Ellipsis Health, pioneer of the first voice vital sign to quantify and manage depression and anxiety at scale

Mainul Mondal, CEO

Problem Tech Solves

Behavioral health conditions are at epidemic proportions with 36 -28 percent of people in the U.S. suffering from depression or anxiety. At the same time, providers are time strapped and overworked. Estimates project a shortage of up to 30 thousand psychiatrists in the U.S. by 2024. These debilitating diseases often go undetected and untreated for years. Data and measurement are fundamental to changing the life journey for those suffering from these diseases. At Ellipsis Health, we have harnessed the revolution in artificial intelligence. Through only 90 seconds of voice, our technology rapidly and consistently identifies and quantifies the severity of depression and anxiety to improve outcomes. Through voice, we see a new vital sign for emotional health. By providing a scalable and objective measurement solution, providers are able to focus their limited bandwidth on patients and ensure that no one falls through the cracks; giving patients the right treatment at the right time. By providing timely reminders and an engagement tool, the platform decreases no-show rates and brings providers closer to their patients. For patients, this tool is more engaging and natural than trying to objectively fill out a survey. It provides an experience free of stigma that can often be very therapeutic. Additionally, because the platform can be used outside of the clinical setting and between appointments, it allows patients to be more in control of their emotional health - knowing, with each use they are providing invaluable insights that will enhance their care journey.

Tech Brief

We use machine-learning algorithms to analyze users’ speech, which is used for screening, quantifying and monitoring the severity of depression and anxiety. Combining the most current deep learning and cutting-edge transfer learning techniques, we develop novel models that detect both acoustic and word-based patterns in voice. The models learn their features directly from data, without reliance on predetermined features, and provide a 20% improvement over the current status quo. We leverage a diverse dataset to ensure our algorithms are not biased and can be deployed globally. We see continual performance gains as our data resources increase. Our models can generalize well to new populations -differing demographics, varying accents, and levels of speaking abilities. We have benchmarked against literature reports of untrained primary care providers and show significantly better performance in sensitivity while holding specificity constant. Our algorithms are robust enough to support real-time, global deployment across different populations with no baseline required. Our HIPAA-compliant platform includes iOS, Android and Web Apps and web-based portals. Users’ speech is collected via an app and fed to our cloud-based ML models to be analyzed. The models produce real-time low-latency analytics and confidence estimates, which are passed to providers, payers, employers, or downstream technology partners that integrate our results. We have invested significant time and focus on integrating into the clinical workflow and supporting the doctor-patient journey. For providers and payers, we deploy secure web-based portals that provide results, integrated alerts and reporting capabilities, which are designed to embed seamlessly into existing workflows.

Tech Differentiators

The status quo for screening and measurement today are Patient Reported Outcomes (PROs) or lengthy conversations with behavioral health professionals. While legacy PROs can be scalable and objective, they are time intensive, unintelligent, non-personalized, and unengaging, and, unlikely to provide an ongoing longitudinal result for monitoring and treatment evaluation. Provider conversations provide intelligent insights and can be reviewed longitudinally, however they are somewhat subjective and are certainly not scalable with the scarcity of resources in today’s world. Our highly scalable technology speeds up time to treatment and recovery with just 90 seconds of speech in English + other languages. The provider receives severity scores, plus word and thematic analysis of the responses. The result is a highly scalable, intelligent solution that identifies and tracks severity over-time and aids in the treatment path for depression and anxiety. While there are other startups in the voice analysis space with mental health applications, we differ in several ways: ~Utilize deep learning to discover features from real-world data, which yields superior performance and generalizability. Many others utilize feature extraction. ~Combine semantics (words) and acoustics (sound), while others use one or the other. By using both, our models demonstrate better performance across populations, including underrepresented populations. ~Awarded blocking patents in the field of vocal analysis for mental health. ~Only company with multiple peer-reviewed published scientific papers on the performance of our models. ~IRB research partners with the likes of Mayo, Vanderbilt, Hartford Healthcare, etc. ~Detect clinical grade anxiety and depression severity while others detect non-clinical emotion


Algorithmic Performance: ~Our publications and submissions validate performance of both acoustic and semantic models against the PHQ-8 and GAD-7. ~Patent granted. Our new algorithm methodology improved performance, and we were awarded our second patent. Usage and Engagement: ~In 2021, Phase II of our IRB-approved study monitoring depression clinic patients began and an IRB proposal at Vanderbilt for monitoring spine surgery patients was approved. ~We began work on a study to validate our algorithms in the adolescent population, and we began tuning our algorithms for Spanish (with positive early results). ~Doctors report the technology is very helpful and easy to use, and 30% of users report that the Ellipsis Health experience is therapeutic, using the app longer than required. ~Engagement measure by speech time and word counts increased rather than decreased over a session (2019 publication). Outcomes: ~400+ general population patients used our solution weekly for six weeks (75% adherence). In 5% of cases, Ellipsis algorithms identified at-risk patients who would have fallen through the cracks. ~On-going study of 150 severely depressed patients is using our solution to monitor patients during treatment to create real-time data for providers. Our depression/anxiety vital signs are integrated into the EHR so clinicians can view results as a part of daily patient reviews. Within the 100 currently recruited subjects, we have already identified and mitigated 30 crisis events over three months. CIGNA PUBLIC LINK - https://www.cignaglobal.com/stress-care/individuals/voice-tool Alleviant Health Centers public link with patient testimonials: https://alleviant.com/rising-higher/ DEMO/PROVIDER PORTAL: https://youtu.be/1aIsba7A_7A

Why Us

Ellipsis Health is strategically applying the power of AI to democratize emotional health and change the understanding and treatment of mental health globally. We provide machine learning solutions that better augment clinicians and promote human dignity, so everyone can have access to better health outcomes.

❮ Return to FULL LIST OF FINALISTSView Next Company❯