Network meta-analysis for an ordinal outcome when outcome categorization varies across trials

Area of Science:

• Biostatistics
• Evidence Synthesis
• Comparative Effectiveness Research

Background:

• Randomized controlled trials often use binary outcomes, but ordinal outcomes (e.g., disease severity) are also common.
• Varying categorization of ordinal outcomes across trials poses challenges for standard network meta-analysis.
• Research synthesizers need methods to address inconsistent outcome categorization in network meta-analysis.

Purpose of the Study:

• To propose a novel network meta-analysis model for ordinal outcomes that accommodates varying categorizations across trials.
• To enable the use of all available data in network meta-analysis, even when outcome levels are combined differently in different studies.

Main Methods:

• Developed a network meta-analysis model for ordinal outcomes allowing multiple categorizations.
• Modified multinomial likelihoods to incorporate partial information from trials with combined levels.
• Employed a Bayesian fixed-effect model with an adjacent-categories logit link to account for ordinality.

Main Results:

• The proposed method was illustrated using a real-world network of trials on antibiotics for preventing liver abscesses in cattle.
• Simulations showed relatively small biases even with varying categorization across trials.
• Large sample sizes resulted in small root mean square errors, indicating good estimation properties.

Conclusions:

• The proposed adjacent-categories logit link method performs well for network meta-analysis with varying ordinal outcome categorizations.
• This method is particularly valuable for research synthesizers dealing with networks containing limited trials for specific outcome categorizations.
• By considering all data in a single estimation, the method enhances the efficiency and reliability of evidence synthesis.