A cross-sectional analysis within a cohort study (34) was conducted in 2022. Data was collected through a pseudo-anonymous online survey where personal information (e.g., names) was collected but linked to the survey responses only via a unique identifier. The personal data and responses were stored separately, ensuring that participants’ identities could not be directly traced without the identifier. Participants were informed that they could withdraw from the study at any point without providing a reason.

Participants were graduates from the Universidad San Francisco Xavier de Chuquisaca (USFX) in Sucre, Bolivia. Participants were recruited in 2018, when they were in their final year of academic training in medicine or nursing. A total of 526 individuals from this previous study, who had agreed to be re-contacted, were invited to participate in this study. Of those who could be reached, 109 participants (23.3%) completed the questionnaire (Figure 1). Of the final sample, 27 participants (24.8%) were men, and 82 participants (75.2%) were women. The average age of participants in 2022 was 27.3 years (standard deviation (SD) = 2.9). All participants provided written informed consent before taking part. A reminder was sent after one, three, and four weeks of initial sending to ensure higher response. To incentivize participation, five Farmacorp vouchers worth 200 Bolivianos (approx. 29 USD) each were offered as prizes in a lottery draw.

Ethical approval for this study was obtained from the Ethical Committee of the Medical Faculty at the Universidad Mayor de San Simón on November 8, 2021, and from the Ethical Committee of the Medical Faculty at Ludwig Maximilian University in Munich on June 11, 2022 (project number 22–0451).

The questionnaire was provided in Spanish via SurveyMonkey (Momentive Europe UC, Dublin, Ireland) and comprised the following components: socio-demographic characteristics, mental distress, employment status, precarious working conditions and perceived social support. All questions and scales included in the questionnaire were sourced from established and validated measurement instruments to ensure reliability and validity of the data collected.

The socio-demographic characteristics included the respondents’ gender (male, female, or diverse), age (continuous), and economic situation (good, neither good nor bad, or bad).

The employment status of respondents was assessed through the question, “Are you currently employed?” Those who responded in the negative were classified as unemployed and were invited to indicate the reasons for their current status. The following response options were provided: “I am looking for a job because I was unable to find one,” or “I am currently engaged in studies or training.” Those who responded in the affirmative were classified as employed and subsequently completed the Employment Precariousness Scale (EPRES) (35). The EPRES consists of the subscales temporariness (in terms of contract duration), disempowerment (in terms of employment conditions such as working hours and pay), vulnerability (in terms of treatment at work), wages (in terms of cost coverage), rights (in terms of benefits such as pensions), and exercise rights (in terms of holidays and sick leave), which sum up to an overall score between 1 and 4. Following a validation study from Chile (35), a (very) high level of precariousness (LoP) was defined as an overall mean ≥ 2.0. Due to the low sample size in the category of none LoP, the moderate and none LoP were combined and defined as an overall mean < 2.0.

Mental distress, the outcome variable, was assessed using the validated 12-item General Health Questionnaire (GHQ-12). For mental distress as continuous variable each item was scored on a scale from 0–3, with positive items coded as 0-1-2-3 and negative items as 3-2-1-0, resulting in a total score ranging from 0 to 36. For the description of the participants, bivariate analysis and for sensitivity analysis, a dichotomous scale (also known as binary scale) of GHQ-12 was applied for mental distress as binary variable, with positive items coded as 0-0-1-1 and negative items as 1-1-0-0, resulting in a score range of 0–12. Since there is no validation study available for Bolivia, a threshold score was adopted from a Chilean validation study (36). Participants with a GHQ-12 score greater than 4 (with mental distress as binary variable) were classified as having clinically relevant mental distress (subsequently referred to as ‘mental distress’). Sensitivity analysis were conducted with a threshold of 5/6.

Perceived social support, the mediating factor, was measured using the validated Multidimensional Scale of Perceived Social Support (MSPSS). This scale assesses the social support an individual perceives from three different sources: family, friends, and a significant other. The MSPSS consists of 12 items ranging from 1 (“very strongly disagree”) to 7 (“very strongly agree”). A higher overall score indicates a higher level of perceived social support. An overall score was used for the mediation analysis, as the subscales (family, friends, and significant others) are highly correlated and measure similar aspects of social support. However, for transparency we also report the subscales.

Gender and economic situation were treated as potential confounders in the analysis, given their likely association with both employment status and mental distress, as supported by evidence (34). Age was not considered a confounder due to the limited age range of participants.

Calculations were done in R (Version 4.1.1) (37). Descriptive statistics were applied to nominal and ordinal variables using absolute and relative frequencies, while metric variables were described as means with standard deviations. Group differences were assessed using Chi-Square tests for categorical variables, Fisher’s Exact Tests for small sample sizes, and t-tests for continuous variables. The Pearson’s correlation was utilized to assess relationships between the main variables. Mediation analysis was performed using the package MEDIATION (38) to explore the mediating effect of perceived social support (mediator) on the relationship between employment status (independent variable) and mental distress (dependent variable), with gender and economic situation as potential confounders. Bootstrapping was used due to its suitability for smaller sample sizes, ensuring robust estimation of effects. Sensitivity analysis were conducted to ensure the robustness of the findings. Multiple imputation (n = 20) was performed using the package MICE (39).

Table 1 presents a description of the characteristics of the participants. A greater proportion of women and medicine graduates completed the questionnaire. The majority of participants described their economic situation as neither good nor bad. A smaller proportion indicated that their economic situation was good, while the fewest described it as bad. About half of the participants indicated that they were employed. Mental distress (as binary variable) was reported by more than two-thirds of the participants. When using a GHQ-12 score threshold of greater than 5 instead of 4, the analysis revealed a prevalence of 56.4% for mental distress among participants.

Participants with mental distress (68.3%) did not differ much from participants without mental distress regarding their gender, career, economic situation, employment status and age (Supplementary Table S1).

More than half of the graduates were employed. Among the unemployed participants, 31.3% were in training, while 68.7% were actively seeking employment. Among those employed, nearly half (44.9%) reported a (very) high level of precariousness at work. In particular, conditions related to disempowerment and exercise rights had a negative impact on the precariousness at work. A detailed table can be found in the supplementary material (Supplementary Table S2).

As shown in Table 2, participants generally perceived their social support as slightly positive, with average scores ranging between 4 (“neither agree nor disagree”) and 5 (“agree”). Support from family was rated the highest, followed by support from significant other. Support from friends was rated the lowest, although it remained within the positive range. A detailed table is provided in the supplementary material (Supplementary Table S3).

Table 3 displays the correlations among the main variables employment status, perceived social support and mental distress (continuous variable). The results indicated that employment was not correlated with perceived social support. There was a negative association between employment and mental distress, indicating that employment was statistically significant linked with lower level of mental distress. Perceived social support was negatively statistically significant associated with mental distress meaning higher scores in perceived social support led to lower scores in mental distress.

As shown in Figure 2, the largest impact of employment on mental distress occurred directly, with a statistically significant effect that resulted in lower scores in mental distress, indicating a positive influence on mental distress. While perceived social support had a statistically significant effect on mental distress (direct effect), it did not statistically significantly mediate the effect of employment on mental distress (indirect effect). Overall, employment was associated with a reduction in mental distress scores, including both direct and indirect effects (total effect). About 2.1% of the total effect of employment on mental distress is mediated through perceived social support.

The sensitivity analysis indicated effects in the same direction with consistent significance levels (Supplementary Figures S1–S3). As shown in Supplementary Figure S1, we performed a complete case analysis without imputation, which resulted in consistent results. In Supplementary Figure S2, we used mental distress as binary variable with a threshold of 4/5, instead of mental distress as continuous variable, which yielded similar results. Additionally, we differentiated between two groups within the unemployed: those undergoing training and those actively seeking work. The results for those actively seeking work were consistent with Table 3, while no effects were observed for those in training (Supplementary Figure S3).

The response rate in our study was comparably low, which means our results cannot be considered fully representative for the target population. This may partly be due to the unique challenges of conducting research during the COVID-19 pandemic and in LMICs like Bolivia. This limitation introduces the potential for selection bias, and the low statistical power may have contributed to the non-detection of relationships, such as the buffering effect, between the study variables. However, a non-responder analysis on this cohort conducted by John, L. et al. (34) found no statistically significant differences in gender distribution (women: 66.5% in the original sample, 75.9% in present sample) or age (original sample: 23.5 years, SD = 0.3; present sample in 2018: 24.0 years, SD = 0.1), among other key factors, suggesting that attrition did not systematically affect these factors.

Misclassification of exposure, mediator, and outcome may have occurred because these were assessed solely through online questionnaires. The GHQ-12, while a validated screening tool with a strong correlation to mental illness and predictive of future psychiatric diagnoses (53), is not a substitute for a clinical diagnosis (3). The MSPSS, also validated in Latin America (42), is limited to assessing emotional support and does not explicitly account for physical proximity as a source of support. This limitation restricted our ability to examine forms of support that rely on physical proximity, which is particularly relevant during a pandemic when physical distancing measures are in place (2). Additionally, we did not differentiate between formal and informal employment, which may have led to misclassification regarding the employment status. Also, we did not distinguish between frontline healthcare workers, which are workers providing direct care for COVID-19 patients, and non-frontline healthcare workers, although there is indication that the prevalence may be higher in frontline workers (9). Consequently, no dose–response relationship could be evaluated.

We acknowledge that not all of Baron and Kenny’s (54) criteria for mediation were fully met in our study. However, the use of bootstrapping offers a robust alternative, as it is particularly suited for detecting effects in smaller samples and under conditions where conventional assumptions may not hold (38).

A bidirectional relationship may arise, as individuals who are unemployed could already have poorer mental health and perceive less social support compared to those who are employed, which complicates causal inference (7). Moreover, since the study cohort consists of young people who typically perceive higher levels of social support (42), the results may not be generalizable to the broader population.

However, overall consistent results were observed in sensitivity analysis (Supplementary Figures S1–S3), reinforcing the reliability of our findings. Potential confounding by gender and economic situation were considered, although we missed other relevant factors like the duration of employment or unemployment and baseline mental health status. The economic situation in our study was assessed using a simple categorical scale (good, neither good nor bad, bad) rather than a continuous scale. While a continuous scale might have provided a more detailed picture of participants’ economic status, the categorical approach was chosen for its clarity and ease of interpretation. This categorization may limit the ability to fully capture the complexity of participants’ economic circumstances.

The high prevalence of clinically relevant mental distress in this population underscores the urgent need for targeted public health interventions. Addressing unemployment as a significant risk factor requires creating stable employment opportunities and improving working conditions in the healthcare sector. Expanding access to professional mental health care, such as integrated mental health services in primary care and telemedicine platforms, can improve accessibility, especially in resource-limited LMICs. Strengthening broader structural support mechanisms beyond social support is essential for improving mental well-being. Public health strategies should focus on providing comprehensive support, including unemployment benefits and access to mental health care, to mitigate the impact of unemployment on mental health.