Skip to main content
DATACC BY DIME PROJECT

EMERGING PHENOTYPING APPLICATIONS

Exploring digital measures for phenotyping in common mental health disorder trials

People with the same common mental health disorder (CMHD) diagnosis often show different patterns of daily functioning, different symptom trajectories, and different responses to treatment. Traditional visit‑based assessments and questionnaires often miss these day-to-day differences.

Digital phenotyping can quantify meaningful aspects of health that are both shared across CMHD and highly individual-specific, in a more continuous and passive manner.

This “emerging application” discussion covers what is already possible with phenotyping today and which new applications are on the horizon.

Where phenotyping adds value

Phenotyping can be used to design trial enrichment strategies and plan subgroup analyses. Digital phenotypes should be simple to explain. Poorly defined or overly complex phenotypes can undermine recruitment, stratification, and enrichment strategies by failing to replicate across settings or studies. If you cannot describe what defines a phenotype in everyday clinical language, treat it as preliminary findings and do not yet let it inform decision-making about trial design, operations, or trial endpoints.

Digital measures capture how measures of sleep, physical activity, variability in daily routines, emotional states, and social interactions actually change across days and weeks, within individuals and across populations, in ways that clinic visits cannot.

Most digital phenotyping studies in common mental health disorders are small and exploratory, using sensor‑derived measures and clinical scales to describe samples, track symptom trajectories, or test simple subgroups. Evidence that these approaches can support reliable clinical decision-making, standalone risk assessments, or population stratification is emerging.

First, define the use case and research question. What decision or design question should this phenotyping work inform? For example, “Can we detect distinct daily‑functioning phenotypes in a population with major depressive disorder?” or “Are there subgroups whose symptom trajectories diverge over time after starting a treatment?”. Next, list measures and data collection approaches: which digital measures (sleep, activity, routines, autonomic) and which clinical assessments will you use, and over what time window (for example, baseline only vs 12‑week follow‑up)?

Once the use case and measures are set, the next decision is how to combine them into a phenotype.

Phenotyping strategies you can use now

Digital phenotypes in CMHD can combine elements from behavioral, physiological, psychological, environmental, and social phenotypes. We describe two approaches with precedent.

cmhd-icon Strategy 1: Define subgroups in your study population
WHY IT MATTERS:

Reveals clinically-meaningful phenotypes from digital measures and COAs in heterogeneous CMHD populations. May provide powerful insights for future study enrichment or stratification strategies.

Digital Mental Health StudyEXAMPLE:

The UCLA Digital Mental Health Study (DMHS) enrolled more than 4,000 participants with depression and anxiety for a full year with iPhone and Apple Watch data plus repeated symptom scales. This kind of long‑term dataset makes it possible to describe longitudinal symptom trajectories.

Read the publication

cmhd-icon Strategy 2: Describe how symptom trajectories evolve over time
WHY IT MATTERS:

Measuring how symptoms evolve over time from baseline establishes intraindividual safety, exacerbation trigger, and responder profiles.

DiMe-Case-Study-Healthesphere-PhenotypingA multimodal digital phenotyping workflow combined ecological momentary assessment (EMA), wearable‑derived sleep and activity trends, smartphone behavioral signals, and clinical/contextual data into longitudinal models for mental health monitoring. Behavioral, physiological, patient‑generated, and clinical data streams were normalized into interoperable structures and summarized as clinically interpretable trends between visits, rather than raw data feeds. In work led by Healthesphere, this proof‑of‑concept approach illustrates how trajectory‑focused phenotyping can inform longitudinal monitoring models, but it remains hypothesis‑generating and does not by itself validate any specific phenotype for individual‑level treatment decisions or stratification.

Read the case study

Behaviour Research and TherapyIn a 12-week trial of an app-based behavioral activation intervention for young adults with elevated depression symptoms, Funkhouser and colleagues modeled each participant’s individual trajectory across six passively sensed behaviors (time at home, walking, stationary time, time in bed, bedtime, waketime) over the first two weeks of treatment, then tested whether those early-change trajectories predicted PHQ-8 improvement from pre- to post-intervention. Greater early decreases in time spent at home (an indicator of reduced behavioral withdrawal) predicted greater symptom improvement, while early trajectories of the other behaviors did not. As a single small secondary analysis in one digital behavioral activation trial, the result is hypothesis-generating, not a validated indicator of who will respond, and should inform future trial design rather than drive individual treatment decisions.

Read the publication

What comes next?

Digital phenotyping performed at a larger scale will significantly improve our evidence-base on important symptom states in CMHD, ultimately accelerating the implementation of these technologies into a care delivery environment to drive clinical-decision making and complement structured assessments and clinician judgment. Making that possible will require careful considerations of clinical workflows, staffing, patient burden and stigma (including unintended behavioral changes due to perceived surveillance), interoperability, escalation logic and liability, consent and privacy.
EXAMPLE

For example, Lightfully Behavioral Health found high engagement with DHT solutions that are transparent and when data is co-interpreted by the clinician and patient. Sharing clear insight dashboards during a session helped clients see exactly what was being tracked and why, turning data into a collaborative tool rather than something done to them, making completion rose to about 90%. In contrast, a post‑discharge text tool that labeled people “high risk” without explaining why, and replied with robotic messages, left patients feeling scrutinized and sometimes stigmatized.

“Implementation needs to be grounded in consent, trust building, and application. Clients need to know what’s collected, how it’s used, what happens if it signals a concern.”

– Nicole Siegfried, PhD, CEDS, Chief Clinical Officer, Lightfully

Similarly to the use case of digital phenotyping to inform clinical research decisions, the use of digital measures in CMHD is motivated by the need to extend clinical understanding beyond episodic assessments into real-world context. First, we must improve our scientific understanding of where these data points can inform thoughtful longitudinal care for CMHD.

What the field needs to answer

Several questions about digital phenotyping in CMHD trials are still unsettled

  • Can baseline sleep and activity-based phenotypes, with appropriate context information, improve detection of meaningful differences in treatment response for one or more specific CMHD?
  • Can digital phenotypes lead to an improvement in trial (operational) efficiencies?
  • Do trajectory and subgroup phenotypes built in one CMHD population hold when tested in similar populations at other sites or in other studies that use comparable measures and protocols?
  • At what point is a digitally-measured subgroup phenotype, replicated across multiple datasets, strong enough to evaluate the definition of a diagnostic subtype?

Conclusion

Digital phenotyping in common mental health disorders is still emerging, but the foundations are within reach today. Start with a clear research question, the right digital measures, and a phenotype simple enough to describe in clinical language. As the field builds replicated evidence, treat early findings as hypothesis-generating, not as drivers of individual treatment decisions.

To position digital measures as endpoints in your trial, see Using digital measures as endpoints in mental health trials. For partner perspectives on how these measures inform longitudinal care, see the Lightfully Behavioral Health example and Videra Health case study.

Back to project homepage