Developing a machine learning-based short form of the positive and negative syndrome scale

Area of Science:

• Psychiatry and Computational Neuroscience
• Development and validation of novel assessment tools
• Application of artificial intelligence in clinical psychology

Background:

• The Positive and Negative Syndrome Scale (PANSS) is a comprehensive psychiatric assessment tool.
• Current PANSS administration and scoring can be time-consuming, taking up to 50 minutes.
• There is a need for efficient and reliable methods to assess psychiatric symptoms.

Purpose of the Study:

• To develop and validate a machine learning-based short form of the PANSS (PANSS-MLSF).
• To ensure the PANSS-MLSF reproduces original PANSS scores accurately.
• To enable the estimation of scores for items removed from the full PANSS scale.

Main Methods:

• Development of the PANSS-MLSF using an artificial neural network.
• Estimation of removed-item scores with the eXtreme Gradient Boosting classifier.
• Assessment of reliability (Cronbach's alpha), concurrent validity (Pearson's r with PANSS), and convergent validity (Pearson's r with CGI-S, MMSE, Lawton IADL).
• Evaluation of removed-item score agreement using Cohen's kappa.

Main Results:

• The study included 573 patients with moderate severity.
• Two PANSS-MLSF versions were proposed: one with 15 items and another with 9 items.
• PANSS-MLSF scores demonstrated high similarity to original PANSS scores (mean squared error 2.6-24.4).
• Good reliability, concurrent validity, and convergent validity were observed for the PANSS-MLSF.
• Moderate to good agreement was found for 60% of estimated removed-item scores.

Conclusions:

• The PANSS-MLSF provides a practical solution for reducing PANSS administration time.
• The short form maintains score comparability and upholds reliability and validity.
• The PANSS-MLSF can effectively estimate scores for removed items, offering further utility.