This systematic review was conducted and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement [23]. Methods of the review process and eligibility criteria were established in advance and documented in a review protocol that can be retrieved from the authors upon request. No changes were made to the protocol during the review process.

Original studies involving healthy individuals older than 18 years measuring self-assessed stress on a smartphone were eligible for review. The language of publication was restricted to English. Papers not meeting eligibility criteria or only describing the technical part of the self-assessment of stress were excluded from review. Where multiple articles were reported on the same study, the article presenting the largest and most detailed dataset was included for review. Only studies in which self-assessed stress was reported on smartphones were eligible for review.

Published studies were identified by conducting a systematic literature search through the electronic databases PubMed, PsycINFO, Embase, IEEE, and ACM. The literature search was supplemented by a hand search of reference lists of retrieved articles. The literature search was conducted by 1 researcher (HP), without time restrictions, using the following keywords: (stress or psychological stress or emotional stress) AND (smartphone or cell phone or cellular phone or mobile phone or mobile application or ecological momentary assessment or experience sampling method) and covered a period from 1980 to May 2016. The last literature search was conducted on May 4, 2016.

A PRISMA flow diagram of the study selection process is presented in Figure 1. All identified titles and abstracts were screened for eligibility by 1 researcher (HP). Potentially relevant articles were retrieved and full-text articles then checked for fulfilling eligibility independently by 2 researchers (HP and MFJ). One researcher extracted data (HP), and a second reviewer (MFJ) independently checked the extracted data. Any disagreements were resolved by a discussion between 3 researchers (HP, MFJ, and LVK).

The literature search identified a total of 1517 articles from the 5 databases, and 11 additional studies were identified by hand search of reference lists. Removing duplicates left 1307 articles for further evaluation. Reviewing abstracts and titles resulted in the exclusion of a total of 1118 articles for not meeting eligibility criteria, the 2 main reasons for exclusion being not including human subjects and not involving stress. Thus, 189 full-text articles were evaluated for eligibility. Of these, 154 articles were excluded from the review for various reasons (Figure 1), with the main reasons being (1) subjective stress not measured on a smartphone (n=81) and (2) subjective stress not measured at all (n=37). A list of excluded articles can be retrieved from the authors upon request. Thus, a total of 35 articles fulfilled the eligibility criteria and were included for review [24-58].

Of the 35 studies, 17 were from the United States and the remainder were from Finland (n=3), Italy (n=4), Germany (n=2), Switzerland (n=2), the United Kingdom (n=3), Australia (n=1), Hong Kong (n=1), Portugal (n=1), and Sweden (n=1). The majority of studies were prospective observational studies [24,28,29,31,33-38,41-45,49-58], 2 were randomized control trials [39,47], 4 were other types of intervention studies [27,40,46,48], 2 were case reports [25,26], and 1 was a cross-sectional study [32]. The study period ranged from 1 hour to 191 days. All the studies were published recently, with the oldest one published in 2007 [44] and more than half of the studies published since 2013 [24,25,27,29-34,36,37,40,42,45,47-54,56-58]. More than half of the studies were published in conference proceedings (n=19), whereas 16 studies were published in peer-reviewed scientific journals (Table 1).

Overall, the studies comprised a total of 1464 healthy adult participants, with sample sizes in individual studies varying from 1 to 198 participants. The mean age of the participants was available for 19 of the studies and ranged from 20.1-52.47 years [27,31-34,37-39,43-46,48-52,55,57]. Gender distribution was available for 25 studies [24,27,30-34,36,37,39-42,45-53,55-57], and of these, 4 studies had equal gender distribution [34,47,51,52], whereas there were 2 large gender-specific studies, 1 male [39] and 1 female [57]. In 9 of the studies, the participants were exclusively students [36,40,41,43,44,49,51-53,55], and in 4 studies, the participants were exclusively employees [31,33,42,58].

The majority of the studies used Android-based smartphones (n=19), 3 used Windows-based smartphones [30,34,41], 3 studies used iPhones [25,49,56], 4 studies [39,43,44,46] used other types of smartphones, whereas the remaining 6 studies did not specify what type of smartphones or operating systems were used [26,28,40,47,51,55]. In 15 of the studies, smartphones were provided for the participants, whereas participants used their own smartphones in 4 studies [32,41,42,49]. Although some participants used their own smartphone in 4 studies, other participants borrowed a smartphone [24,37,53,56]. In total, 12 studies did not specify ownership of the smartphones used [25,26,28,30,39,43,44,50,51,55,57,58].

Overall, the included studies used many different methods to measure smartphone-based self-assessed stress. The most common method (n=11) was using a Likert scale (from a 5-point scale to a 10- or 100-point scale) [25-28,32,34,40,48,50,52,54-56,58]. Five studies used a yes or no answer question to measure self-assessed stress [41,43-45,57], and 5 studies used questionnaires [24,49,50,53]. Two studies did not specify how smartphone-based self-assessed stress was measured [36,38].

The frequency of smartphone-based self-assessed stress reports varied . In most of the studies, participants were asked to report their stress levels multiple times per day: from twice a day [39,49,50] to up to once every half hour [24]. In 1 study, participants reported self-assessed stress on a weekly basis [47], whereas in 3 studies, self-assessed stress was reported once per day [29,41,46]. In 4 studies, the frequency of self-assessment was not specified [26,32,35,36].

Six studies investigated self-assessed stress in the context of the workplace [31,33,35,38,42,58], and 1 study in relation to rehabilitation after work-related stress [46]. Two studies measured self-assessed stress in relation to smoking cessation [27,48], 2 in relation to smoking behavior [37,56], and 1 in relation to concurrent smoking and drinking [55]. Two studies investigated self-assessed stress levels in relation to vitamin intake [39,47], 1 in relation to emotional eating [30], and another in relation to snack-food intake [57]. Three studies looked at self-assessed stress in the context of studying [40,43,44]. One study was done in a laboratory context [32], another in relation to driving [52], and a third looked at stress levels in epidemiological behavior context [41]. The remaining studies (n=13) reported no specific context, and participants registered self-assessed stress during their everyday life. About half (n=16) of the studies investigated stress as the primary objective [24-26,28,29,31-33,35,38,42,50,52,54,58].

In 5 studies, validated stress scales in addition to smartphone-based self-assessed stress were reported. Four of these studies used the Perceived Stress Scale (PSS) [24,49,50,53,59], and 1 used Derogatis Stress Profile (DSP) [46,60]. In 2 of the studies, participants filled out the PSS at baseline only [24,50], and in 2 studies, participants filled out the PSS at both baseline and follow-up [49,53]. The study using DSP was an interventional study, and participants filled out the scale 4 times during the study period [46].

Three studies investigated the correlation between smartphone-based self-assessed stress and validated stress scales [24,46,53]. Adams et al reported a statistically nonsignificant correlation (r=.562, P=.11) between smartphone-based self-assessed stress levels and PSS score [24]. Another study by Parkaa et al reported a statistically nonsignificant correlation (ρ=.07, P=.64) between smartphone-based self-assessed stress and DSP score [46]. Finally, a study by Wang et al reported a statistically significant positive moderate correlation between smartphone-based self-assessed stress and PSS score both pre- (r=.458, P=.003) and poststudy (r=.412, P=.009) [53].

A total of 13 studies collected smartphone generated objective data [24,26,29,31,33,41,42,48-53]. Six studies investigated the association between smartphone generated objective data and smartphone-based self-assessed stress [24,29,31,33,41,49]. Among these, 2 studies investigated the association between smartphone-based self-assessed stress and verbal data [24,33]; Adams et al reported a statistically positive correlation (r=.59, P value not specified) between smartphone-based self-assessed stress and voice-stress, whereas Ferdous et al reported a significant positive correlation between smartphone-based self-assessed stress and duration of verbal interaction for 17 of their 28 participants (r=.06-.55, P<.005).

A study by Madan et al reported that communication diversity was reduced for participants who often assessed themselves as being stressed, and the authors interpreted this as a tendency to isolate [41]. A study by Sano et al reported that higher self-assessed stress levels were statistically significantly correlated with lower activity level in the evening, fewer and shorter text messages sent, and less screen activity in the evening [49].

Two studies investigated the association between self-assessed stress and smartphone generated objective data in order to detect stress [29,31]. A study by Ceja et al looked at smartphone generated objective data from the accelerometer and “achieved a maximum overall accuracy of 71% for user-specific models and an accuracy of 60% for the use of similar-users models” to classify self-assessed stress levels [31]. A study by Bogomolov et al collected both social features (phone calls and text messages) and proximity features (Bluetooth) and obtained “the accuracy score of 72.28% for a 2-class daily stress recognition problem” [29].

This was the first systematic review on smartphone-based self-assessment of stress in healthy adult individuals. A total of 35 published articles involving a total of 1464 participants were included for review. Overall, the study designs were highly heterogeneous, using various methods of self-assessment in different contexts. Most of the studies were conducted in the United States or Western Europe. Android-based smartphones were most commonly used for measuring self-assessed stress, many participants borrowed smartphones during the studies, and often stress was reported multiple times per day.

Regarding the validity of smartphone-based self-assessed stress levels, stress levels measured using validated stress scales were available in 5 studies, but only 3 of these studies investigated the correlation between smartphone-based self-assessed stress and validated stress scales. Among these 3 studies, only 1 study found a statistically significant positive correlation between self-assessed stress and a validated stress scale (PSS) [53]. It should be noted that the study by Wang et al included a larger sample (n=48) compared with the other 2 studies combined (n=7; n=17) [24,46], suggesting a low statistical power of the other 2 studies. In addition, the study by Wang et al included university students on a university campus, limiting the generalizability of the study findings. The validity of smartphone-based self-assessment of stress may be different across populations and should be investigated further in future studies. Thus, findings from this systematic review suggest that the validity of smartphone-based self-assessed stress has been sparingly investigated and is unknown. The studies included described convergent validity of smartphone-based self-assessment of stress. Other parameters such as sleep, mood, and activity level may correlate with validated stress scales; however, content validity was not investigated in this review. In addition, the reliability and predictive validity of smartphone-based self-assessment of stress were also not investigated.

Smartphone generated objective data were collected in 13 studies and 6 studies investigated the association between smartphone-based self-assessed stress and these objective data. Two studies found a positive correlation between self-assessed stress and verbal data, whereas another 2 studies found associations between self-assessed stress and communication diversity, activity levels, text messages, and screen on or off patterns. The last 2 studies found smartphone generated data to be a predictor (accuracy up to 72.28%) for detecting self-assessed stress. Overall, regarding smartphone generated objective data, the studies collected various smartphone generated data and the results seem exploratory, with a tendency to report statistically significant positive correlations with self-assessed stress only.

A majority of the included studies collected objective data alongside the self-assessed data. Some of them used physiological measures collected from various worn sensors, but others only used objective data collected from sensors embedded within the smartphones. Seven studies collected all 3 kinds of stress measures. Collecting physiological measures such as heart rate requires participants to carry additional sensors (user burden), whereas smartphone generated objective data are collected from a smartphone that is most likely already being carried around. Smartphone generated objective data can usually be collected automatically, eliminating attrition due to monitoring. Objective smartphone data are behavioral data that can reflect behavior related to stress. Different people react differently to stress, and combined with self-assessed data on stress, smartphone generated objective data might be used for detecting stress. Early stress detection in healthy populations such as students and employees could help to prevent stress-related diseases. Thus, the use of smartphone generated objective data as a marker of stress in healthy individuals has been sparingly investigated and future well-designed studies investigating this would be interesting.

Stress levels were assessed from self-reported data, both from smartphones and from validated scales. PSS was developed in 1983. It has 10 questions and is widely used within psychological and psychiatric sciences. It has shown good internal reliability (Cronbach alpha=.78-.91 [61]) and is correlated with various self-report and behavioral criteria [59]. DSP is a 77-item self-report inventory developed in 1980 and has also shown good internal reliability (Cronbach alpha=.83-.88 [62]). It should be emphasized that the different methods for self-assessment of stress, smartphone-based and validated scales, do not necessarily measure the same thing. Validated stress scales measure more long-term stress levels, whereas self-assessment on smartphones is more about current stress levels. Validated scales such as PSS have a somewhat clear definition of stress, as they have several items that the participants have to answer. Many of the smartphone-based self-assessment measures of stress were not explicit in their definition of stress, and participants often only answered 1 question about their level of stress. Stress is a popular term and can mean different things to different people; some people might only register stress that they experience as a negative thing, whereas others might also register the kind of stress (eustress) that is positive and can be motivating. As noted in a study by Muaremi et al, stress was not necessarily a negative event or feeling for some of the participants [42].

Registering self-assessed stress multiple times a day can be a tool to help people self-monitor stress levels. In this way, self-monitoring may play a role in helping people to manage stress. Self-monitoring brings awareness of stress levels and encourages behavioral change according to a situation [63]. However, being asked to self-assess one’s stress level up to multiple times a day could introduce a negativity bias. This could result in participants assessing their stress to be higher than it actually is and even potentially cause more stress per se. It may be that measurements in themselves are stressful, but also the situation to have the self-assessed results of chronic stress constantly at hand and to be unable to cope with a given stressful situation. In this case, people may be constantly reminded that they are unable to cope with stress, which may be the reason they are measuring self-assessed stress in the first place. Investigating the effect of introducing coaching or coping elements to the self-assessment apps would be interesting. It should be stressed that we identified no study that investigated whether the use of smartphone to continuously monitor stress—subjectively reported or objectively assessed—per se had a reducing effect on stress level. Whether self-assessments multiple times a day would be a threat to the reliability and validity is unknown and should be investigated further. Most studies looked at self-assessed stress in everyday life, either without context or in the context of work or studying. Many people carry their smartphones with them during most of the day and therefore smartphones are a device well suited for this type of data collection. Registering stress multiple times a day, in different situations, can shed light on where and when people are experiencing stress.

A study by Wang et al looked at stress in students over a whole semester and revealed how their self-assessed stress level increased as their workload increased, with the peak being during final examinations [53]. Following a group of people prospectively over time could help distinguish between the normal stresses that come and go and the chronic, potentially health-damaging kind of stress. Being aware of chronic stress is the first step toward eradicating or minimizing it.

Most studies measured self-assessed stress on Android-based smartphones, and many participants were provided with smartphones during the study period. Allowing participants to use their own smartphones to collect self-assessment of stress would be the least disruptive for participants, as they are already familiar with the device. Using one’s own smartphone would also be likely to more accurately reflect real life, especially in regards to the objective smartphone data. It is possible that participants did not, in all cases, own smartphones. It is also possible that the study smartphones were specially programmed for the study or that participants’ smartphones were different from the ones that were required for the study.

Smartphones constitute a new and an exciting research tool within psychological well-being and health care. Nevertheless, the majority of the identified studies have been published in proceedings from technological conferences. In general, many of these studies focused primarily on the technical side of the smartphone system, and a number of these did not present data on population characteristics such as age [24,26,28-30,35,36,41,44,53,54,58], gender [26,28,29,35,38,43,44,54,58], or employment status of participants [24-30,32,34,35,38,39,45,48,50,54,56].

Limitations at a study level: Several concerns regarding the individual studies and outcomes limited the overall findings of this study. The included studies had highly heterogeneous designs and used various methods to measure smartphone-based self-assessed stress. In addition, in many cases studies did not include clear descriptions of the recruitment process. The studies included were at risk of selection bias, and at an individual study level, there was a lack of information on potential confounding factors such as age, gender, and educational level, which possibly could have affected self-assessed stress level. A large part of the studies included a relatively small sample of participants and reported unadjusted statistical analyses. Validated stress scales were only used in 5 studies out of the 35 studies included. More than half of the included studies did not investigate stress as their primary objective, and information was therefore limited: only 1 out of the 4 largest studies had stress as their primary objective. In general, studies focusing on stress had fewer participants (mean n=24.7) compared with the studies not focusing on stress (mean n=56.3). Self-assessed stress was investigated in selected groups, often recruited through convenience sampling at a university or a workplace. In many of the studies, participants were provided with a smartphone to use during the study period, and some participants received economic incentives to fill out the self-assessments of stress. The generalizability of these studies was therefore limited, but findings could be relevant for more narrow populations such as university students. Overall, methodological limitations related to study designs, self-assessments of stress, as well as statistical analyses of the included studies were observed. There is a need for studies investigating the use and validity of smartphone-based self-assessed stress in more general populations.

Limitations at a review level: Some limitations to this review should be mentioned. Research using smartphones is expanding, and due to the intersectionality of this research (medicine, psychology, and information technology), studies are being published in very diverse forms and places. Our review shows that many of these kinds of studies are being published in conference proceedings. Therefore, conducting a search strategy that is able to capture all relevant scientific articles is a challenge. The review process was restricting among healthy smartphone users and articles published in English, which might have reduced the global acceptance.

Stress has become a major health problem in the Western world. Awareness of one’s own stress level is important, and smartphones are potentially a proper minimally intrusive tool for self-assessment of stress.

Self-assessment of stress using smartphones in everyday life is a step toward stress awareness. Looking at self-reported stress levels in relation to other more objective data from smartphones, such as geolocation and physical activity, could help to further understanding of stress and stress-related behavior. However, well-designed studies using strict methodology investigating the validity of smartphone-based self-assessment of stress are warranted. Future studies should investigate how to validly measure subjective stress using smartphones, which by nature is accurate in time and place, in contrast to a self-reported scale on stress administered once a day or less frequently. They should also collect information on and address possible confounding factors in the statistical analyses. In addition, and of even more paramount importance, they should investigate in a randomized controlled trial whether the use of smartphone to monitor stress—subjectively or objectively assessed—per se has a beneficial or detrimental effect on stress level.

This review included only studies with healthy adult participants. Smartphones can and are also being used to measure self-assessed stress in various patient populations, especially within the mental health field, where stress is a risk factor. However, addressing this aspect was beyond the scope of this review.

This systematic review identified 35 studies using smartphones to measure self-assessed stress in healthy adults. The studies were from different countries and used different self-assessment methods in varying contexts, such as in the workplace, in relation to smoking cessation, and on university campuses. Android-based smartphones were most commonly being used, and the validity of smartphone-based self-assessed stress compared with validated stress scales was limited by low statistical power of the individual studies and small number of studies reporting on validated scales. Some smartphone generated objective data, including voice, activity, and general usage data, were associated with self-assessed stress measured on smartphones. Smartphone generated objective data could represent a potential tool for predicting stress levels. There is a need for further studies investigating the validity of smartphone-based self-assessed stress and smartphone generated objective measures of stress using validated stress scales, and studies investigating the beneficial or detrimental effects of smartphone-based monitoring stress, both subjectively and objectively, on stress levels per se.