Pelago, a virtual substance use disorder (SUD) provider, has expanded to mental health and behavioral addiction care.
The new Pelago Behavioral Health Platform means co-occurring conditions can be assessed and treated in a single plan. Behavioral addiction programming includes gambling, gaming and binge eating. The consolidated offering is available to employers and payers, with member access starting in November. The announcement coincided with the annual Behavioral Health Tech (BHT) conference in Nashville.
Since its founding, Pelago has focused on treating more patients with SUD without inflating costs. That requires identifying people with untreated SUD, engaging them in treatment and doing so at a much lower cost than what a residential treatment or intensive outpatient program require, Pelago executives say. The company claims it achieves nearly $12,000 in annual savings per SUD patient.
Pelago’s employer clients face escalating healthcare costs and “a feeling of misaligned incentives” in the traditional behavioral health system, which prioritizes session-based care, Pelago Co-founder and CEO Yusuf Sherwani, M.D., told Fierce Healthcare at BHT. They had requested that Pelago expand its services using a clinical model that is “acuity-based,” per Sherwani.
Since this summer, each Pelago member has begun intake and triage with Sona, the company’s clinical AI agent (available via voice and text). People with higher-acuity needs get routed to a clinician right away, as opposed to someone who may be able to get their needs met through Sona. Whether a member ultimately uses Sona, a human clinician or both for support is their choice, though the company sees the best clinical outcomes for members using both.
There is always a clinician in the loop with Sona, but for lower-acuity cases, human providers are less involved in care. It’s the opposite for high-acuity cases, where Sona is involved, but the clinician is driving the care. Sona’s model draws from a repository of over 800,000 conversations, and it takes into account individual user’s Sona chat history. And, if a patient is getting clinical care through Pelago and has consented to sharing data, it also draws on summaries of therapy sessions.
So far, the company has found some promising outcomes among members who use Sona and report on outcome measures. Two sequential observational studies that analyzed more than 9,000 Sona conversations found that among participants who began with clinical-range depression or anxiety scores and completed repeat assessments, nearly half improved by at least five points on the PHQ9 or GAD7 assessments within about a month. Participants in both studies reported 20% to 25% fewer poor mental health days each month.
Safety is a top consideration for Pelago when building Sona, per Sherwani. The AI agent has conversational guardrails to pick up on things like suicide risk that get escalated as needed. In fact, Sona captures suicide risk at higher rates than the PHQ9. On that measure, Pelago sees about 10% of patients answering affirmatively to the question on suicidal ideation. Via Sona, that increases to about 20% of patients.
“Even people who appear low-acuity … actually do often end up having much more complex symptoms, which traditional measurement-based care approaches wouldn’t catch,” Sherwani said. “It kind of shows you the limitation of a traditional questionnaire-based approach versus increasingly sophisticated systems that can cut through some of that noise.”
Sherwani acknowledged that while Sona may flag cases that are ultimately not of concern, “we would prefer to err on the side of caution and over-escalate than under-escalate.”
In addition to drawing on chat history and session summaries, Sona can tailor its recommendations to an individual’s specific plan benefits. Sona can also generate CBT lessons and help patients put them into practice between sessions, reporting progress back to the clinician. “It feels like an extension to the clinician,” Sherwani said.
While Sona is available as a voice agent and via text, Pelago sees better outcomes with voice. Patients tend to talk for longer with the voice agent, as it’s easier than typing, and it more closely resembles the experience of talking to a human therapist, Sherwani noted. The voice agent can also pick up on important patient intonations, unlike a text-based chat.
Sherwani believes payers today are asking if covering therapy for everyone, regardless of their clinical acuity, makes sense. And there are only two potential choices, in his view. One is a return to more aggressive utilization management, which could harm access. The other, Sherwani believes, is using tech to match people to the right level of care.