Between 19 November 2025 and 14 January 2026, data was collected via an anonymous online questionnaire. Medical students (from their second study year onward) at two Austrian private medical universities were asked to participate. Students were invited via the Austrian National Union of Students mailing list (KL [Karl Landsteiner Privatuniversität für Gesundheitswissenschaften]: n = 733; PMU: 607). The survey was available in both English and German and participation was voluntary.

Of 257 students starting the survey, answers from 168 were included in the final dataset after excluding incomplete responses and students not meeting the inclusion criteria, yielding a response rate of 12.5% (168/1340). The inclusion criteria were as follows: medical students enrolled at KL or PMU, age 18+, completion of at least one academic year, provision of informed consent.

Well-being was assessed using the World Health Organization-Five Well-Being Index (WHO-5) [14], covering positive mood, calmness, and vitality (e.g., “I have felt cheerful and in good spirits”). The five items were rated on a six-point Likert scale (0–5), and raw scores (0–25) were multiplied by four to obtain a percentage score (0–100%).

Perceived stress was measured using a single-item visual analogue scale (VAS) from 0 (no stress) to 100 (maximum stress). Despite its limited depth, this measure has been widely used in epidemiological research [15, 16].

Using the Maslach Burnout Inventory—Student Survey (MBI-SS) [17, 18], burnout symptoms were assessed on the three subscales emotional exhaustion (EE), cynicism (CY), and academic efficacy (AE). Academic efficacy was reverse-coded (eight-mean score) to represent reduced AE (redAE), so that higher scores represent a stronger reduction in AE. Items were rated on a seven-point Likert scale (1 = never to 7 = always). Internal consistency was high: EE α = 0.87, CY α = 0.92, AE α = 0.79. Burnout prevalence was determined through norm-based cut-offs (EE ≥ 3.3, CY ≥ 2.5, redAE ≥ 1.9) [19], and burnout was defined by exceeding all three thresholds, following Thun-Hohenstein et al. [8].

Six study-related stressors were assessed on a five-point Likert scale, with 1 being “not stressful at all” and 5 being “very stressful.” Based on existing literature [20,21,22], the following stressors were included: academic workload, examination pressure, financial pressure, work–life balance, time management, and a perceived lack of mental health support.

The questionnaire contained demographic data for age, gender, year of study, semester, and university.

For the three MBI-SS subscales EE, CY, and redAE, means and standard deviations were calculated for the total sample as well as separately by gender (Table 1, 2). To contextualize subscale levels, mean scores were compared with Austrian reference samples reported by Unterholzer ([19]; Table 1) as well as with pooled PMU data from Thun-Hohenstein et al. ([8]; Table 2). Standardized differences (z-score) were calculated using the reference standard deviations to compare this sample with the normative data. The one-sample tests shown are meant as benchmarks, not as formal inferential tests. Burnout prevalence, based on norm-based cut-offs (EE ≥ 3.3, CY ≥ 2.5, redAE ≥ 1.9), was calculated for each subscale individually and for combined burnout (all three exceeding the cut-offs) as the primary definition (Table 3). Additionally, a less strict criterion requiring at least two of three subscales to exceed the cut-offs was calculated as a sensitivity analysis to allow comparison with studies using different thresholds. In order to test how the six stressors relate to burnout and well-being (WHO-5), a structural equation model (SEM) was used. Emotional exhaustion, CY, and redAE were set up as latent variables based on their MBI-SS items and well-being as a latent factor based on the WHO-5 items. All six stressors were included as predictors at the same time, with age and gender as covariates. Stressor correlations and covariances among the latent factors were freely estimated. The model was estimated using robust maximum likelihood (MLR). Guided by modification indices (MI > 10), nine residual covariances were added within the same MBI-SS subscale (EE: items 2–3; CY: items 1–2, 1–3, 2–4, 3–4; redAE: items 1 & 5, 3–4, 3 & 6, 4 & 6), reflecting shared method variance among similarly worded items. No cross-construct covariances were freed. Model fit was evaluated using chi-square, the Comparative Fit Index (CFI), the Tucker–Lewis Index (TLI), the Root Mean Square Error of Approximation (RMSEA), and the Standardized Root Mean Square Residual (SRMR). For readability, only statistically significant paths are shown in Fig. 1. Perceived stress was additionally examined in a multiple linear regression (Table 4). Regression was chosen over SEM because it was measured on a single item. While the regression analysis used only cases with complete data (N = 166), the SEM utilized all available data using FIML estimation (N = 168). Age and gender acted as covariates in the SEM, while age, semester, and gender were entered into the regression equation. Due to the small male subsample (n = 51), gender was modeled as a covariate rather than via multi-group SEM.

Structural equation model of stressors predicting EE, CY, redAE, and psychological well-being (WHO-5). Standardized path coefficients (β) are shown. Only statistically significant paths are displayed. Line thickness indicates statistical significance (thin line: p < 0.05; medium line: p < 0.01; thick line: p < 0.001). All stressors were allowed to covary; covariances among latent burnout dimensions were freely estimated but are not depicted for clarity. Age and gender were included as covariates. Model fit: χ2(295) = 493.98, p < 0.001; CFI = 0.92; TLI = 0.89; RMSEA = 0.055 (90% CI 0.046–0.063); SRMR = 0.062. Explained variance: EE R2 = 0.59; CY R2 = 0.36; redAE R2 = 0.37. Well-being R2 = 0.33