Digital Phenotyping for Stress, Anxiety, and Mild Depression: Systematic Literature Review

Background Unaddressed early-stage mental health issues, including stress, anxiety, and mild depression, can become a burden for individuals in the long term. Digital phenotyping involves capturing continuous behavioral data via digital smartphone devices to monitor human behavior and can potentially identify milder symptoms before they become serious. Objective This systematic literature review aimed to answer the following questions: (1) what is the evidence of the effectiveness of digital phenotyping using smartphones in identifying behavioral patterns related to stress, anxiety, and mild depression? and (2) in particular, which smartphone sensors are found to be effective, and what are the associated challenges? Methods We used the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) process to identify 36 papers (reporting on 40 studies) to assess the key smartphone sensors related to stress, anxiety, and mild depression. We excluded studies conducted with nonadult participants (eg, teenagers and children) and clinical populations, as well as personality measurement and phobia studies. As we focused on the effectiveness of digital phenotyping using smartphones, results related to wearable devices were excluded. Results We categorized the studies into 3 major groups based on the recruited participants: studies with students enrolled in universities, studies with adults who were unaffiliated to any particular organization, and studies with employees employed in an organization. The study length varied from 10 days to 3 years. A range of passive sensors were used in the studies, including GPS, Bluetooth, accelerometer, microphone, illuminance, gyroscope, and Wi-Fi. These were used to assess locations visited; mobility; speech patterns; phone use, such as screen checking; time spent in bed; physical activity; sleep; and aspects of social interactions, such as the number of interactions and response time. Of the 40 included studies, 31 (78%) used machine learning models for prediction; most others (n=8, 20%) used descriptive statistics. Students and adults who experienced stress, anxiety, or depression visited fewer locations, were more sedentary, had irregular sleep, and accrued increased phone use. In contrast to students and adults, less mobility was seen as positive for employees because less mobility in workplaces was associated with higher performance. Overall, travel, physical activity, sleep, social interaction, and phone use were related to stress, anxiety, and mild depression. Conclusions This study focused on understanding whether smartphone sensors can be effectively used to detect behavioral patterns associated with stress, anxiety, and mild depression in nonclinical participants. The reviewed studies provided evidence that smartphone sensors are effective in identifying behavioral patterns associated with stress, anxiety, and mild depression.


Background
Digital phenotyping is "the moment-by-moment quantification of the individual level human phenotype in situ using data from personal digital devices" [1].Digital phenotyping applies the concept of phenotypes, in other words, the observable characteristics resulting from the genotype and environment, to conceptualize observable patterns in individuals' digital data.In the last decade, digital phenotyping studies have been able to compare typical and atypical patterns in daily activities to correlate atypical behavior with negative emotions [2,3].Behavioral patterns include variations in mobility, frequency of being in various locations, and sleep patterns.In smartphones, user data can be stored, managed, interpreted, and captured in enormous amounts [1,4,5].This can be done actively or passively.Active data collection requires the user to self-report and complete surveys, whereas passive sensing collects data automatically without user input [5].Most studies combine active and passive sensing to more accurately detect and predict behavioral abnormalities.Modern smartphone analytics can be used for the discovery of commonalities and abnormalities in user behavior.The ease of using passive sensing makes it an ideal data gathering method for mental health studies [6][7][8] and an ideal technique for assessing mental health [9].
Digital phenotyping has been successful in the early detection and prediction of behaviors related to neuropharmacology [10]; cardiovascular diseases [11]; diabetes [12]; and major severe injuries, such as spinal cord injury [13], motivating further adoption.Digital phenotyping has also proven useful for the detection of severe mental health issues, such as schizophrenia [14,15], bipolar disorder [16], and suicidal thoughts [17].Digital phenotyping has been so successful for specialized, clinical populations that it is increasingly considered for mass market use with nonclinical populations.Digital phenotyping applications and software tools have been used to capture employee information, such as their screen time and clicking patterns [18].However, there are not many digital phenotyping studies that have specifically examined the detection or prediction of stress, anxiety, and mild depression.
Individuals with stress, anxiety, and mild depression can develop chronic mental health symptoms that impact their mobility, satisfaction with life, and social interaction [19,20].When these symptoms are not detected early, they worsen, and the impact is more significant [21][22][23], increasing the need for medication and hospitalization.This makes mild mental health symptoms a valid target for digital phenotyping, as its goal is to enable early detection and, subsequently, early treatment.Smartphones are increasingly ubiquitous [24], which makes them an optimal platform for digital phenotyping.We constrained our systematic literature search to the more challenging problem of the detection of mild mental health symptoms using only smartphone sensors and excluded studies that used additional wearable sensors.In general, we believe that additional wearables might increase the effectiveness of digital phenotyping in detecting stress, anxiety, and mild depression.Given the ubiquity of smartphones, we aimed to answer the following question: what is the effectiveness of digital phenotyping using smartphone sensors in detecting stress, anxiety, and mild depression?

Objectives
The objective of this systematic literature review was to better understand the current uses of digital phenotyping and results of using digital phenotyping for the detection and prediction of mild behavioral patterns related to stress, anxiety, and mild depression.The 2 research questions this review sought to answer were as follows: 1.What is the evidence of the effectiveness of digital phenotyping using smartphones in identifying behavioral patterns related to stress, anxiety, and mild depression? 2. In particular, which smartphone sensors are found to be effective, and what are the associated challenges?
For these research questions, we considered statistically significant associations between sensor patterns and behavioral patterns as evidence of effectiveness.

Type of Studies
This review followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [25] (Multimedia Appendix 1). Figure 1 shows the reviewing process and search results.In the first round of screening studies, 1 author excluded studies that were not relevant to the research questions.Another author reran the queries for confirmation.Studies were included in this review if they were conducted to measure and detect stress, anxiety, or mild depression, even if they included other variables, such as job performance, promotion, or discrimination.We included studies in which data were collected through smartphones with an iOS (Apple Inc) or Android (Google LLC) operating system.Data collected through wearable devices were excluded.We included studies in which the participants were adults aged ≥18 years and were from a nonclinical population.Studies conducted with nonadult participants (eg, teenagers and children) were excluded.Given our research questions, if the studies' participants had or had had any severe mental health disorder, such as schizophrenia, bipolar disorder, or psychosis, they were not included.We also excluded personality and character measurement and phobia studies.The primary research language was English.The studies included were conducted from September 2010 to September 2023.Peer-reviewed conference articles and journal articles were included.The data we wished to extract were the study aim, data collected, operating system in the smartphone used for data collection, behavioral patterns identified, surveys used for verification, and sample size.A total of 3 authors reviewed the studies independently to extract data and confirm the extracted data.After the first round of data extraction, 1 author re-examined the studies to extract the predictive modeling used.These data are presented in the Results section.We noticed that participants in the included studies fell into 1 of 3 major groups (ie, students, adults, and employees).We refer to the participants of the studies that recruited adults enrolled in universities as "students," participants of the studies that recruited adults

XSL • FO
RenderX unaffiliated to any particular organization as "adults," and participants of the studies that recruited adults employed at a particular organization as "employees."

Search Strategy
A total of 3 databases were queried: Web of Science, ACM, and PubMed.PubMed is a medicine-based database, ACM is a technology-based database, and Web of Science is a cross-domain database.The search query was the same for the 3 platforms: "digital phenotyping" OR "passive sensing" AND (stress OR anxiety OR ((mild OR moderate) AND depression)).

Duration
The study length varied from 10 days [26] to 3 years [27].One study [28] conducted in-depth interviews with students lasting an average of 4.5 hours per person, and another study was a controlled laboratory study [29].These 2 studies are not presented in Table 1.In the studies conducted with students, a semester or spring or winter term was a common duration.The studies with general nonclinical adult populations were typically longer than those with students.

RenderX
Table 1.Duration of the reviewed studies (N=38; 2 studies are excluded, as 1 [28] is interview based and the other [29] is a controlled laboratory study).

Number of Participants
The number of participants ranged from a minimum of 7 adults [26] to a maximum of 18,000 adults [27].Apart from the 3-year longitudinal study with 18,000 participants [27], the average number of participants was 129.4 (SD 184.01).We observed a pattern of attrition, where the number of participants who completed the study was lower than the number of the participants recruited.The number of participants reported in this review is the final sample size.For example, one of the studies [52] recruited 112 participants, of whom 84 (75%) completed the study.In the study by Pratap et al [55], there was a drastic drop in participants, with only 359 (30.42%) of the 1180 enrolled participants completing the study.Another significant drop was seen in the study by Nepal et al [40], where 750 participants were interested in the research, whereas only 141 (18.8%) of them completed the study.Some studies were less affected; for example, 86 participants started the study by Rhim et al [49], and 78 (91%) completed it.

Publication Years of the Studies
Although the query started with the year 2010, the earliest publication was from 2014 [26], extending to articles published as of April 2023 [35].Over the years, the interest in detecting and predicting stress, anxiety, and mild depression in the nonclinical population has increased (Table 2).

Studies With Students
Table 3 presents the data extracted from the studies that were conducted with student populations.The average length of the studies with students was 158.6 (SD 176.4) days.The average number of participants was 137.3 (SD 152.1).There were significantly more studies with students than studies with employees or general adults.The sample sizes of the studies with students were similar to those of the studies with adults but smaller than those of the studies with employees.In the studies with students, various passive sensors were used, and some were found to be effective for detection, prediction, or both.
Students with severe stress spent significantly less time on campus and were less involved in workrelated activities than students with normal stress.Students with severe stress were more involved in these activities at the start of the semester, but the involvement decreased over time.

iOS and Android
Wi-Fi data Detect depression and stress Zakaria et al [38]

Studies With Adults
Table 4 presents the data extracted from the studies conducted with the general adult population.The average study duration was 201.6 (SD 367) days.Apart from a 3-year longitudinal study with 18,000 participants, the average number of participants was 123.4 (SD 139.8).Of the 8 studies with adults, 2 (25%) [32,52] were conducted with the same set of participants.A total of 3 (38%) studies used predictive modeling, with regression-based models being the most common [34,36,52], and 1 (12%) study identified gender differences in behavioral patterns [27].Overall, the research with adults showed that GPS, accelerometer, ambient audio, and illuminance data related to individuals' emotional state.Adults with depression were less likely to leave home and were less physically active, whereas adults who were socially anxious were more active and left their home more often but avoided going to places where they needed to socially interact.

Studies With Employees
Table 5 presents the data extracted from the studies that were conducted with employees.Among the 4 studies with employees, 1 (25%) study recruited its own participants [56], and the other 3 (75%) studies [40][41][42] used the Tesserae data set [63].Compared with students and adults, the employee population was the least studied, with the fewest articles.However, the studies with employees had the largest number of participants, with a mean of 427.3 (SD 280.3).All 4 studies used regression-based predictive modeling, and 2 (50%) of them [40,56] evaluated a variety of models, with logistic regression, support vector machine, and random forest being the most common methods.Detecting and predicting employees' stress in workplaces were examined in tandem with employees' work performance.The research goal for these studies was to understand the underlining reasons for lowered work-related productivity.In contrast to the other 2 populations (ie, students and adults), less mobility was seen as positive for employees because less mobility in workplaces was associated with more positivity and higher performance.

Overview
Table 6 provides an overview of the range of sensors used to detect patterns related to mild mental health symptoms and summarizes the evidence of the effectiveness of the various sensors.The first column lists the sensor, and the second column presents how the data from that sensor are interpreted; in other words, it presents the behavior-related information that the sensor data are intended to represent.The third column indicates which articles found significant associations between the specific sensor and stress, anxiety, or mild depression.The fourth column indicates which articles found no significant associations between the specific sensor and mental health outcomes (ie, explicitly stated so in the articles).In the subsequent sections, we discuss the types of activities detected by the sensors.

Social Interaction: Call and Text Logs, Audio, Microphone, and Bluetooth
The social interaction of an individual is reflective of their current mood and mental state [44,64,65,66].Individuals with depression and stress may be expected to decrease their social interactions.This is measured through the frequency of receiving texts and calls, how fast individuals respond, and the frequency of being around others.Among the 40 included studies, 18 (45%) [27,28,30,31,[33][34][35]37,[44][45][46][47][48]51,53,55,60] examined call logs to understand social interaction patterns, mainly through the number of incoming and outgoing calls, the number of missed calls, and the duration of calls.Individuals who experience depression and stress may engage in longer outgoing calls [51].Evening communications were predictive of depression [47], anxiety, and loneliness [54].Students who experienced discriminations [53] and anxious participants had more evening communications [54].Metadata on SMS text messages were examined in 10 (25%) [27,28,30,31,33,37,44,45,48,55] of the 40 studies, including the frequency of receiving SMS text messages and the average time of responses.People who are socially anxious were found to take different amounts of time to respond to SMS text messages and calls [33].Increases in the number of calls were associated with increased social anxiety [48].Those who experienced social anxiety were less likely to call or text in public [44].For students, fewer conversations were associated with more stress [39] and more mood instability [42].One of the studies found that more emotionally unstable individuals tended to text more than emotionally stable individuals [27].

Location: GPS, Bluetooth, and Wi-Fi
Location data can provide insights into individuals' mental health state in terms of the normal or abnormal variety and frequency of locations visited [67].As presented in Table 6, GPS has been one of the most commonly used passive sensors for stress, anxiety, and mild depression research.The findings regarding location consistently demonstrate that students and adults who experienced depression, anxiety, or stress tended to visit fewer places [39,44,50,[58][59][60].One of the studies [48] found that location data are highly inversely correlated with mild depression severity.The main way in which this is measured is through the frequency of exiting the house, the variety of locations visited, and mobility.The frequency of exiting the house is less for individuals who are depressed, and there is less variety in the visited locations for individuals who are socially anxious.Individuals who are feeling depressed often experience being less energetic [68,69].Overall, negative emotions were associated with time spent at specific locations, but this is also affected by personal routines and preferences [30].For students, stress and lower subjective well-being were associated with more time spent on campus [39,49] and less time spent at campus food locations [39].Students who experienced depression spent more time at home [60], whereas individuals at higher risk of psychosis spent less time at home [51].Time spent at exercise locations was positively correlated with changes in depressive symptoms [48].Another study [38] distinguished between students experiencing severe stress and those with normal stress levels, revealing that students with severe stress spent significantly less time on campus and were less involved in work-related activities compared with their counterparts with normal stress levels.As for employees, higher performers were found to visit fewer locations on weekday evenings but more locations during weekends [56].

Voice Recognition: Audio
The microphone is used to measure audio data of speech and ambient noises.One of the studies [26] examined how people with stress speak by analyzing their voice, including the speed of speech, how energetic their vocality is, and the pitch.One caveat is that the study by Adams et al [26] used audio captured within laboratory environments and found that stress could be recognized from the absence of speech.In variable environments, it will be harder to recognize the changing voice patterns.One study found that generalized anxiety and depression related to reward-related words in ambient speech, and social anxiety related to vision-related words [32].Another study [52] identified that people with depression tend to speak less and use more death-related words.

Sleep: Accelerometer, Audio, and Illuminance
Sleep is highly correlated with individuals' mental state [26,35,36,42,[45][46][47]59,60].Among the 40 included studies, 5 (13%) [35,46,52,60] found that more disturbed sleep correlated with more depressive symptoms.However, occasional sleep disturbance is not necessarily predictive.For example, for those with social anxiety, sleep disturbance might be positive because it suggests night-time activity and social interactions.Metadata on the time spent in darkness can be indicative of sleep patterns.The study by Fukazawa et al [36] stated that anxiety levels increase when the time spent in darkness increases.The study by Di Matteo et al [52] found that individuals with symptoms related to social anxiety and depression spent less time in darker environments.Another study [39] stated that stress changed students' sleep patterns, where they became less likely to move around between 6 PM and midnight.Of the 40 studies, 6 (15%) found that shorter sleep duration was correlated with more mood instability [42], more depressive symptoms [59,60], and more stress [36,44].One of the studies [45] also found that the student population, in general, tended to sleep less during examination periods and slept more during breaks, and they felt more stressed during both breaks and examination periods.

Phone Use: On and Off Screen, Lock and Unlock, and App Use
Today, smartphones are used for self-regulated "distractions," such as the use of social media [38].This type of self-regulated distraction can temporarily reduce stress.The study by Chikersal et al [47] showed that depression can impact concentration levels, so if distraction by phone can be measured, this could be a potential predictive marker.Several studies found that increasing phone use was correlated with more depressive symptoms [46,47,50,52,[58][59][60], anxiety [52,59], impulsivity [34] and lower subjective-wellbeing [49].The study by Morshed et al [42] outlined that for postsemester depression, phone use at night is not predictive, whereas another study [47] summarized that phone use during the day is predictive of depression.More frequent phone locks or unlocks correlated with higher levels of depressive symptoms [60] and impulsivity [34].Higher performing employees tended to unlock their phones less frequently in the evenings [56].Additionally, individuals who were promoted spent more time on their phones during early mornings and late evenings, with more unlocks occurring during nighttime compared with their nonpromoted counterparts [40].

Physical Activity and Mobility: Accelerometer
According to Table 6, along with GPS, accelerometer is one of the most widely used passive sensors in digital phenotyping research to monitor participant's mobility, activity, and sedentary periods.Increased sedentary time was correlated with increased depressive symptoms [47,48,50,[58][59][60], increased mood instability [27,42], increased stress [36] and decreased subjective well-being [49].Exercise duration was positively correlated with changes in anxiety [36] and depressive symptoms [48].The study by Mirjafari et al [56] found that the amount of movement and physical activity was related to employee's stress level and highlighted that if the activity is regular, it should reduce stress.Different occupations require different levels of physical activity, social interactions, and mobility.For instance, developers spend most of their time at their desks, and their tasks might require less social interaction and mobility at work, but this does not mean they are more stressed.Project managers have more mobility during the day, and this may be because they need to move around to meet with the stakeholders [56].Several studies have observed variations in mobility and gait consistency.The study by Boukhechba et al [44] reported that individuals with high social anxiety exhibited a narrower range of activities, whereas the study by Xu et al [60] revealed that students experiencing depression demonstrated more consistent mobility patterns.Additionally, accelerometer data indicated that individuals with low social anxiety maintained a steady walking pace, whereas those with high social anxiety tended to walk more rapidly and with greater irregularity [33].

Muscle Activity: Keyboard
Stress can cause muscle tension [70,71].One of the studies [29] collected the data of users with stress via a keyboard in a laboratory environment and found that typing pressure significantly increased under stressful conditions.

Challenges
Digital phenotyping for mild mental health symptoms in nonclinical participants can present ethical challenges, limitations to the research, and technical challenges.We review the challenges that were stated in the literature.

Ethical Challenges
Among the 40 included studies, 7 (18%) specifically mentioned privacy-related ethical concerns [28,31,35,36,40,41,43].A major concern for participants across several studies was whether authorities, such as employers or teachers, will have access to their data.One of the studies [28] conducted in-depth interviews with 15 students to understand their perspectives on digital phenotyping through app prototypes.They found that the students' core concerns were whether the acquainted university staff had access to the data.They also found that students' acceptability of such apps depends on the perceived relevancy of the data collected and the effects on students' devices.The study by Nepal et al [40] with employees reported a similar privacy concern of whether the employees' data would be leaked to their boss; if the boss is aware of a potential mental health issue, it may impact their work performance ratings.
The methods of collecting and storing passive sensing data also present privacy concerns [28,70,72], particularly when the tracked data involve sensitive topics, such as mental health [72].Sensors that infer individuals' social interactions provide insights into their mental health status [26,36,53].However, these types of data were less likely to be shared by participants XSL • FO RenderX because of privacy concerns.In the study by Rooksby et al [28], students identified camera, microphone, call log, and keyboard data as highly unacceptable types of data to capture.
Location data were associated with privacy and security concerns.In the study by Wen et al [34], participants felt uncomfortable with location tracking because it might breach their privacy and were hesitant to log their location when they moved from one place to another.Some studies excluded specific sensors to protect the participant's privacy.Location data were not recorded owing to security concerns, even though they could provide valuable insights into the mental state [36,38].In the study by Adams et al [26], the microphone was disabled to capture calls and conversations while individuals were talking to their family members.Another ethical concern was regarding the misuse of data.The main focus in studies of digital phenotyping using smartphones was on tracking participants' usual behavioral patterns and identifying whether they behaved unusually.There were concerns regarding secondary uses.For example, participants' leaked data can be used for advertising purposes or to create content [34,41].

Limitations to the Research
Coping mechanisms related to stress and anxiety vary among individuals [22].Individual differences can make it challenging to label individuals as stressed, anxious, or depressed, particularly nonclinical participants.Certain behavioral patterns can be generally expected; however, not all individuals will follow the same pattern.To make generalizable and powerful analyses and understand behavioral patterns associated with mild mental health concerns, it is recommended to study diverse groups for longer than a 2-week period.Of the 40 included studies, 2 (5%) [33,39] focused on a particular demographic subset, namely, undergraduate students.Therefore, the generalizability of the studies is limited.In the studies by Rooksby et al [28], Exposito et al [29], and Wang et al [50], limited variation in representation was seen as a major limitation.The studies by Rhim et al [49], DaSilva et al [39], and Fukazawa et al [36] stressed the importance of selecting a wider age group, as younger people use their smartphones proactively, whereas older people's behavioral patterns might show differences when they are experiencing mild mental health symptoms.The study by Nepal et al [40] suggested that diverse population testing is required for more reliable results, considering interindividual differences.Furthermore, the accuracy and effectiveness of machine learning models are highly affected by data set quality.We noticed that over the last 4 years [38,46,57,60], there has been increased focus on the generalizability of machine learning models, with the goal of assessing generalizability across students from various years, classes, and institutions.

Technical Challenges
Digital phenotyping studies on mild symptoms related to mental health with nonclinical participants presented technical challenges.A main concern was the accuracy of the sensor data collected from smartphones.The study by Fukazawa et al [36] sought to understand the time spent in darkness and its effects on the relationship between stress and anxiety patterns and sleep.However, when individuals carried their smartphone in their pockets or bags, the smartphone could not detect the darkness of the environment.This presented a challenge because illuminance data were captured even when the phone was not used actively.Similar concerns were raised in the study by Di Matteo et al [52].The time spent in darkness feature did not distinguish whether the device was in a dark room or a dark location (ie, in the pocket).The study by Melcher et al [35] stated that the captured accelerometer data may not accurately represent daily activity, as not all participants constantly carried their phones throughout the day.In the study by Di Matteo et al [32], environmental audio did not produce clear transcripts in louder environments.This study mentioned that transcripts were produced based on dictionaries, so language analysis of complex speech, such as metaphors and sarcasm, was ignored.Therefore, the entire content of the conversation might not be correctly interpreted.In the study by Di Matteo et al [52], similar challenges were identified, as the speech data produced from smartphones were not clear.The recorded voices of the participants were masked by those of the people around them or even sound from other sources such as television or radio.Moreover, it was not possible to identify whether the death-related words came from the participants or from the people they interacted with.
Another technical challenge identified was battery life [47].As expected, moment-by-moment data collection requires high power use, which might shorten the battery life.Participants had to charge their phones more often, which was inconvenient, and altered their usual behavior because they could not carry their phones as usual when the phones were charging.The study by Chikersal et al [47] mentioned another technical limitation: the transfer rate was affected if the app stopped working randomly.During these times, data were not transferred or collected.With the increase in the use of 5G technology, Wi-Fi data for indoor locations may cease to be relevant.In the study by Zakaria et al [38], some users were on their 5G indoors rather than their Wi-Fi, and this may point to a future trend of the use of 5G.We now turn to the discussion.

Principal Findings
This literature review examined digital phenotyping studies that detected and predicted stress, anxiety, and depression in their mild states in nonclinical populations using data collected from smartphones.The primary objective of digital phenotyping in the context of mild mental health was similar among the 3 participant cohorts: students, adults, and employees.However, notable distinctions emerged among these groups.Among university students, the geographical proximity and relevance of the university campus were discerned as influential factors.Moreover, academic pursuits, particularly coursework and study-related activities, assumed significance within this demographic.Conversely, among employees, work aspects held salience, accompanied by the workplace environment.The remaining studies encompassed a general population cohort, delineated by undisclosed characteristics.Overall, we found that identifying behavioral abnormalities related to stress and anxiety was possible but raised certain challenges.Generalized XSL • FO RenderX stress and anxiety symptoms vary largely among individuals, whereas serious diagnoses, such as bipolar disorder or schizophrenia, have well-documented behavioral changes.Sleep was a strong predictor variable, yet some individuals tended to sleep more while they were stressed, whereas others lacked sleep under stress.This may be one of the reasons why there are fewer studies and reviews completed on stress and anxiety compared with studies on serious conditions such as bipolar disorder, severe depression, and schizophrenia.Another reason is that clinical psychologists and psychiatrists who are familiar with clinical populations are leading the digital phenotyping research.
Studies tended to use self-report to categorize nonclinical populations as stressed, depressed, or anxious.It was not always clear whether the identified patterns of the passive sensor data would effectively discriminate among groups.Most studies used prestudy and poststudy surveys to identify participants' mental state.There were concerns raised regarding the accuracy of the categorization of self-report surveys.For instance, the study by Sefidgar et al [53] stated that students with stress may not report themselves as very stressed.Melcher et al [70] conducted a review and found that students were concerned regarding their professors learning about their data [71].Thus, the accuracy of self-report remains an issue for passive sensing studies that use self-report labels, especially when there are privacy concerns.This may be related to the high dropout rates in the studies.
Many types of data sensors were used in the reviewed studies.Few articles related sensor patterns to specific symptoms validated by relevant psychological evidence.One of the studies [46] extracted interpretable rules (such as intermittent sleep episodes or number of bouts of being asleep or number of outgoing calls during weekends) through association rule mining to distinguish the behavioral patterns between students who were depressed and students who were not.However, although the behavioral patterns were identified, they were not validated to be exclusive to the addressed mental health issue; for example, high mobility and physical activity do not necessarily mean that the person is not stressed.In the study by Tseng et al [45], students were more mobile during the examination week, despite being under high pressure and stress.In the same study, some students were less mobile when studying for their examinations, which we cannot necessarily be interpreted as being under stress.Of the 40 included studies, 4 (10%) [35,58,59,61] explored the effects of the COVID-19 pandemic on behavioral and mental health.Additional recent investigations, which independently gathered their own data sets during the COVID-19 pandemic, have shown that quarantine measures have influenced individual behavioral patterns.For the purpose of making precise predictions in digital phenotyping, it is imperative to consider contextual and environmental factors.
Privacy and secondary data uses were the main concerns identified for digital phenotyping.Individuals using digital phenotyping systems have the right to provide informed consent.This means that they should be made aware of how all their data will be used, who will have access to their data, where their data will be stored, and for how long their data will be stored, and they have the right to decline to participate.We urge researchers and medical practitioners to carefully consider the system design and requirements because data transferred to the cloud and other services may fall under various service agreements.To empower end users and improve the quality of digital phenotyping systems, we recommend that transparent algorithms and explainable artificial intelligence be combined with user-accessible and understandable displays so that adults can engage in the process of identifying and categorizing patterns related to mild mental health symptoms.
The digital phenotyping research focused on in this review may enable the design of tailored intervention programs for nonclinical participants who are showing symptoms of stress, anxiety, and mild depression.Most of the studies included in this review were conducted within a restricted timeline and limited scope of detection and prediction.Only 4 (10%) of the 40 studies mentioned potential intervention programs upon predicting stress, anxiety, and mild depression [31,38,47,53].
Our review has some limitations.We excluded studies conducted with teenagers, children, and adults who were clinically diagnosed.Thus, we missed studies that focused on the detection and prediction of stress, anxiety, and mild depression in these populations.These populations are likely to show different patterns than those in adults who are not clinically diagnosed.Further, we excluded studies conducted using technologies other than smartphones.We chose this more limited subset of technologies to scope findings related to widely available technologies.The availability of technologies is changing rapidly, and wearables such as smartwatches are becoming more common.As wearable technologies become ubiquitous, we recommend including them in future systematic reviews.This literature review is unique in that it examines studies focused on the behavioral patterns of nonclinical populations, namely students, employees, and adults who are stressed, anxious, or mildly depressed.We examined each type of sensor and indicated when it was significantly associated with mild mental health symptoms.We identified commonalities in the studies in terms of ethical challenges, limitations to the research, and technical challenges.

Conclusions
This systematic literature review found that digital phenotyping can be an effective way of identifying certain behavioral patterns related to stress, anxiety, and mild depression.A range of passive sensors was used in the studies, such as GPS, Bluetooth, ambient audio, light sensors, accelerometers, microphones, illuminance, and Wi-Fi.We found that location, physical activity, and social interaction data were highly related to participants' mental health and well-being.The surveyed literature discussed the ethical and technical challenges that limit the accuracy and generalizability of results.One of the greatest challenges was privacy concerns, and these were primarily related to camera, location, SMS text message, and call log data.Another challenge was the significant variation among individuals and their unique behaviors related to mental health.Finally, technical limitations have not been fully resolved, with issues such as the sensor for illuminance still capturing data while not in use reducing the accuracy of the XSL • FO RenderX collected data.It is hoped that this overview of digital phenotyping and mental health studies conducted in the last decade, including the common privacy and technical concerns, can help move this area of research forward, ultimately improving the quality of passive sensing, and provide benefits in terms of the early detection of relevant mild mental health phenomena.

Figure 1 .
Figure 1.Systematic literature reviewing process and search results with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) diagram.

Table 2 .
Number of reviewed reports (N=36) by year.

Table 3 .
Summary of the reviewed studies with student participants.

Table 4 .
Summary of the reviewed studies with adult participants.

Table 5 .
Summary of the reviewed studies with employee participants.

Table 6 .
Sensor summary of the reviewed studies.