Defining Affirmation: Generating an Affirmative Clinical Framework Utilizing Artificial Intelligence

Background and Purpose: The applications of artificial Intelligence (AI) in social work is understudied (Ameen et al., 2023), with emerging studies highlighting AI’s potential in research with LGBTQ+ youth, who face persistent mental health disparities (Bragazzi et al., 2023), and in analyzing effective clinical interactions in therapeutic settings (Imel et al., 2019). LGBTQ+ affirmative practice promotes resilience and addresses discrimination’s psychological impacts, contributing to stronger therapeutic alliances and better mental health outcomes (Alessi et al., 2019), yet there is a paucity of research that explores clinicians’ presentation of affirmation. The aim of this study is to explore affirmative LGBTQ+ group therapy to identify “affirming” clinical language and group processes and elaborate on the process of utilizing AI for qualitative analysis.

Methods: A pragmatic qualitative approach (Ramanadhan et al., 2021) integrating constructivist grounded theory (Charmaz, 2014) and artificial intelligence (AI) assisted natural language processing (Creed et al., 2022) was used to iteratively analyze affirmative cognitive behavioral therapy (CBT) groups. The dataset comprised 64 group sessions conducted between March 2020 and April 2022, totaling 5,760 minutes of recordings. Sessions involved over 600 LGBTQ+ youth aged 12-24 in age-appropriate groups with a range of LGBTQ+, racial, and ethnic identities. The utterances of the 18 group co-facilitators are the analytic focus of this study. Recordings were AI-transcribed using Otter.ai, cleaned, and half were manually coded inductively by a trained team of six coders with a range of LGBTQ+ identities and clinical experience, including one coder who participated in the AFFIRM intervention, and two AFFIRM facilitators. Coding analyzed the facilitators’ verbal expressions of affirmation and group facilitation process using NVivo 14. After coding saturation was reached, the inductive coding framework and recording excerpts were input into a clinical AI platform, Lyssn, and the remaining transcripts were uploaded to generate an AI-assisted clinical evaluation of affirmative group therapy co-facilitation, replicating the protocols of a similar study on CBT effectiveness (Creed et al., 2022). AI-generated qualitative findings were compared to the human-coded data using constant comparison in a member checking session with AFFIRM facilitators, and interrater reliability estimates were calculated (Tortt et al., 2023).

Results: This research yielded good interrater reliability (87% agreement) between the coders and Lyssn, ICC (2,1) = .80, 95% CI [.72, .86], p < .001. The analysis produced six themes of affirmative practice utterances and actions: promoting LGBTQ+ client agency while acknowledging minority stressors, actively building and drawing on community, referencing the affirmative nature of the group therapy process, enhancing client cognitive flexibility, identifying participant strengths, and emphasizing hope and queer joy. Effective group facilitation strategies included using the self, redirecting, using humour, and fostering member connections. Non-affirming processes included displaying facilitator tension, invalidating participants, and interrupting.

Conclusions and Implications: This study highlights affirmative clinical skills that can strengthen social work practice during a critical time for LGBTQ+ youth mental health. The pragmatic approach to systematically integrating AI into a rigorous qualitative study will be detailed and may reduce the gap between research and practice to strengthen similar social work research efforts.