Skip to content
Sleep & circadian healthCohort study3 min read

Wearable sleep and activity data predicted mood episodes in 55 patients, but only modestly overall

MoveVitals Editorial Collective · Study published · Reading updated

Original paper:Mood Prediction of Patients With Mood Disorders by Machine Learning Using Passive Digital Phenotypes Based on the Circadian Rhythm: Prospective Observational Cohort Study. · Journal of medical Internet research · 2019 · Original paper ↗

Key takeaway

Using circadian-rhythm features from wearable and smartphone data in 55 patients with mood disorders, a random forest model predicted mood state for the next 3 days with 65% accuracy (AUC 0.7). Accuracy was higher for episode types, but the sample is tiny and the abstract does not describe testing on separate patients.

Study design

Cohort study

Reading basis

Published abstract

Reading focus

Association & confounding

The research question

Mood disorders are closely tied to disturbed circadian rhythms, and wearables record activity, sleep, light and heart rate continuously. The authors tested whether passive digital data organized around circadian rhythm could predict a patient's coming mood state or episode.

Who and how

Prospective observational cohort of 55 patients with mood disorders (major depressive disorder, bipolar disorder type 1 and type 2) followed for 2 years. A smartphone app recorded self-rated daily mood scores and detected light exposure, and daily worn activity trackers collected activity, sleep and heart rate. The passive data were turned into 130 features based on circadian rhythms, and a mood prediction algorithm was developed with random forest. The abstract does not describe how the model was validated.

What it found

The authors’ reported results. Read each estimate with its population and uncertainty.

  1. 1
    Predicting mood state for the next 3 days, accuracy was 65% in all patients, 65% in major depressive disorder, 64% in bipolar I and 65% in bipolar II, with AUC values of 0.7, 0.69, 0.67 and 0.67.
  2. 2
    For specific episodes in all patients, accuracy was 85.3% for no episode, 87% for depressive, 94% for manic and 91.2% for hypomanic episodes, with AUC values of 0.87, 0.87, 0.958 and 0.912.
  3. 3
    In bipolar II patients, accuracy was 82.6% for no episode, 74.4% for depressive and 87.5% for hypomanic episodes (AUC 0.919, 0.868 and 0.949), with generally good sensitivity and specificity.

Limits & uncertainty

  1. 1
    Only 55 patients were studied, with even fewer in each diagnosis group, so the accuracy figures are unstable.
  2. 2
    The abstract does not say whether the model was tested on patients it had not seen, so the accuracy may be optimistic.
  3. 3
    Mood scores were self-recorded by patients, and accuracy for next-3-day mood was only 65%, which would produce many wrong predictions in practice.
  4. 4
    The authors expect clinical benefit but the abstract shows no test of whether using predictions improves outcomes, and this reading is based on the abstract only.

What to take from it

This is early research suggesting that sleep, activity, light and heart rate patterns hold some signal about coming mood shifts, not a tool you can use to predict your own mood. If your mood or sleep changes sharply, talk to a clinician rather than relying on app predictions.

Reading your own data

Sleep duration, sleep timing, daily activity and resting heart rate are the kinds of data used here. Watching your own regularity (consistent sleep and wake times) over weeks on the same device is reasonable, but a wearable cannot diagnose or predict a mood episode.

How to read this evidence

Exposure precedes follow-up outcomes, but confounding and reverse causation can remain. A hazard ratio is not an individual absolute risk or a demonstrated intervention effect.

Authors’ structured abstract English

The attributed English abstract is reproduced under the recorded open licence. It is source material, not a MoveVitals interpretation. The prepared analysis above separates the authors’ findings from our interpretation.

Background

Virtually, all organisms on Earth have their own circadian rhythm, and humans are no exception. Circadian rhythms are associated with various human states, especially mood disorders, and disturbance of the circadian rhythm is known to be very closely related. Attempts have also been made to derive clinical implications associated with mood disorders using the vast amounts of digital log that is acquired by digital technologies develop and using computational analysis techniques.

Objective

This study was conducted to evaluate the mood state or episode, activity, sleep, light exposure, and heart rate during a period of about 2 years by acquiring various digital log data through wearable devices and smartphone apps as well as conventional clinical assessments. We investigated a mood prediction algorithm developed with machine learning using passive data phenotypes based on circadian rhythms.

Methods

We performed a prospective observational cohort study on 55 patients with mood disorders (major depressive disorder [MDD] and bipolar disorder type 1 [BD I] and 2 [BD II]) for 2 years. A smartphone app for self-recording daily mood scores and detecting light exposure (using the installed sensor) were provided. From daily worn activity trackers, digital log data of activity, sleep, and heart rate were collected. Passive digital phenotypes were processed into 130 features based on circadian rhythms, and a mood prediction algorithm was developed by random forest.

Results

The mood state prediction accuracies for the next 3 days in all patients, MDD patients, BD I patients, and BD II patients were 65%, 65%, 64%, and 65% with 0.7, 0.69, 0.67, and 0.67 area under the curve (AUC) values, respectively. The accuracies of all patients for no episode (NE), depressive episode (DE), manic episode (ME), and hypomanic episode (HME) were 85.3%, 87%, 94%, and 91.2% with 0.87, 0.87, 0.958, and 0.912 AUC values, respectively. The prediction accuracy in BD II patients was distinctively balanced as high showing 82.6%, 74.4%, and 87.5% of accuracy (with generally good sensitivity and specificity) with 0.919, 0.868, and 0.949 AUC values for NE, DE, and HME, respectively.

Conclusions

On the basis of the theoretical basis of chronobiology, this study proposed a good model for future research by developing a mood prediction algorithm using machine learning by processing and reclassifying digital log data. In addition to academic value, it is expected that this study will be of practical help to improve the prognosis of patients with mood disorders by making it possible to apply actual clinical application owing to the rapid expansion of digital technology.

Sources & attribution

Mood Prediction of Patients With Mood Disorders by Machine Learning Using Passive Digital Phenotypes Based on the Circadian Rhythm: Prospective Observational Cohort Study.

Cho CH, Lee T, Kim MG, In HP, Kim L, Lee HJ.

Journal of medical Internet research · 2019-04-17

DOI: 10.2196/11029 ↗PubMed · 30994461 ↗Europe PMC · Recorded licence: cc by ↗

Abstract © the original authors, reproduced under the recorded open licence. Translations support reading; use the English record to check terminology and numerical results. Our methodological interpretation is separate from the authors’ conclusions.

MoveVitals editorial interpretation · educational content · no individual clinical review recorded

Sources checked 2026-11-01

Practical guides

Read in context

·Trial meta-analysis

Exercise improved sleep-quality scores in adults 45 and older with sleep disorders

Pooling 23 randomized trials with 1,478 adults aged 45 to 77 who had sleep disorders, exercise improved Pittsburgh Sleep Quality Index scores (SMD 0.62, 95% CI 0.40–0.83), and aerobic exercise showed the largest effect in the network analysis (SMD 1.34, 95% CrI 0.86–1.82). Certainty was low to moderate, the outcomes were questionnaires, and the best weekly dose of about 539 MET-min/week is a model estimate, not a prescription.

Journal of Exercise Science & Fitness

In-depth explainersRead study →
·Cohort study

Sleep regularity tracked mortality more closely than sleep duration in 60,977 UK adults

Among 60,977 UK Biobank adults who wore a wrist accelerometer for a week, the most regular fifth of sleepers had a fully adjusted all-cause mortality HR of 0.70 (95% CI 0.59–0.83) against the least regular fifth, over about 6.3 years and 1859 deaths. Regularity fitted the data better than sleep duration, but the study is observational and rests on one week of sleep data.

Sleep

In-depth explainersRead study →
·Randomised trial

Sleep extension cut daily energy intake by about 270 kcal in adults with overweight

In 80 adults with overweight who usually slept under 6.5 hours, two weeks of sleep-extension counselling added about 1.2 hours of sleep and lowered measured energy intake by 270.4 kcal/d versus control (95% CI −393.4 to −147.4). The trial was short, single-centre and in a narrow group, so it does not show lasting weight loss or any benefit for people who already sleep enough.

JAMA Internal Medicine

In-depth explainersRead study →
·Cohort study

Six hours or less of sleep at 50 and 60 linked to higher dementia risk

In 7959 British civil servants followed for about 25 years, sleeping six hours or less at age 50 (HR 1.22, 95% CI 1.01–1.48) or age 60 (HR 1.37, 1.10–1.72) was linked to a higher dementia rate than sleeping seven hours, and persistent short sleep gave an HR of 1.30 (1.00–1.69). Sleep was mostly self-reported and the study is observational, so early brain disease or other factors may explain part of the link.

Nature Communications

In-depth explainersRead study →
Sleep & circadian health →