Today psychiatrists recognize that people seeking care rarely present as the well-circumscribed conditions defined in textbooks. Rather, they present with a range of complex, intersecting problems that vary with time, treatment, and other contextual factors. As a result, most modern clinicians adopt eclectic approaches to clinical diagnosis, formulation and treatment selection, guided by an idiosyncratic combination of established nosology and symptom-based considerations.
Indeed, the nosology of psychiatric disorders remains an unsolved problem for the medical field, with current definitions and diagnoses for conditions like schizophrenia and bipolar disorder have evolved over the years but retain many elements similar to those when developed 100 years ago.1 Although it was careful observation of patients' longitudinal symptoms via notecards that enabled psychiatrist Emil Kraepelin to create those definitions in the 1890s, today digital tools and methods offer the potential to refine our characterization of mental illnesses across a much wider sample of individuals and cultural contexts. These new tools are the practical smartphones currently owned by over 2.5 billion2 people around the world. And the new class of methods is the ability to quantify individual-level human behavior over time, moment-to-moment, in situ (ie, the person's natural environment), using the data from smartphones and other personal digital devices like smartwatches, which has been termed digital phenotyping.3
In its essence, digital phenotyping represents the neo-Kraepelinian approach to history taking and longitudinal assessment, but without having to follow someone around with notecards. Like the rest of the world population, people with mental illnesses increasingly use smartphones as part of their daily lives.4,5 The same features that make smartphones useful tools in everyday life, such as their tools for communication, navigation, social networking, and productivity, may also offer useful data, as a byproduct of the use of these tools, toward quantifying the lived experiences of people with mental health conditions. As a result, smartphone-based digital phenotyping offers distinct clinical and practical advantages over the paper-and-pencil approaches taken by Kraepelin and even modern-day psychiatrists with astute observational skills, because it enables nuanced, individual-level observations with far greater precision and standardization than previously possible, all by using personal digital devices.
The rationale for interpreting sensor readings from a smartphone as a “digital phenotype” is grounded in the technological sophistication of today's smartphones and the fact that we have come to rely on them in many ways for communication, navigation, and organization. That is, because many people often have their smartphone nearby or on their person, digital phenotyping captures longitudinal information across the diverse locations and environments that people experience. (Indeed, some have even proposed the term “nomophobia,”6 as a fear of being out of phone contact, to highlight the strong attachment many people now feel toward their digital devices.) As a result, interactions with smartphones may be viewed as a proxy for underlying states of behavior, emotion, and cognition, at least in some people.
For example, it would not be unreasonable to interpret a sudden increase in text messages, phone calls, travel distance, and physical activity as a possible episode of hypomania or mania, particularly in someone with bipolar disorder or appropriate family history.7 Aggregate changes in digital sensor and smartphone interaction patterns, particularly when considered together and with baseline information, might be enough to raise concerns significant enough to warrant a just-in-time clinical intervention. And yet, the hypothetical benefits of remote-monitoring strategies have not yet proven digital phenotyping offers reproducible or reliable digital biomarkers, given the early stages of research efforts. The clinical value of this data will emerge with more research and time.
The ability of smartphones to quantify these moment-by-moment changes at the person-level and to record and synthesize them—unobtrusively, and thus without unnecessary interactions with their clinician—offers a wealth of new information that can be used to characterize fluctuations in a patient's condition and environmental exposures experienced daily outside of the clinic or hospital setting.
The application of digital phenotyping is probably most relevant to today's busy psychiatrist in its potential to streamline the error-prone process of initiating and maintaining patients on successful therapeutic strategies. Once a treatment has been started, digital phenotyping is likely to offer tremendous potential—beyond mere nosological considerations—to address numerous pragmatic considerations for how to initiate and maintain a patient safely on a treatment regimen that could include pharmacological or behavioral care. Although eclectic approaches to treatment selection eventually work in most patients seeking care, they rely on an individualized, trial-and-error process that carries significant risk and uncertainty for the patient and provider and may be both costly and inefficient. For most patients, few reliable prognostic signs are evident at the outset of treatment, and psychiatrists must resort to hopeful supportive comments (eg, “this will probably help, and if not we'll try something else”), and then wait for the person receiving care to return and report unintended side effects to decide whether it may need more time, whether the treatment is working, or a change in dosage is required. Because digital phenotyping is likely to enable strategies to sense and synthesize this information with minimal patient or provider burden, future psychiatrists are likely to employ digital therapeutic management strategies that look different from the bread-and-butter clinical psychiatry practiced today. Of course, the future of digital phenotyping in psychiatry is far from certain; and yet, there are many signs of early promise.
Exploring the Clinical Promise of Digital Phenotyping
Digital phenotyping has already demonstrated potential to augment a wide range of mental and physical ailments. Although promotion of physical activity is one of the most effective interventions for many illness conditions, quantifying the relationship between activity levels and mental health has been a challenge. One study8 used continuous physical movement data to identify activity patterns that distinguished between mood states in bipolar disorder. This study found that among participants with bipolar disorder, depressive days could be distinguished from inter-episode days by a specific pattern that included overall low activity levels and a delayed activity onset.8 Another study explored mobility patterns recorded by the smartphone GPS over a 2-week period in 40 participants and concluded that movement and location variance were significantly different between participants with low versus high depressive symptom severity scores.9 Thus, digital phenotyping may potentially be used to classify mood states on the basis of activity patterns.
Another potential use of digital phenotyping is to anticipate and prevent relapses of behavioral conditions. Our group recently explored this in patients with schizophrenia by capturing mobility patterns from GPS, sociability patterns from anonymized call/text logs, and self-reported symptoms from smartphone surveys. The resulting data, when analyzed using anomaly-detection methods derived from physics, highlighted the potential of digital phenotyping to help predict psychotic relapses up to 2 weeks early, thus offering a window for early and perhaps even preventive interventions.10 For those with bipolar disorder, the Monitoring, Treatment and Prediction of Bipolar Disorder Episodes smartphone platform has been used in several studies to show how patterns related to the duration and number of incoming/outgoing phone calls may help classify manic versus depressive episodes,11 which likewise could be leveraged to detect and prevent episodes as they are just emerging. These results will need to be replicated and explored in further detail to fully understand their clinical implications.
Digital phenotyping can also be used to provide quantitative data pertaining to specific aspects of mental and behavioral health. For instance, abnormal sleep is a central feature in many mental health conditions, and one that data from phones and wearables is well suited to measure from multiple sources. One study12 explored the role of voice data collected from smartphones to quantify global sleep patterns, with data collected from over 8,000 participants. The authors concluded that social pressures weaken and conceal biological drives toward sleep, suggesting ways to quantify how our modern world may contribute to poorer health.12 Likewise, social functioning can also be quantified with digital phenotyping methods; researchers have used proximity data from smartphone Bluetooth signals to estimate a person's real-world social networks.13 As part of ongoing research, our group is now seeking to quantify cognition using smartphone apps that capture the temporal processes of taking a cognitive test on the phone screen itself while also capturing relevant environmental factors, such as where and when that test was taken. These examples of quantifying movement, sleep, social networks, and cognition highlight how digital phenotyping methods are already creating novel windows into understanding the fundamental constructs underlying many mental illnesses.
In addition to enabling remote assessments, digital phenotyping methods can support just-in-time adaptive interventions, or JITAIs14 (sometimes pronounced “jedis”), a new wave of digital interventions with the potential of delivering optimal dosing of therapeutics based on real-time information about the person's state or environment.15 Imagine just the right dose of mindfulness being delivered when a person is about to experience a panic attack, or a personalized medication reminder to help a patients with adherence at the location that they take their pills, or a perfectly timed call or text message from a trusted friend when a patient is at risk of abusing a substance. These examples and many others are likely to drive development of smartphone-based intervention apps to augment mental health care. Today, over 10,000 mental health apps are available for immediate download on commercial app marketplaces.16 And yet, few of these 10,000 are supported by actual evidence, let alone evidence of effectiveness, and digital interventions that adapt to the context and condition of the patient are only beginning to emerge. Linking digital phenotyping methods to digital interventions will be an important next step for the field; although research in this area is still nascent,17 we anticipate context-aware intervention apps will be an exciting growth area over the next few years.
Research efforts in digital phenotyping are also not limited to traditional academic settings. Large well-known technology companies have indirectly sponsored digital phenotyping related mental health efforts,18 and are conducting research on digital phenotyping for mood disorders.19 Given that digital phenotyping exists at the intersection of technology to capture continuous real-time data, statistical and computational methods to analyze that data, and health care to use the results, new stakeholders will continue to emerge in this space. Although there is no US Food and Drug Administration (FDA)-approved mental health digital phenotyping app or wearable device today, Apple's (Cupertino, CA) FDA clearance letters20 for both the electrocardiogram and irregular rhythm notification functions on the Apple Watch highlight new regulatory pathways for monitoring technologies. Issues of data safety/privacy as well as patient safety in using new digital tools like smartphone apps and wearables remain on ongoing topics of concern with no clear answers today, although currently the FDA is considering redoing its approval process for software as a medical device in its ongoing pre-certification pilot program.21 Professional organizations like the American Psychiatric Association have also started offering guidance on smartphone apps to help psychiatrists and the public make more informed decisions regarding selecting a mental health app.22
Potential Pitfalls of Remote Monitoring
The potential of remote monitoring to greatly expand the future psychiatrist's toolkit cannot be understated, particularly as smartphone sensors, analytical methods, and clinical use cases improve. And yet, this flexibility and potential also carry risks and formidable challenges that have yet to be overcome.
Some of these challenges are technical and will be worked through by the research and technical communities. For instance, mental health digital phenotyping studies today have yet to converge on standard methodologies required to perform systematic meta-analyses on the association between sensors and mental health status.23 Groups that maintain global technology standards are now working to establish such standards for mobile health tools.24 Reproducing individual digital phenotyping studies remains a challenge with few efforts undertaken and those that have often reporting differing results.25 This challenge will be mitigated as the field moves toward open source platforms that are more transparent than proprietary ones, making it easier to audit and reproduce studies.
Appropriate statistical tools to make sense of the vast amounts of data are still emerging from the research pipeline. Most current approaches to the analysis of smartphone sensor data use a two-step approach. For simple variables (eg, number of daily phone calls), raw data signals can be processed into daily summary statistics that are interpretable without much signal processing. However, many smartphone data streams are noisy or have gaps where data is missing, requiring model-derived statistics that jointly consider multiple streams of data simultaneously (eg, classification of physical mobility status from location, accelerometer, and gyroscope data) and/or fill in missing values using mathematical approximation methods (eg, imputation) prior to computing summary statistics. One article comparing methods to estimate how far someone has traveled on a given day demonstrated that appropriate imputation can result in a 10-fold improvement in accuracy across all mobility features,15 supporting cautious interpretation of digital phenotyping results as the field continues to optimize methods for data analysis.
The other set of challenges are sociocultural and involve gaining the trust of patients and providers and require clinicians to inform and guide the appropriate use cases where digital phenotyping might be useful. It remains unclear under what circumstances mental health patients will be eager to share their smartphone sensor data outside of the context of clinical studies in which they are compensated for their participation.26,27 Few studies report adherence rates or assess the factors contributing to nonadherence in their data.28 There is a clear need for an improved understanding of patient engagement and efforts to ensure that the people these technologies are designed for are truly represented in the research and development process.
Aside from the unknown willingness of patients to use digital phenotyping as a tool in their treatment, the ethics of digital phenotyping itself raise critical issues regarding patient autonomy and patient privacy among a host of other issues. Misuse of this type of data could cause serious harm to patients, and unanticipated consequences are important to consider. Resources like the Connected and Open Research Ethics initiative offer an open forum for all stakeholders to share best practices, pose questions, and collaboratively learn how to ensure that digital phenotyping evolves in a safe and ethical manner.29 More work is needed to ensure that patient perspective is included in this ethical discussion to ensure an equitable balance of privacy and accessibility that meets real-world concerns of patients and their families, not just providers and payers, who have different liability concerns. Currently mental health apps have a poor track record for protecting data and a recent study suggested that the majority of popular apps for depression and smoking cessation may transmit data without proper disclosure to users.30
Finally, not all digital phenotyping studies have successfully identified associations between sensor data and mental health status. Some variability of findings from study-to-study is of course to be expected in any field, but at this early stage it seems prudent to dedicate a reasonable amount of effort to replicate promising studies and to reproduce study findings from previously collected data. Like any emerging field, it is important to recognize the importance of these negative studies in shaping tool development.
Digital Mental Health for an Open Digital Age
As digital phenotyping and related methods mature, new efforts will continue to fuel the potential for remote monitoring and intervention strategies to gain traction among practicing clinicians. Because of the complexity of the collected data, such as its high volume, high temporal density, and the substantial heterogeneity and large amount of noise present in the data, we forecast the importance of open science, including both software and statistical methods development, in the next phase of digital phenotyping advances. Closed or propriety digital phenotyping efforts are inherently less reproducible than open ones and are thus of lesser value to the clinical community and patient populations. Open digital phenotyping platforms can support the truly scalable nature of this approach needed to establish normative benchmarks and standardized, interpretable methods. Recent examples of such efforts include the University of California, Los Angeles Grand Challenge,31 which aims to capture mental health- related digital phenotyping data from over 100,000 participants, the National Institutes of Health All of Us Study,32 which aims to capture broad physical and mental health data from 1 million participants in the United States, and the Remote Assessment of Disease and Relapse – Central Nervous System project, which aims to capture similar data across numerous European countries.33
In conclusion, the rapid increase in smartphone ownership and use over the past decade has opened new possibilities for collecting and synthesizing digital data in the service of improving our understanding and treatment of psychiatric conditions. Many efforts are already underway across academic and industry settings with promising findings that may transform the everyday practice of clinical psychiatry. And yet, these technological advancements introduce new challenges for the treating psychiatrist as with any new technology, including an expected evolution in both the evidence basis and ethical standards. Although digital phenotyping and associated intervention strategies hold great promise to facilitate real-time monitoring and treatment of patients, it will be critical for research to continue to evaluate the utility of these efforts through open, transparent development among a diverse community of stakeholders.34
- Torous J, Keshavan M. The future of psychoses as seen from the history of its evolution. Curr Behav Neurosci Rep. 2014;1(2):94–99. doi:. doi:10.1007/s40473-014-0011-4 [CrossRef]
- The Statistics Portal. Number of smartphone users worldwide from 2014 to 2020 (in billions). https://www.statista.com/statistics/330695/number-of-smartphone-users-worldwide/. Assessed April 17, 2019.
- Torous J, Kiang MV, Lorme J, Onnela JP. New tools for new research in psychiatry: a scalable and customizable platform to empower data driven smartphone research. JMIR Ment Health. 2016;3(2):e16. doi:. doi:10.2196/mental.5165 [CrossRef]
- Gay K, Torous J, Joseph A, Pandya A, Duckworth K. Digital technology use among individuals with schizophrenia: results of an online survey. JMIR Ment Health. 2016;3(2):e15. doi:. doi:10.2196/mental.5379 [CrossRef]
- Glick G, Druss B, Pina J, Lally C, Conde M. Use of mobile technology in a community mental health setting. J Telemed Telecare. 2016;22(7):430–435. doi:. doi:10.1177/1357633X15613236 [CrossRef]
- King AL, Valença AM, Nardi AE. Nomophobia: the mobile phone in panic disorder with agoraphobia: reducing phobias or worsening of dependence?Cogn Behav Neurol. 2010;23(1):52–54. doi:. doi:10.1097/WNN.0b013e3181b7eabc [CrossRef]
- Faurholt-Jepsen M, Busk J, Þórarinsdóttir H, et al. Objective smartphone data as a potential diagnostic marker of bipolar disorder. Austral N Z J Psychiatry. 2019;53(2):119–128. doi:. doi:10.1177/0004867418808900 [CrossRef]
- Gershon A, Ram N, Johnson SL, Harvey AG, Zeitzer J. Daily actigraphy profiles distinguish depressive and well periods in bipolar disorder. Clin Psychol Sci. 2016;4:641–650. doi:. doi:10.1177/2167702615604613 [CrossRef]
- Saeb S, Zhang M, Karr CJ, et al. Mobile phone sensor correlates of depressive symptom severity in daily-life behavior: an exploratory study. J Med Internet Res. 2015;17(7):e175. doi:. doi:10.2196/jmir.4273 [CrossRef]
- Barnett I, Torous J, Staples P, Sandoval L, Keshavan M, Onnela JP. Relapse prediction in schizophrenia through digital phenotyping: a pilot study. Neuropsychopharmacology. 2018;43(8):1660–1666. doi:. doi:10.1038/s41386-018-0030-z [CrossRef]
- Faurholt-Jepsen M, Vinberg M, Frost M, Christensen EM, Bardram JE, Kessing LV. Smartphone data as an electronic biomarker of illness activity in bipolar disorder. Bipolar Disord. 2015;17(7):715–728. doi:. doi:10.1111/bdi.12332 [CrossRef]
- Walch OJ, Cochran A, Forger DB. A global quantification of “normal” sleep schedules using smartphone data. Sci Adv. 2016;2(5):e1501705. doi:. doi:10.1126/sciadv.1501705 [CrossRef]
- Boonstra TW, Larsen ME, Townsend S, Christensen H. Validation of a smartphone app to map social networks of proximity. PloS One. 2017;12(12):e0189877. doi:. doi:10.1371/journal.pone.0189877 [CrossRef]
- Nahum-Shani I, Smith SN, Spring BJ, et al. Just-in-time adaptive interventions (JITAIs) in mobile health: key components and design principles for ongoing health behavior support. Ann Behav Med. 2017;52(6):446–462. doi:. doi:10.1007/s12160-016-9830-8 [CrossRef]
- Barnett I, Onnela JP. Inferring mobility measures from GPS traces with missing data [published online ahead of print October 26, 2019]. Biostatistics. doi:10.1093/biostatistics/kxy059 [CrossRef].
- Torous J, Roberts LW. Needed innovation in digital health and smartphone applications for mental health: transparency and trust. JAMA Psychiatry. 2017;74(5):437–438. doi:. doi:10.1001/jamapsychiatry.2017.0262 [CrossRef]
- Myin-Germeys I, Klippel A, Steinhart H, Reininghaus U. Ecological momentary interventions in psychiatry. Curr Opin Psychiatry. 2016;29(4):258–263. doi:. doi:10.1097/YCO.0000000000000255 [CrossRef]
- The Mood Challenge. http://www.moodchallenge.com/. Accessed April 24, 2019.
- Verily. Tackling mental health at Verily. https://blog.verily.com/2017/05/tackling-mental-health-at-verily.html. Accessed April 17, 2019.
- US Food and Drug Administration. Letter: Re: ECG app. https://www.accessdata.fda.gov/cdrh_docs/pdf18/den180044.pdf. Accessed April 24, 2019.
- US Food and Drug Administration. Digital health software precertification (pre-cert) program. https://www.fda.gov/medicaldevices/digitalhealth/digitalhealthprecertprogram/default.htm. Accessed April 17, 2019.
- Torous JB, Chan SR, Gipson SY, et al. A hierarchical framework for evaluation and informed decision-making regarding smartphone apps for clinical care. Psychiatr Serv. 2018;69(5):498–500. doi:. doi:10.1176/appi.ps.201700423 [CrossRef]
- Rohani DA, Faurholt-Jepsen M, Kessing LV, Bardram JE. Correlations between objective behavioral features collected from mobile and wearable devices and depressive mood symptoms in patients with affective disorders: systematic review. JMIR MHealth UHealth. 2018;6(8):e165. doi:. doi:10.2196/mhealth.9691 [CrossRef]
- IEEE Standards Association. P1752 - Standard for mobile health data. https://standards.ieee.org/project/1752.html. Assessed April 17, 2019.
- Asselbergs J, Ruwaard J, Ejdys M, Schrader N, Sijbrandij M, Riper H. Mobile phone-based unobtrusive ecological momentary assessment of day-to-day mood: an explorative study. J Med Internet Res. 2016;18(3):e72. doi:. doi:10.2196/jmir.5505 [CrossRef]
- Torous J, Wisniewski H, Liu G, Keshavan M. Mental health mobile phone app usage, concerns, and benefits among psychiatric outpatients: comparative survey study. JMIR Ment Health. 2018;5(4):e11715. doi:. doi:10.2196/11715 [CrossRef]
- Mueller NE, Panch T, Macias C, Cohen BM, Ongur D, Baker JT. Using smartphone apps to promote psychiatric rehabilitation in a peer-led community support program: pilot study. JMIR Ment Health. 2018;5(3):e10092. doi:. doi:10.2196/10092 [CrossRef]
- Waalen J, Peters M, Ranamukhaarachchi D, et al. Real world usage characteristics of a novel mobile health self-monitoring device: results from the Scanadu Consumer Health Outcomes (SCOUT) Study. PLoS One. 2019;14(4):e0215468. doi:. doi:10.1371/journal.pone.0215468 [CrossRef]
- Torous J, Nebeker C. Navigating ethics in the digital age: introducing Connected and Open Research Ethics (CORE), a tool for researchers and institutional review boards. J Med Internet Res. 2017;19(2):e38. doi:. doi:10.2196/jmir.6793 [CrossRef]
- Huckvale K, Torous J, Larsen ME. Assessment of the data sharing and privacy practices of smartphone apps for depression and smoking cessation. JAMA Netw Open. 2019;2(4):e192542. doi:. doi:10.1001/jamanetworkopen.2019.2542 [CrossRef]
- UCLA Grand Challenges. Understanding, preventing and treating the world's greatest health problem. https://grandchallenges.ucla.edu/depression/. Accessed April 17, 2019.
- National Institutes of Health. The future of health begins with you. https://allofus.nih.gov/. Accessed April 17, 2019.
- RADAR-CNS. Remote assessment of disease and relapse – central nervous system. https://www.radar-cns.org/. Accessed April 17, 2019.
- Torous J, Baker JT. Why psychiatry needs data science and data science needs psychiatry: connecting with technology. JAMA Psychiatry. 2016;73(1):3–4. doi:. doi:10.1001/jamapsychiatry.2015.2622 [CrossRef]