Can phone sensors flag how depressed someone is?
In 28 adults carrying a sensor phone for 2 weeks, less regular GPS movement and more phone use tracked higher PHQ-9 scores; normalized entropy classified any vs no depressive symptoms at 86.5% accuracy.
Source
Mobile Phone Sensor Correlates of Depressive Symptom Severity in Daily-Life Behavior: An Exploratory Study
Study at a glance
- Design
- Cross-sectional — 2-week exploratory sensor study: community adults carried Purple Robot; PHQ-9 at baseline correlated with GPS/usage features
- N
- N=28 · 40 recruited; 28 with sufficient sensor data (18 GPS, 21 phone usage); 12 excluded for <50% data coverage
- Population
- US community adults (19–58 years; 20 women, 8 men in the analyzed set) recruited via Craigslist
- Outcome
- Correlation of GPS/phone-usage features with PHQ-9 severity; classification of PHQ-9 ≥5 vs <5
Structured fields used in claim comparison tables when every cited study has a complete layer.
Inspect source passages
Can phone sensors flag how depressed someone is?
These passages occur verbatim in the stored paper text. They support the points listed below; they do not verify every statement in the explanation.
40 recruited; 28 analyzed.
A total of 40 adult participants were recruited from the general community to carry a mobile phone with a sensor data acquisition app (Purple Robot) for 2 weeks.
Normalized entropy classifier accuracy 86.5%.
Using the normalized entropy feature and a classifier that distinguished participants with depressive symptoms (PHQ-9 score ≥5) from those without (PHQ-9 score <5), we achieved an accuracy of 86.5%.
What they did
Recruited 40 community adults to carry Purple Robot for 2 weeks (GPS every 5 minutes; screen on/off). Baseline PHQ-9 was correlated with GPS features (circadian movement, normalized entropy, location variance) and phone-usage duration/frequency in the 28 people with enough data.
What they found
Circadian movement r=−.63 (P=.005), normalized entropy and location variance each r=−.58 (P=.012). Usage duration r=.54 (P=.011) and frequency r=.52 (P=.015). A normalized-entropy classifier of PHQ-9 ≥5 vs <5 reached 86.5% accuracy; PHQ-9 regression error was 23.5%.
The limits
What it doesn't show
n=28 with a one-time self-report PHQ-9 cannot diagnose clinical depression or prove sensors should replace interviews; 12/40 lacked usable data, so selection bias is possible.
Key terms
- PHQ-9
- Patient Health Questionnaire-9; self-report depressive-symptom score from 0–27; ≥5 marked any symptoms here.
- Circadian movement
- How strongly GPS traces follow a 24-hour rhythm; lower values tracked higher PHQ-9 (r=−.63).
- Normalized entropy
- How evenly time is split across location clusters (0–1); lower values meant sticking to fewer places and tracked depression.
- Location variance
- Log of GPS latitude+longitude variance; a mobility marker independent of which places were visited.
- Purple Robot
- Android sensing app that sampled GPS every 5 minutes and logged screen-on usage.
- NRMSD
- Normalized root-mean-square deviation; percent error when estimating PHQ-9 from a sensor feature (23.5% here).
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
Analyzed n was:
Common questions
How many people were analyzed?
28 of 40 recruits (18 GPS, 21 usage).
What PHQ-9 split was used?
≥5 vs <5; 14 people in each group among the 28.
Best GPS correlation?
Circadian movement r=−.63, P=.005.
Classification accuracy?
86.5% using normalized entropy.
More on mHealth