Bayesian Networks for Prescreening in Depression: Algorithm Development and Validation

Area of Science:

• Computational psychiatry
• Health informatics
• Machine learning in healthcare

Background:

• Prompt identification of depressive symptomatology (DS) is crucial for effective treatment.
• Existing machine learning models often lack practical application and real-world benefits.

Purpose of the Study:

• Develop a novel methodology for identifying individuals likely to exhibit DS.
• Utilize probabilistic measures for explainable identification of influential features.
• Propose tools for real-world application in DS screening.

Main Methods:

• Employed three datasets: PROACTIVE, PNS 2013, and PNS 2019.
• Used Bayesian networks for feature selection and machine learning for DS prediction.
• Analyzed the trade-off between sensitivity, specificity, and interview reduction.

Main Results:

• Achieved high sensitivity and specificity across datasets, with AUCs up to 0.809.
• Identified key features like postural balance, shortness of breath, and age perception.
• Demonstrated a potential reduction in screening interviews by up to 52% while maintaining 0.80 sensitivity.

Conclusions:

• A novel methodology for DS identification using Bayesian networks was developed.
• The approach effectively identifies significant features and reduces screening burden.
• Facilitates improved early identification and intervention strategies for individuals with DS.