How accurately can supervised machine learning model predict a targeted psychiatric disorder?

Area of Science:

• Psychiatry
• Artificial Intelligence
• Machine Learning

Background:

• Hoarding disorder (HD) is a psychiatric condition characterized by compulsive collecting and distress over discarding items.
• Misdiagnosis of HD is common due to lack of patient/family recognition and professional expertise.
• This study assessed if a supervised machine learning (ML) model could replicate psychiatrist diagnostic accuracy for HD.

Purpose of the Study:

• To evaluate the diagnostic capability of a supervised ML model in identifying hoarding disorder (HD).
• To compare the ML model's diagnostic performance against experienced psychiatrists using self-report data.
• To explore the potential of AI in psychiatric disorder risk assessment.

Main Methods:

• 500 online participants completed the Hoarding Rating Scale-Self Report (HRS-SR) and GAD-7.
• A decision tree classification model predicted clinical HD based on HRS-SR and GAD-7 scores.
• Model performance was evaluated against diagnoses made by three experienced psychiatrists.

Main Results:

• The ML model identified 93% of clinician-diagnosed hoarding disorder cases.
• The Hoarding Rating Scale-Self Report (HRS-SR) alone detected approximately 60% of cases.
• Key ML performance metrics included 55% Matthews Correlation Coefficient and 79% Area Under Curve (AUC).

Conclusions:

• Machine learning shows significant potential for psychiatric disorder risk assessment prior to clinical consultation.
• AI-driven analysis of questionnaire data can potentially reduce wait times and improve patient prognoses.
• This study highlights the utility of ML in complementing psychiatric diagnostic processes.