Published on 13.08.18 in Vol 6, No 8 (2018): August
Preprints (earlier versions) of this paper are available at http://preprints.jmir.org/preprint/9691, first published Dec 19, 2017.
Correlations Between Objective Behavioral Features Collected From Mobile and Wearable Devices and Depressive Mood Symptoms in Patients With Affective Disorders: Systematic Review
Background: Several studies have recently reported on the correlation between objective behavioral features collected via mobile and wearable devices and depressive mood symptoms in patients with affective disorders (unipolar and bipolar disorders). However, individual studies have reported on different and sometimes contradicting results, and no quantitative systematic review of the correlation between objective behavioral features and depressive mood symptoms has been published.
Objective: The objectives of this systematic review were to (1) provide an overview of the correlations between objective behavioral features and depressive mood symptoms reported in the literature and (2) investigate the strength and statistical significance of these correlations across studies. The answers to these questions could potentially help identify which objective features have shown most promising results across studies.
Methods: We conducted a systematic review of the scientific literature, reported according to the preferred reporting items for systematic reviews and meta-analyses guidelines. IEEE Xplore, ACM Digital Library, Web of Sciences, PsychINFO, PubMed, DBLP computer science bibliography, HTA, DARE, Scopus, and Science Direct were searched and supplemented by hand examination of reference lists. The search ended on April 27, 2017, and was limited to studies published between 2007 and 2017.
Results: A total of 46 studies were eligible for the review. These studies identified and investigated 85 unique objective behavioral features, covering 17 various sensor data inputs. These features were divided into 7 categories. Several features were found to have statistically significant and consistent correlation directionality with mood assessment (eg, the amount of home stay, sleep duration, and vigorous activity), while others showed directionality discrepancies across the studies (eg, amount of text messages [short message service] sent, time spent between locations, and frequency of mobile phone screen activity).
Conclusions: Several studies showed consistent and statistically significant correlations between objective behavioral features collected via mobile and wearable devices and depressive mood symptoms. Hence, continuous and everyday monitoring of behavioral aspects in affective disorders could be a promising supplementary objective measure for estimating depressive mood symptoms. However, the evidence is limited by methodological issues in individual studies and by a lack of standardization of (1) the collected objective features, (2) the mood assessment methodology, and (3) the statistical methods applied. Therefore, consistency in data collection and analysis in future studies is needed, making replication studies as well as meta-analyses possible.
JMIR Mhealth Uhealth 2018;6(8):e165
Recently, there has been an increasing body of research investigating the use of mobile and wearable devices as a treatment intervention for depression . Several mobile solutions have been proposed to utilize a self-monitoring and intervention-based treatment of depression [ - ]. One particular research approach adopted by many research groups has been to investigate how objectively measured behavioral features such as “location” and “social interaction” correlate with depression; using this approach, they have tried to differentiate euthymic and depressed states [ - ]. For example, using a mobile phone app passively recording information from sensors in the phone, Saeb et al [ ] could show a statistically significant correlation between 6 different objective features, including mobile phone usage frequency and self-assessed mood using the Patient Health Questionnaire-9 (PHQ-9) scale [ ] in nonclinical samples. Similarly, Faurholt-Jepsen et al [ ] found 5 different objective features, including the number of outgoing short message service (SMS) text messages, which had a statistically significant positive correlation with depression severity as assessed using the Hamilton Depression Rating Scale (HDRS) in patients with bipolar disorder (BD).
The diagnostic process, as well as the process of symptom severity assessment in affective disorder, is based upon a combination of clinical evaluations and patient information, and there is a lack of objective markers of, for example, trait and state.
Digital behavioral markers have been defined as higher-level features reflecting behaviors, cognitions, and emotions, which are measured using low-level features and sensor data collected from digital technology, including mobile and wearable computing devices . Many studies have found statistically significant correlations between objective behavioral features collected from mobile and wearable devices and mood symptoms in nonclinical samples of participants without psychiatric illnesses [ - ] as well as in clinical samples of patients diagnosed with psychiatric disorders [ , - ].
The discovery of such significant correlations between objective features and depressive mood symptoms has raised great enthusiasm regarding using mobile and wearable devices in the treatment and monitoring of depression and other affective disorders. It has been argued that such an approach may provide an easy and objective way to monitor illness activity and could serve as a digital marker of mood symptoms in affective disorders [, ]. Thus, if there is a well-established correlation between a specific digital marker—such as the number of steps taken and depressive mood symptoms—it would, in practice, be possible to develop an entirely automatic monitoring system. When, for example, the measured objective feature deviates from healthy behavior, an alarm or trigger could be raised in the clinic, which then could contact the patient [ ].
However, when looking across individual studies, it is not easy to identify which objective features consistently correlate with depressive mood symptoms and in what way. Some studies have shown similar results, while others have shown contradicting results. For example, Beiwinkel et al  found a statistically significant negative correlation between the number of outgoing SMS text messages and the HDRS, whereas Faurholt-Jepsen et al [ ] found a statistically significant positive correlation. Asselberg et al [ ] found a negative correlation with mobile phone usage frequency and depressive symptoms, while Saeb et al [ ] found the opposite.
No prior work has presented a comprehensive quantitative overview of objectively collected mobile features and how they relate to depressive mood symptoms. A more qualitative overview has recently been provided by Dogan et al , which highlights different mobile systems that have been developed to record subjective and objective features of individuals with affective disorders. They describe the findings of 29 different studies divided into different feature categories, such as physical activity, location, and phone usage, in a study-by-study evaluation.
Hence, a relevant question arises: to what degree studies show similar or different correlations between objective features and depressive mood symptoms, and how strong these correlations are? The purpose of this paper is to provide a systematic review of the available studies investigating the correlation between objectively collected features from mobile and wearable devices and depressive mood symptoms measured using various methods. Our systematic review aims to answer the following questions: (1) Which objective features have been collected? (2) What is the correlation between objective features and depressive mood symptoms? (3) Are the correlations similar across studies collecting the same features? Answering these questions could help us identify which objective features have shown most consistency across multiple studies and assist in designing future studies using technologies for objective assessment of depressive mood symptoms.
Systematic Review Process
We initiated the systematic review by following the PICO (Patient problem Intervention, Comparison, and Outcome) worksheet guidelines . Then, we conducted and reported the systematic review according to the preferred reporting items for systematic reviews and meta-analyses statement [ ].
Inclusion and Exclusion Criteria
The following inclusion criteria were met with the included original papers: (1) The study involved any type of objectively measured features; (2) the data were collected via a mobile phone or other nonintrusive consumer-based mobile or wearable device; (3) participants were assessed on a mood scale, which included self-reported scales (eg, PHQ-9) or clinical diagnostic scales (eg, HDRS) used within psychiatry to quantify abnormal depressed mood either prior, during, or within the poststudy period; (4) comparisons of the objective features and the assessed depression scales between or within subjects were available or provided upon request from the respective corresponding author; (5) and as per the PICO Search Strategy, the following publication types were included: Meta-Analysis, Cohort study, Systematic Review, Case-Control Study, Randomized Controlled Trial, and Case series or report.
To ensure a broad inclusion of studies investigating the relationship between objective features and mood symptoms, the third statement was deliberately chosen to reflect a broad selection of clinical and nonclinical participants rated on different mood scales. This included both commonly used and clinically verified rating scales, such as the HDRS and PHQ-9, as well as nonstandard scales designed for a specific usage or technology, such as the 7-point (−3 to 3) scale used in the MONARCA (MONitoring, treAtment and pRediCtion of bipolAr Disorder Episodes) project [, ].
We excluded original papers on the following premises: (1) nonquantitative studies or studies where only subjective features were collected; (2) if no English version of the paper was available; (3) studies that included participants with disorders other than mood disorders; (4) studies with nonhuman participants; (5) studies within social media since this topic has been thoroughly investigated elsewhere ; (6) studies with participants <18 years of age [ ], to keep the focus on behavioral objective features collected on adults; (7) studies conducted before January 1, 2007; (8) studies that have not been published through peer review; and (9) the following publication types: trial protocols, in vitro or lab research, animal research, and editorials or letters or opinions.
The corresponding author (DAR) searched the following databases on November 25, 2016 to target both clinical and technical scientific literature: IEEE Xplore, ACM Digital Library, Web of Sciences, PsychINFO, PubMed, DBLP computer science bibliography, HTA, DARE, Scopus, and Science Direct. Systematic reviews and meta-analysis publications were included in the search for a subsequent cited reference search, which was conducted on April 27, 2017.
A broad database-specific search string was designed to target all studies that investigated mood disorders within a mobile setting. The specific search string for PubMed was as follows:
(smartphone OR mobile OR wearable OR “smart phone” OR app OR apps) AND (depression OR bipolar OR unipolar OR “affective disorder” OR “mental health” OR “mood disorder”) AND (“2007/01/01”[Date–Publication]: “2017/01/01” [Date–Publication]) AND English[Language]
The search strings for the other databases can be found in.
The resulting publications were combined to one large spreadsheet, using an in-house Matlab script, with header information: database, title, author, publication year, publication type, and publisher.
After removal of duplicates, studies were screened for eligibility in two phases. In phase 1, one author (DAR) excluded the studies based on the title. The title revealed several exclusion criteria, including different disorders (Alzheimer, schizophrenia, diabetes, chronic pain, autism, Parkinson, PTSD, or anorexia nervosa); nonhuman experiments; mobile phone addiction topics; focuses on diary methods, which only involve subjective data; use of internet-based interventions; and nonmedical-related topics such as bipolar electricity. In phase 2, one author (DAR) went through the abstract. If eligible, the full text was retrieved and reviewed. We excluded studies in which no objective features were collected, studies that only used self-assessment, and studies concerning emotion.
The resulting list, together with review papers from phase 1, were then used in a cited reference search by two authors (DAR, JEB) to produce the final list. The final list was critically investigated by all authors, which led to the exclusion of 16 papers due to outcome measures that did not represent mood assessment (eg, happiness scales [- ], Quality of Life [ ], or Satisfaction With Life Scale [ ], as these do not reflect abnormal depressed mood) or wearables that were not consumer based (eg, a Holter monitor [ ] or multisensory clothing [ - ]).
Several studies only reported correlation strengths or did not include correlation results between the objective features and the outcome assessment [, , , ]. For these studies, we contacted the corresponding author via email and acquired the relevant data in all cases. The results of the study selection process are outlined in .
Data were extracted from the final list by one author (DAR) in a predetermined format validated by a second author (JEB). The data were extracted into 2 separate tables; one for nonclinical samples of participants without psychiatric illnesses () and one for clinical samples of patients diagnosed with Unipolar Disorder (UD) or BD ( ). The division into 2 tables was reviewed by all authors. Both tables listed the following data for each study; first author, year of publication, the specification of the mobile device, number of participants, participant age, days of the study, and the outcome depression scale. included a diagnosis column. The supplementary material contains expanded versions of and (found in and respectively), which also include information about the method of recruitment and the method of assessing the relation between objective features and mood symptoms (eg, Pearson correlation, two-sample t test). The tables in and provide a detailed overview of the different features for each study, classified into a feature category, the sensor used, a small description, and the results with respect to the mood assessment.
We were interested in investigating the correlation between behavioral objective features and depressive mood symptoms across all the included individual studies. To do this, we first identified all types of objective features, which have been applied in the eligible studies. The features were presented in a nomenclature list to create a standardized definition across all studies. Second, we investigated the strength of the correlation between objective features and depressive mood symptoms (ie, the correlation coefficient) across the included studies.
The investigation was performed by combining the directionality of the correlation values for identical objective features, weighted by the respective sample size and visualized as the x-axis and total sample size (log-transformed) on the y-axis. This was done in two separate graphs: one presenting nonclinical samples of participants (presents data from ) and the other presenting clinical samples of patients diagnosed with either UD or BD ( presents data from ; shows patients with BD only). The two groups would most likely display different behaviors, and the separation was done on this premise. However, a combined result is displayed in for the convenience of the reader.
A positive directionality indicates that a larger quantity of the respective feature tends to give a higher depression score (eg, lower mood score, indicating a positive correlation with the depression score), while a negative directionality indicates that a larger quantity of the feature value tends to give a lower depression score (eg, a larger mood score, indicating a negative correlation with the depression score). All correlation values with outcome measures that represented larger values with better mood outcomes were multiplied by −1 to achieve the same weighted correlation directionality across studies.
|Reference||Technology used||Participants (N=1189), n||Participant age (years), mean (SD)||Study duration (days)||Mood scale|
|Asselbergs et al, 2016 ||Android; Funf||5||22||21.1 (2.2)||36||10p mood|
|Baras et al, 2016 ||Android; EmotionStore||9||1||N/Aa||14||BRUMSb|
|Becker et al, 2016 ||Android; Funf||5||22||N/A||42||Mood|
|Ben-Zeev et al, 2015 ||Android||37||10||22.5||70||PHQ-9c|
|Berke et al, 2011 ||Multisensor (waist)||4||4||85.3 (4.1)||10||CES-Dd|
|Canzian and Musolesi, 2015 ||Android; MoodTraces||15||13||31||71||PHQ-8e|
|Cho et al, 2016 ||Phone records||234||298||57||N/A||BDI-21f|
|Chow et al, 2017 ||Android||35||37||19.8 (2.4)||17||DASS-21g|
|DeMasi et al, 2016 ||Android||17||27||N/A||56||BDI-21|
|Edwards and Loprinzi, 2016 ||Digi-Walker Pedometer||16||23||21.82||7||PHQ-9|
|Farhan et al, 2016 ||Android or iOS; |
|Mark et al, 2016 ||Fitbit flex||20||20||N/A||12||Affect balance|
|Matic et al, 2011 ||Windows M. 6.5; |
|Mehrotra et al, 2016 ||Android||25j||N/A||N/A||30||PHQ-8|
|Mestry et al, 2015 ||Android||1||1||22||34||DASS21|
|Pillai et al, 2014 ||Actigraph||10||29||19.55 (3.2)||7||BDI-21|
|Saeb et al, 2015 ||Android; Purple robot||8||20||28.9 (10.1)||14||PHQ-9|
|Saeb et al, 2016 ||Android; Studentlife||38||10||N/A||70||PHQ-9|
|Wang et al, 2014 ||Android; Studentlife||38||10||N/A||70||PHQ-9|
|Wang et al, 2015 ||Android; Studentlife||37j||N/A||N/A||70||PHQ-9|
aN/A: not applicable.
bDepression subscale of Brunel Mood Scale.
cPHQ-9: Patient Health Questionnaire-9
dCES-D: The Center for Epidemiological Studies Depression Scale.
ePHQ-8: Patient Health Questionnaire-8
fBDI-21: Becks depression inventory.
gDASS-21: Depression Anxiety Stress Scales.
hStudy reported participant age as a range, rather than mean.
irPOMS: reduced Profile of Mood States.
jTotal number of participants; number of male and female participants not specified.
|Reference||Technology used||Participants (N=3094), n||Clinical diagnosis||Participant age (years), mean (SD)||Study duration (days)||Mood scale|
|Abdullah et al, 2016 ||Android; MoodRhythm||2||5||BD||25-64a||28||SRM II-5b|
|Alvarez-Lozano et al, 2014 ||Android; Monarca||18c||N/Ad||BD||N/A||150||7p mood|
|Beiwinkel et al, 2016 ||Android; SIMBA||8||5||BD||47.2 (3.8)||365||HDRSe|
|Berle et al, 2010 ||Actigraph||10||13||UD||42.8 (11)||14||Group difference|
|Dickerson et al, 2011 ||iOS; Empath||0||1||UD||83||14||10p mood|
|Doryab et al, 2016 ||Android||3||3||UD||>18f||20||CES-Dg|
|Faurholt-Jepsen et al, 2012 ||Actiheart||8||12||UD||45.2 (12)||3||Group difference|
|Faurholt-Jepsen et al, 2015 ||Actiheart||7||11||UD||45.6 (11.1)||3||HDRS-17|
|Faurholt-Jepsen et al, 2016 ||Android; Monarca||9||19||BD||30.3 (9.3)||84||HDRS-17|
|Faurholt-Jepsen et al 2014 ||Android; Monarca||5||12||BD||33.4 (9.5)||90||HDRS-17|
|Faurholt-Jepsen et al, 2015 ||Android; Monarca||20||41||BD||29.3 (8.4)||182||HDRS-17|
|Faurholt-Jepsen et al, 2016 ||Android; Monarca||11||18||BD||30.2 (8.8)||84||HDRS-17|
|Gershon et al, 2016 ||Actigraph||14||23||BD||34.4 (10.4)||46||Group difference|
|Gonzales et al, 2014 ||Actigraph||15||27||BD||41.0 (11.2)||7||IDS-C-30h|
|Grünerbl; 2015 ||Android||2||8||BD||33-48||84||7p mood|
|Guidi et al, 2015 ||Android||0||1||BD||36||98||mood state|
|Hauge et al, 2011 ||Actigraph||14||11||UD||42.9 (10.7)||14||Group difference|
|Krane-Gartiser et al, 2014 ||Actigraph||5||7||BD||39.9 (15.6)||1||Group difference|
|Loprinzi and Mahoney, 2014 ||Actigraph (hip)||1261||1313||UD||46.3||7||Group difference|
|Miwa et al, 2007 ||Armband; SenseWear Pro||5||0||UD||35.1||87||Group difference|
|Muaremi et al, 2014 ||Android||6c||N/A||BD||18-65||76||7p mood|
|O’Brien et al, 2016 ||Actigraph||16||43||UD||74 (6)||7||MADRSi|
|Osmani et al, 2013 ||Android||0||5||BD||N/A||90||−3:3 moodj|
|Palmius et al, 2016 ||Android; AMoSS||9||27||BD||44 (14)||60||QIDS-SR16k|
|St-Amand et al, 2013 ||Actigraph||7||7||BD||44.6 (11)||14||Group difference|
|Todder et al, 2009 ||Actigraph||14||13||UD||49 (13)||7||Group difference|
aStudy reported participant age as a range, rather than mean.
bSRM II-5: Social Rhythm Metric II-5.
cTotal number of participants; number of male and female participants not specified.
dN/A: not applicable.
eHDRS: Hamilton Depression Rating Scale.
fAll participants in study above 18 years of age.
gCES-D: The Center for Epidemiological Studies Depression Scale.
hIDS-C-30: Inventory for Depressive Symptomatology, Clinical-rated.
iMADRS: Montgomery-Åsberg Depression Rating Scale.
j−3:3 mood: 7-point mood scale ranging from −3 to 3.
kQIDS-SR16: Quick Inventory of Depressive Symptomatology-Self Reported.
A meta-analysis of the specific correlation values was not considered for this systematic review. The heterogeneity across the studies was too substantial to perform any valid meta-analysis of correlations.
Not only were different analytical methods applied (eg, some using within-subject correlation others between subjects, some using day-averaged others week-averaged data) but also different apparatus and mood assessments were used. However, there is a clear correlation directionality invariance shown by studies comparing different analysis methods [, ] and studies replicating same analysis methods on different datasets [ , ], which puts forth the argument that the directionality is a stable metric. Regarding the specific correlation values, we still encourage the reader to look at the results across studies using , , , and as a reference.
Of 3507 potentially eligible studies, 46 met the criteria of the review. A flowchart of the screening process is shown in. Characteristics of the included studies are summarized in and .
lists studies including nonclinical samples of participants (n=20), and lists studies including clinical samples of patients diagnosed with either UD or BD (n=26). A more detailed overview of the included studies is listed in , , , and .
We identified 7 overall behavioral feature categories, which we denoted as “Feature Categories.” These categories used 17 unique data inputs to analyze 85 different objective features. The same features were used across studies, yielding 176 investigated features, with information about directionality with respect to the mood score on 155/176 (88%) of the cases. The other cases (n=21) report on accuracy and weightings by combining objective features into single evaluations, which was mostly observed in research papers with classification models [, , , ].
The 7 feature categories are defined and described in. An overview of the studies that contributed to each of the categories is provided in . The supplementary files also include a graph illustration of the data inputs and how they contributed to the different category ( ).
An in-depth analysis of each feature occurring in more than 2 studies is shown inand for nonclinical and clinical samples of participants, respectively. and were constructed as follows.
The x-axis is a weighted directionality of the correlation between the feature and mood symptoms. Positive values represent a larger depressive score and vice versa.(wD) is defined as:
Fx is the correlation value of a unique feature such as SMS text message sent. M is the total number of Fx across all studies where N is the combined total number of participants. “sgn” denotes the sign operation which is −1 for values below zero, and 1 for values above zero. As an example, when considering the correlation between screen active frequency and mood symptoms, according to the table in, this is analyzed in 2 studies; one study with N=28 shows a positive correlation, whereas one study with N=27 shows a negative correlation. This yields a wD as follows:
A wD value of 1 would indicate that all studies have a positive correlation between the measured feature and the mood assessment. This means that consistency across studies would place the feature on either +1 (consistent positive) or −1 (consistent negative) on the x-axis.
The y-axis is log-transformed values, to accommodate the large diversity, of the total number of participants on which the feature is measured. Nonclinical samples of participant studies measuring call frequency (n=370) had the highest average study participants, while clinical samples of participants measuring humidity (n=6) had the lowest.
The size of the feature pie chart represents M, which is the total number of studies of that particular feature. Similarly, the pie charts are divided into statistically significant (green), statistically nonsignificant (or lack of reporting; red) correlations, and missing information on statistical significance (gray).
In total,and provide an overview of the correlation between statistically significant features and depressive mood symptoms. Each feature is followed by the result reported in the figure, which is the wD value, the number of studies that included the feature (n), the percentage of statistically significant cases (s), and the mean (SD) of the participants included in the “n” studies.
For nonclinical samples of participants (), we observed the following:
Most studies, excluding call duration (wD=−0.04, n=4, s=25%, mean 278.50 [SD 293.32]), call frequency (wD=−0.04, n=3, s=33.33%, mean 370.67 [SD 279.44]), screen active frequency (wD=0.02, n=2, s=50%, mean 27.5 [SD 0.71]), and transition time (wD=−0.26, n=2, s=0%, mean 38.00 [SD 14.14]), agree on the correlation direction because most features are either at −1 or +1. Home stay (wD=1, n=4, s=75.00%, mean 56.75 [SD 23.32]), circadian rhythm (wD=−1, n=2, s=100%, mean 38.00 [SD 14.14]), and entropy (wD=−1, n=6, s=83.33%, mean 51.67 [SD 22.98]) have the largest number of statistically significant studies, whereas distance (wD=−1, n=4, s=0%, mean 45.75 [SD 24.10]), movement speed (wD=−1, n=2, s=0%, mean 63.50 [SD 21.92]), and transition time have no statistically significant studies.
|Social (n=38), with statistically significant correlation reported for 26% (10/38) results and statistical evaluation missing for 16% (6/38) results. Features describing social behavior, including activity related to phone calls, texting, social network size, and other people in the user\'s context.|
|Physical activity (n=48), with statistically significant correlation reported for 46% (22/48) results and statistical evaluation missing for 6% (3/48) results. Features describing physical activity, including movement and step count.|
|Location (n=38), with statistically significant correlation reported for 50% (19/38) results and statistical evaluation missing for 8% (3/38) results. Features describing mobility, including GPS tracking, clustering of location (eg, home stay), and transition time.|
|Device (n=24), with statistically significant correlation reported for 54% (13/24) results and statistical evaluation missing for 0% (0/24) results. Features describing device (mobile phone or wearable) usage, including app usage, lock or unlock events, and classification of app usage.|
|Subject (n=24), with statistically significant correlation reported for 50% (12/24) results and statistical evaluation missing for 21% (5/24) results. Features capturing the subject\'s physical state, including sleep and voice.|
|Environment (n=2), with statistically significant correlation reported for 0% (0/2) results and statistical evaluation missing for 0% (0/2) results. Features collected from the physical surroundings of the user.|
|Bio (n=2), with statistically significant correlation reported for 50% (1/2) results and statistical evaluation missing for 0% (0/2) results. Biometric features related to the subjects body.|
aSMS: short message service.
bGPS: global positioning system.
cGSM: Global System for Mobile communication.
eLED: light-emitting diode.
fEDA: electrodermal activity.
Similarly, for the clinical sample of patients (), we observed the following:
Most studies, excluding distance (wD=0.30, n=2, s=0%, mean 10.00 [SD 4.24]), humidity (wD=0, n=2, s=0%, mean 6.00 [SD 0.00]), SMS text message sent (wD=−0.50, n=4, s=25%, mean 30.00 [SD 21.76]), and activity (wD=−0.81, n=9, s=66.67%, m 23.33 [SD 16.10]), agree on the correlation direction because they are at either −1 or +1 on the wD axis. Cell tower ID (wD=−1, n=3, s=66.67%, mean 19.67 [SD 8.33]), screen active duration (wD=1, n=3, s=66.67%, mean 21.33 [SD 6.66]), and activity have the largest statistically significant percentage, whereas distance, SMS text message received (wD=−1, n=2, s=0%, mean 45.00 [SD 22.63]), and humidity have the lowest.
Several objective features were only included in a single study. Therefore, their relationship to a depressive mood scale cannot be compared across studies as done inand . Some of these features are quite creative and worth mentioning. The most promising results for the nonclinical samples include the time spent in break rooms (ρ=−0.21, nonsignificant) [ ], and less SD of stillness amount, which can be interpreted as a more uniform activity pattern (beta=−3.3, P<.001) [ ]. For the clinical samples, it includes the increased amount of time with no sound detection (speech pauses; beta=0.34, P=.004) [ ], increased number of calls missed (beta=0.05, P=.006) [ ], and fewer incidences of quick or sudden movements (jerk; t=4.06, P<.001).
Data and Methods Reporting
In the 46 eligible studies, 19 different mood assessment methods were used. The most common assessment method was the PHQ (n=9), whereas assessment methods like the Montgomery-Åsberg Depression Rating Scale and the Brunel Mood Scale , which are patient-reported outcome measures, were only used in a single study.
Seven different technologies were used for collecting the objective features. The most frequent one was mobile phone (n=30); mostly Android phones were used (n=27), with iOS (n=2) and Windows (n=1) phones also being used. Wearable devices were reported to be located on various areas on the body, including the upper arm , wrist (nondominant hand [ ], right hand [ ]), waist [ ], hip [ ], and chest [ ].
When analyzing the relation between the objective features and depressive mood assessments, we identified several regression-, machine learning–, correlation-, and group-difference methods. In total, across the 46 eligible studies, 12 different methods were used, with Pearson correlation being the most used (n=17).
Details on the analysis method were, in general, not well documented. This especially applied to studies where correlation analysis was secondary to the main hypothesis [, ]. Important details that were mentioned in few of the studies included possible confounding variables such as age, sex, and body mass index (BMI) [ , ]; data sampling methods such as global positioning system (GPS) polling strategies [ ]; the window-length in days or averaging methods of objective data that were correlated with the outcome [ , ]; and within-subject or between-subject analysis [ , ]. Full transparency, by providing the data, was seen in only 2 of the studies [ , ], with 5 studies using existing public data [ , , - ].
In this paper, we present the results of the first systematic review on the correlation between objective behavioral features collected via mobile and wearable devices and the assessment of depressive mood symptoms as measured by different rating scales and questionnaires. This was possible due to the increased research on mobile and wearable computing devices in the context of mental health [, , ], yielding 46 included studies in this review. We found that 57% (26/46) studies (a small majority) were performed on clinical samples of participants. However, when analyzing the number of participants included in these studies, they constituted a majority (3094/4283, 72%). We separated these two groups since nonclinical samples of participants are, by definition, healthy and, most likely, will display different behaviors than clinical samples of patients diagnosed with UD or BD.
We want to emphasize, for the subsequent discussions, that correlation assessments do not imply causality, but rather simple associations. The correlation between two measures could also be mediated through one or several covariates, which were not explored in any of the included studies . For instance, Disabato et al were able to validate a correlation by including a statistical mediation model [ ]. They concluded that the presence of positive life events mediated the correlation between gratitude and depression. A simple correlation assessment also does not provide knowledge on the clinical utility of these data in the classification of affective episodes in UD or BD because sensitivity, specificity, positive predictive values, and negative predictive values were not investigated in most studies. However, discovering and understanding the relationships between objective features and their relation to mood symptoms may be relevant in a clinical setting because it may provide an easy and objective way to monitor illness activity outside the clinical settings and could serve as a digital marker for mood symptoms [ ].
The social category had the lowest percentage of statistically significant correlations, by vote counting, across studies (10/38, 26%). Social included features such as call duration and number of conversations, which can be accessed on Android phones, contrary to iPhones . We did not find any research article that explains how social patterns change with depression, but the review article by Baker et al [ ] on online social networks suggests a complex relation involving factors that mediate or moderate the correlation and increase the variability in the findings. Furthermore, Cho et al [ ] found a direct opposite correlation between genders (male negative, female positive) in the call duration and call frequency features. This suggests that social-based features should be treated as a highly personalized feature that should be assessed in a within-subject analysis.
The feature category with the highest percentage of statistically significant correlation features across studies was device (13/24, 54%). As an example, using data provided by the corresponding author , we observed statistically significant results in communication app usage (r=−0.33, P=.007) calculated using a within-subjects analysis of covariance. The low variability with device-based features could indicate that there is a general tendency for participants to use their phones more, but at the same time, withdraw from the social context by lowering the communication app usage.
The feature category subject is similar to Device, investigated less but with a high percentage of statistically significant correlations across studies (12/24, 50%). This includes features within sleep and voice. In particular, sleep duration was the most investigated feature (n=6), with statistically significant correlations in 4 studies Furthermore, subject was one of the less included categories, which could be due to the second-level processing required to achieve features of voice  or sleep durations through multiple sensors [ ].
Nonclinical Samples of Participants
As seen in, we found two features that have a strong positive (ie, close to 1 wD) correlation with depression: home stay and screen active duration; both of these showed a large proportion of statistically significant correlations across studies. Moreover, all 4 studies with a positive correlation between home stay and depression level also had a large average participant number. Individual studies have shown that the degree to which a person stays at home is associated with depression [ ], and it is a general hypothesis that this relation is positive. We were able to verify this hypothesis by combining the results across the included studies in this review.
On the other hand, no prior hypothesis has been formulated regarding the relationship between general phone usage and depressive mood symptoms. However, studies have shown a statistically significant positive correlation between depressive symptoms and the feature screen active duration . Similarly, subjective-based mobile phone use has been studied in relation to depression, where Thomée et al found that high mobile phone use was associated with symptoms of depression [ ]. These findings were replicated in this review, with only a single statistically nonsignificant contradictive result from a two-sample study by Mestry et al [ ] (r=−0.03, P=.79).
On the left side in, we see several features that have a strong negative correlation to depression, including location clusters, entropy, and sleep duration. A majority of these features indicates that enhanced physical activity and more movement outside of the house are observed when participants score lower on the depression scale. This is consistent with the Actigraph systematic review papers by Scott et al [ ], who revealed a consensus of lower mean activity levels associated with bipolar depression, and Burton et al [ ], who revealed a pattern of lower daytime activity but higher nighttime activity in depression.
Entropy is the most prominent feature in the figure with many studies (n=6), all yielding a negative correlation and a high statistically significant proportion. The only case of nonsignificance was reported by Saeb et al  (r=−0.42, P=.082), who, however, did show a high negative correlation. Entropy is a measure that captures the distribution of time spent at the different location clusters registered. Thus, a high entropy would indicate that the participant spends time more uniformly across different location clusters. Because all studies consistently showed a negative correlation, this implies that a higher entropy correlates with a better mood. If a participant stays home for a longer time than usual, the entropy will drop. Hence, there is a dependency between entropy and home stay, which is also evident in the figure where they are almost mirrored, both with a large proportion of statistically significant findings. Both features can be collected via the location Application Programming Interface, which uses the GPS sensor typically embedded in all mobile phones or wearables.
Features with less consistent findings across studies regarding positive or negative correlations are located closer to 0 wD; these include features such as screen active frequency, call duration, call frequency, and transition time. At first look, it seems that these features are not related to mood symptoms and, hence, exhibit random correlation values. However, another explanation could be gender or cultural differences. In a cross-cultural study with people from Switzerland and Turkey, Hernández et al  found different correlation directions between the two groups in screen active frequency and number of running apps. Furthermore, several device-based features such as browser app usage and reading app usage have different correlation directions between genders (male positive, female negative) [ ], and the two social features call duration and call frequency also exhibit different correlation directions between genders (male negative, female positive) [ ].
Transition time has been currently only investigated by the research group of Saeb et al [, ], who conducted a study to replicate previous findings of the same features. The first study showed a positive correlation (r=0.21, P=.40), while a second study showed negative correlation (r=−0.32, nonsignificant). The feature then yields a low negative wD due to the latter including more participants and placed more centrally due to the contradictive results.
Clinical Samples of Patients Diagnosed With Unipolar or Bipolar Disorder
The feature screen active duration is similar to the nonclinical samples, with a high proportion of statistically significant studies and a consensus on positive correlation among the studies. Note, however, that this feature was the only one within the Device category that was investigated for both nonclinical and clinical samples of participants.
The features of sleep duration and distance have switched to a positive wD incompared with . Only 2 studies have investigated distance for clinical samples of patients. Beiwinkel et al [ ] reported a negative correlation in a between-subject analysis, but the within-subject analysis that we reported had almost zero correlation (r=0.03, P=.66). In contrast, Abdullah et al [ ] showed a negative correlation direction by the negative weighting coefficient (w=−1.56 × 10−2) using the Support Vector Machine analysis. However, with a small number of total participants, only 2 studies, both nonsignificant, distance was found to be weakly represented in the literature. Sleep duration, on the other hand, had statistically significant findings in both groups. This feature is a good example of the reasoning in analyzing depressed symptoms in clinical samples separately from the nonclinical sample. In clinical samples of nonseasonal depression, patients often suffer from abnormal sleep patterns with problems falling asleep, interrupted sleep, and early morning waking, while such a sleep pattern not is seen among healthy subjects.
Social-based features were more extensively investigated with clinical samples of patients. The two features incoming call duration and incoming call frequency reveal a strong tendency that participants tended to receive more calls and talk longer during these calls when depressed. On the other hand, the features outgoing call duration and outgoing call frequency tend to suggest that patients make more and longer calls when they are less depressed. This difference between incoming and outgoing calls highlights that these features should be kept separate, and it raises concerns with some of the results on Call duration with nonclinical samples of participants as in a study by Wang et al , who measured call duration and frequency across incoming and outgoing calls.
The feature SMS text messages sent was found to have a lower wD, showing inconsistencies across the 4 studies. We did not find any results in the literature that could explain the lower wD on SMS text messages sent, although the use of internet- and app-based chat and video communication platforms has been increasing, while SMS text message communication has fallen drastically. In Denmark, there has been a drop of 19.6% in SMS text messages sent from 2015 to 2016 . This suggests that SMS text message logging should be used with caution and should be extended to include other relevant messaging technologies.
The clinical sample consisted of both unipolar and bipolar patients. Optimally we would have liked to analyze data separately for these two patient groups due to findings that show psychomotor activity and sleep discrepancies between unipolar and bipolar depression . However, the focus here is on the level of depressive mood symptoms as a function of objective features, where bipolar and unipolar patients show same directionality compared with healthy controls [ ]. Nevertheless, we repeated the analysis of correlation directionalities, including patients with BD only, and the results remained unchanged. See .
Data Collection and Analysis Method
When combining the studies investigating objective features and their relation to mood symptoms, it became apparent that a meta-study on the exact correlation values would be misleading. The lack of detailed reporting on analysis methods was clearly demonstrated in a study by Beiwinkel et al , where a between-subject (cross-sectional analysis) relationship yielded a statistically nonsignificant (P=.82) regression coefficient of −0.04, while a within-subject (longitudinal analysis) relationship yielded a statistically significant (P=.03) regression coefficient of −0.11, on the feature of cell tower ID. Data aggregation length was also a concern because the duration of studies included in this review spans from 7 days [ , ] to 12 months [ ]. Canzian & Musolesi [ ] presented results on the correlation between PHQ-8 and different mobility features for 1 to 14 days of aggregation. The absolute correlation value increased from .152 (−.016 not absolute) to .432 on the feature maximum distance. The change was most likely due to a larger data pool, which lowered the variance toward “outlier” days or even noise in the data stream. It might also be related to the day of the week. For instance, Saeb et al [ ] found variations in the objective feature home stay between work days and weekends. Furthermore, the lack of reporting confounding variables in the analysis was a concern. Faurholt-Jepsen et al [ ] have demonstrated the effect of adding confounding variables to the analysis, where an unadjusted model without confounding variables on, for example, screen active duration (beta=194.8, P=.06) becomes statistically significant when controlling for age and sex (beta=209.6, P=.04).
To investigate depressive severity, many studies measured mood pre-, during, and poststudy, and there were correlation differences depending on when the mood assessment was done . In a more detailed study [ ], we saw a gradual lowering in correlation between various objective features and prestudy PHQ-9, which was not that surprising because the PHQ-9 questionnaire captures symptoms of the last 2 weeks and not future behavior. However, interestingly, the correlation was stable in the 8 weeks when the features were assessed using the poststudy questionnaire. Ben-Zeev et al [ ] looked even closer on a day-based sample resolution. Here we see a directionality switch with some of the objective features. Sleep duration was modeled with a positive regression coefficient with pre-post change PHQ-9 (eg, higher sleep duration modeled a worse PHQ-9 change score) almost throughout the study period, but during the last quarter, it changed to a negative regression coefficient, which is consistent with the literature, as depicted in [ , ]. These findings highlight the importance of transparency regarding the analysis methods. The implications regarding the results presented in this systematic review are minimal because the induced correlation differences remain invariant to the correlation directionality, which is the focus here.
Limitations are also associated with the different technologies, including hardware and software, used to collect objective features from mobile phones and wearables. Studies have shown statistically significant differences that need to be accommodated within the study design . For example, Farhan et al [ ] developed a mobile phone-based sensing app with the PHQ-9 assessment on both iOS and Android. The study showed that the feature movement duration changed from a correlation of r=0.06 (P=.43) on Android to r=−0.13 (P=.07) on iOS. They argued that the difference was due to technical details regarding whether data was pooled or sent from the sensor. This example demonstrates a change in correlation directionality, which could have had an impact on our results if more studies were reporting on movement duration. Even though we reported the Android results, to be consistent with the other mobile phone-based studies, there was no impact on our results because the directionality of the remaining features was identical.
The result on weighted directionality inincludes 9 studies that reported on group differences between clinical subjects and healthy controls. They compared the mean value of the feature between the two groups to understand what directionality the feature has concerning the disorder. If the study reported longer sleep duration in the clinical group than in healthy controls, it indicates that the directionality is positive. This could be problematic in nonlinear cases, such as the observation by McKercher et al [ ]. They found that male participants with depression were taking 7500-9999 steps per day, contrary to healthy controls who were in the lower or upper levels of respectively <7500 or >9999 steps per day.
The included studies used different ways of measuring mood symptoms, which undoubtedly had an impact on the correlation value, while the directionality of the correlation stayed intact. Several studies have shown a high correlation between different mood assessment methods. Simple mood scales for self-assessment, such as a 7-point selection from −3 till 3, have shown statistically significant correlations with clinically validated rating scales such as the HDRS . For example, there is a high correlation between the commonly used assessments methods of depression; PHQ-9, Becks depression inventory (BDI), and HDRS (lowest PHQ-9 vs HDRS: r=0.73; ) [ ]. The Center for Epidemiological Studies Depression Scale and BDI have also been shown to be highly correlated (r=0.84, P<.001) [ ]. Patient-based outcome measures such as PHQ-9 and BDI have the benefit of being conducted outside the clinic, target very specific symptoms, exclude clinician bias, and facilitate the doctor-patient communication. However, they have some drawbacks such as a biased response depending on the recipient and a lack of meaningful interpretation of the changes to the outcome value [ ].
As previously mentioned, we have chosen to include a broad definition of mood-based assessments in this review. However, a limitation is that some of them are questionable in the assessment of mood and depression. For instance, studies assessing “happiness,” “well-being,” and “quality of life” have been excluded in this review [- ], even though it has been shown that happiness scores correlated moderately with depression, measured using BDI (r=−0.57, P<.001) [ ].
The heterogeneity of the included studies also limits implementations in future studies. Faurholt-Jepsen et al  presented a new feature calls missed, which is statistically significantly correlated with HDRS-17 (beta=0.05, P=.006). The result was presented in 2016, but not replicated in any of the later studies, such as the comprehensive study on phone records with 532 subjects [ ].
Absolute Valued Correlations
Several research groups chose to present their correlation results in absolute values [, , ]. This is a problem because the directionality of the correlation is lost, and the only information left is a measure of the strength of the relation. Canzian & Musolesi [ ] clearly visualized this problem in several histogram plots representing each subject correlation values; these plots almost resemble a normal distribution around zero, but with a tail toward one of the directions. Raw correlation values were provided when requested from 1 of 3 studies [ ]. The other 2 commented on their choice of reporting absolute values of the correlation:
We observed very different behaviors among users, having in some cases positive correlations, in others, negative ones and in others no correlation at all .
However, asand show, there are several consistent correlations between features and mood assessments. Therefore, because this systematic review has revealed several features with common correlations across multiple studies, we hope to encourage future studies to present raw correlation values. This will make cross-study comparisons more valid. Further discussion on the use of absolute correlation values can be found elsewhere [ ].
The analysis provided in this paper has shown that it is time consuming and difficult to compare and analyze data across studies due to a high level of heterogeneity. To provide more systematic and automatic analyses, a significant degree of standardization is needed in three areas:
- Standardized data collection and feature extraction. The way that physical activity, social activity, and mobility features based on accelerometer and GPS data are extracted should be standardized across studies. For example, the feature location entropy seems like a promising feature and could be collected and calculated consistently across studies.
- Standardized mood assessment tools. The review revealed that a wide range of clinical (n=11) and nonclinical (n=9) mood rating scales were used. This makes it hard to compare correlations across studies when such different scales are used. We suggest that future studies include a clinician-based rating scale of severity of depression such as the HDRS as well as a self-reported questionnaire of depression such as the PHQ-9 or the BDI-21.
- Standardized statistical correlation methodology. The reviewed papers applied more than 11 different methods for correlation values, with different time windows. We suggest that raw correlation values are presented in addition to associations adjusted for relevant demographic variables, including sex and age, and clinical variables, such as BMI.
We also invite future systematic reviews to focus on classification models. They include accuracy measures and weightings that assist in the understanding of the individual objective features to classify mood and can investigate nonlinear interactions between multiple features and mood scores. As an example, Muaremi et al  used microphone features to classify mood; they achieved an F1 accuracy of 82% and discovered speaking time as the best-performing feature. By expanding to include GPS and accelerometer-related features, Abdullah et al [ ] achieved an F1 accuracy of 85.5%, with the GPS feature distance achieving largest weighting. A Naïve Bayes Classifier, to predict mood based on a combination of location features, achieved an accuracy of 81.7% [ ].
In our search, we came across several studies with sensor systems that are not currently fully mobile. This includes electroencephalogram (EEG) systems [- ]. For instance, Li et al [ ] achieved a 99.1% accuracy discriminating depressed and nondepressed participants based on EEG. Other systems monitoring body temperature [ ], saliva [ ], autonomic nerve balance [ ], and facial muscle activities [ ] could also be relevant. However, because these sensor modalities are not mobile or wearable to any great extent, they were excluded. These sensor modalities could, however, potentially be included in a ubiquitous mobile system for mood disorders in the future.
Mobile and wearable devices provide a unique platform for continuous collection of behavioral data from patients in real-time and within naturalistic settings. Many researchers have used this to investigate the relationship between behavior and mood disorder symptoms, as recorded by mobile or other wearable devices. In this systematic review, we identified a total of 46 eligible papers of such studies, of which 26 involved clinical samples.
We found 7 feature categories () that were investigated across the studies. Subject-based and device interaction features represented the largest percentage of statistically significant relationships. In a detailed analysis of the 85 objective features that were identified, we were able to find strong consistencies between several behavioral features across the studies. For example, in the nonclinical sample, there was a consistent positive correlation between the features home stay and mobile phone screen active duration with mood symptoms (eg, more time at home and longer phone usage indicated a more depressed mood). Furthermore, several behavioral features had a coherent negative correlation with mood symptoms, including amount of vigorous activity, location variance, and distance moved. In the clinical samples, mobile phone screen active duration was replicated as a constant positive correlating feature together with incoming call frequency and duration. Similarly, a coherent negative correlation was found, including the amount of visible GSM cell towers (reflecting mobility), SMS text messages received, and outgoing call frequency and duration.
This research was funded by the Innovation Fund Denmark through the RADMIS project and the Copenhagen Center for Health Technology (CACHET).
Conflicts of Interest
LVK has within the preceding 3 years been a consultant for Sunovion. JEB is a shareholder in Monsenso ApS and has been a consultant for Lundbeck A/S. Other authors report no financial activities.
Multimedia Appendix 1
Search string used in all the database searches.PDF File (Adobe PDF File), 48KB
Multimedia Appendix 2
Extended information of the included study for the nonclinical samples.PDF File (Adobe PDF File), 57KB
Multimedia Appendix 3
The reported results between the measured objective feature and the outcome measure for the nonclinical sample.PDF File (Adobe PDF File), 66KB
Multimedia Appendix 4
Extended information of the included study for the clinical sample.PDF File (Adobe PDF File), 61KB
Multimedia Appendix 5
The reported results between the measured objective feature and the outcome measure for studies with a medical diagnose of either Unipolar Disorder (UD) or Bipolar Disorder (BD).PDF File (Adobe PDF File), 72KB
Multimedia Appendix 6
Features collected from at least two studies using clinical samples with Bipolar disorder only. The x-axis (wD) represents a weighted directionality of the correlation between the feature and mood symptoms. Positive values represent a larger depressive score and vice versa. The y-axis represents the logarithm of the total number of participants across all studies for this feature. The size of each pie chart represents the number of studies that recorded the feature, while the green, red, and grey areas represent statistically significant, statistically nonsignificant correlations, and missing statistical significance respectively.PNG File, 62KB
Multimedia Appendix 7
Features collected from at least two studies using clinical and nonclinical samples of participants. The x-axis (wD) represents a weighted directionality of the correlation between the feature and mood symptoms. Positive values represent a larger depressive score and vice versa. The y-axis represents the logarithm of the total number of participants across all studies for this feature. The size of each pie chart represents the number of studies that recorded the feature, while the green, red, and grey areas represent statistically significant, statistically nonsignificant correlations, and missing statistical significance respectively.PNG File, 128KB
Multimedia Appendix 8
Table of the 46 included studies showing the feature category that they report on.PDF File (Adobe PDF File), 56KB
Multimedia Appendix 9
Graphical linkage showing which sensor types are used within the seven feature categories.PNG File, 47KB
- Callan JA, Wright J, Siegle GJ, Howland RH, Kepler BB. Use of Computer and Mobile Technologies in the Treatment of Depression. Arch Psychiatr Nurs 2017 Dec;31(3):311-318. [CrossRef] [Medline]
- Grist R, Porter J, Stallard P. Mental Health Mobile Apps for Preadolescents and Adolescents: A Systematic Review. J Med Internet Res 2017 May 25;19(5):e176 [FREE Full text] [CrossRef] [Medline]
- Zhao J, Freeman B, Li M. Can Mobile Phone Apps Influence People's Health Behavior Change? An Evidence Review. J Med Internet Res 2016 Oct 31;18(11):e287 [FREE Full text] [CrossRef] [Medline]
- Huguet A, Rao S, McGrath PJ, Wozney L, Wheaton M, Conrod J, et al. A Systematic Review of Cognitive Behavioral Therapy and Behavioral Activation Apps for Depression. PLoS One 2016;11(5):e0154248 [FREE Full text] [CrossRef] [Medline]
- Dogan E, Sander C, Wagner X, Hegerl U, Kohls E. Smartphone-Based Monitoring of Objective and Subjective Data in Affective Disorders: Where Are We and Where Are We Going? Systematic Review. J Med Internet Res 2017 Jul 24;19(7):e262 [FREE Full text] [CrossRef] [Medline]
- Faurholt-Jepsen M, Vinberg M, Frost M, Debel S, Margrethe CE, Bardram JE, et al. Behavioral activities collected through smartphones and the association with illness activity in bipolar disorder. Int J Methods Psychiatr Res 2016 Dec;25(4):309-323. [CrossRef] [Medline]
- Saeb S, Zhang M, Karr CJ, Schueller SM, Corden ME, Kording KP, et al. Mobile Phone Sensor Correlates of Depressive Symptom Severity in Daily-Life Behavior: An Exploratory Study. J Med Internet Res 2015;17(7):e175 [FREE Full text] [CrossRef] [Medline]
- O'Brien JT, Gallagher P, Stow D, Hammerla N, Ploetz T, Firbank M, et al. A study of wrist-worn activity measurement as a potential real-world biomarker for late-life depression. Psychol Med 2017 Jan;47(1):93-102 [FREE Full text] [CrossRef] [Medline]
- Canzian L, Musolesi M. Trajectories of Depression: Unobtrusive Monitoring of Depressive States by means of Smartphone Mobility Traces Analysis. : UbiComp '15; 2015 Presented at: Proceedings of the 2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing; September 07-11, 2015; Osaka, Japan p. 1293-1304. [CrossRef]
- Faurholt-Jepsen M, Frost M, Vinberg M, Christensen EM, Bardram JE, Kessing LV. Smartphone data as objective measures of bipolar disorder symptoms. Psychiatry Res 2014 Jun 30;217(1-2):124-127. [CrossRef] [Medline]
- Alvarez-lozano J, Frost M, Osmani V, Bardram JE, Kessing LV, Mayora O, et al. Tell me your apps and I will tell you your mood: Correlation of apps usage with bipolar disorder state. 2014 Presented at: 7th Int Conf Pervasive Technol Relat to Assist Envotinments - PETRA; May 27-30, 2014; Rhodes, Greece p. 1-7. [CrossRef]
- Kroenke K, Spitzer RL, Williams JB. The PHQ-9: validity of a brief depression severity measure. J Gen Intern Med 2001 Sep;16(9):606-613 [FREE Full text] [Medline]
- Mohr D, Zhang M, Schueller S. Personal Sensing: Understanding Mental Health Using Ubiquitous Sensors and Machine Learning. Annu Rev Clin Psychol 2017 Dec 08;13:23-47. [CrossRef] [Medline]
- Mestry M, Mehta J, Mishra A, Gawande K. Identifying associations between smartphone usage and mental health during depression, anxiety and stress. 2015 Presented at: Proceedings - 2015 International Conference on Communication, Information and Computing Technology, ICCICT 2015; January 16-17, 2015; Mumbai, India. [CrossRef]
- Asselbergs J, Ruwaard J, Ejdys M, Schrader N, Sijbrandij M, Riper H. Mobile Phone-Based Unobtrusive Ecological Momentary Assessment of Day-to-Day Mood: An Explorative Study. J Med Internet Res 2016 Mar 29;18(3):e72 [FREE Full text] [CrossRef] [Medline]
- Matic A, Osmani V, Popleteev A, Mayora-Ibarra O. Smart phone sensing to examine effects of social interactions and non-sedentary work time on mood changes. 2011 Presented at: CONTEXT'11 Proceedings of the 7th international and interdisciplinary conference on Modeling and using context; September 26-30, 2011; Karlsruhe, Germany p. 200-213. [CrossRef]
- Farhan A, Yue C, Morillo R, Ware S, Lu J, Bi J, et al. Behavior vs. introspection: Refining prediction of clinical depression via smartphone sensing data. 2016 Presented at: IEEE Wireless Health, 2016; October 25-27, 2016; Bethesda, Maryland, USA p. 30-37. [CrossRef]
- Doryab A, Min J, Wiese J, Zimmerman J, Hong J. Detection of behavior change in people with depression. 2014 Presented at: AAAI; July 27-31, 2014; Quebec, Canada p. 12.
- Osmani V, Maxhuni A, Grünerbl A, Lukowicz P, Haring C, Mayora O. Monitoring activity of patients with bipolar disorder using smart phones. 2013 Presented at: Proceedings of International Conference on Advances in Mobile Computing & Multimedia; 2013; Vienna, Austria p. 85. [CrossRef]
- Guidi A, Vanello N, Bertschy G, Gentili C, Landini L, Scilingo E. Automatic analysis of speech F0 contour for the characterization of mood changes in bipolar patients. Biomedical Signal Processing and Control 2015 Mar;17:29-37. [CrossRef]
- Frost M, Doryab A, Faurholt-Jepsen M, Kessing L, Bardram J. Supporting disease insight through data analysis. 2013 Presented at: UbiComp '13 Proceedings of the 2013 ACM international joint conference on Pervasive and ubiquitous computing; 2013; Zurich, Switzerland. [CrossRef]
- Beiwinkel T, Kindermann S, Maier A, Kerl C, Moock J, Barbian G, et al. Using Smartphones to Monitor Bipolar Disorder Symptoms: A Pilot Study. JMIR Ment Health 2016;3(1):e2 [FREE Full text] [CrossRef] [Medline]
- Schardt C, Adams M, Owens T, Keitz S, Fontelo P. Utilization of the PICO framework to improve searching PubMed for clinical questions. BMC Med Inform Decis Mak 2007 Jun 15;7:16 [FREE Full text] [CrossRef] [Medline]
- Moher D, Liberati A, Tetzlaff J, Altman D, PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med 2009 Jul 21;6(7):e1000097 [FREE Full text] [CrossRef] [Medline]
- Doryab A, Frost M, Faurholt-Jepsen M, Kessing LV, Bardram JE. Impact factor analysis: combining prediction with parameter ranking to reveal the impact of behavior on health outcome. Pers Ubiquit Comput 2014 Sep 21;19(2):355-365. [CrossRef]
- Faurholt-Jepsen M, Vinberg M, Frost M, Christensen E, Bardram J, Kessing L. Smartphone data as an electronic biomarker of illness activity in bipolar disorder. Bipolar Disord 2015 Nov;17(7):715-728. [CrossRef] [Medline]
- Baker DA, Algorta GP. The Relationship Between Online Social Networking and Depression: A Systematic Review of Quantitative Studies. Cyberpsychol Behav Soc Netw 2016 Nov;19(11):638-648. [CrossRef] [Medline]
- Pachucki MC, Ozer EJ, Barrat A, Cattuto C. Mental health and social networks in early adolescence: a dynamic study of objectively-measured social interaction behaviors. Soc Sci Med 2015 Jan;125:40-50. [CrossRef] [Medline]
- Hernández N, Yavuz G, Eşrefoğlu R, Kepez T, Özdemir A, Demiray B, et al. Thought and Life Logging: A Pilot Study. 2015 Presented at: International Conference on Ubiquitous Computing and Ambient Intelligence; December 1-4, 2015; Puerto Varas, Chile p. 26-36. [CrossRef]
- Jaques N, Taylor S, Azaria A, Ghandeharioun A, Sano A, Picard R. Predicting students' happiness from physiology, phone, mobility, and behavioral data. In: Int Conf Affect Comput Intell Interact Workshops. 2015 Sep Presented at: 2015 International Conference on Affective Computing and Intelligent Interaction (ACII); 2015; Xi’an, China p. 222-228 URL: http://europepmc.org/abstract/MED/28515966 [CrossRef]
- Kamdar MR, Wu MJ. PRISM: A DATA-DRIVEN PLATFORM FOR MONITORING MENTAL HEALTH. Pac Symp Biocomput 2016;21:333-344 [FREE Full text] [Medline]
- Ono E, Nozawa T, Ogata T, Motohashi M, Higo N, Kobayashi T, et al. Relationship between social interaction and mental health. 2011 Presented at: IEEE/SICE Int Symp Syst Integr SII; December 20-22, 2011; Kyoto, Japan p. 246-249. [CrossRef]
- Gao Y, Li H, Zhu T. Predicting Subjective Well-Being by Smartphone Usage Behaviors. 2014 Presented at: International Conference on Health Informatics; March 3-6, 2014; Loire Valley, France p. 317-322. [CrossRef]
- Zhang Z, Tian X, Lim J. New algorithm for the depression diagnosis using HRV: A neuro-fuzzy approach. 2011 Presented at: Proc Int Symp Bioelectron Bioinformatics; 2011; Suzhou, China p. 283-286. [CrossRef]
- Gentili C, Valenza G, Nardelli M, Lanatà A, Bertschy G, Weiner L, et al. Longitudinal monitoring of heartbeat dynamics predicts mood changes in bipolar patients: A pilot study. J Affect Disord 2017 Feb;209:30-38. [CrossRef] [Medline]
- Valenza G, Nardelli M, Lanata' A, Gentili C, Bertschy G, Kosel M, et al. Predicting Mood Changes in Bipolar Disorder through Heartbeat Nonlinear Dynamics. IEEE J Biomed Health Inform 2016 Apr 20. [CrossRef] [Medline]
- Minassian A, Henry B, Geyer M, Paulus M, Young J, Perry W. The quantitative assessment of motor activity in mania and schizophrenia. J Affect Disord 2010 Jan;120(1-3):200-206 [FREE Full text] [CrossRef] [Medline]
- Ishimaru S, Kise K. Quantifying the Mental State on the Basis of Physical and Social Activities. 2015 Presented at: Adjun Proc ACM Int Jt Conf Pervasive Ubiquitous Comput Proc ACM Int Symp Wearable Comput Internet ;1217?1220; September 7-11, 2015; Osaka, Japan p. 1217-1220. [CrossRef]
- Saeb S, Lattie EG, Schueller SM, Kording KP, Mohr DC. The relationship between mobile phone location sensor data and depressive symptom severity. PeerJ 2016;4:e2537 [FREE Full text] [CrossRef] [Medline]
- Baras K, Soares L, Paulo N, Barros R. ‘Smartphine': Supporting students' well-being according to their calendar and mood. 2016 Presented at: Int Multidiscip Conf Comput Energy Sci Split; July 13-15, 2016; Split, Croatia. [CrossRef]
- Becker D, Bremer V, Funk B, Asselbergs J, Riper H, Ruwaard J. How to Predict Mood? Delving into Features of Smartphone-Based Data. 2016 Presented at: Proc 22nd Am Conf Inf Syst AMCIS; August 1-10, 2016; San Diego, California, USA p. 1-10.
- Ben-Zeev D, Scherer E, Wang R, Xie H, Campbell A. Next-generation psychiatric assessment: Using smartphone sensors to monitor behavior and mental health. Psychiatr Rehabil J 2015 Sep;38(3):218-226 [FREE Full text] [CrossRef] [Medline]
- Berke EM, Choudhury T, Ali S, Rabbi M. Objective measurement of sociability and activity: mobile sensing in the community. Ann Fam Med 2011;9(4):344-350 [FREE Full text] [CrossRef] [Medline]
- Cho YM, Lim HJ, Jang H, Kim K, Choi JW, Shin C, et al. A cross-sectional study of the association between mobile phone use and symptoms of ill health. Environ Health Toxicol 2016;31:e2016022 [FREE Full text] [CrossRef] [Medline]
- Chow PI, Fua K, Huang Y, Bonelli W, Xiong H, Barnes LE, et al. Using Mobile Sensing to Test Clinical Models of Depression, Social Anxiety, State Affect, and Social Isolation Among College Students. J Med Internet Res 2017 Mar 03;19(3):e62 [FREE Full text] [CrossRef] [Medline]
- DeMasi O, Aguilera A, Recht B. Detecting change in depressive symptoms from daily wellbeing questions, personality, and activity. 2016 Presented at: IEEE Wirel Heal WH; October 25-27, 2016; Bethesda, Maryland, USA p. 22-29. [CrossRef]
- Edwards M, Loprinzi P. Effects of a Sedentary Behavior-Inducing Randomized Controlled Intervention on Depression and Mood Profile in Active Young Adults. Mayo Clin Proc 2016 Aug;91(8):984-998. [CrossRef] [Medline]
- Mark G, Czerwinski M, Iqbal S, Johns P. Workplace Indicators of Mood: Behavioral and Cognitive Correlates of Mood Among Information Workers. 2016 Presented at: Proc 6th Int Conf Digit Heal Conf; April 11-13, 2016; Montréal, Quebec, Canada p. 29-36. [CrossRef]
- Mehrotra A, Hendley R, Musolesi M. Towards multi-modal anticipatory monitoring of depressive states through the analysis of human-smartphone interaction. 2016 Presented at: Proc ACM Int Jt Conf Pervasive Ubiquitous Comput Adjun - UbiComp; September 12-16, 2016; Heidelberg, Germany p. 1132-1138.
- Pillai V, Steenburg LA, Ciesla JA, Roth T, Drake CL. A seven day actigraphy-based study of rumination and sleep disturbance among young adults with depressive symptoms. J Psychosom Res 2014 Jul;77(1):70-75. [CrossRef] [Medline]
- Wang R, Chen F, Chen Z, Li T, Harari G, Tignor S, et al. StudentLife: assessing mental health, academic performance and behavioral trends of college students using smartphones. 2014 Presented at: Proc ACM Int Jt Conf Pervasive Ubiquitous Comput - UbiComp; September 13-17, 2014; Seattle, Washington, USA p. 3-14. [CrossRef]
- Wang R, Campbell A, Zhou X. Using opportunistic face logging from smartphone to infer mental health: challenges and future directionsUsing opportunistic face logging from smartphone to infer mental health. 2015 Presented at: ACM Int Jt Conf Pervasive Ubiquitous Comput, ACM Int Symp Wearable Comput - UbiComp; September 7-11, 2015; Osaka, Japan p. 683-692. [CrossRef]
- Abdullah S, Matthews M, Frank E, Doherty G, Gay G, Choudhury T. Automatic detection of social rhythms in bipolar disorder. J Am Med Inform Assoc 2016 May;23(3):538-543. [CrossRef] [Medline]
- Berle JO, Hauge ER, Oedegaard KJ, Holsten F, Fasmer OB. Actigraphic registration of motor activity reveals a more structured behavioural pattern in schizophrenia than in major depression. BMC Res Notes 2010 May 27;3:149 [FREE Full text] [CrossRef] [Medline]
- Dickerson R, Gorlin E, Stankovic J. Empath: a continuous remote emotional health monitoring system for depressive illness. 2011 Presented at: Proc 2nd Conf Wirel Heal; October 10-13, 2011; San Diego, California, USA p. 5. [CrossRef]
- Faurholt-Jepsen M, Brage S, Vinberg M, Christensen EM, Knorr U, Jensen HM, et al. Differences in psychomotor activity in patients suffering from unipolar and bipolar affective disorder in the remitted or mild/moderate depressive state. J Affect Disord 2012 Dec 10;141(2-3):457-463. [CrossRef] [Medline]
- Faurholt-Jepsen M, Brage S, Vinberg M, Jensen HM, Christensen EM, Knorr U, et al. Electronic monitoring of psychomotor activity as a supplementary objective measure of depression severity. Nord J Psychiatry 2015 Feb;69(2):118-125. [CrossRef] [Medline]
- Faurholt-Jepsen M, Busk J, Frost M, Vinberg M, Christensen E, Winther O, et al. Voice analysis as an objective state marker in bipolar disorder. Transl Psychiatry 2016 Dec 19;6:e856 [FREE Full text] [CrossRef] [Medline]
- Gershon A, Ram N, Johnson SL, Harvey AG, Zeitzer JM. Daily Actigraphy Profiles Distinguish Depressive and Interepisode States in Bipolar Disorder. Clin Psychol Sci 2016 Jul;4(4):641-650 [FREE Full text] [CrossRef] [Medline]
- Gonzalez R, Tamminga CA, Tohen M, Suppes T. The relationship between affective state and the rhythmicity of activity in bipolar disorder. J Clin Psychiatry 2014 Apr;75(4):e317-e322 [FREE Full text] [CrossRef] [Medline]
- Grünerbl A, Muaremi A, Osmani V, Bahle G, Ohler S, Tröster G, et al. Smartphone-based recognition of states and state changes in bipolar disorder patients. IEEE J Biomed Health Inform 2015 Jan;19(1):140-148. [CrossRef] [Medline]
- Hauge ER, Berle J, Oedegaard KJ, Holsten F, Fasmer OB. Nonlinear analysis of motor activity shows differences between schizophrenia and depression: a study using Fourier analysis and sample entropy. PLoS One 2011 Jan 28;6(1):e16291 [FREE Full text] [CrossRef] [Medline]
- Krane-Gartiser K, Henriksen TEG, Morken G, Vaaler A, Fasmer OB. Actigraphic assessment of motor activity in acutely admitted inpatients with bipolar disorder. PLoS One 2014;9(2):e89574 [FREE Full text] [CrossRef] [Medline]
- Loprinzi PD, Mahoney S. Concurrent occurrence of multiple positive lifestyle behaviors and depression among adults in the United States. J Affect Disord 2014 Aug;165:126-130. [CrossRef] [Medline]
- Miwa H, Sasahara S, Matsui T. Roll-over detection and sleep quality measurement using a wearable sensor. 2007 Presented at: Conf Proc IEEE Eng Med Biol Soc - EMBC; August 23-26, 2007; Lyon, France.
- Muaremi A, Gravenhorst F, Grünerbl A, Arnrich B, Tröster G. Assessing bipolar episodes using speech cues derived from phone calls. In: Lect Notes Inst Comput Sci Soc Telecommun Eng LNICST. 2014 Presented at: MindCare: Pervasive Computing Paradigms for Mental Health; May 8-9, 2014; Tokyo, Japan p. 103-114. [CrossRef]
- Palmius N, Tsanas A, Saunders K, Bilderbeck A, Geddes J, Goodwin G, et al. Detecting Bipolar Depression From Geographic Location Data. IEEE Trans Biomed Eng 2017 Dec;64(8):1761-1771 [FREE Full text] [CrossRef] [Medline]
- St-Amand J, Provencher MD, Bélanger L, Morin CM. Sleep disturbances in bipolar disorder during remission. J Affect Disord 2013 Mar 20;146(1):112-119. [CrossRef] [Medline]
- Todder D, Caliskan S, Baune BT. Longitudinal changes of day-time and night-time gross motor activity in clinical responders and non-responders of major depression. World J Biol Psychiatry 2009;10(4):276-284. [CrossRef] [Medline]
- Mok W, Sing R, Jiang X, See S. Proposal of a depression detector. 2014 Presented at: Asia-Pacific Signal Inf Process Assoc Annu Summit Conf APSIPA; December 9-12, 2014; Chiang Mai, Thailand. [CrossRef]
- Shannon T, Annie D, Lan S. Speech analysis and depression. 2016 Presented at: Asia-Pacific Signal Inf Process Assoc Annu Summit Conf APSIPA; December 13-16, 2016; Jeju, Korea p. 1-4. [CrossRef]
- Horwitz-Martin R, Quatieri T, Godoy E, Williamson J. A vocal modulation model with application to predicting depression severity. 2016 Presented at: BSN th Annu Body Sens Networks Conf; June 14-17, 2016; San Francisco, USA p. 2016-2013. [CrossRef]
- Glenn T, Monteith S. New measures of mental state and behavior based on data collected from sensors, smartphones, and the Internet. Curr Psychiatry Rep 2014 Dec;16(12):523. [CrossRef] [Medline]
- Donker T, Petrie K, Proudfoot J, Clarke J, Birch M, Christensen H. Smartphones for smarter delivery of mental health programs: a systematic review. J Med Internet Res 2013;15(11):e247 [FREE Full text] [CrossRef] [Medline]
- Bewernick BH, Urbach AS, Bröder A, Kayser S, Schlaepfer TE. Walking away from depression-motor activity increases ratings of mood and incentive drive in patients with major depression. Psychiatry Res 2017 Dec;247:68-72. [CrossRef] [Medline]
- Disabato DJ, Kashdan TB, Short JL, Jarden A. What Predicts Positive Life Events that Influence the Course of Depression? A Longitudinal Examination of Gratitude and Meaning in Life. Cogn Ther Res 2016 May 30;41(3):444-458. [CrossRef] [Medline]
- Ben-Zeev D, Schueller S, Begale M, Duffecy J, Kane J, Mohr D. Strategies for mHealth research: lessons from 3 mobile intervention studies. Adm Policy Ment Health 2015 Mar;42(2):157-167 [FREE Full text] [CrossRef] [Medline]
- Burns MN, Begale M, Duffecy J, Gergle D, Karr CJ, Giangrande E, et al. Harnessing context sensing to develop a mobile intervention for depression. J Med Internet Res 2011;13(3):e55 [FREE Full text] [CrossRef] [Medline]
- Thomée S, Härenstam A, Hagberg M. Mobile phone use and stress, sleep disturbances, and symptoms of depression among young adults--a prospective cohort study. BMC Public Health 2011 Jan 31;11:66 [FREE Full text] [CrossRef] [Medline]
- Scott J, Murray G, Henry C, Morken G, Scott E, Angst J, et al. Activation in Bipolar Disorders: A Systematic Review. JAMA Psychiatry 2017 Feb 01;74(2):189-196. [CrossRef] [Medline]
- Burton C, McKinstry B, Szentagotai TA, Serrano-Blanco A, Pagliari C, Wolters M. Activity monitoring in patients with depression: a systematic review. J Affect Disord 2013 Feb 15;145(1):21-28. [CrossRef] [Medline]
- Energistyrelse. 2016. Telestatistik Første halvår 2016 URL: https://ens.dk/sites/ens.dk/files/Tele/telestatistik_-_foerste_halvaar_2016.pdf [accessed 2018-07-24] [WebCite Cache]
- de Araújo TC, Silva L, Moreira A. Data Quality Issues on Environmental Sensing with Smartphones. 2017 Presented at: Proc 6th Int Conf Sens Networks SENSORNETS; February 19-21, 2017; Porto, Portugal p. 59-68.
- McKercher CM, Schmidt MD, Sanderson KA, Patton GC, Dwyer T, Venn AJ. Physical activity and depression in young adults. Am J Prev Med 2009 Feb;36(2):161-164. [CrossRef] [Medline]
- Jeon S, Han C, Ko Y, Yoon S, Pae C, Choi J, et al. Measurement-based Treatment of Residual Symptoms Using Clinically Useful Depression Outcome Scale: Korean Validation Study. Clin Psychopharmacol Neurosci 2017 Feb 28;15(1):28-34 [FREE Full text] [CrossRef] [Medline]
- Park S, Lee SW, Kwak J, Cha M, Jeong B. Activities on Facebook reveal the depressive state of users. J Med Internet Res 2013;15(10):e217 [FREE Full text] [CrossRef] [Medline]
- Greenhalgh J, Meadows K. The effectiveness of the use of patient-based measures of health in routine practice in improving the process and outcomes of patient care: a literature review. J Eval Clin Pract 1999 Nov;5(4):401-416. [Medline]
- Cheng H, Furnham A. Personality, self-esteem, and demographic predictions of happiness and depression. Personality and Individual Differences 2003 Apr;34(6):921-942. [CrossRef]
- Servia-Rodríguez S, Rachuri K, Lathia N, Mascolo C, Rentfrow P, Sandstrom G. Mobile Sensing at the Service of Mental Well-being: a Large-scale Longitudinal Study. 2017 Presented at: Proc 26th Int Conf World Wide Web; April 3-7, 2017; Perth, Australia p. a. [CrossRef]
- Evans J. Straightforward statistics for the behavioral sciences. In: Straightforward Statistics for the Behavioral Sciences. N. Scituate, MA: Duxbury Press; 1995.
- Zhang X, Hu B, Zhou L, Moore P, Chen J. An EEG based pervasive depression detection for females. In: Lect Notes Comput Sci (including Subser Lect Notes Artif Intell Lect Notes Bioinformatics) Internet Springer, Berlin, Heidelberg. 2013 Presented at: ICPCA/SWS'12 Proceedings of the 2012 international conference on Pervasive Computing and the Networked World; November 28-30, 2012; Istanbul, Turkey p. 848-861.
- Peng H, Hu B, Liu Q, Dong Q, Zhao Q, Moore P. User-centered depression prevention: An EEG approach to pervasive healthcare. User-centered depression prevention: An EEG approach to pervasive healthcare. 2011 5th Int Conf Pervasive Comput Technol Healthc Work Internet IEEE; 2011 Presented at: 5th International Conference on Pervasive Computing Technologies for Healthcare (PervasiveHealth); May 23-26, 2011; Dublin, Ireland p. 325-330.
- Li X, Hu B, Shen J, Xu T, Retcliffe M. Mild Depression Detection of College Students: an EEG-Based Solution with Free Viewing Tasks. J Med Syst 2015 Dec;39(12):187. [CrossRef] [Medline]
- Butca C, Suciu G, Ochian A, Fratu O, Halunga S. Wearable sensors and cloud platform for monitoring environmental parameters in e-health applications. 2014 Presented at: 11th Int Symp Electron Telecommun ISETC; November 14-15, 2014; Timisoara, Romania p. 1-4.
- Zhang L, Yang W, Yang Y, Liu H, Gu Z. Smartphone-based point-of-care testing of salivary α-amylase for personal psychological measurement. Analyst 2015 Nov 07;140(21):7399-7406. [CrossRef] [Medline]
- Oyama-Higa M, Wang W, Kaizu S, Futaba T, Suzuki T. Smartphone-based device for checking mental status in real time. 2016 Presented at: BIOSTEC 2016 Proceedings of the International Joint Conference on Biomedical Engineering Systems and Technologies; February 21-23, 2016; Rome, Italy p. 137-142. [CrossRef]
- Thulkar D, Bhaskarwar T, Hamde S. Facial electromyography for characterization of emotions using LabVIEW. 2015 Presented at: Int Conf Ind Instrum Control ICIC; May 28-30, 2015; Pune, India p. 683-686.
|BD: bipolar disorder|
|BDI: Becks depression inventory|
|BMI: body mass index|
|CBT: cognitive behavioral therapy|
|GPS: global positioning system|
|HDRS: Hamilton Depression Rating Scale|
|MONARCA: MONitoring, treAtment and pRediCtion of bipolAr Disorder Episodes|
|PHQ: Patient Health Questionnaire|
|PICO: Patient problem Intervention, Comparison, and Outcome|
|SMS: short message service|
|UD: unipolar disorder|
Edited by G Eysenbach; submitted 19.12.17; peer-reviewed by D Hidalgo-Mazzei, R Grist, J Zulueta; comments to author 18.03.18; revised version received 13.05.18; accepted 18.06.18; published 13.08.18
©Darius A Rohani, Maria Faurholt-Jepsen, Lars Vedel Kessing, Jakob E Bardram. Originally published in JMIR Mhealth and Uhealth (http://mhealth.jmir.org), 13.08.2018.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR mhealth and uhealth, is properly cited. The complete bibliographic information, a link to the original publication on http://mhealth.jmir.org/, as well as this copyright and license information must be included.