New Sentio Collaboration Study: NLP Sentiment Analysis Tracks Client Distress and Treatment Progress

Read the full peer-reviewed study: The association of transformer-based sentiment analysis with symptom distress and deterioration in routine psychotherapy care (Faust, Awad, Vaz, & Rousmaniere, 2026, Frontiers in Digital Health).

What the study found

A new peer-reviewed study from researchers at Sentio University and the Sentio Counseling Center demonstrates that a generic, open-source AI language model can produce useful, objective measurements of psychological distress directly from psychotherapy session transcripts. Across 751 telehealth psychotherapy sessions, the team applied a fine-grained BERT-based Transformer sentiment model to client and therapist speech, then compared the resulting scores to the Outcome Questionnaire 45 (OQ-45), one of the most widely used patient-reported outcome measures in routine clinical practice.

Client sentiment scores showed moderate negative correlations with the overall OQ-45 score (r = -0.31) and the Symptom Distress subscale (r = -0.34), with weaker but directionally consistent correlations to the Interpersonal Relations and Social Role subscales. Critically, therapist sentiment was not correlated with client OQ-45 scores, which is evidence that the client sentiment measure reflects an intrinsic feature of the client's emotional state rather than the topic of conversation. The session sentiment feature also significantly distinguished between OQ-45 alert categories under both the Rational and Empirical models, including the Red alert flag for clients at risk of premature dropout or negative outcomes.

How AI sentiment analysis predicts therapy outcomes

This work validates some long-held conjectures about the relationships between how clients use language in psychotherapy sessions and other markers of emotional distress. Specifically, we see that even a very generic AI language model can provide useful measurements of distress as well as the client's progress or deterioration under care. The verbal sentiment measured from session transcripts is more objective than the self-reported questionnaire instruments that are widely used in clinical practice. This may ultimately lead to stand-alone indicators of client health that complement or even relieve the burden of lengthy surveys, an outcome with real implications for the kind of routine outcome monitoring built into Sentio's MFT program.

Democratized and private AI in a real clinical setting

This research program is, by design, a demonstration of the democratized and private use of advanced AI tools in a real-world clinical setting. Every step of the data analysis pipeline that we built at Sentio's clinic, from transcribing the session videos to the foundational sentiment model, is open source. The transcription used WhisperX, the speaker diarization used pyannote, and the sentiment classification used a fine-grained BERT model hosted on the Hugging Face model hub.

The computational analysis ran on off-the-shelf commodity hardware, which means that any clinic could set up such an AI-powered monitoring system on its own premises. This also opens up the exciting possibility of clinics running customized diagnostic tools tailored to their own practice. Most importantly, running everything on-prem meant that sensitive client data was never exposed to third-party models, computing, or storage. This kind of privacy-preserving, low-cost, reproducible methodology is also a core principle of our AI certification for therapists.

About Douglas Faust, PhD

Douglas Faust holds a PhD in theoretical and mathematical physics from the University of Washington, where his dissertation was titled "On the Splitting of a Quantum Degenerate Gas of Identical Bosons." He also holds a Master's degree in physics from the University of Washington. In addition to his collaboration with Sentio, Doug currently holds an appointment as a lecturer in the Department of Mathematics at Western Washington University and works as a data scientist in Intel Corporation's Logic and Technology Department.

Doug has worked in machine learning and AI across a wide range of domains, from natural language processing to finance to physics, including prior roles as a data scientist at Read AI and a lecturer in the Physics Department at Seattle University. His combination of deep mathematical training and applied ML experience makes him an ideal collaborator for the kind of careful, validated, clinically grounded AI research that Sentio is committed to producing. You can connect with Doug on LinkedIn.

Research collaboration at Sentio

As machine learning technology advances rapidly, Sentio is intentionally partnering with researchers like Doug Faust to ensure that the AI tools touching mental health care are validated, transparent, and grounded in real clinical practice. Sentio University, together with the Sentio Counseling Center, offers a distinctive research environment: a fully remote training clinic with continuous routine outcome monitoring, structured deliberate practice supervision, and a graduate-level MFT training program that produces some of the richest psychotherapy process data available. We expect this to be the first of many studies in our ongoing research program characterizing the use of modern NLP and large language models in psychotherapy.

References

Faust, D. K., Awad, P., Vaz, A., & Rousmaniere, T. (2026). The association of transformer-based sentiment analysis with symptom distress and deterioration in routine psychotherapy care. Frontiers in Digital Health, 8, 1792536. https://doi.org/10.3389/fdgth.2026.1792536

Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics, 4171–4186. https://doi.org/10.18653/v1/N19-1423

Lambert, M. J. (2013). Outcome in psychotherapy: The past and important advances. Psychotherapy, 50(1), 42–51. https://doi.org/10.1037/a0030682

Munikar, M., Shakya, S., & Shrestha, A. (2019). Fine-grained sentiment classification using BERT. 2019 Artificial Intelligence for Transforming Business and Society (AITB), 1–5. https://doi.org/10.1109/AITB48515.2019.8947435