What Is Digital Phenotyping? The Concept Explained

Digital phenotyping is an approach that uses the data personal digital devices such as smartphones and wearables leave in daily life to quantify a person’s behavior and state moment by moment, in real-world settings. By analogy with the phenotype produced by genes, the name means “the phenotype produced by device use.”

This concept drew particular attention in mental health research. A person’s state changes continuously throughout daily life rather than in the brief moment of a clinic visit, and digital devices open a window onto that flow of daily life. This article lays out the origin of the concept, the kinds of signals, its meaning in mental health, and the ethics and privacy issues that must be addressed alongside it.

---

01. Where the Concept Came From

Digital phenotyping was established through two strands of discussion.

• Onnela & Rauch (2016) proposed digital phenotyping as a methodology for strengthening research on behavior and mental health using smartphone-based data. The core is the use of data that occurs in-situ, at high frequency, and naturally.
• Insel (2017) framed it as “technology for a new science of behavior,” discussing it as a way to complement the limits of the subjective reports and one-off observations that clinical psychiatry has long relied on.

In other words, digital phenotyping is not a particular product but closer to a research paradigm that seeks to measure state from the everyday digital traces.

02. Passive Signals and Active Signals

Digital phenotyping data is broadly divided into two kinds.

Passive data

• Information collected without the user specifically being conscious of it or inputting it.
• Examples: movement and activity patterns, rhythms of device use, temporal patterns of screen interaction, and so on.
• The advantage is that it imposes little burden and is continuous; the disadvantage is that it is hard to interpret without context and is highly privacy-sensitive.

Active data

• Information obtained only when the user directly responds or inputs.
• Examples: surveys, mood check-ins, intentionally recorded voice, and so on.
• The advantage is that it carries context and self-report together; the disadvantage is that it creates participation burden and response bias.

Voice spans these two axes. Voice a user leaves intentionally is closer to active, while voice that occurs naturally during interaction is closer to passive. What signals voice carries is covered in detail in the Complete Guide to Voice Biomarkers.

03. Why It Matters in Mental Health

Traditional mental health assessment relies heavily on brief interviews in the clinic and subjective recall. The problem is that a person’s state keeps changing by day, by week, and by situation. A single snapshot in time easily misses this variability.

Digital phenotyping seeks to complement this point.

• High-frequency, continuous observation: Instead of rare interviews, it can observe many time points in daily life.
• Ecological validity: Data is collected in real-life context rather than in a laboratory.
• Capturing within-person change: Instead of comparing to others, it can observe changes from an individual’s usual state.

This last point is especially important. Because each person has a different baseline, the real value of digital phenotyping lies not in absolute-value comparison but in observing change relative to a personal baseline. This principle was also explained as the core of early detection in Early Detection of Employee Burnout.

04. Limits You Must Know

Digital phenotyping is appealing, but overconfidence is dangerous.

1. Variability in reproducibility and effect size. The association between signal and state can vary greatly by dataset, population, and environment.
2. Lack of context. Passive signals often cannot explain why a given pattern appeared.
3. Bias. Training on data skewed toward particular demographic groups leads to unbalanced performance.
4. It is not diagnosis. Digital phenotyping is a method for observing and researching state, not a medical diagnosis.
5. Confounders. Voice in particular is heavily affected by colds, sleep, noise, device, and language, so it cannot be confirmed from a single sample.

05. Ethics and Privacy: The First Issue to Address

Digital phenotyping deals with a person’s most private everyday data. That is why ethics must be discussed before technical performance.

• Consent: Users must clearly understand and consent to what is collected, why, and how.
• Minimized collection and purpose limitation: Use only as much as needed, only for the defined purpose.
• De-identification and access control: Minimize exposure in individually identifiable form.
• Preventing power asymmetry: Especially in employment contexts, design so that data is not used against the individual.

SYMPLE takes these principles as the starting point of its design. It aids self-understanding for individuals, while providing organizations only with group-level signals that cannot identify individuals. We do not aim for approaches that point to a specific individual’s state.

06. Summary

• Digital phenotyping = an approach that seeks to quantify behavior and state in real-world settings from the everyday data of personal devices.
• Insel (2017) and Onnela & Rauch (2016) established the concept.
• Signals are divided into passive and active, and voice spans the two.
• It has limits of reproducibility, context, and bias, and is closer to observing change than to diagnosis.
• Above all, privacy and consent come first.

What SYMPLE builds on these principles can be found in About SYMPLE.

---

References

1. Insel TR. Digital phenotyping: technology for a new science of behavior. JAMA. 2017;318(13):1215–1216.
2. Onnela JP, Rauch SL. Harnessing smartphone-based digital phenotyping to enhance behavioral and mental health. Neuropsychopharmacology. 2016;41(7):1691–1696.
3. Low DM, Bentley KH, Ghosh SS. Automated assessment of psychiatric disorders using speech: A systematic review. Laryngoscope Investigative Otolaryngology. 2020;5(1):96–116.
4. Cummins N, Scherer S, Krajewski J, et al. A review of depression and suicide risk assessment using speech analysis. Speech Communication. 2015;71:10–49.