Machine-actionable criteria chart the symptom space of mental disorders

Area of Science:

• Computational psychiatry
• Clinical informatics
• Medical decision support

Background:

• Diagnostic criteria in manuals like DSM-5 are narrative and not computationally accessible.
• Existing probabilistic models rely on large text corpora, lacking transparency.
• Assessing conceptual coherence of diagnostic constructs is challenging.

Purpose of the Study:

• To develop a deterministic framework for translating diagnostic criteria into a machine-actionable format.
• To enable computational interrogation and systematic analysis of the full symptom space.
• To provide a transparent and reproducible method for evaluating diagnostic constructs.

Main Methods:

• Developed a deterministic framework to represent diagnostic criteria computationally.
• Applied the framework to chart schizophrenia-spectrum disorders.
• Evaluated the National Academies' definition of Long COVID against existing disorders.

Main Results:

• The framework successfully translated narrative diagnostic criteria into a computable format.
• Schizophrenia-spectrum disorders were charted, highlighting conceptual distinctions despite symptom overlap.
• The National Academies' Long COVID definition was found to be largely subsumed by depressive and anxiety disorders.

Conclusions:

• Making diagnostic consensus computable offers a reproducible foundation for evaluating diagnostic constructs.
• The framework supports the development of interpretable, regulatory-compliant diagnostic support tools.
• This approach enhances the transparency and systematic analysis of clinical diagnostic criteria.