Skip to main content
. 2023 Dec 13;25:e46778. doi: 10.2196/46778

Table 4.

Mood disorders: retrieved studies and their main outcomes.

Study Country (region) Data Psychiatric disorder Symptoms investigated Assessment technology Sample, n Data collection time Statistics Synthesis of main results
Bipolar disorder

Busk et al [19], 2020 Denmark 2020 Bipolar disorder Activity, alcohol use, anxiety, irritability, cognitive difficulties, medication, mood, sleep, and stress Active: self-report questionnaire on mood, HDRS (Hamilton Depression Rating Scale), YMRS (Young Mania Rating Scale) 84

  • Period of collection: 9 months

  • Frequency of collection: active (5 times, expert ratings and daily, self-ratings)

 Hierarchical Bayesian regression model
History of 4 days of self-assessment:

  • R2=0.51;

  • RMSEa=0.32

 The proposed method can predict mood for up to 7 days with low error compared with classical machine learning methods.

Ebner-Priemer et al [20], 2020 Karlsruhe, Germany 2020 Bipolar disorder Sleep, activity, and sociality Passive (smartphone): frequency and length of calls, screen illumination, speed of transmitted and received data, distance traveled, frequency of different classes of activity, speed of movement and number of steps
Active: expert ratings and self-ratings		 29

  • Period of collection: 12 months

  • Frequency of collection: passive (continuously) and active (biweekly, expert ratings and daily, self-ratings)

 Day-to-day association:

  • mania status and activity (β=.123; 95% CI [0.075 to 0.170]).

  • mania and sleep (β=−.098 95% CI [−0.157 to −0.040])

 Results showed that shorter sleep duration and increased activity were associated with higher levels of mania, whereas higher than average activity correlated with lower levels of depression.

Faurholt-Jepsen et al [21], 2021 Copenhagen, Denmark 2021 Bipolar disorder Movement (place, stops, moves), location entropy, routine (average distances for each day) Passive (smartphone): GPS, Wi-Fi, mobile repeater signals
Active: daily self-report of mood

  • bipolar=46




  • controls=31

 Period of collection: 9 months
Frequency of collection: passive (continuously) and active (daily)		 Location entropy:
BDb vs HCc (β=−.14; 95% CI=−0.24 to −0.034; P=.009)		 This study shows how alterations in location data, reflecting patterns of mobility, may prove to be a promising measure of illness and disease fluctuations in patients with bipolar disorder, and can be used to monitor the effects of treatments.

Tseng et al [22], 2022 Taiwan 2020-2021 Bipolar disorder Sleep and emotional status Passive (smartphone): GPS data
Active: YMRS, HAMD, ASRMd, and DASSe-21; daily mood, walking time, and bed time		 159

  • Period of collection: 13 months

  • Frequency of collection: passive (continuously) and active (daily)

 Associations
Daily mood:

  • Week-to-week:

  • Movement (r=0.213; P<.001)

  • Month-to-month

  • Movement (r=0.199; P<.001)

Sleep duration:

  • Week-to-week:

  • Movement (r=0.344; P<.001)

  • Daily mood (r=0.073; P<.001)

  • Month-to-month

  • Movement (r=0.676; P<.001)

Movement:

  • Week-to-week:

  • Sleep (r=0.264; P<.001)

  • Daily mood (r=0.295; P<.001)

  • Month-to-month

  • Sleep (r=0.663; P<.001)

  • Daily mood (r=0.211; P<.001)

 The smartphone app has the potential to provide an informative and reliable means for real-time tracking of BD status.
Major depressive disorder

Abbas et al [23], 2021 Canada 2021 Major depressive disorder Digital Markers of Major Depressive Disorder (facial and vocal characteristics) Passive (smartphone): internal camera, microphone
Active: Montgomery-Asberg Depression Rating Scale (MADRS)		 18

  • Period of collection: 4 weeks

  • Frequency of collection: 3 time points (baseline, 2 weeks, 4 weeks)

 Repeated measures ANOVA for markers from baseline to 4 weeks of antidepressant treatment.
Neutral stimuli:

  • Voice percentage (F2,26=5.6; P<.009)

  • Overall expressivity (F2,28=32.6; P<.001)

  • Head movement mean (F=8.9; P<.007)

  • Head pose change mean (F=5.01; P<.033)

Positive stimuli:

  • Voice percentage (F2,26=3.59; P<.004)

  • Overall expressivity (F2,28=40.67; P<.001)

  • Head movement mean (F=3.58; P<.041)

Negative stimuli:

  • Voice percentage (F2,26=4.66; P<.019)

  • Overall expressivity (F2,28=36.95; P<.001)

 Digital markers of motor functioning associated with Major Depressive Disorder demonstrate validity as measures of response to antidepressant treatment

Bai et al [24], 2021 Beijing, China 2021 Major depressive disorder Mood status and emotional stability Passive (wearable) through smartphone app: sleep, heart rate, steps count
Passive (smartphone): GPS, messages, calls, screen lock and unlock, app use
Active: VASf PHQ-9g, GAD-7h 		334

  • Period of collection: 12 weeks

  • Frequency of collection: passive (continuously) and active (VAS: daily, PHQ-9: biweekly, GAD-7: 5 times)

 The highest accuracy of classification between steady and mood swing=76.67% (SD 8.47%) The results showed that the best model used was one in which measurements of sleep, heart rate, step count, and call logs were considered.

Jacobson et al [25], 2020 Lebanon, New Hampshire (United States) 2020 Major depressive disorder Position, movement, light exposure, heart rate, cardiac variability Passive (smartphone): GPS, Wi-Fi, Google Places, finger pressure on rear camera
Active: DASS-14 (depression scale), PANAS-Xi 		31

  • Period of collection: 7 days

  • Frequency of collection: passive (continuously) and active (hourly)

 Correlation between predicted depressed mood levels and observed depressed mood r=0.587, 95% CI (0.552-0.621) Passively collected smartphone data can accurately predict future depressed mood in a sample reporting clinical level of depression.

Laiou et al [26], 2022 London, United Kingdom, 2022 Major depressive disorder Movement, socially relevant activities, environmental factors (noise and light) Passive (smartphone): GPS, incoming and outgoing calls
Active: PHQ-8		 164

  • Period of collection: 2 years

  • Frequency of collection: passive (continuously) and active (every 14 days)

 Association between stay home and symptoms severity:

  • Weekdays (95% CI 0.023-0.178; median 0.098; home stay: 25th-75th percentiles 17.8-22.8; median 20.9 h a day)

  • Weekends (95% CI −0.079 to 0.149, median 0.052; home stay: 25th-75th percentiles 19.7-23.5; median 22.3 h a day)

 The results suggest that staying at home is associated with the severity of major depressive disorder symptoms and illustrate that passive detection of individuals with depression is possible and may provide important clues for monitoring the course of major depressive disorder symptoms.

Pedrelli et al [27], 2020 United States 2020 Major depressive disorder Smartphone use, activity levels, skin conductance, heart rate variability, sleep, social interaction Passive (2 wristbands) through smartphone app: electrodermal activity, peripheral skin temperature, heart rate, motion from IMUj, and sleep from actigraphy
Active: Hamilton Depression Rating Scale 17 items (HDRS-17)		 31

  • Period of collection: 8 weeks

  • Frequency of collection: passive (continuously,22 h a day/7 day a week) and active (6 times)

 Correlation between estimate model and clinical-rated assessment = (0.46; 95% CI 0.42-0.74 to 0.7; 95% CI 0.66 to 0.74) Monitoring patients with major depressive disorder via smartphones and wrist sensors is feasible and can provide an estimate of changes in the severity of depressive symptoms.

Cho et al [28], 2020 Seoul, Republic of Korea 2020 Major depressive disorder Circadian rhythm (heart rate, activity, sleep, light exposure), mood Passive (wearable) through smartphone application: activity, sleep, and heart rate
Active: daily self-report via eMoodChart

  • With app CMRk=14

  • Without app CMR=59

  • Period of collection: 12 months

  • Frequency of collection: passive (continuously) and active (daily)

 CRM group vs non-CRM group:

  • Total depressive episodes (n/year; exp β=.033; P=.03), 96.7% fewer

  • Shorter depressive episodes (total; exp β=.005; P<.001), 99.5% fewer

  • Shorter manic or hypomanic episodes (exp β=.039; P<.001), 96.1%

  • Total mood episodes (exp β=.026; P=.008), 97.4%

  • Shorter mood episodes (total; exp β=.011; P<.001), 98.9%

 The results of the study confirmed that providing daily circadian rhythm–based mood prediction feedback through the CRM app with a wearable activity tracker, analyzing the life patterns of individual patients with mood disorders, significantly reduced the number and duration of mood episodes compared with those in the control group.
Mood disorder

Chikersal et al [29], 2021 United States 2021 Mood disorder Depressed mood, movement, communication, smartphone use, sleep, physical activity Passive (smartphone): Bluetooth, calls, GPS, screen
Passive (wearable): steps and sleep
Active: BDI-IIl 		138

  • Period of collection: 16 weeks

  • Frequency of collection: passive (continuously) and active (twice, at the beginning and at the end)

 Detection accuracy:

  • Post semester depressive symptoms=85.7%

  • Symptom severity=85.4%

 Results showed that the best feature model for predicting depression was the one using a 7-feature set, including Bluetooth, calls, campus map, location, phone use, steps, and sleep.

Cho et al [30], 2019 Seoul, Republic of Korea 2019 Mood disorder Mood, activity, sleep, light exposure, heart rate Passive (smartphone): light exposure
Passive (wearable) through smartphone app: sleep, activity, and heart rate
Active: eMoodchart app		 55

  • Period of collection: 2 years

  • Frequency of collection: passive (continuously) and active (daily)

 Mood state prediction accuracy for the next 3 days:

  • All patients: 65%, AUCm=0.7

  • Major depressive disorder: 65%, AUC=0.69

  • Bipolar I: 64%, AUC=0.67

  • Bipolar II: 65%, AUC=0.67

Accuracy for all patients predictions:

  • No episode: 85.3%, AUC=0.87

  • Depressive episode: 87%, AUC=0.87

  • Manic episode: 94%, AUC=0.958

  • Hypomanic episode: 91.2%, AUC=0.912

 The study authors obtained a machine learning model capable of estimating the degree of accuracy of each patient’s mood status over a 3-day period, demonstrating that it is possible to develop effective learning models for predicting mood in patients with mood disorders.

Mehrotra et al [31], 2016 Heidelberg, Germany 2016 Mood disorder Notification management, smartphone use, depressed mood Passive (smartphone): number of notifications clicked, average time to view notifications, notification response time, number of apps started, time spent on apps, phone unlock count
Active: PHQ-8		 25

  • Period of collection: 30 days

  • Frequency of collection: passive (continuously) and active (every day for 14 days)

 Correlation between depression and all notification metrics of the last 14 days ranged from 0.4 to 0.6 The results suggest that using data from the last 14 days of monitoring can improve the accuracy of predicting the depressive score a user will have on the current day.

Wahle et al [32], 2016 Switzerland 2016 Mood disorder Position, movement, mood Passive (smartphone): phone use, accelerometry, Wi-Fi, GPS
Active: PHQ-9		 126

  • Period of collection: 9 months

  • Frequency of collection: passive (continuously) and active (biweekly)

 Binary classification performance for biweekly PHQ-9:

  • Random Forest model=60.1%

  • Support vector machine=59.1%

 Results showed that participants with clinical levels of depression and who had adherence of 8 weeks or more (N=12), had lower PHQ-9 scores at the end of the study (P=.01). Participants who used the app for an extended period showed significant reduction in self-reported symptom severity.

aRMSE: root mean square error.

bBD: Bipolar Disorder

cHC: Healthy Controls

dASRM: Altman Self-Rating Mania.

eDASS: Depression, Anxiety and Stress Scale.

fVAS: Visual Analog Scale.

gPHQ: Patient Health Questionnaire.

hGAD-7: Generalized Anxiety Disorder-7.

iPANAS-X: Positive and Negative Affect Schedule-Expanded version.

jIMU: Inertial Measurement Unit

kCMR: Circadian Rhythm of Mood

lBDI-II: Beck Depression Inventory-II.

mAUC: Area Under the Curve.