This study used data from a 3-year longitudinal follow-up study of the Hong Kong Mental Morbidity Survey (HKMMS). The baseline HKMMS was a population-representative survey consisting of 5719 Chinese adults aged between 16 and 75 living in Hong Kong (21, 22). After 3 years, a follow-up study was conducted to investigate the long-term mental health outcomes. In this exploratory case-control study, invitations were randomly sent to a total of 45 participants (23 cases and 22 control subjects), out of which 33 of them (17 cases and 16 control subjects) agreed to participate in the study. One particular case was unable to be matched with control subjects due to the challenges encountered in finding a suitable match. A comparison between those who agreed to participate and those who declined was conducted, revealing no significant differences between the two groups in terms of age, gender, education level, marital status and employment status (p > 0.05). Participants with a score of 12 or higher on the Revised Clinical Interview Schedule (CIS-R) (23) were recruited as cases. Depression diagnosis was further ascertained using the Chinese-bilingual version of the Structured Clinical Interview for DSM-IV, Axis I, patient version (SCID) (24, 25). Control subjects were age-, gender- and education-matched to the cases, and had no lifetime history of mental disorders as defined by the SCID. Exclusion criteria included intellectual disability, dementia or significant cognitive impairment, lifetime history of psychotic disorders, major visual problem that affected perception of the surrounding environment, and inability to use a smartphone-based program. The study was carried out in accordance with the latest version of the Declaration of Helsinki. Ethical approval was obtained from the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster (Reference No. UW15-304). Written informed consent was obtained from all participants at the time of recruitment.

Potential participants were contacted by phone to explain the aim and objective of this study. After obtaining participants’ consent, face-to-face interviews were conducted at their homes for an initial assessment and detailed explanation of the study procedures. Participants were given a smartphone (Samsung Galaxy J5) with the ESM application installed. The device was programmed to emit random electronic signals eight times a day between 10:00 am and 10:00 pm for seven consecutive days. At least one signal was scheduled within each 90-minute span using a random number table. The timing of the signals was designed to balance the participant burden while ensuring the collection of valid data at random timepoints throughout the day. On receiving each electronic signal, participants were prompted to complete a self-reported questionnaire within a 15-minute window. Any data submitted beyond this time were discarded. The same set of questionnaires was used throughout the study period, with each taking approximately 2 minutes to complete. The ESM procedure was explained to participants during a briefing session, and a practice trial was administrated prior to the initial data collection to ensure that the participants understood the instructions. Previous studies showed that the application of ESM was feasible and valid among both the general population (26) and people with depression (27).

At each signal, participants were first instructed to take a photograph of their current location, which was used to verify the surrounding environment. They were then asked to respond to a list of 20 ESM items ( Table 1 ), most of which were rated on a 7-point Likert scale (1 = not at all; 7 = very much). Additional context-based questions regarding their location and social environment were answered by selecting from a list of options. Emotional states were assessed with four PA (items 2, 4, 6 and 9: happy, relaxed, satisfied and excited) and four NA (items 3, 5, 7 and 8: anxious, lonely, irritated and sad). PA and NA scores were derived from the average of their respective items. With reference to previous ESM studies on measuring activity- or event-related stress (5, 28, 29), we assessed ES by asking the extent to which participants currently experience discomfort with the environment. ES was considered any of the three self-reported items measuring arousal (item 10: I feel uncomfortable my current environment), frustration (item 11: I am frustrated with my current environment) and stress perceived from the surrounding environment (item 12: I would like to leave this place). ES score was calculated as the average of these three items. Interaction, personal control and social engagement in the environment were measured with items 13 to 18. A general statement of ‘my mood is affected by the surrounding environment right now’ was asked at the end of assessment. In addition to the ESM procedure, the CIS-R total scores were retrieved from the follow-up study in assessing psychological distress. The score was derived from 14 sections of non-psychotic symptoms, with higher scores indicating greater psychological distress (23, 30).

During the 7-day assessment, each participant generated eight time-point entries per day, resulting in a maximum of 56 observations. Assuming full compliant from 30 participants, the dataset would comprise a total of 1680 observations. Data processing and cleaning were performed according to the data management guide for experience-sampling studies (31). Before data analysis, the ESM data were checked to identify any problematic entries. Specifically, we classified entries that took less than 10 seconds to complete (less than 0.5 seconds per item) as non-compliant responses, indicating the participant tapped the answers without considering the options. Furthermore, trials with over 90% of identical entries were considered as invalid and excluded from the analysis (32).

The psychometric properties of the ESM assessment were first examined. The four PA items, four NA items and three ES items were assessed for internal consistency using Cronbach’s alpha coefficient. The convergent validity of the affect items was evaluated using correlational analysis of CIS-R total scores with PA and NA scores. The ESM data followed a hierarchical structure in which multiple momentary observations are nested within participants. Due to the highly correlated observations within each individual, multilinear regression modeling was employed to analyze ESM data variability at individual and group levels simultaneously (33). Prior research has proposed compliance rate cut-offs between 30% and 60% of prompts (28, 34–36). Given the limited exploration of momentary environmental stress, our study employed a higher compliance rate of 70% to effectively capture a comprehensive picture of participants’ daily experiences in this understudied area. Participants who completed less than 70% of the data points (fewer than 40 self-reports out of 56) were excluded from the analysis. Prior to analysis, the individual-level covariates (e.g., the ES) at different data points were centred around the individuals means (37). Multilevel linear regression analysis was first conducted with PA and NA as the dependent variables and ES as independent variables in the model. These models were then further adjusted for age, gender and the CIS-R total score. Additionally, in the separate ES-sensitivity models, ES, group status (0 = control, 1 = depressed) and their interaction term (ES × GROUP) were included as independent variables to examine whether the group status (healthy controls versus depressed participants) moderated emotional reactivity to the daily environment. The analysis of interaction and personal control with the environment between groups were conducted using same procedure. All statistical analyzes were carried out using SPSS Statistics version 20.0 (38).

Seventeen participants with depression and sixteen healthy matched control subjects were recruited. Two individuals from the depression group and one from the control group failed to complete at least 40 entries (compliance rate ranging from 11% to 39%) and were excluded from the analyzes. The final sample consisted of 15 depressed cases and 15 control subjects. Women comprised two thirds of the sample. The mean age of the sample was 49.4 years (SD 8.9 years; range 31 to 66 years), and 80% of the participants had a secondary education or above. There were no significant differences between two groups regarding marital status (p = 0.256) or employment status (p = 0.143). The median CIS-R scores for the control subjects and depressed cases were 0 (range 0 to 7) and 23 (range 12 to 33), respectively. The sociodemographic and clinical characteristics of the sample are depicted in Table 2 .

Table 3 presents the ESM characteristics of the sample. Of the 1680 signals sent, we received 1307 completed momentary assessments (675 from the control group and 635 from the depressed group). All of the participants completed at least 70% of the entries. The overall compliance rate was 77.8% (range 71.4% to 89.3%). The control group had a higher compliance rate (78.6%) than the depressed group (75%) (p = 0.008). Compared to the control group, the depressed group displayed a lower PA (p < 0.001), higher NA (p < 0.001) and increased ES (p < 0.001). The frequency distribution of the responses on PA, NA and ES are presented in Supplementary Figures 1.1 , 1.2 and 1.3 , respectively. A comprehensive view of ESM responses on PA for 15 cases and 15 control subjects is shown in Supplementary Figure 2 . Participants with depression reported spending more alone time (p < 0.001), had more time at home (p < 0.001) and were less engaged in social events (p < 0.001) than the control participants. They also indicated less impact of the environment on their mood (p < 0.001).

The Cronbach’s alpha for momentary assessments was 0.96 for PA, 0.80 for NA and 0.98 for ES, indicating good internal consistency among the items in the three subscales (39). The correlation coefficient between PA and NA was moderately high (-0.64), suggesting a significant negative correlation between the two opposite affect items. Furthermore, a correlation of -0.68 of PA and a correlation of 0.51 of NA with the CIS-R total score suggested moderate to good convergent validity of the affect items.

Multilevel linear regression analysis revealed significant associations between ES and PA (B = -1.21; 95% CI -1.64 to -0.78; p < 0.001) and between ES and NA (B = 0.45; 95% CI 0.26 to 0.63; p < 0.001) in the full sample of 30 participants. These associations persisted after adjustment for age, gender and the CIS-R total score ( Table 4 ). A significant interaction effect of ES × GROUP was found for both PA (B = 0.71; 95% CI 0.35 to 1.08; p < 0.001) and NA (B = -0.34; 95% CI -0.54 to -0.13; p = 0.003), indicating that the group modified the positive and negative emotional reactivity toward ES ( Table 5 ). When stratified by group, ES was associated with a decreased PA for the control group (B = -1.77; 95% CI -2.39 to -1.16; p < 0.001), while no association was observed for depressed group (B = -0.12; 95% CI -0.80 to 0.57; p = 0.723). On the other hand, ES was associated with an increased NA for the depressed group (B = 0.35; 95% CI 0.03 to 0.68; p = 0.031), but this was not seen in the control group (B = 0.08; 95% CI -0.18 to 0.34; p = 0.527) ( Supplementary Table 1 ). Moreover, participants with depression reported less interaction with the environment (B = -1.20; 95% CI -2.29 to -0.11; p = 0.032) and perceived less personal control over their environment compared to those in the control group (B = -1.37; 95% CI -2.52 to -0.22; p = 0.021) ( Supplementary Table 2 ).

Our study identified a significant relationship between ES and emotional reactivity. We noted that stress from the environment was associated with lower PA and higher NA, regardless of participants’ demographic characteristics or levels of psychological distress (42). However, emotional responses involving changes in various response systems such as feelings, perceptions, and behavior, may vary between healthy individuals and those with depression (43). We therefore examined whether participants’ health status (healthy versus depressed) altered emotional reactions to daily environment factors. We observed a notable increase in NA towards environmental stressors in the depressed group, while no such increase was found in control group. These results aligned with previous studies that individuals with depression tend to experience negative emotional reactions to daily stress across different contexts, such as events, activities and social interactions (28, 44, 45). For instance, Myin-Germeys et al. (28) examined emotional reactivity in different patient groups with severe mental illnesses and found that the group with major depressive disorder displayed greater stress-induced NA compared to healthy control subjects. In another ESM study of 279 twin pairs, Wichers et al. (46) suggested that the tendency to respond to daily life stressors with NA could be considered as a risk of depression. Moderately stressful events induced minimal effect to healthy people but greatly triggered NA in those with depression or a strong familial predisposition to depression. Our study built upon previous findings and provided further evidence that daily life stressors are not limited to specific events or activities, but also encompass the influence of physical environment.

Negative emotional reactions to daily ES are a potential marker for depression, as supported by other observational studies linking stress induced by adverse environment to an increased depression risk (47, 48). This mechanism could be explained by stress sensitivity, where individuals with higher sensitivity to stress are more prone to developing affective symptoms (49). People with history of depression exhibit greater sensitivity to daily stresses, and even modest stressors can trigger NA. Our findings suggest that daily hassles in environment could heighten stress sensitivity in those with depression, thereby intensifying their negative reactions. Despite facing more ES, the depressed participants perceived their mood to be less affected by the environment. These findings suggest that people with depression may encounter stressful experience with limited awareness of how their mood is influenced by the environment.

Notably, healthy control subjects in our study showed no changes in NA but exhibited a significant decrease in PA when exposed to environmental stressors. However, we could not observe this trend in depressed participants, which contradicts our hypothesis. Several factors could explain this discrepancy. First, the depressed group consistently had low PA levels throughout the study period, reflecting the pathological state of depression and made it difficult to detect emotional reactivity (28). Second, previous studies have shown that naturalistic acute stressors reduce hedonic capacity or the ability to experience pleasure in healthy subjects (50, 51). Perceived stress manifests as decreased PA rather than increased NA in non-clinical populations (52). Third, the healthy participants may have been more conscious of the surrounding environment than the depressed subjects. Many of them expressed concerns about the environment and its potential impact, which might have increased their sensitivity to emotional changes associated with their surroundings. Lastly, the effect of ES on affect may differ among individuals with different personality traits like agreeableness, neuroticism and extraversion (53, 54). Future studies should account for these factors while examining the affective response to ES.

Our findings indicate a difference in emotional response to ES between people with depression and those without. Depressed individuals tend to manifest stress through NA, while healthy individuals exhibit stress through PA. These results suggest that the psychopathological pathway of depression may be associated with stress in urban environments. Ecological factors contribute to depression and may be influenced by other psychosocial mediators such as perceived control and interactions with physical environment. Individuals with a lower sense of control over their environment are less receptive to positive experiences and more sensitive to daily environmental stressors, may make them more susceptible to depression. In addition, our findings suggest that underlying vulnerability modifies how individuals respond to stressors in different situations. Intervention strategies should consider the environmental contextual factors and the psychosocial mechanism connecting everyday stressful environments to symptomatology.

This study has several strengths worthy of note. First, the methodological design using ESM offered not only strong ecological validity, but also good reliability, improved accuracy and reduced retrospective bias. By collecting real-time data on daily affect, environmental stress and other relevant information for a week, we assessed moment-to-moment experiences in participants’ everyday environment. To ensure a comprehensive understanding of these experiences, we placed great emphasis on achieving a high compliance rate among participants, ensuring a robust and diverse set of information for analysis. The ESM technique proves particularly beneficial for people with depression, who are frequently reported to have a negative retrospective bias (55, 56). Second, we focused on the stressful experiences the participants perceived from their environment instead of assessing the quality of the environment. This approach allowed us to unambiguously capture the adverse psychological effects (i.e., emotional reactivity) caused by suboptimal environments (57). Third, we incorporated multiple information sources, which enhanced the accuracy of the ESM measures. For example, participants were asked to provide photographs for location verification and CIS-R scores were retrieved from interview for between-subject measures. Finally, the use of semi-structured SCID interviews ensured adequate diagnostic verification and confirmation of symptom severity to support the participants’ eligibility.

Several methodological limitations must be addressed. Despite employing an advanced experience sampling method, this study was limited by its modest sample size and potentially limited generalizability of the findings. Further investigation with a larger sample size is warranted. In addition, a practice effect was commonly reported in ESM studies, especially among patients with major depressive disorder (27, 58). Although we did not observe a decrease in the response latencies over the course of the study (practice effect not supported), the depressed participants had a lower compliance rate compared to the controls, which could potentially contribute to a reduced statistical power. Nevertheless, the compliance rate of 75% in the depressed group was comparable with, and somewhat higher than the rates reported in previous studies (59, 60). Besides, it is important to note that while negatively appraised environments are not necessarily the same as stressful environments, there is often an overlap in daily life. Participants who are predisposed to high NA might report elevated ES. Similarly, the more intense an environment is, it tends to be perceived as negative (61). The response bias should be acknowledged. Another limitation was the self-reported nature of the ESM, which predominantly assessed the subjective affective states. Future studies should consider incorporating objective physiological measures, such as heart rate or cortisol level, to gain a more comprehensive understanding of the stress response to the environment. Finally, the concept of environmental stress and its associated contextual factors like location or social situation, have yet to be fully explored. Future research should explore the impact of ES on emotions in different contexts.