This study used a longitudinal follow-up design with repeated measurements collected at 6-month intervals between March 2022 and September 2022. The full study protocol included 4 waves; however, at the time of analysis, data from the first 2 waves were available and included in this study. Each measurement wave consisted of a 14-day ambulatory monitoring period, during which participants provided continuous physiological data and mEMA, followed by a postmeasurement questionnaire. The study was conducted in naturalistic work environments to examine real-world patterns of occupational well-being. The primary objective was to investigate associations between EDA, valence, burnout, and engagement. The analytic sample included participants with sufficient paired physiological and self-report data across the first 2 waves (wave 1: n=57; wave 2: n=51; total observations=108). The study design reflects a digital health-monitoring approach combining wearable sensing and mobile-based self-report in real-world settings.

Participants were recruited from 3 Finnish knowledge-intensive companies. A total of 123 employees enrolled in the study, of whom the majority were Finnish-speaking (n=119, 96%), while a small proportion were of other nationalities (n=4, 3%) with English as their working language. Participants were invited via email and provided with detailed study information in Finnish and English. Eligibility criteria included being employed by one of the participating organizations at the start of the study and having no significant self-reported health conditions that would prevent participation. Individuals with known nickel allergy were advised not to participate due to potential skin reactions to the wearable device. Symptoms of anxiety and depression were not exclusion criteria but were assessed and included as covariates in the analyses.

Participation was voluntary, and all participants provided informed consent prior to data collection. No financial compensation was provided. The average duration of work experience in the current position was 7.1 (SD 6.8) years in wave 1 and 6.1 (SD 5.7) years in wave 2. Detailed demographic, occupational, behavioral, and psychological characteristics of the analytic sample are reported separately for each measurement wave in Table S2 in Multimedia Appendix 1, including age, sex, working hours, family commitments, sleep duration, medication affecting sympathetic nervous system activity, managerial position, work experience, nicotine use, BMI, anxiety symptoms, and depressive symptoms.

Data collection was conducted in naturalistic work settings and followed a systematic sequence within each measurement wave. Prior to the first measurement wave, participants were provided with a wearable device and access to a mobile app that was used to collect both physiological data from the device and mEMA responses. Participants also received instructions for using both the device and the app. For subsequent waves, participants retained the devices and were informed by email about the upcoming data collection period and instructed to resume the same measurement protocol.

Each measurement wave consisted of a 14-day monitoring period. During this time, participants continuously wore the wearable device to record EDA during both daytime activities and nighttime sleep. Participants were instructed to wear the device for at least 16 hours per day to ensure sufficient data coverage. In parallel with physiological monitoring, participants completed mEMA of valence using a mobile app. Prompts were delivered every 3 hours during waking hours, and participants were asked to report their overall emotional experience during the preceding interval. Responses could be entered retrospectively within a 12-hour window, allowing participants to maintain normal sleep patterns without interruption. Following the 14-day monitoring period, participants were invited to complete an online questionnaire assessing burnout, engagement, and relevant background variables. The questionnaire was designed to capture experiences over the same 2-week period as the physiological and mEMA data collection. This timing was intentional, as the validated self-report measures used in the study assess experiences retrospectively over approximately 2-week periods, allowing alignment between the temporal scope of the questionnaire data and the physiological monitoring window. Data quality criteria were predefined to ensure reliable physiological measurement. Days with less than 16 hours of wearable data were excluded, and participants with fewer than 6 valid days of data within a measurement wave were excluded from analyses. This multimodal data collection approach enabled the integration of continuous physiological monitoring, repeated momentary self-reports, and retrospective questionnaire data within the same real-world observation period. Figure 2 illustrates the data collection workflow within each measurement wave.

The study was conducted in accordance with the principles of the Declaration of Helsinki. Ethical approval was obtained from the Regional Ethics Committee of the Expert Responsibility Area of Tampere University Hospital (identifier R21149). Participation was voluntary, and all participants provided informed consent prior to enrollment. Participants received no financial or other compensation for their participation. Participants were informed about the purpose of the study, the types of data collected (including continuous physiological data and self-reported measures), and their right to withdraw from the study at any time without consequences.

Given the use of wearable devices and mobile apps in a workplace context, particular attention was paid to data privacy and participant autonomy. Physiological and self-report data were collected for research purposes only and were not shared with employers at the individual level. Data were deidentified prior to analysis, and all results are reported at an aggregated level. Data-handling procedures complied with applicable data protection regulations, including the General Data Protection Regulation.

Continuous EDA data were processed using the device manufacturer’s software to derive minute-level DNE time series. A 12-hour initial usage period was completed prior to each measurement period, as recommended by the manufacturer. Data quality control followed predefined wear-time criteria: days with less than 16 hours of valid data were excluded, and participants with fewer than 6 valid days per measurement period were excluded from the analyses.

From the DNE time series, summary variables were derived to characterize patterns of autonomic activation. These included indicators of daily average activation, between-day variability, and activation levels during waking and rest periods, as well as within-period variability (Table 1). For each valid 24-hour period (6 AM to 6 AM), daily DNE was calculated as the mean of the minute-level DNE values, and these daily values were averaged across the measurement period. Between-day variation was defined as the standard deviation of these daily means. Nighttime (11 PM to 7 AM) and daytime (7 AM to 11 PM) DNE levels were calculated analogously. Within-nighttime and within-daytime variations were defined as the average across days of the interquartile range of minute-level DNE values within the corresponding time interval. These derived variables were used as predictors in the statistical models. As a physiological plausibility check, diurnal patterns of the DNE index were examined, including expected reductions during nighttime periods. These patterns are reported in the Results section.

Adequate paired wearable, EMA, and questionnaire data were available for 57 participant-wave observations in wave 1 and 51 participant-wave observations in wave 2, resulting in 108 analyzable observations from 75 distinct participants (Figure 3). Complete datasets required valid wearable data, EMA responses, and completed surveys within the same measurement wave. Detailed data availability for each data source is presented in Table S3 in Multimedia Appendix 1. The analytic sample had a mean age of 39.2 (SD 7.0) years in wave 1 and 39.7 (SD 7.3) years in wave 2; the corresponding female/male distributions were 32/25 and 26/25. Mean GAD-7 scores were 5.4 (SD 4.1) and 5.8 (SD 3.7), and mean BDI-II scores were 8.7 (SD 6.7) and 9.4 (SD 7.5), respectively. Additional demographic, occupational, behavioral, and psychological characteristics are reported in Table S2 in Multimedia Appendix 1.

The distribution of burnout (BAT-12) and engagement (UWES-9) scores across the analytic sample (n=108 observations) is presented in Table 2. Burnout and engagement were strongly negatively correlated at both time points (wave 1: r=−0.69, 95% CI −0.81 to −0.52, t₅₅=−7.08, P<.001; wave 2: r=−0.79, 95% CI −0.87 to −0.65, t₄₉=−8.94, P<.001).

To examine overall patterns of autonomic activation, average hourly levels of the DNE index were analyzed across the 24-hour cycle (Figures 4A and 4B). DNE levels increased in the morning (around 7 AM), remained elevated during daytime hours, and decreased in the late evening (around 11 PM), with an overall mean level of approximately 50.

Regression results for burnout and engagement are summarized in Table 1. Daily DNE was negatively associated with burnout (estimate=−0.77, 95% CI −1.52 to −0.02; P=.045). Valence showed a negative association with burnout that did not reach statistical significance (estimate=−1.19, 95% CI −2.49 to 0.12; P=.07). The interaction between daily DNE and valence was not statistically significant (estimate=0.15, 95% CI −0.00 to 0.31; P=.052), although the positive interaction coefficient suggested that the negative association between daily DNE and burnout was more pronounced at lower levels of valence and weaker at higher levels of valence. Greater within-nighttime variability in DNE was independently associated with higher burnout scores (estimate=0.02, 95% CI 0.01-0.03; P<.001). Higher daytime DNE levels showed a positive association with burnout that did not reach statistical significance (estimate=0.44, 95% CI −0.06 to 0.95; P=.08). In addition, higher depressive symptom scores (BDI-II) were strongly associated with higher burnout scores (estimate=0.05, 95% CI 0.03-0.07; P<.001).

In the engagement model, higher daily DNE levels were associated with higher engagement scores (estimate=1.82, 95% CI 0.18-3.45; P=.03). Nighttime DNE levels showed a negative association with engagement that did not reach statistical significance (estimate=−0.57, 95% CI −1.19 to 0.05; P=.07), and daytime DNE levels also showed a negative association with engagement that did not reach statistical significance (estimate=−1.09, 95% CI −2.19 to 0.01; P=.052). Greater within-nighttime variability in DNE was associated with lower engagement scores (estimate=−0.03, 95% CI −0.05 to −0.01; P=.009). The interaction between daily DNE and valence was statistically significant (estimate=−0.35, 95% CI −0.69 to −0.00; P=.047), indicating that the positive association between daily DNE and engagement was stronger at lower levels of valence and weaker at higher levels of valence. In addition, higher depressive symptom scores (BDI-II) were strongly associated with lower engagement scores (estimate=−0.08, 95% CI −0.12 to −0.04; P<.001).

Model comparison analyses in Table 3 further supported the relevance of DNE- and valence-related variables for burnout. Removing all DNE and valence variables from the model significantly worsened model fit (χ²₁₀=28.5; P=.002). Similarly, excluding all DNE variables (χ²₉=28.0; P=.001) or valence alone (χ²₄=13.6; P=.009) significantly reduced model fit, whereas excluding daytime DNE variables alone did not significantly worsen model fit (χ²₃=6.2; P=.10).

A broadly similar pattern was observed for engagement. Removing all DNE and valence variables from the model significantly worsened model fit (χ²₁₀=23.7; P=.009). Similarly, excluding all DNE variables (χ²₉=23.5; P=.005), daytime DNE variables alone (χ²₃=9.1; P=.03), or valence alone (χ²₄=10.3; P=.04) significantly reduced model fit.

Illustrative model-based profiles for hypothetical combinations of arousal and valence are presented in Table 4 and Figure 5. Across both burnout and engagement outcomes, valence shifted predicted values in the expected direction, with negative valence associated with higher predicted burnout scores and lower predicted engagement scores compared with positive valence.

Under high arousal conditions, predicted burnout scores were higher when valence was negative (BAT-12=3.84) than when it was positive (BAT-12=2.04). A similar pattern was observed under low arousal conditions, although the differences were smaller (BAT-12=3.50 vs 1.88). Correspondingly, predicted engagement scores were lowest under combinations of negative valence and high arousal (UWES-9=0.01) and highest under positive valence conditions (UWES-9=4.37‐4.80). Prediction intervals were wide and partially overlapping, indicating uncertainty in the magnitude of the estimated effects. These profiles illustrate the fitted model relationships and were not intended to assess predictive accuracy or individual-level classification.

Time-resolved analyses showed that hourly DNE profiles differed across burnout and engagement subgroups (Figures 6A and 6B; Tables S4 and S5 in Multimedia Appendix 1). For burnout, the at-risk subgroup showed higher DNE levels than the no-risk subgroup mainly from midday to midnight. The at-risk subgroup also showed higher DNE levels than the likely burnout subgroup from late morning to late evening. In contrast, differences between the no-risk and likely burnout subgroups were not statistically significant in the hour-level comparisons. For engagement, subgroup differences varied across the 24-hour cycle. The low or very low engagement subgroup tended to show higher daytime DNE levels than the moderate or high engagement subgroups, although statistically significant differences were limited to selected hours. The very high engagement subgroup also differed from the moderate and high engagement subgroups during late-evening and early-night hours. No statistically significant differences were observed between the low or very low and very high engagement subgroups in the hour-level comparisons.

This study examined the use of continuous wearable EDA together with mEMA in a naturalistic occupational setting and evaluated associations between the proprietary DNE index, valence, burnout, and engagement. The protocol yielded analyzable multimodal data from repeated 14-day monitoring periods, providing practical evidence that this type of data collection can be implemented in real-world settings.

EDA-derived indices were associated with burnout and engagement but do not provide construct-specific measurement of these states. Although several DNE-related associations were only near-significant, these associations emerged despite adjustment for a large number of potentially correlated covariates, including depressive symptoms. This suggests that DNE-related variables may capture aspects of burnout not fully explained by conventional psychological or behavioral measures. However, because the model included multiple correlated DNE variables and interaction terms, the observed interaction patterns should be interpreted cautiously. Including these variables improved model fit relative to models without them, indicating that ambulatory physiological signals and momentary self-reports provide complementary information. In both outcomes, the association between EDA-derived arousal and burnout or engagement depended on valence, suggesting that arousal alone is insufficient to distinguish between these states.

Time-resolved analyses suggested that occupational well-being was not reflected merely in static average levels of autonomic activation but also in the timing and shape of daily DNE profiles. Burnout-related differences were most evident from midday to late evening, with the at-risk group showing the highest activation. Engagement-related differences appeared nonlinear: moderate and high engagement were associated with lower daytime DNE than both low or very low and very high engagement. This suggests that engagement may not follow a simple “more engagement–less arousal” gradient but may instead be linked to a more balanced daily activation profile. Overall, the results support the feasibility of integrating wearable EDA sensing with mEMA to capture real-world patterns of occupational well-being and highlight the importance of considering both arousal and valence in interpreting these patterns.

The present findings extend prior psychophysiological research by demonstrating that continuous, real-world monitoring of autonomic activation can provide meaningful insights into occupational well-being. The observed diurnal variation in the DNE index—characterized by higher daytime and lower nighttime levels—was consistent with previously described temporal variation in EDA [41-43]. This pattern provides physiological plausibility for the DNE signal but does not by itself establish construct-specific validity or full biomarker validation, as behavioral activity, sleep timing, environmental conditions, wear behavior, and features of the proprietary normalization procedure may also contribute to the observed variation.

The association between EDA-derived arousal and both burnout and engagement aligns with the circumplex model of occupational well-being, which conceptualizes these states along dimensions of arousal and valence [21-23]. Importantly, the findings indicate that both burnout and engagement may involve elevated levels of arousal, differing primarily in their associated valence. This supports a view that high-arousal states in occupational contexts are not inherently positive or negative but must be interpreted in relation to subjective experience. The moderating role of valence in the regression models further underscores the importance of combining physiological and self-reported data to differentiate between qualitatively distinct states that may share similar levels of autonomic activation.

The time-of-day–specific patterns observed in subgroup analyses provide additional insight into the dynamic nature of occupational well-being. Elevated evening and early-night arousal among individuals at higher risk of burnout may reflect prolonged activation and insufficient recovery, consistent with theories of allostatic load [44] and the effort-recovery model [45]. Similarly, the nonmonotonic relationship between engagement and arousal suggests that both very low and very high engagement may be associated with elevated activation, whereas moderate levels of engagement may reflect a more balanced physiological state. These findings are in line with previous research indicating that engagement, while generally considered beneficial, may also involve increased physiological demands under certain conditions [16].

Contrary to our initial expectation, higher levels of burnout were not associated with reduced autonomic activation. Instead, individuals in the likely burnout subgroup showed elevated DNE levels, particularly during the late-evening and early-night hours. Although findings regarding physiological correlates of burnout remain inconsistent, several studies have reported autonomic dysregulation characterized by elevated sympathetic activity and reduced parasympathetic functioning among individuals with burnout [46,47]. In contrast, a systematic review by de Looff et al [18] concluded that while higher job stress was consistently associated with increased sympathetic activity, evidence regarding burnout remained inconclusive because of methodological heterogeneity and inconsistent findings across studies. Our findings add to this literature by suggesting that elevated autonomic activation may also characterize individuals at higher risk of burnout in naturalistic occupational settings.

An alternative explanation relates to differences in measurement methodology. As noted by de Looff et al [18], psychophysiological burnout studies vary considerably in recording duration, measurement context, analytical procedures, and physiological outcome measures, making comparisons difficult. Most previous studies relied on resting assessments, laboratory stress protocols, or monitoring periods lasting only hours or a few days. Continuous ambulatory monitoring across 14 consecutive days may capture temporal patterns of autonomic activation and recovery that shorter assessments cannot detect. Notably, the most pronounced differences between burnout subgroups in this study emerged during late-evening and early-night hours, suggesting that alterations in recovery-related autonomic regulation may be particularly relevant to burnout risk.

The absence of reduced activation is also noteworthy because our original expectation was partly informed by literature linking burnout and depression, including evidence of overlapping symptoms and shared underlying mechanisms [48,49]. Reduced physiological reactivity and blunted stress responses have been reported in depression and in some clinical burnout populations, particularly among individuals with severe exhaustion or comorbid affective disorders [50-52]. However, the present sample consisted of actively working knowledge workers rather than patients with clinically diagnosed burnout, and symptoms of depression and anxiety were statistically controlled in the regression analyses. It is therefore possible that the physiological profile observed here reflects an earlier stage of the burnout process characterized by sustained sympathetic activation rather than physiological hypoactivation. Future studies including clinically burned-out populations are needed to determine whether autonomic activation changes across different stages of burnout development.

Taken together, these findings suggest that continuous physiological monitoring captures patterns of autonomic activation that are meaningfully associated with occupational well-being but are not specific to individual psychological constructs. Instead, EDA appears to reflect general arousal dynamics, which, when combined with contextual or self-reported information such as valence, can provide a more nuanced understanding of burnout and engagement in real-world settings.

The present findings highlight the potential of integrating wearable physiological sensing and mEMA into digital health approaches for monitoring occupational well-being in real-world settings. Continuous EDA provided a low-burden means of capturing patterns of autonomic activation over extended periods, while EMA-based assessments of valence offered complementary contextual information. Together, these modalities enabled a more nuanced characterization of burnout and engagement than either approach alone. Our findings align with recent work suggesting that wearable and EMA-based monitoring may capture meaningful longitudinal variation in occupational well-being in naturalistic settings [3].

An important methodological aspect of this study relates to the use of a proprietary, device-derived EDA index. Rather than evaluating a fully transparent physiological signal-processing pipeline, this study examines the behavior of a commercially implemented EDA-derived arousal metric under naturalistic conditions. This distinction reflects a broader consideration in digital health research, where many scalable and user-facing technologies rely on proprietary algorithms. In this context, this study provides preliminary convergent evidence from the observed behavior of the DNE index and its associations with established self-report measures of occupational well-being. The observed diurnal pattern provides physiological plausibility but does not by itself establish construct-specific validity or full biomarker validation. More broadly, the findings illustrate both the potential and the limitations of evaluating proprietary sensing systems when the underlying processing pipeline is not fully accessible.

At the same time, this approach entails important trade-offs. Limited computational transparency constrains direct comparability with studies using open EDA processing methods and highlights the need for continued validation of wearable-derived indices across populations and contexts. Future research should seek to bridge this gap by combining proprietary systems with complementary validation approaches, including multimodal sensing and open physiological benchmarks.

From a digital health perspective, such approaches may enable more timely and proactive monitoring of occupational well-being compared with traditional metrics, such as sick leave or employee turnover, which typically reflect deterioration after it has already occurred. Continuous physiological and self-reported data may help identify patterns of sustained activation and insufficient recovery that are not readily captured by periodic self-report surveys.

However, the use of wearable and mobile technologies in occupational contexts raises important ethical considerations. Physiological data are inherently sensitive, and their use in workplace settings may introduce risks related to privacy, perceived surveillance, and potential misuse of data for performance evaluation or discrimination. To mitigate these risks, the implementation of such systems should be guided by principles of transparency, voluntary participation, and clear separation between health monitoring and organizational decision-making processes.

In particular, future applications could include individual feedback and self-management approaches. However, the present findings are based on group-level associations and do not establish the reliability or diagnostic usage of these measures at the individual level. Future research with larger samples is needed before such applications can be validated. Aggregated and anonymized data may also be used to inform organizational practices, provided that appropriate safeguards are in place.

The findings also suggest that high levels of engagement may be associated with elevated physiological activation. Although engagement is generally considered beneficial, sustained high-arousal states may also require adequate recovery. These findings highlight the importance of balancing performance and recovery in digital health interventions. Rather than focusing solely on maximizing engagement, future systems should support sustainable patterns of work by integrating physiological signals with contextual and subjective information.

Overall, wearable and mobile technologies offer promising opportunities for workplace contexts, but their responsible use requires careful consideration of ethical, legal, and social implications to ensure that benefits to individuals and organizations are achieved without compromising individual privacy, autonomy, and trust.

This study has several strengths. First, it used continuous, long-term physiological monitoring in real-world settings, enabling the assessment of autonomic activation patterns over extended periods rather than relying on laboratory-based or short-term measurements. Second, the combination of wearable EDA and mEMA allowed for the integration of physiological and subjective data, providing a more comprehensive view of occupational well-being. Third, the longitudinal design, including 2 measurement waves, enhanced the robustness of the findings and allowed examination of repeated observations within individuals.

Several limitations should also be considered. The final analytic sample was smaller than initially anticipated due to incomplete data across the study components, including insufficient EDA wear time, missing mEMA responses, and noncompletion of the postmeasurement questionnaire. Given the number of fixed-effect terms and the inclusion of related DNE summaries and interaction terms, individual coefficient-level findings should be interpreted cautiously and together with their confidence intervals and the model comparison results. Because burnout and engagement were assessed after the monitoring period, these measures were unavailable for participants who did not complete the questionnaire, and it was not possible to determine whether attrition was systematically related to the outcomes of interest. However, the observed distribution of burnout levels in the sample was broadly consistent with reported prevalence estimates in the Finnish working population, suggesting that substantial bias due to selective attrition is unlikely, although it cannot be ruled out.

The self-report measures of burnout and engagement were designed to capture experiences over the same 2-week period as the physiological monitoring and were completed shortly after the measurement period. However, some participants completed the questionnaires with a delay (median 3 days, mean 5 days, maximum 17 days), introducing a potential temporal mismatch between physiological data and self-reported outcomes. Given that burnout and engagement are typically conceptualized as relatively stable constructs rather than day-to-day states, this delay is unlikely to have materially affected the observed associations, although some attenuation of relationships cannot be ruled out.

The administration of mEMA every 3 hours was reported as burdensome by some participants, which may have reduced compliance and limited variability in valence responses. In addition, the 3-hour retrospective reporting window may have attenuated short-lived fluctuations in affect. Future studies could explore less burdensome mEMA designs, such as reduced prompt frequency or simplified response formats, to improve adherence and sensitivity to momentary changes. The study relied on assumptions regarding typical daily schedules, including fixed working hours and sleep periods, which may not accurately reflect individual variations, particularly in occupations involving flexible or irregular work patterns. This may have introduced some imprecision in distinguishing between rest and activity periods.

Importantly, the DNE index is derived from a proprietary algorithm and cannot be independently reconstructed from raw EDA signals. The archived research data contain DNE outputs from the minute level through the measurement period summary variables, together with the R code used for study-specific data processing and statistical analyses. Thus, the derivation of the study-specific DNE summaries and the statistical analyses can be reproduced from the archived DNE output, whereas the proprietary transformation from sensor-level EDA to DNE cannot be independently reproduced. Exact replication of the original data acquisition pipeline is further limited because the specific device and software version used in this study is no longer supported. However, the device and DNE-processing configuration remained unchanged during the study, reducing the possibility that differences between measurement waves were attributable to device or software updates.

The proprietary nature of the DNE algorithm also highlights the importance of longitudinal validation across device generations. Future studies should evaluate whether newer hardware and software versions produce comparable DNE profiles and maintain similar associations with occupational well-being outcomes. Future research should aim to replicate these findings in larger and more diverse samples, including clinical populations with severe burnout. Longitudinal studies with finer temporal alignment between physiological and self-report measures would help clarify the temporal dynamics of autonomic activation in relation to occupational well-being. Furthermore, continued development and validation of wearable-derived indices, along with integration of multimodal data sources, may enhance the interpretability and applicability of digital health approaches for monitoring mental well-being in real-world contexts.

This study provides exploratory evidence that continuous wearable EDA monitoring can be combined with mEMA in naturalistic occupational settings. DNE-derived measures of autonomic arousal were associated with burnout and engagement, while mEMA-based valence provided complementary information for interpreting these associations.

The findings support further investigation of proprietary EDA-derived arousal metrics as candidate digital biomarkers but do not by themselves establish DNE as a fully validated, construct-specific biomarker of burnout or engagement. The study was not designed to establish prospective predictive performance, screening or diagnostic validity, sensitivity to change, or individual-level decision usage.

More broadly, the findings illustrate both the potential and the limitations of evaluating proprietary wearable-derived measures in real-world digital health research. Further studies using independent physiological reference measures, larger and more diverse populations, and prospective validation designs are needed before clinical or occupational applications can be considered.