Developing a Machine Learning-Based Automated Patient Engagement Estimator for Telehealth: Algorithm Development and...

Area of Science:

• Psychology and Machine Learning
• Behavioral Health Care Technology
• Telehealth Innovations

Background:

• Patient engagement is crucial in behavioral health but challenging in telehealth due to limited nonverbal cues.
• Existing training for telehealth patient engagement is scarce, necessitating new methods for assessment.
• Machine learning offers a potential solution for estimating patient engagement during virtual therapy sessions.

Purpose of the Study:

• To evaluate machine learning models' ability to estimate patient engagement levels in tele-mental health.
• To determine if machine learning can support and enhance therapeutic engagement between clients and psychotherapists.
• To introduce a novel dataset for advancing telehealth engagement detection research.

Main Methods:

• A multimodal learning approach was developed, utilizing latent vectors for affective and cognitive engagement features.
• A semisupervised learning solution was explored due to labeled data constraints in healthcare.
• The Multimodal Engagement Detection in Clinical Analysis (MEDICA) dataset, comprising 1229 video clips, was created and used for experiments.

Main Results:

• The proposed algorithm achieved a 40% improvement in root mean square error for engagement estimation compared to state-of-the-art methods.
• Real-world tests showed positive correlations between the model's engagement estimates and psychotherapists' Working Alliance Inventory scores.
• The findings suggest the model's potential to provide patient engagement estimations that align with clinical assessments.

Conclusions:

• Machine learning can accurately and reliably estimate patient engagement in telehealth, supporting therapeutic alliance.
• The developed algorithm integrates psychological theories with machine learning for enhanced telehealth patient engagement assessment.
• The creation of the MEDICA dataset and the proposed method open new research avenues for telehealth tools.