The statistical analysis of matched data in psychiatric research

Area of Science:

• Psychiatry
• Biostatistics
• Clinical Research Methodology

Background:

• Clinical studies in psychiatry often utilize matched-pair designs, pairing patients with their clinicians.
• Dichotomous response variables in these studies are typically analyzed using McNemar's chi-squared test for correlated proportions.
• A key assumption of McNemar's test is the statistical independence of matched pairs, which is violated when clinicians are matched with multiple patients.

Purpose of the Study:

• To address the limitations of McNemar's test when the independence assumption is violated in psychiatric matched-pair studies.
• To present and discuss alternative statistical methods for analyzing correlated proportions in such scenarios.
• To provide researchers with valid analytical tools for complex matched-pair data in psychiatric research.

Main Methods:

• Review and adaptation of existing statistical tests for correlated binary data.
• Exploration of methods that account for non-independent matched pairs in a clinical setting.
• Comparative analysis of proposed alternatives against McNemar's test under specific conditions.

Main Results:

• Identified specific statistical tests that accommodate non-independent matched pairs.
• Demonstrated the conditions under which these alternative tests provide accurate results.
• Highlighted the potential biases and errors introduced by applying McNemar's test inappropriately.

Conclusions:

• McNemar's test is not suitable for all matched-pair designs in psychiatric research, particularly when clinicians are matched with multiple patients.
• Alternative statistical approaches are necessary to ensure the validity of results in these complex data structures.
• The proposed methods offer robust solutions for analyzing correlated proportions, enhancing the reliability of psychiatric clinical study findings.