DigiHero study and data assessment

We analyzed data from the ongoing cohort study on digital health research in Germany (DigiHero, DRKS Registration-ID: DRKS00025600). DigiHero was initiated in January 2021 in Halle (Saale) and has since then expanded to a nationwide digital cohort study with currently more than 120,000 participants. Addresses for approximately one-third of households were randomly selected from resident registries across Germany. Residents aged 18 and older were invited by letter. In cities with populations over 100,000 inhabitants, the invitation sample was limited to 30,000 individuals. Regional response proportions ranged from two to eight percent. Additionally, all German residents aged 18 and older could self-register via the DigiHero website, resulting in a convenience sample based on voluntary participation. Participation is fully digital, using the LimeSurvey® platform, with four scheduled contacts annually.

On November 3, 2023, 81,498 DigiHero participants were invited via e-mail to take part in a survey examining employment, work conditions, and well-being (‘Work and Well-Being’ questionnaire, see supplement). Through December 15, 2023, 42,577 participants completed the questionnaire (52% response).

All participants provided written electronic informed consent by actively selecting designated checkboxes confirming informed consent and agreement to the privacy policy within the online survey. Ethical approval from the Ethics Committee of the Martin Luther University Halle-Wittenberg, Germany, was obtained for the DigiHero study and the ‘Work and Well-Being’ questionnaire (Reg. No. 2020−076). We followed STROBE reporting guidelines for observational studies (S1 File).

The ‘Work and Well-Being’ questionnaire captured professional activity (currently working; seeking employment; not working), psychological and occupational stress at the current time (t1), and a retrospective assessment of the Omicron wave (t0). Currently working participants answered additional questions regarding their primary occupation; multiple occupations were noted. Participants selected their current occupation from a drop-down menu based on the hierarchical structure of the five-digit coding scheme German Classification of Occupations (Klassifikation der Berufe (KldB), [24]). This survey was limited to the third hierarchical level, comprising 10 occupational areas (1-digit KldB), 37 occupational main groups (2-digit KldB), and 144 occupational groups (3-digit KldB). Furthermore, the 37 occupational main groups were aggregated in five occupational sectors and 14 occupational segments. Additionally, participants’ current job titles were obtained. Based on power calculations analogous to a study from the UK Biobank [25], 353 randomly selected job titles were subjected to independent re-coding by an expert to assess the reliability of participants’ self-coded occupations. Cohen’s Kappa statistic indicated comparable reliability to the UK study (1-digit KldB κ = 0.63, 2-digit KldB κ = 0.55, 3-digit KldB κ = 0.48).

Occupational SARS-CoV-2 infection risk was assigned based on the subjects’ occupational information. In general, workers having contact with SARS-CoV-2-infected individuals at work face a significantly elevated risk [26]. Therefore, workers in medical and non-medical healthcare occupations, including healthcare support, cleaning in healthcare facilities or retail sales of medical and health products, as well as those in theology and church community services, were classified as having a very high occupational risk. The assignment of non-healthcare workers into high or probable risk categories followed a previous study’s methodology [22], which was primarily based on whether the occupation involved frequent and likely contact with the public during professional activities (high risk) or probable contact (probable risk). The majority of participants in the high-risk group were employed in education, social work, law enforcement, correctional services, grocery retail, gastronomy, postal and delivery services, transportation, and cleaning services. Those at probable occupational infection risk included employees in public administration, financial services, sales, and occupational safety and health. All other occupations were not assigned an elevated risk of occupational SARS-CoV-2 infection (classified as “none”).

Additionally, status of employment (permanent employee with no predefined end date of employment; fixed-term employee; temporary/agency worker; civil servant; self-employed/freelancer) and job duration in years were recorded. If participants reported different occupations at t0 and t1, their t0 job titles were also recorded. Weekly working hours (full-time ≥35 hours; 30–35 hours; 20- < 30 hours; < 20 hours), and frequency of personal customer contact (daily; occasionally; rarely; none) were recorded for both time points.

Psychological and occupational stress were assessed with validated instruments. The brief four-item Patient Health Questionnaire (PHQ-4) [27], comprising the two-item PHQ-2 and the two-item Generalized Anxiety Disorder (GAD)-2 scales, was used to rate depressive and anxiety symptoms. PHQ-4 scores were categorized as normal (0–2), mild (3–5), moderate (6–8), and severe (9–12) symptoms. Work-privacy conflicts were rated analogously to the German version of the Copenhagen Psychosocial Questionnaire using the question: “To what extent do the demands of your work interfere with your private and family life?” [28]. The middle category “somewhat” served as reference against high (”to a very large or large extent”) and low (“to a small or very small extent”) work-privacy conflicts. Loneliness at work was assessed using the three-item UCLA Loneliness Scale [29] with a workplace-specific focus. For example, the first adapted item was: ”I lack companionship at work.” The three questions on loneliness utilized a 3-point Likert scale (often = 3, sometimes = 2, rarely or never = 1), allowing the total score to range between 3 and 9. A score of 6 or higher indicated moderate to severe loneliness at work [30].

Chronic work-related stress was assessed using the short version of the effort-reward imbalance (ERI) questionnaire comprising three effort and seven reward items [31]. A higher effort sum score indicates that more demands are perceived as stressful, while a higher reward sum score reflects a greater level of reward. Therefore, an effort-reward ratio close to zero suggests a favorable state (i.e., relatively low effort and relatively high reward), whereas values greater than 1 indicate a high level of effort that is not adequately compensated by the rewards received or expected. Intrinsic effort was assessed as overcommitment to work using the six-item overcommitment questionnaire [32]. The overcommitment score ranges from 6 to 24, with higher scores indicating more likely excessive engagement. As both instruments capture long-term parameters, these two instruments were inquired exclusively at t1, unlike the other measures, which were assessed at both t0 and t1.

Sociodemographic characteristics of the study population, including age, sex (male; female; other), and education (low: ≤ 10 years of schooling; medium: > 10 years of schooling; high: university degree) were obtained during the initial recruitment between January 21, 2021 and December 12, 2023. S2 File provides a summary of the study variables from the ‘Work and Well-Being’ questionnaire and the demographic characteristics used in the analyses.

Study sample

For the current analysis, eligibility was restricted to participants with complete PHQ-4 scores for t0 and t1, employment status data, and not enrolled as students. From the currently working individuals, participants were excluded due to missing or insufficient job titles (n = 478), employees on temporary leave (n = 9), career changers (n = 1,847), and those without information on their employment status at t0 (n = 978). As a result, data from 34,303 subjects were analyzed.

Statistical analysis

Continuous variables were characterized by median and interquartile range (IQR) or mean and standard deviation. Categorical variables were summarized by frequencies and percentages. The outcome, psychological distress, as assessed using the four-category PHQ-4 variable, was analyzed with ordinal regression (proportional odds models), and results reported as odds ratios (OR) with 95% confidence intervals (95% CI) for t0 and t1. Models with occupational sector, segment, and main group as risk factors present ORs for each category, with reference to all other occupations. For other factors, reference categories are specified. A directed acyclic graph (DAG, http://www.dagitty.net/) was used to determine minimally sufficient adjustment sets of covariates to estimate total effects (TE) of occupational exposure on depressive and anxiety symptoms (S3 File). Accordingly, controlling for sex, age, education, employment status, and weekly working hours adequately accounts for confounding to estimate TE. Further adjustments for work-related psychosocial stressors, including chronic work-related stress, loneliness at work, overcommitment to work, and work-privacy conflicts, were applied to reveal controlled direct effect (CDE) estimates. Analyses were stratified by sex, age group, and educational attainment. Sensitivity analyses were conducted by combining the very high and high occupational SARS-CoV-2 risk categories to address potential misclassification arising from changing risks at different pandemic time points. All models were analyzed using complete-case analysis. Statistical analyses were performed in R (Version 4.5.0) using the ‘ordinal’ package. Sankey graphs were prepared with SankeyMATIC.

Study population

Table 1 depicts the characteristics of the overall study population. Over 60% were employed at the time of the survey, whereas only 1.4% identified themselves as job seekers. Most non-working participants were of retirement age (median 68 years, IQR 64–73 years). A notably higher proportion of women than men participated in the survey (59.6% vs. 39.2%), along with 39 individuals identifying as ’other’. More than 60% of the participants held a university degree.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Sample description of all study participants according to depression and anxiety symptoms and results of univariable proportional odds models at t0 and t1 (N = 34.303).

https://doi.org/10.1371/journal.pone.0346871.t001

After the pandemic (t1), the median PHQ-4 score was 2 (IQR 1−4), with 4% of participants experiencing severe (PHQ-4 ≥ 9) and 13% substantial (PHQ-4 ≥ 6) symptoms of depression and anxiety. Symptom severity reported retrospectively for the Omicron wave (t0) were similar (Table 1). The proportion of participants with more severe symptoms was lowest among non-working subjects, followed by currently working participants. Job seekers reported the most severe symptoms (Fig 1). This is further demonstrated by the comparison with working participants, showing elevated odds at t1 (OR=2.23, 95% CI 1.89–2.64) and at t0 (OR=1.62, 95% CI 1.37–1.92). While symptoms improved among working and non-working participants post-pandemic, the proportion of distressed job seekers increased from 60.9% to 67.1%. Other risk factors for more severe symptoms included lower educational attainment, female sex, and identifying as ‘other’ sex (the latter based on a very small sample). In contrast, increasing age conferred a protective effect (Table 1 and S4 File).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Mental distress in early 2022 (t0, retrospectively assessed) and late 2023 (t1) measured with the 4-item Patient Health Questionnaire (PHQ-4) among workers (A), job seekers (B), and non-working participants (C).

https://doi.org/10.1371/journal.pone.0346871.g001

Among working participants (n = 20,948), the proportion of individuals with high work-privacy conflicts or feeling lonely at work decreased, whereas personal customer contact increased after the pandemic. Most working subjects were permanently employed (78.1%). Potentially more stressful employment states, such as fixed-term or temporary employment, or multiple jobs were reported by less than 5%. Overall, 3,479 workers (16.6%) were classified as having a very high occupational SARS-CoV-2 infection risk, 1,585 workers (7.6%) as having high risk, and 2,571 workers (12.3%) as having probable risk (Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Description of occupational characteristics and strain among 20,948 workers.

https://doi.org/10.1371/journal.pone.0346871.t002

Risk assessment of depressive and anxiety symptoms by occupational classification

Risk factors for more severe depressive and anxiety symptoms observed in the overall study population were also present among working participants. These included younger age, being female, and lower educational attainment (Table S5.1 in S5 File). Furthermore, employees with fixed-term or temporary contracts were at higher risk compared to permanent employees at both time points, as were part-time workers.

Occupations with a higher risk of more severe depressive and anxiety symptoms after the pandemic, compared to before, include those in traffic and logistics without vehicle driving (t1: OR=1.24, 95% CI 1.04–1.48; t0: OR=1.04, 95% CI 0.87–1.24, Table 3). At t0, an increased risk was observed in the personal services sector (OR=1.08, 95% CI 1.02–1.15), which was not elevated at t1 (OR=0.95, 95% CI 0.89–1.00). The effect was especially seen in medical and health care professionals (t0: OR=1.08, 95% CI 1.01–1.16; t1: OR=0.88, 95% CI 0.82–0.95) but attenuated after additional adjustment for occupational factors (CDE, Table S5.2 in S5 File). Occupations with consistently elevated odds, e.g., software development and programming, public administration, management assistants in transport and logistics) or reduced odds (e.g., professionals in human medicine and dentistry, teaching and training professionals), irrespective of time, were primarily observed in the CDE models or at occupational group level (Tables S5.2 and S5.3 in S5 File).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Modeling the risk of elevated depressive and anxiety symptoms (PHQ-4 categories) among 20,948 working participants by occupational sector, segment and main group.

https://doi.org/10.1371/journal.pone.0346871.t003

Separate models were calculated for each occupational sector, segment, or main group, with all others excluding the specific category serving as reference group.

Work-related psychosocial factors influencing distress

Post-pandemic (t1), occupational SARS-CoV-2 infection risk was not associated with elevated depressive and anxiety symptoms (TE, Table 4). At t0, individuals with high occupational SARS-CoV-2 infection risk exhibited increased odds of depressive and anxiety symptoms (e.g., very high risk OR=1.10, 95% CI 1.02–1.19), as the largest share of the very high occupational risk group consisted of medical health professionals (92%) whose infection risk declined sharply after the pandemic (Table 3). At t1, occupational customer contact emerged as a protective factor for depressive and anxiety symptoms, while at t0, daily customer contact was associated with more severe depressive and anxiety symptoms (TE, Table 4). However, these effects reversed after adjustment for additional occupational factors (CDE, Table S5.4 in S5 File). Models including only one additional factor at a time indicated that work-privacy conflicts and overcommitment primarily mediated the association between occupational infection risk, customer contact, and depressive and anxiety symptoms. Conversely, loneliness at work amplified risks at t0. At both time points, factors such as loneliness at work, chronic work-related stress, overcommitment to work, and high work-privacy conflicts were associated with more severe symptoms.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Odds ratios for elevated depressive and anxiety symptoms by occupational risk factors.

https://doi.org/10.1371/journal.pone.0346871.t004

Stratified analyses

Elevated odds for certain occupations at t0 and t1 were observed predominantly in women (Tables S5.5 to S5.10 in S5 File). Higher occupational SARS-CoV-2 infection risk and daily customer contact at t0 increased odds in the TE models exclusively among women and elderly. Stratification by age and education showed elevated odds in medical and non-medical healthcare occupations at t0 among older individuals and those without a university degree (OR=1.21, 95% CI 1.09–1.34; OR=1.14, 95% CI 1.02–1.27, respectively). Conversely, among those with a university degree, elevated risks were observed in service occupations within the social sector and cultural work at t0 (OR=1.12, 95% CI 1.03–1.22). However, these associations did not persist in the CDE models (Table S5.10 in S5 File). In contrast, the CDE models showed high odds for younger individuals and those with lower education in IT-sector and natural sciences service roles at both time points, and in goods production, particularly manufacturing, at t1. The odds related to other occupational factors (i.e., loneliness at work, chronic work-related stress, overcommitment to work, and work-privacy conflicts) showed similar patterns across sexes, age groups, and educational levels at both time points.

Strengths and limitations

The large convenience sample represents a strength, as it enabled a detailed examination of factors for psychological distress across specific occupational groups, rather than relying solely on generalized occupational categories. Furthermore, validated scales were used to assess symptoms and occupational stressors at two time points. However, the findings should be interpreted with caution due to limitations. First, the lack of representativeness inherent to convenience sampling, along with self-selection, limits the generalizability of the findings to the general population. Participants with better health and digital literacy appear to have participated more often in DigiHero. Compared to both, the 2022 German census [44] and the non-respondents of this survey, the study sample included a higher proportion of women (60% versus 51% and 53%, respectively) and more individuals holding a university degree (64% versus 20% and 37%). The study population was also selective in respect to occupational composition, with health and legal professions being overrepresented while occupations in cleaning or transport logistics were underrepresented. Consequently, findings primarily reflect the study population and should be interpreted cautiously, with limited generalizability to less healthy, less digitally literate, or underrepresented occupational groups. Furthermore, overall participation in DigiHero was low, ranging from 2% to 8% across regions. Second, data from t0 were collected retrospectively, which may introduce recall bias, a common limitation to retrospective studies. This could have influenced how participants recalled symptoms during the early 2022 Omicron wave, potentially leading to under- or over-reporting of PHQ-4 symptoms. Third, although it became evident that infection risks for workers in different occupations varied over time [6], the classification of occupations according to SARS-CoV-2 infection risk was based on considerations from the beginning of the pandemic [22,26]. This approach was chosen to provide a consistent reference point, facilitating comparability across studies. Furthermore, occupational differences in COVID-19 outcomes became less pronounced over time [6], Consequently, occupational infection risks appear to have been less differentiated in early 2022 and late 2023 compared with the early stages of the pandemic. To account for this convergence of occupational risks, sensitivity analyses were conducted in which the very high and high occupational SARS-CoV-2 risk categories were combined. These analyses similarly indicated a decline in the impact of occupational SARS-CoV-2 risk on severe symptoms from t0 (OR=1.09, 95% CI 1.02–1.16) to t1 (OR=0.93, 95% CI 0.87–0.99).