2 articles · Updated · Endocrinology Advisor · Jun 6
Summary
A meta-analysis of 46 studies covering 8,146 adults found standardized diagnostic interviews for common psychiatric conditions had pooled test-retest reliability of κ=0.69, lower than many clinicians assume.
Reliability differed by diagnosis: substance use disorders scored higher than mental disorders overall—κ=0.72 versus 0.65—with opioid disorders reaching 0.81 and nonaffective psychoses falling to 0.55.
The review also found substantial variation across studies, with heterogeneity at I²=93%; diagnostic criteria partly explained differences for substance use disorders, while sample size and retest interval did not.
The findings suggest structured interviews work better for conditions with clearer behaviors or timelines than for disorders relying more on subjective experience, raising questions about a tool widely used in psychiatric classification.
With psychiatry's 'gold standard' failing, will objective biomarkers finally deliver reliable diagnoses or create new uncertainties?
As AI models show diagnostic promise, can they navigate the deep subjectivity of human mental states safely and reliably?
If we abandon rigid diagnostic labels for a spectrum, how will treatment and insurance systems adapt to this new reality?
Standardized Diagnostic Interviews Under Scrutiny: 2026 Meta-Analysis Finds Moderate Reliability, Urges Diagnostic Reform in Psychiatry
Overview
A major new meta-analysis published in May 2026 in JAMA Network Open is challenging the long-held belief that Standardized Diagnostic Interviews (SDIs) are the gold standard for mental health diagnosis. The study found that SDIs show only moderate consistency when the same person is assessed more than once, raising concerns about their reliability. As a result, researchers are urging caution and a thorough reevaluation of current diagnostic practices. Experts now emphasize that a truly reliable diagnosis should combine standardized tools with a deeper understanding of each individual's unique circumstances, moving beyond reliance on SDIs alone.