‘The Extended Phenotype’ is a book published by the evolutionary biologist Richard Dawkins way back in 1982. In the book he introduces a concept by the same name, suggesting that a phenotype shouldn’t just be limited to biological processes like protein synthesis or tissue growth. Instead it should be extended to ‘include all effects that a gene has on its environment, inside or outside the body of the individual organism.’
On the 9th January 2007, Steve Jobs unveiled the the first Apple iPhone, providing a way for phenotypes to extend beyond what even old man Dawkins initially envisaged…
Let’s dive in.
🤷♂️ Problem
Poor mental health is a huge problem. Up to 10 million people in England will require new or additional mental health support as a result of the COVID-19 pandemic.
Waiting times are increasing. Of those on a waiting list for mental health services, nearly two thirds wait more than four weeks between their initial assessment and second appointment (one in four wait more than three months).
Monitoring is sporadic. In mental health (as for many other areas of healthcare), a formal assessment at a physical appointment doesn’t provide a true reflection of the trajectory of the condition. Patient recall about their condition during the intervening time between appointments is also poor.
💡 Solution
Use passive and active digital biomarkers, derived or facilitated by smartphone use, to monitor mental wellbeing, detect deterioration, and intervene when necessary.
📖 Terms
Digital phenotyping. The moment-by-moment quantification of the individual-level human phenotype in situ using data from personal digital devices, in particular smartphones.
Ecological momentary assessments (EMAs). Basically a type of questionnaire or assessment. EMAs study people’s thoughts and behaviour in their daily lives by repeatedly collecting data in an individual’s normal environment, at or close to the time they carry out that behaviour. Often more accurate than older data collection methods that require people to recall their behaviour or feelings days, weeks or months later.
Active data. Involves the distribution of EMAs to a user’s mobile device for them to complete. Frequency can vary from multiple times a day to just once a week or less.
Passive data. The data streams that can be collected from a user’s device without their input. Includes location, accelerometer activation, screen time, screen locks/unlocks, bluetooth use, microphone use, app usage, browser history and light sensing (to name just a few).
Digital phenotyping therefore combines active questionnaires (‘Rate how you're feeling today on a scale of 1-10’) with passive data collection (reduced accelerometer activation) to build a comprehensive picture of an individual’s mental wellbeing.
📚 History
Passive sensing, digital phenotyping, whatever you want to call it - it’s barely a decade old. But the commercial prospects were clear from the start…
2008. Saul Shiffman - a world-renowned researcher in the fields of behaviour change and relapse coins the term ecological momentary assessment recognising that clinical psychology typically relies on global retrospective self-reports collected at research or clinic visits, which are limited by recall bias and are not well suited to address how behaviour changes over time.
2011. A team from the MIT Media Lab's Human Dynamics group develop the ‘Funf Open Sensing Framework’ - an extensible sensing and data processing framework for mobile devices with the aim of providing an open source, reusable set of functionalities around the collection, uploading, and configuration of data signals accessible via mobile phones.
2012. The same team spin-off Funf into Behav.io - a company that develops software and services to allow mobile devices to sense, understand, and react to human behaviour and context.
2013. Behav.io is acquired by Google. A press release describes the sale as an opportunity to help Google develop improvements centred on the belief "that our digital experiences should be better connected with the way we experience the world."
2015. American Physician (and one of America’s 100 most influential leaders) Sachin H. Jain coins the term ‘digital phenotype’ - recognising the need for a framework for how digital technologies will be integrated into the patient journey and play a role in precision medicine.
2016. Digital phenotyping comes to mental health. Finnish network scientist Jukka-Pekka Onnela publishes a paper exploring the application of digital phenotyping to enhance behavioural and mental health.
💼 Use cases
Beyond mental health, digital phenotyping has been explored in other areas of healthcare:
👥 Players
Investment in mental health companies has quadrupled since 2015. Venture capital funding of US mental health startups in 2020 outpaced the $1.06 billion invested in 2019. Some of these guys got a slice of that:
Beiwe. Founded by J-P Onnela (see history above) and the Harvard T.H. Chan School of Public Health, Beiwe is a research platform providing high-throughput digital phenotyping infrastructure. They have the ‘Beiwe’ app to collect digital phenotyping data and ‘Forest’ - a data analytics package.
Mindstrong. A US based health-tech company providing virtual therapy, psychiatric care and leveraging digital phenotyping from smartphones to support their services. Scored a $100M series C in 2020 in order to help them scale the company.
Syndi*. A UK startup using digital phenotyping to provide personalised support and recommendations within existing healthcare pathways. Currently focussing on mental health pathways with an eye on university and college students.
Monsenso. A Danish company providing a mental healthcare solution including patient app (for questionnaire completion and passive data collection), a clinician portal and a data analytics platform.
CompanionMX. Born out of the MIT Media Lab and clinically validated at Harvard Medical School teaching hospitals and multiple Veterans Affairs clinics, the Companion system uses active monitoring of voice and passive monitoring of other smartphone metadata to continuously produce acoustic and behavioral biomarkers that predict core symptoms of mood and anxiety disorders.
Behavidence. A US startup developing an app that utilises digital phenotyping and machine learning algorithms in order to give the users, care givers and clinical therapists high accuracy tools to evaluate onset, progression and relapse of mental health conditions. Interesting idea but check out their Google Play store reviews…
LivNao. Mental health management platform. Supposedly 100% passive with patent-pending technology that can infer answers to validated questionnaires like GAD-7 and PHQ-9. We’re sceptical.
Ksana. University of Oregon spinout, Ksana have developed the Effortless Assessment Research System (EARS) - (contrived much?). Allows for the collection of EMAs and mobile sensing data.
Aido. A Finnish outfit focussing on schizophrenia. Their platform monitors medication adherence alongside other digital phenotyping data (mood, sleep, social activity).
*Author is a paid contractor.
🤔 Challenges
C’mon Apple. Passive data collection requires software to run on the down-low while you’re using your phone. The iOS operating system prohibits apps from running constantly in the background and the app approval process to the Apple App Store is more challenging.
Phone alone. Left your phone at home? Taking a break from social media? Accurate phenotyping requires consistent use of the hardware in question in order to derive meaningful insights. Reduced mobility as a result of work from home trends can’t be helpful either.
Standardisation. Or lack of. The way data is collected, analysed and reported varies widely among studies investigating the benefit of digital phenotyping in mental health. Patients, users and clinicians need to have confidence in the evidence base, standardisation will help that.
Privacy. A big one. Allowing an app to monitor things like call frequency, location etc will be uncomfortable for many. Particularly in an age where privacy is at the forefront of people’s minds. Convincing people that their data is secure, anonymised and not misused will be no mean feat.
Clinical integration. Who’s monitoring the users of these platforms? What’s the threshold for intervention? How does this fit into the technical and clinical workflow of mental health service providers? A sociotechnical approach to digital phenotyping will ensure it integrates with existing services and doesn’t cause a headache for mental health practitioners.
Cautionary tales. Proteus health made waves in 2017 with their digital therapeutic for schizophrenia. Heralded as a major breakthrough for improving medication adherence, the company raised hundreds of millions on a multi-billion dollar valuation. In May 2020, the company filed for bankruptcy. The undoing of Proteus was probably multi-faceted, but perhaps there’s something in the big-brother style monitoring of mental health diseases and therapeutics. Is digital phenotyping any different?
🌅 Opportunities
Students and smartphones. Smartphone ownership is high amongst university and college students (96% of US adults ages 18–29 years own and operate smartphones regularly). They’re also seeing a rise in poor mental health as campuses remain shut. The opportunity for digital phenotyping amongst this cohort is huge.
Just in time. Digital phenotyping allows for timely and adaptive interventions. Tackling a deterioration in mental health at an early stage can lead to better outcomes and lower costs.
Wearables. We’ve focussed mainly on smartphones in this article, but wearables and other connected devices can be leveraged too. Scaling these technologies require using whatever digital and electronic resources a user has at their disposal.
Outcomes. Plenty of pilot studies show that digital phenotyping is acceptable to people, but does it improve outcomes? Bigger, longer and more thorough trials are required to provide a definitive answer
Relapse. Relapse rates for mental health conditions are high. Digital phenotyping can not only help at the onset of a mental health condition, but also at the other end - once talking therapy is over or an anti-depressant has been tapered off.
Shared data repositories. Digital phenotyping studies typically generate rich datasets which may be exploited for multiple analytical purposes. There is an opportunity to consider how these might be structured as reusable resources (like the UK Biobank).
🔮 Predictions
Fitness tracking. It might feel strange now, but we’ll eventually track our mental wellbeing in a similar way to how we track our steps, workouts and sleep. The passive nature of it will encourage people to do so.
Closed loop. A contentious prediction, but future platforms might start deploying different interventions (chatbots, app recommendations) based on digital phenotype data in order to improve someone’s mental state. Do we need humans in the loop, or can digital mental health care be a closed circuit? (or light touch at the least).
Insurance interest. Insurance companies already use wearables like Fitbits to tailor their insurance premiums. No doubt insurers will be sniffing around digital phenotyping platforms as a further input for their premium calculators. What are the implications of someone’s activity levels dropping off or their texting frequency reducing? High insurance premium for a looming mental health problem?
🔗 Links
This Nature paper on the opportunities of clinical digital phenotyping
This Vox article on digital phenotyping using smartphones and wearables
That’s it for this week - catch ya next time 👋