixed- and random-effects models for summarizing ordinal data in meta-analysis of diagnostic staging...

Area of Science:

• Biostatistics
• Medical Diagnostics
• Clinical Research

Background:

• Accurate disease staging is critical for treatment decisions in conditions like rectal cancer and Crohn's disease.
• Existing meta-analysis models often treat staging data as nominal, overlooking its inherent ordinal nature.
• Evaluating diagnostic test accuracy and summarizing findings in meta-analyses requires appropriate statistical methods.

Purpose of the Study:

• To extend existing multinomial meta-analysis models to accommodate the ordinal characteristics of disease staging data.
• To develop and compare fixed- and random-effects ordinal models for summarizing staging accuracy.
• To provide statistical tools for meta-analyses supporting evidence-based clinical guidelines.

Main Methods:

• Developed three ordinal models: adjacent-category logits, continuation-ratio logits, and proportional odds logits.
• Extended a previous multinomial model to incorporate ordinality in meta-analysis.
• Applied both fixed- and random-effects approaches for model comparison.
• Utilized data from rectal cancer staging using endoluminal ultrasonography and magnetic resonance imaging.

Main Results:

• The proposed ordinal models effectively summarize staging data, capturing its inherent order.
• The models provide proportions of correctly staged, understaged, and overstaged patients per stage.
• Comparison of fixed- and random-effects approaches was conducted.

Conclusions:

• The developed ordinal models offer a more appropriate method for meta-analyzing disease staging data compared to nominal models.
• These statistical approaches can enhance the accuracy and utility of meta-analyses for clinical practice and guideline development.
• The study demonstrates the application of these models using real-world rectal cancer staging data.