Can phone movement data flag changes in depression before they happen?
When people's phones showed they were moving around less and visiting fewer places, their depression scores tended to rise afterwards, but changes in mood did not predict later changes in movement.
Source
Evaluation of Changes in Depression, Anxiety, and Social Anxiety Using Smartphone Sensor Features: Longitudinal Cohort Study
Study at a glance
- Design
- Cohort — Prospective 16-week cohort with passive Android sensor collection and symptom questionnaires every 3 weeks; lagged repeated-measures correlations between changes in sensor features and later changes in symptoms, overall and within 4 symptom clusters.
- N
- N=282 · 282 participants overall; k-means symptom clusters contained 88 (minimal symptoms), 71 (depression and social anxiety), 69 (depression and anxiety) and 54 (multiple comorbidities).
- Population
- US adults with Android phones recruited through volunteer research registries, deliberately oversampled so at least half had moderate or worse depressive symptoms; people with bipolar or psychotic disorders were excluded.
- Outcome
- Changes in depression (PHQ-8), generalised anxiety (GAD-7) and social anxiety (SPIN) scores, correlated with prior 2-week changes in GPS, semantic location, communication and app-use features.
Structured fields used in claim comparison tables when every cited study has a complete layer.
What they did
The researchers had 282 US adults install an app that passively logged GPS location, calls, texts and app use for 16 weeks, and every 3 weeks participants completed questionnaires on depression, generalised anxiety and social anxiety. They grouped participants into four symptom profiles and correlated changes in sensor features over two weeks with the next change in symptom scores, and also tested the reverse direction.
What they found
Increases in GPS-measured location variety, time patterns and travel were followed by decreases in depression scores, with modest correlations (for example r = -0.17 in the full sample and -0.36 in the group with multiple comorbidities). More active app use was linked to falling depression in the multiple-comorbidity group, and more phone calls to rising social anxiety in the depression and social anxiety group. No sensor features tracked later changes in generalised anxiety, and symptom changes did not predict later sensor changes.
The limits
What it doesn't show
The study was exploratory and observational, so it shows a time ordering, not that moving less causes depression to worsen; the authors say their hypotheses need confirmatory testing. Correlations were modest, and the authors state phone data alone are not good enough to monitor symptoms. Only Android users who volunteered through research registries took part, people with missing surveys had somewhat higher symptoms, and findings apply only to the 2-week lag examined. An unexpected positive link between time at exercise locations and depression was likely an artefact of many zero values.
Key terms
- Passive sensing
- Collecting behavioural data from device sensors such as GPS without the user having to do anything.
- PHQ-8
- An eight-item self-report questionnaire measuring the severity of depressive symptoms.
- Location entropy
- A measure of how evenly a person spreads their time across different places; low values mean time is concentrated in few locations.
- Repeated measures correlation
- A correlation that estimates the within-person association between two variables measured several times, rather than differences between people.
- k-means clustering
- A data-driven method that sorts people into a chosen number of groups with similar scores, here on baseline symptom items.
- Benjamini-Hochberg correction
- A procedure that adjusts p-values to control the false discovery rate when many tests are run.
Flashcards
0 of 10 answers reviewed
Research intelligence for this paper
See its role on concept claims, tensions it is part of, placement history, and related discoveries.
Quiz yourself
How often did participants complete symptom questionnaires?
Common questions
Does this show that getting out more will reduce depression?
No. The data show that changes in movement came before changes in depression scores, which is consistent with that idea, but the study did not manipulate movement, so another factor could drive both.
Why group participants into symptom clusters?
People with depression often also have anxiety or social anxiety, and different symptom combinations may show up in different behaviours; clustering let the authors see, for example, that call frequency mattered mainly when social anxiety was present.
Could this be used for clinical monitoring now?
The authors say not on its own: effect sizes were modest. They suggest sensor data might help time supportive messages in digital interventions or be combined with wearables.
More on mHealth