Digital Phenotyping

Digital phenotyping (also known as behavioral sensing or personal sensing) is a moment-by-moment quantification of individual-level phenotypes using data from sensors embedded in personal digital devices, specifically smartphones, wearable devices, and computers (1). Data is often collected passively without requiring active input from the user. A phenotype is generated based on a collection of various behavioral data: geospatial data via GPS, movement behaviors via accelerometer, voice samples via microphones, and social interactions via call and text logs and Bluetooth (2). Some mobile applications can prompt users to generate inputs, such as completing symptom rating scales or diaries. The passive data collection can provide information relevant to psychiatric, aging, and other illness phenotypes.

Digital phenotyping

Digital phenotyping is an emerging area of research and have a wide range of potential clinical utilities. Adolescents and young adults are the most commonly studied populations because personal smartphone ownership occurs at such high rates. A potential opportunity with digital phenotyping relates to applying it to the mechanisms and behaviors underlying psychiatric disorders rather than outcomes alone.

Reasons for digital phenotyping

Reasons for mental health digital phenotyping may include:

1. Prevention – Digital phenotypes allow for detection of high-risk alcohol use by using passive identification of binging episodes using activity and phone utilization data in order to trigger prevention interventions (3).

2. Screening and diagnosis – Digital phenotypes can be used to identify individuals who are at-risk or undiagnosed for certain conditions such as mood disorders (4).

3. Monitoring – Voice analysis offers a tool to predict treatment outcomes in illnesses where predicting clinical response is low (5).

4. Treatment – Opiate overdose detection is an example of combining digital phenotypes with wearable devices. A harm reduction intervention aided by smartphones uses digital phenotypes to detect signs of respiratory distress, which can lead to early interventions to administer naloxone or connect with medication assisted treatment once medically stable (6).

Barriers and limitations

There are several barriers to using digital phenotyping in clinical practice. Since the field of digital phenotyping is new, early studies often include small samples, short study timelines, and occasionally used inappropriate cross-validation techniques. Smaller studies risk being inadequately powered to create reliable and generalizable results to the broader population of interest. While these proof-of-concept, feasibility, and pilot studies are appropriate to studying digital phenotyping, failure to replicate studies has already been observed (7). Digital phenotyping will need to show cost-effectiveness and efficacy. Equity is an issue for implementing digital phenotyping. It will be important for technologies to developed on multiple platforms instead of capturing data on the newest devices. At the same time, there must be a balance of updating the underlying algorithms used to generate the digital phenotype. Machine learning is used as the technology for interpreting signals into meaningful and actionable information. However, some machine learning is trained in limited populations, which can lead to bias in the population of interest. Privacy and ethical issues should be considered because many of the machine learning algorithms rely of passively collected data. Data collected from passive GPS traces can be personally identifiable and pose risks of identification.

References

1. Torous J, Kiang M, Lorme J, Onnela JP (2016). New tools for new research in psychiatry: A scalable and customizable platform to empower data driven smartphone research. JMIR Mental Health 3, e16 (2016).

2. Huckvale, K., Venkatesh, S. & Christensen, H. Toward clinical digital phenotyping: a timely opportunity to consider purpose, quality, and safety. npj Digital Medicine 2, 88 (2019).

3. Santani, D. et al. DrinkSense: characterizing youth drinking behavior using smartphones. IEEE Transaction Mobile on Computing 17, 2279–2292 (2018).

4. Faherty, L. J. et al. Movement patterns in women at risk for perinatal depression: use of a mood-monitoring mobile application in pregnancy. Journal of the American Medical Informatics Association. 24, 746–753 (2017).

5. Mundt, J. C., Vogel, A. P., Feltner, D. E. & Lenderking, W. R. Vocal acoustic biomarkers of depression severity and treatment response. Biological Psychiatry 72, 580–587 (2012).

6. Nandakumar, R., Gollakota, S. & Sunshine, J. E. Opioid overdose detection using smartphones. Scientific Translational Medicine. 11, eaau8914 (2019).

7. 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. Journal of Medical Internet Research. 18, 68–68 (2016).

Submitted by (Samuel Lindner)