AI-Based Fidelity Feedback to Enhance CBT (AFFECT)

Enhancing the Quality of CBT in Community Mental Health Through AI-generated Fidelity Feedback

This study is being conducted together by researchers at the University of Pennsylvania and Lyssn.io, Inc., ("Lyssn"), a technology start-up developing digital tools to support evidence-based psychotherapies (EBPs) for mental health disorders and addiction. This study will implement a technology to assess and enhance the quality of EBPs like Cognitive Behavioral Therapy (CBT) that includes a user interface geared to clinical, supervision, and administrative workflows and needs, and then assess this technology for effectiveness in comparison to usual care.

There is a tremendous global burden of mental illness: Over 50 million American adults have a diagnosable mental health disorder, and major depression on its own is the leading cause of disability worldwide. In the face of this burden, clinical research has documented a variety of effective EBPs (e.g. CBT), and these psychotherapies are utilized on a massive scale. Systems have invested over $2 billion in training providers in specific EBPs. Once trained, however, therapists' adherence to the EBP, also called fidelity, is both crucial for effectiveness and difficult to assess. There is no scalable method to assess the fidelity and quality of EBPs in community practice settings. This is a foundational problem for healthcare systems.

Advances in speech processing and machine learning make technology a promising solution to this problem. The use of technology - instead of humans - to evaluate EBPs means that objective, performance-based feedback can be provided quickly, efficiently, cost-effectively, and without human error. If successful, the present research will be among the first examples of a method for building, monitoring, and assessing the quality of therapy that can scale up to large, real-world healthcare settings.

In this study, the investigators will implement an existing, fully-functional prototype (LyssnCBT) that includes a user interface geared to community mental health (CMH) clinical, supervision, and administrative workflows and needs, and then assess for effectiveness of psychotherapy supported by LyssnCBT in comparison to usual care.

This study will implement LyssnCBT in 5 community mental health agencies, beginning with a single-arm pilot field trial to identify and address any specific barriers to implementing the tool in a community mental health context. The study team will then conduct a larger study in community mental health agencies comparing LyssnCBT to services as usual.

Study Overview

• Status: Completed
• Conditions: Therapy; Cognitive Behavioral Therapy
• Intervention / Treatment: Other: LyssnCBT

Detailed Description

The research team will first recruit 10 therapists and two supervisors from one CMH clinic in order to conduct a single-arm field trial of LyssnCBT. Each therapist will use the LyssnCBT platform with two clients over the course of two weeks (four total sessions per therapist). In addition, supervisors will conduct a supervision session with each therapist using the LyssnCBT tool's fidelity feedback. Next, participants will complete brief Likert assessments on technical reliability, functional reliability, and experiences integrating the system into the daily workflow. Usage data from LyssnCBT will be captured automatically by the system, including: which software features were used, time spent reviewing sessions and transcripts, and time spent reviewing artificial intelligence (AI) generated CBT fidelity feedback. This data and feedback will be used for a final refinement of the LyssnCBT software and related clinical and supervision protocols prior to the main trial.

Following the pilot study, the research team will recruit 4 additional CMH clinics to participate in a type 2 hybrid implementation-effectiveness, randomized stepped-wedge study comparing LyssnCBT to SAU. 50 therapists and their supervisors will be recruited from the participating clinics, and each therapist will be asked to have at least 5 clients participating in the study at any given time, from among their regular caseload. Across 18 months of planned data collection (~75 weeks), the investigators expect a minimum of 1,875 clients for 50 therapists (i.e., 50 therapists x 5 sessions per week x 75 weeks = 18,750 sessions, with an average of 10 sessions per client). All 5 clinics will start with SAU, and clinics will be randomized to begin LyssnCBT sequentially over time.

The primary data being collected throughout the project are recordings of therapy sessions, which are also collected as part of the typical operating procedures of the Penn Beck Community Initiative (BCI). CBT fidelity will be assessed by AI-generated Cognitive Therapy Rating Scale (CTRS) scores for every recorded therapy session, which will be recorded via the Lyssn platform during both SAU and LyssnCBT phases of the study. For client outcomes, the PHQ-9 and GAD-7 will be collected at each session and client drop-out will be assessed via a brief monthly survey sent out to participating therapists. Finally, after three months of engagement with the LyssnCBT tools, each participating therapist and supervisor will complete our battery of implementation measures, including the system usability scale, AIM, IAM, and FIM.

• Study Type: Interventional
• Enrollment (Actual): 449
• Phase: Not Applicable

Contacts and Locations

Study Locations

• United States
  • Pennsylvania
    • Philadelphia, Pennsylvania, United States, 19104
      • The Penn Collaborative for CBT and Implementation Science

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: 18 years and older (Adult, Older Adult)
• Accepts Healthy Volunteers: Yes

Description

Inclusion Criteria:

Therapists

• Able to participate in therapy sessions conducted in English
• Employed at a Philadelphia CMH treatment center that allows the recruitment and participation of therapists in research-related activities
• Willing to allow their session recordings to be used for research purposes
• Computer and internet access

Supervisors

• Oversee participating therapists
• Computer and internet access

Clients

• Able to participate in therapy sessions conducted in English
• Willing to allow the team to collect data and use their session recordings for research purposes

Exclusion Criteria:

• Unwilling to allow the research team to access their therapy session recordings

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Health Services Research
• Allocation: Randomized
• Interventional Model: Crossover Assignment
• Masking: None (Open Label)

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Active Comparator: LyssnCBT; Therapists will use the LyssnCBT tool with clients for recording and session-sharing functionalities. Therapists and supervisors will also have access to LyssnCBT features like speech-to-text transcription, annotation tools, and AI-generated metrics.	Other: LyssnCBT; LyssnCBT is a technology that allows therapists and supervisors access to tools that assist with assessing CBT session fidelity, including speech-to-text transcription, annotation tools, and AI-generated metrics.
No Intervention: SAU (services-as-usual); Therapists will use the LyssnCBT tool with clients for recording and session-sharing functionalities. No other LyssnCBT features will be available for therapist or supervisor review.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Change in therapist CBT fidelity as assessed by CTRS scores	CBT Fidelity will be assessed by AI-generated Cognitive Therapy Rating Scale (CTRS) scores for every recorded therapy session, which will be recorded via the Lyssn platform during both SAU and LyssnCBT phases of the study. CBT sessions are scored on different session components (agenda, feedback, understanding, etc.) with the CTRS on a scale of 0-6, from "poor" to "excellent", where higher scores indicate a greater degree of CBT fidelity. CTRS deviation scores will be created.Mean CTRS scores during SAU will be estimated for each therapist and for each CTRS item. These means will be subtracted from each corresponding CTRS score during the LyssnCBT phase.	Phase 2; Immediately after every recorded therapy session
Acceptability of LyssnCBT tools (Acceptability of Intervention Measure; AIM)	A brief, 4-item measure of a mental health professional's perception of how acceptable an innovation would be for use in their context. Each statement about the acceptability of the innovation is rated on a 5-point ordinal scale ranging from "completely disagree" to "completely agree," where higher scores indicate higher levels of acceptability. This measure has demonstrated good test-retest reliability and sensitivity to change among community mental health providers.	After three months of engagement with the LyssnCBT tools
Appropriateness of LyssnCBT tools (Intervention Appropriateness Measure; IAM)	A brief, 4-item measure of a mental health professional's perception of how appropriate an innovation would be for use in their context. Each statement about the appropriateness of the innovation is rated on a 5-point ordinal scale ranging from "completely disagree" to "completely agree," where higher scores indicate higher levels of appropriateness. This measure has demonstrated good test-retest reliability and sensitivity to change among community mental health providers.	Phase 2; Three months after the site moves from SAU to LyssnCBT
Feasibility of LyssnCBT tools (Feasibility of Intervention Measure; FIM)	A brief, 4-item measure of a mental health professional's perception of how feasible an innovation would be for use in their context. Each statement about the feasibility of the innovation is rated on a 5-point ordinal scale ranging from "completely disagree" to "completely agree," where higher scores indicate higher levels of feasibility. This measure has demonstrated good test-retest reliability and sensitivity to change among community mental health providers.	Phase 2; Three months after the site moves from SAU to LyssnCBT
Usability of LyssnCBT tools (System Usability Scale; SUS)	A brief, 10-item questionnaire to evaluate an end-user's perceptions of an innovation's usability. Each statement is rated on a 5-point ordinal scale ranging from "strongly disagree" to "strongly agree," with every other question reverse-coded. Scores are converted to a common scale, added together and then multiplied by 2.5 to convert the original scores to a scale of 0-100. "Average" usability is defined as a total score >68	Phase 2; Three months after the site moves from SAU to LyssnCBT
Change in client depression symptoms (Patient Health Questionnaire; PHQ-9)	The PHQ-9 (Patient Health Questionnaire) is a brief, widely used depression inventory assessing frequency of symptoms with 9 total items rated on a likert scale of 0-3. Higher scores indicate a greater frequency of symptoms of depression and presence of a clinically significant depressive disorder. This measure has demonstrated good test-retest reliability and can be completed within 2-3 minutes. This measure will be collected at every therapy session.	Immediately before weekly therapy sessions for the duration of treatment or through study completion, a maximum of 18 months
Change in client anxiety symptoms (General Anxiety Disorder; GAD-7)	The GAD-7 (General Anxiety Disorder) is a brief, widely used anxiety inventory assessing frequency of symptoms with 7 total items rated on a likert scale of 0-3. Higher scores indicate a greater frequency of symptoms of anxiety and presence of a clinically significant anxiety disorder. This measure has demonstrated good test-retest reliability and can be completed within 2-3 minutes. This measure will be collected at every therapy session.	Immediately before weekly therapy sessions for the duration of treatment or through study completion, a maximum of 18 months
Client drop-out/premature termination from therapeutic engagement	Dropout is collected via a brief monthly survey sent out to participating therapists which will ask, for each participating client, whether the client has been discharged and if that discharge was planned.	Immediately before weekly therapy sessions for the duration of treatment or through study completion, a maximum of 18 months

Collaborators and Investigators

• Sponsor: University of Pennsylvania
• Collaborators: National Institute of Mental Health (NIMH); Lyssn.io, Inc.
• Investigators: Principal Investigator: Torrey A Creed, PhD, Director, The Penn Collaborative for CBT and Implementation Science; Principal Investigator: David Atkins, PhD, CEO, Lyssn

Publications and helpful links

General Publications

• Creed TA, Salama L, Slevin R, Tanana M, Imel Z, Narayanan S, Atkins DC. Enhancing the quality of cognitive behavioral therapy in community mental health through artificial intelligence generated fidelity feedback (Project AFFECT): a study protocol. BMC Health Serv Res. 2022 Sep 20;22(1):1177. doi: 10.1186/s12913-022-08519-9.

Study record dates

Study Major Dates

• Study Start (Actual): March 9, 2023
• Primary Completion (Actual): October 31, 2025
• Study Completion (Actual): January 31, 2026

Study Registration Dates

• First Submitted: February 18, 2022
• First Submitted That Met QC Criteria: April 15, 2022
• First Posted (Actual): April 22, 2022

Study Record Updates

• Last Update Posted (Actual): March 24, 2026
• Last Update Submitted That Met QC Criteria: March 23, 2026
• Last Verified: March 1, 2026

More Information

Terms related to this study

• Keywords: depression; anxiety; cognitive behavioral therapy; community mental health; training; therapy; evidence-based psychotherapy; CBT fidelity
• Additional Relevant MeSH Terms: Mental Disorders; Behavioral Symptoms; Behavior; Anxiety Disorders; Depression
• Other Study ID Numbers: R42MH128101 (U.S. NIH Grant/Contract)

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No