OR WAIT null SECS
Among the innovations presented at the 2019 APA meeting: voice analytics for detecting and monitoring mood, and smartphone and web-based passive data as a digital biomarker for mental health disorders.
A popular topic at the 2019 American Psychiatric Association (APA) Annual Meeting was digital mental health. In this two-part series, we asked presenters at the session “Revitalizing Psychiatry Through Engaging with Innovation to Increase Access and Inclusion with Care” to share some in-depth comments and thoughts with Psychiatric Times.
Part 1 of this series covered social media use for youth, and app privacy and efficacy claims. In this article, the presenters discuss voice analytics for detecting and monitoring mood, and smartphone and web-based passive data as a digital biomarker for mental health disorders.
Professor Julien Epps addresses “Can We Diagnose Mental Illnesses Using Voice Data Collected From Phones?”
Speech production is not only the most complex coordination of neuromuscular activity in the entire body, but is sensitive to a vast range of influencing factors, including cognitive function, affective state, motor function, fatigue, and social context. Speech can also be conveniently collected non-invasively and non-intrusively at low cost via smartphone and has attracted research attention as a behavioral signal that is indicative of many psychiatric disorders.
Speech comprises two main components: linguistic-what is said-and acoustic-how it is said. To date, most research has focused on acoustic approaches. For example, insights about depressed speech include reductions in prosodic variation, spectral variability, vowel space area and speech rate, and increases in phone duration and motor incoordination. However, even features as simple as the duration of speech detected by a smartphone over a fixed period, as a proxy for social interaction, have been shown to be indicative of depression level.
Automatic speech-based assessment systems are likely to be applied to screening of the general population or to the monitoring of mental state within an individual (eg, in response to treatment), rather than to diagnosis. Because of the richness of information conveyed by speech, it contains many forms of unwanted variability: due to speaker identity, spoken content, and other factors unrelated to the disorder. Mitigation of these sources of variability is an active research area.
Many speech feature extraction approaches can run in a fraction of real time on a smartphone processor, or extracted features can be efficiently sent via network to a server to process. Similarly, many machine learning approaches are computationally feasible for smartphone platforms. Smartphones offer a unique opportunity to either elicit speech (via app) or process speech passively during everyday use. Elicitation methods include sustained vowels, read speech, diadochokinetic stimuli (eg, “pa-ta-ka”), interview, virtual human interview, and mood induction. However, tasks controlled for affect, articulatory effort, and linguistic complexity have been shown to help discriminate disorders more effectively based on speech.
Opportunities abound, including the possibility for regular minimal-cost assessment, longitudinal within-individual analysis, and integration with smartphone-based interventions or automatically cued therapy (eg, social rhythm therapy). Smartphones also offer the opportunity to crowdsource very large databases and hence develop a stronger evidence base. For more information, please contact firstname.lastname@example.org.
Abhishek Pratap addresses “Can the Digital Data From Smartphone and Online Web Searches Indicate Early Signs of Severe Mental Health Issues, Including Suicide?”
The large-scale adoption of smartphones and online search engines (>5 billion queries a day) offers a unique opportunity for researchers to engage, assess, and monitor social, emotional, and behavioral states at the population level, in real time, with limited user burden, and at low cost. Estimates show that adults in the US are spending more than 150 minutes on their phones daily, with more than 2500 average screen touches. There is growing evidence for utilizing this high-volume, high-velocity smartphone usage data (passive sensing) to assess mood disorders at the population level and on an individualized (N-of-1) level.
Continuous passive data streams (eg, GPS) could offer the means to assess, contextualize, and trigger need-based survey assessments objectively without requiring the user to share sensitive location data. Geospatial contextual analysis pipeline (gSCAP) is a recent effort that is piloting ways to process and generate geospatial contextual features (eg, visiting a park, having coffee, spending time at home) onboard the user’s device without knowing the exact location.
Online information-seeking behavior (web searches) is another rich source of passively collected data that can potentially uncover health-related behavior based on the proximity and type of information sought by the user. In an on-going pilot study as part of Aftercare Focus Study (AFS) at the University of Washington, Seattle, search data with true labels (prior suicide attempts) show distinct trends before and after a confirmed suicide attempt event. This could offer an unparalleled opportunity to understand proximal risk factors for adults who are severely distressed or thinking of killing themselves and potentially provide an early intervention opportunity. You can find more work by Abhishek Pratap at: https://scholar.google.com/citations?user=qt2AE_4AAAAJ&hl=en
Dr Torous is Director of the Digital Psychiatry Division, Department of Psychiatry at Beth Israel Deaconess Medical Center, Boston; Editor in Chief of JMIR Mental Health; Web Editor of JAMA Psychiatry; and Digital Psychiatry Editor for Psychiatric Times. Twitter: @JohnTorousMD. Dr Epps is Professor in Signal Processing and Deputy Head of School (Education), School of Electrical Engineering and Telecommunications, at UNSW Sydney (University of New South Wales) in Australia. Abhishek Pratap is Principal Scientist, Digital Health, at Sage Bionetworks in Seattle, WA.