Voice-based AI to scale evaluation of crisis counseling in 988 rollout

The national rollout of 988-a single, easy-to-access crisis line anywhere in the US modeled after 911-will provide streamlined access to crisis counseling services to millions of Americans, resulting in a massive expansion of services, projected to be as high as 40 million annual calls by 2026. Yet, at present, there is no scalable method for evaluating the quality of crisis counseling services. Call centers participating in 988 are required to review 3% of calls for quality assurance (QA), where manual review is slow, expensive, and human resource intensive, and thus, the vast majority of crisis counseling quality and effectiveness is unmeasured and unknown. The current, fast-track SBIR proposal will develop and evaluate an Al-based software system (LyssnCrisis) for automated crisis counseling QA from an audio recording of a crisis counseling call. Importantly, this work builds from Lyssn's previous, successful work in developing and evaluating automated systems for Motivational interviewing (Ml) and cognitive behavioral therapy (CBT) quality monitoring. Lyssn.io, Inc., ("Lyssn") is a start-up developing Al-based technologies to support training, supervision, and quality assurance of evidence-based behavioral health counseling interventions. Our goal is to develop innovative health technology solutions that are objective, scalable, and cost efficient. Lyssn offers a HIPAA-compliant, cloud-based platform for secure recording, sharing, and reviewing of conversations, which includes Al-generated metrics for Ml and CBT The proposed LyssnCrisis tool will build from - and be integrated into - this core platform. Lyssn is partnering with ProtoCall Services, Inc., which is a national backup center for 988 and runs 988 for the state of New Mexico. ProtoCall has a 20-year track record of providing high quality crisis counseling services to more than 500 community and university customers. Their expertise, relationships, and amassed data form the clinical foundation for the current research. Phase I builds from an existing Lyssn prototype to develop and evaluate LyssnCrisis. Core activities include behavioral coding of crisis counseling calls (N=200 calls / 30,000 utterances) and development of machine learning models that can reliably and accurately evaluate core aspects of crisis counseling (e.g., empathic listening, suicide risk assessment). Phase II will conduct a field-based usability trial to optimize LyssnCrisis and then a hybrid implementation-effectiveness randomized trial (N = 50 call-takers, 13,600 calls) to evaluate the effectiveness of LyssnCrisis to improve crisis counseling skills and caller outcomes. Analyses will also examine the hypothesized mechanism underlying LyssnCrisis. The research is strongly aligned with IMH's 2022 Strategic Plan and its emphasis on a computational approach to scaling up treatment delivery and monitoring. Successful execution will provide automated, scalable crisis counseling QA for the first time ever, supporting access to qualirty crisis counseling services as the US invests in these critical, just-in-time services in 988.