Burnout and Engagement Within Medical Education: A Repeated Measures Study on Their Evolution and Main Determinants

Study Design and Ethics Approval

This study is part of the WeMeds research project (www.wemeds.be) and received approval from the Ethics Committee Research UZ/KU Leuven in April 2021 (S64150). All participants provided signed informed consent prior to enrollment. Participation was voluntary, uncompensated, and conducted in accordance with the Declaration of Helsinki’s ethical principles for medical research involving human subjects. The present study focuses on the quantitative part of the Wemeds study, namely a prospective cohort design with three repeated measures collected via a self-administered online survey.

Population and Recruitment

Medical students and residents from all five medical faculties in Flanders (Belgium) – University of Leuven (KU Leuven), Free University of Brussels (VUB), University of Hasselt (UH), University of Ghent (UG), and University of Antwerp (UA) – were invited to participate. Eligible participants were Dutch-speaking, at least 18 years old, and enrolled as first-year bachelor students (cohort 1), first-year master students (cohort 2), or first-year residents (cohort 3) at T0. In Belgium, first-year bachelor students are equivalent to first-year undergraduates, while first-year master students are fourth-year medical students with more specialized knowledge. Residents were divided into two groups based on their professional setting: GP residents, working in GP settings, and other residents, primarily based in hospitals.

The survey was administered annually over three consecutive academic years: 2021–2022 (T0, baseline), 2022–2023 (T1), and 2023–2024 (T2). Recruitment utilized both direct and indirect online channels, including emails, the WeMeds website, event flyers, and various social media platforms. The recruitment period ran from October to February each year, with monthly reminders (i.e. four in total) sent to the universities and relevant organisations. Interested individuals could register online immediately upon invitation, enabling survey completion and automatic notification for subsequent measurement moments. An open cohort design was implemented, allowing registered participants to be invited annually for follow-up measurements, while also permitting the enrollment of new participants at each time point.

Data Collection

Statistical Analysis

All analyses and visualizations were conducted using R, version 4.3.0 [56]. Missing data resulted from participant drop-out and the open-cohort design. At the item level, when less than 5% of responses were missing, mean imputation was applied to retain usable cases, consistent with recommended practice for small proportions of missing data [14, 57]. If more than 5% of an item was missing, the item was omitted. Linear mixed-effects models were then estimated using maximum likelihood estimation, which appropriately accommodates remaining missing outcomes data [14].

Data were analyzed as two subsets: a full dataset (participants with ≥1 time point) and a complete dataset (participants with all 3 time points). The full dataset maximized statistical power, while the complete dataset was used for robustness checks. To test whether data were Missing Completely At Random (MCAR), t-tests compared participants who responded at 2 time points with those who responded only once. Descriptive statistics were calculated for both datasets, including means, standard deviations, and Cronbach’s alpha for reliability. Normality was assessed using histograms, Q-Q plots, skewness, and kurtosis [58]. Pearson’s correlations (p < .05; r ≥ .30 considered meaningful) were used to examine relationships between variables [59].

LMMs were used to model changes in emotional exhaustion, cynicism, professional efficacy, and engagement within learning phases (Hypothesis 1a, b), accounting for data clustering. Separate LMMs were estimated for each outcome and for each cohort (bachelor students, master students, and residents). Fixed effects included cohort, measurement year, gender, and university; random effects included participant ID and measurement year. Differences between GP and other residents were assessed using t-tests with Benjamini-Hochberg correction. Additional LMMs tested Hypotheses 2a,b, assessing associations between demands, resources, personality traits, and outcomes, including multiple fixed effects (measurement year, cohort, gender, university, demands, resources, and personal traits) and random effects (participant ID and measurement year). Independent variables were centered and scaled for interpretability. Cross-lagged panel analyses examined whether variables at T0 predicted outcomes at T1, and T1 predicted T2 outcomes. These analyses were conducted on the full dataset to ensure power. Statistical significance was set at p < .05 with Benjamini-Hochberg correction for multiple comparisons.

Sociodemographic Information of Participants and Descriptives

In total, 790 participants completed 1257 responses in the full dataset, resulting in an overall response rate of 14%. The mean age of the participants was 22.46 (SD = 3.96), 72.15% were female (N = 917), and 96.87% (N = 1.227) had no children. Of all participants, 465 participants completed one measurement only, 183 participated twice, and 142 three times. T-tests showed no significant differences between participants with one versus two measurements, suggesting random attrition (Supplementary Table S.1). Furthermore, the highest item-level missingness was 2.35%, which was below the commonly accepted threshold of 5%.

Although the response rate was relatively low, the composition of our sample closely reflects that of the target population. Female participants constituted the majority of our sample, which aligns with their high representation within the medical student and resident population. For residents, official numbers from the Annual Statistics on Health Care Practitioners 2024 of the Federal Public Service Health, Food Chain Safety and Environment, indicate that women comprise 68% of GP residents and 60% of other residents [60]. For medical students, data provided by the five participating faculties indicated that, on average, 67% are female.

In addition, Supplementary Table S.2 shows the participant characteristics of the full dataset. In this dataset, cohort 3 included 151 GP residents and 174 other residents. Similar participant characteristics of the complete dataset can be found in Supplementary Table S.3, while Supplementary Table S.4 details GP residents and other residents in full and complete datasets. Supplementary Table S.5 provides the outcome variables per cohort in the full dataset, including means, standard deviations, score ranges, skewness, kurtosis and Cronbach’s alpha. The same details for the complete dataset are in Supplementary Table S.6. Additionally, correlation tables, presenting Pearson correlation plots for all study variables per cohort, are shown in Supplementary Table S.7.

Evolution of Burnout Complaints and Engagement

LMMs were used to examine changes within the three cohorts for emotional exhaustion, cynicism, professional efficacy, and engagement (Hypothesis 1a), and to compare GP residents with other residents (Hypothesis 1b). The reference groups were first-year bachelor students, first-year master students and first-year residents. For emotional exhaustion, results indicate a significant increase in bachelor year 3 (β = 0.267, SE = 0.105, p < .05), resident year 2 (β = 0.345, SE = 0.116, p < .01) and resident year 3 (β = 0.468, SE = 0.135, p < .01) (Supplementary Table S.8). Figure 2 displays the predicted means. The t-test results with Benjamini-Hochberg correction show differences between GP and other residents, though none are statistically significant (Supplementary Table S.9, Supplementary Figure S.1).

Figure 2

For cynicism, results indicate a significant increase in bachelor year 3 (β = 0.394, SE = 0.108, p < .001), master year 2 (β = 0.408, SE = 0.094, p < .001) and master year 3 (β = 0.409, SE = 0.133, p < .01) (Supplementary Table S.10). Figure 3 displays the predicted means pattern. T-test results with Benjamini-Hochberg correction showed a statistically significant difference between GP residents year 2 and other residents in year 2 (β = 0.528, SE = 0.199, p < .05) (Supplementary Table S.11, Supplementary Figure S.2).

Figure 3

For professional efficacy, results indicate a significant decrease in bachelor year 2 (β = –0.195, SE = 0.072, p < .01), bachelor year 3 (β = –0.391, SE = 0.076, p < .001), master year 2 (β = –0.388, SE = 0.072, p < .001) and master year 3 (β = –0.343, SE = 0.089, p < .001) (Supplementary Table S.12). Figure 4 displays the predicted means pattern. The t-tests with Benjamini-Hochberg correction show a significant difference between GP residents year 3 and other residents year 3 (β = –0.360, p < .05) (Supplementary Table S.13, Supplementary Figure S.3).

Figure 4

For engagement, results indicate a significant decrease in bachelor year 3 (β = -0.276, SE = 0.077, p < .01), resident year 2 (β = –0.340, SE = 0.093, p < .001) and resident year 3 (β = –0.423, SE = 0.106, p < .001) (Supplementary Table S.14). Predicted means are illustrated in Figure 5. The t-tests with Benjamini-Hochberg correction showed no significant differences between GP residents and other residents (Supplementary Table S.15, Supplementary Figure S.4).

Figure 5

Main demands, resources and personality traits

Additional LMMs were used to examine which demands (Hypothesis 2a) and resources (Hypothesis 2b) were most strongly associated with emotional exhaustion, cynicism, professional efficacy and engagement within each cohort. For emotional exhaustion, workload, work-home conflict and meaningfulness were significant determinants for all three cohorts, and for cohort 3 meaningfulness was a fourth significant determinant (Supplementary Table S.16). These results are visualized in Figure 6, which presents bee swarm plots with corresponding confidence intervals. Cross-lagged analyses (Supplementary Table S.17) showed that for cohort 1 and cohort 3, neuroticism and work-home conflict remained significant, while for cohort 2, only neuroticism remained significant.

Figure 6

For cynicism, among cohorts 1 and 2, workload, work-home conflict, meaningfulness, learning opportunities and neuroticism were identified as significant determinants; in cohort 2, the learning environment also remained a significant determinant. For cohort 3, work-home conflict and meaningfulness remained the only significant determinants (Supplementary Table S.18). Figure 7 displays bee swarm plots illustrating these associations along with their corresponding confidence intervals. Cross-lagged analyses (Supplementary Table S.19) indicated that perfectionism was a significant predictor in cohort 1, whereas meaningfulness and neuroticism were significant in cohort 2; in cohort 3, only work–home conflict remained a significant predictor over time.

Figure 7

For professional efficacy, among cohorts 1 and 2, workload, meaningfulness, learning opportunities, neuroticism and perfectionistic strivings were significant determinants. In cohort 2, work-home conflict, learning environment and perfectionistic concerns additionally emerged as significant, whereas in cohort 3, only meaningfulness remained significant (Supplementary Table S.20). An unexpected positive association with work-home conflict prompted the inclusion of an interaction term between workload and work–home conflict, given their high intercorrelation; this interaction was also significant (β = 0.150, SE = 0.067, p < .05). Figure 8 presents bee swarm plots depicting these associations with confidence intervals. Cross-lagged analyses (Supplementary Table S.21) identified no significant predictors for cohorts 1 and 3. In cohort 2, however, workload, learning opportunities and neuroticism were positively associated with changes in professional efficacy.

Figure 8

For engagement, in cohorts 1 and 2, workload, meaningfulness, learning opportunities, neuroticism, and perfectionistic strivings were significant determinants (Supplementary Table S.22). In cohort 3, work-home conflict replaced workload as a significant demand. Figure 9 displays these results in bee swarm plots. Cross-lagged analyses (Supplementary Table S.23) indicated that perfectionistic strivings significantly predicted engagement in cohort 1, learning opportunities in cohort 2, and no significant predictors were identified in cohort 3.

Figure 9

Modest Increases in Burnout Complaints and Decreases in Engagement

Hypothesis 1a was supported, with modest but significant increases in burnout complaints and decreases in engagement observed within medical education learning phases. These findings align with former studies indicating a progressive rise in burnout complaints accompanied by a declining trend in engagement during medical education [13, 14, 17, 18, 19, 61, 62]. Interestingly, bachelor students showed deterioration mainly in the third year, reflected by increases in emotional exhaustion and cynicism, suggesting either an accumulation of chronic stress in the first two years or a substantial escalation of stress during the third year. Among master students, emotional exhaustion remained stable, whereas cynicism increased and professional efficacy decreased; while among residents, emotional exhaustion rose again and engagement declined. These phase-specific patterns underscore the nuanced and complex trajectories of burnout complaints and engagement in medical education, demonstrating that burnout complaints do not necessarily uniformly decrease or increase over time [63].

GP Residents Versus Other Residents

Furthermore, Hypothesis 1b was supported, with GP residents reporting significantly lower levels of cynicism and higher professional efficacy compared to other residents. These differences are in line with meta-analytic evidence indicating that family medicine is among the specialties with lower burnout prevalence rates, in contrast to hospital specialties such as radiology and surgery [20]. A potential explanation may lie in differences in work context: GP residents often manage relatively stable patients with chronic conditions in community settings, enabling sustained patient relationships that may protect against burnout [64]. In contrast, hospital-based residents more frequently care for acutely ill and unstable patients with complex complications and face greater exposure to litigation risk, factors that may contribute to higher burnout levels [65]. Additionally, GP training programs in Belgium incorporate designated supervisors, regular mentoring sessions, and allocate one day per week for study, factors that may contribute to enhanced well-being among GP residents [66, 67].

Workload and Work-Home Conflict as Key Demands

Hypothesis 2a was partially supported. As hypothesized, workload was identified as a key demand during preclinical phases, where it was significantly associated with emotional exhaustion, cynicism, and engagement. In the clinical phase, workload remained relevant for emotional exhaustion. However, work–home conflict was a more influential demand for emotional exhaustion and cynicism within all learning phases compared to workload, exceeding our expectation that it would be primarily relevant during residency years. In addition, and contrary to our expectation, cognitive demands were not significant in either the preclinical or residency phases, with workload and work-home conflict proving more salient. Cross-lagged analyses further confirmed the predictive role of work–home conflict for subsequent increases in emotional exhaustion and cynicism, suggesting a potentially causal influence, which is consistent with prior research [68, 69, 70, 71, 72, 73, 74].

In order to understand the importance of work-home conflict within all learning phases, it may be important to situate it within broader societal transformations, including evolving gender roles, generational trends, and the erosion of work-life boundaries, amplified in part by technological advancements [75, 76, 77, 78, 79]. As both women and men now increasingly engage in both earning and caregiving roles, the traditional separation between gender roles becomes less clear [77, 80]. Further, the generational composition of our samples, largely Generation Z, should also be taken into consideration. While Gen Z is often characterized as prioritizing a healthy work-life balance, working hours flexibility and well-being, studies have shown that life-stage variables, such as parenting status, and societal changes, have a more substantial impact on work-home dynamics than generational effects alone [78, 79]. Lastly, digital technologies, despite enabling flexibility, also stimulate constant connectivity, increasing stress and complicating efforts to maintain healthy work–life boundaries [76, 81].

Meaningfulness and Learning Opportunities a Key Resources

Hypothesis 2b received partial support. As anticipated, meaningfulness was the most important resource for residents, whereas for students, meaningfulness and learning opportunities were similarly influential. Specifically, among students, meaningfulness and learning opportunities were strongly associated with cynicism, professional efficacy, and engagement, but not with emotional exhaustion. For residents, meaningfulness was the primary resource linked to all outcomes, with learning opportunities only related to engagement. These findings echo recent evidence demonstrating that meaningful work buffers against burnout and promotes engagement [42, 82, 83, 84, 85]. For example, Chênevert et al. (2021) reported that physicians value meaningful work and patient recognition more than general organizational support [85]. They found that time spent on meaningful tasks constitutes a strong protective factor against burnout and also reduces dropout intentions [86, 87].

Contrary to expectations, the learning environment was only significantly associated with cynicism and professional efficacy in master students and not with emotional exhaustion or engagement. This finding seems to diverge from existing evidence [36, 43, 44, 45, 88, 89]. However, this ambiguous pattern may also reflect the complexity of the learning environment in a medical educational context, which comprises multiple distinct components, including academic learning and learning in a clinical context. Each component is characterized by specific factors, such as the educational climate, supervisory presence and style, and alignment of tasks with competency development [36]. These contextual elements were not accounted for in the present analysis. The learning environment is further characterized by less clearly defined teams and fluid relationships among students, residents, and supervisors [90]. Third year master students and residents, for instance, frequently rotate between departments and engage in short-term collaborations with multiple supervisors, which may complicate perceptions about the learning environment. Additionally, uncertainty regarding the appropriate timing or circumstances for approaching supervisors, coupled with concerns about potential negative repercussions related to evaluations may further attenuate the perceived impact of environmental support [90].

Perfectionism and Neuroticism

Consistent with former studies, neuroticism was associated with higher burnout complaints and lower engagement, a pattern largely confirmed by the cross-lagged analyses, except for engagement. This aligns with prior research presenting neuroticism as an important determinant for poor well-being [24, 25, 38, 61, 88]. For instance, Bianchi et al. (2018) reported neuroticism to be more influential than work-related or social factors. Perfectionism showed a more nuanced pattern, being more related to professional efficacy and engagement, particularly through perfectionistic strivings, reflecting a motivation to achieve high standards [91, 92]. However, due to insufficient statistical power within the cohorts, we could not examine potential mediating or moderating effects of these personality traits. Therefore, consistent with Robins et al. (2018), caution is necessary when interpreting the direct influence of personality traits, as their mediation analyses suggest a more complex relationship in which personality also shapes demands and resources that subsequently affect burnout and engagement [93].

Implications for Practice

In a medical education context, strategies to address these key demands and resources, namely work–home conflict, workload, meaningfulness, and learning opportunities, should be tailored to the distinct realities of students and residents. For medical students, in an academic learning environment, interventions may focus on flexible curricular scheduling, protected study time, manageable deadlines, and access to supportive infrastructure such as green spaces and sports facilities, alongside educational designs that emphasize meaningful learning, professional identity formation, and participatory teaching approaches [66, 94, 95, 96]. For residents in clinical settings, structural measures such as flexible and self-scheduling, limits on duty hours supported by adequate staffing, task reallocation through skill-mix teams, and access to sabbaticals can help mitigate work–home conflict and excessive workload, while reducing administrative burden and stimulating interprofessional collaboration and shared decision-making may strengthen meaningfulness in clinical work [66, 97, 98]. Together, these context-specific strategies acknowledge the differing demands of academic and clinical training environments while addressing core determinants of burnout and engagement across medical education.

Strengths and Limitations

One of the key strengths of this study is the repeated measures design, which enables the examination of the evolution of burnout complaints and engagement within learning phases of medical education, namely bachelor years, master years, and resident years. This approach allows for a more comprehensive understanding of how burnout complaints and engagement evolve over time, capturing evolutions within learning phases and fluctuations that cross-sectional studies might miss. Another strength is the study’s focus on a broad range of contributing factors, including demands, resources, and personality traits, which provides a nuanced view of which factors are most important. Additionally, the inclusion of multiple cohorts from various medical faculties strengthens the generalizability of the findings, offering insights that can be applied within diverse educational settings. Nevertheless, we should also note several limitations of our research. First, the response rate was low at 14%, despite our continuous recruitment efforts. This may have limited generalizability and suggest that the findings primarily reflect the views of a self-selected sample rather than the entire population. Participants suggested that high workload, time constraints, the sensitive nature of the topic, and survey fatigue may have contributed to limited participation, introducing the potential for participation bias. Sensitivity analyses comparing participants who completed a single survey with those completing multiple surveys using t-tests revealed no significant differences, indicating that attrition did not substantially bias the findings. In addition, our sample did not significantly differ from the population with regard to gender composition. Nevertheless, future studies could mitigate this limitation by integrating survey assessments into the medical curriculum, which would both normalize participation and reduce the risk of selection bias by reaching the full population. Further, the incorporated validated questionnaires (i.e., MBI, COPSOQ) relied on self-report which is prone to social desirability bias, as participants might under- or overreport their burnout complaints and engagement levels. Nonetheless, the use of an anonymous online survey might mitigate this bias to some extent. In addition, it is important to note that our analyses do not diagnose clinical burnout, therefore, our discussions revolve around burnout complaints rather than burnout as defined by clinical criteria. Next, while the design provides valuable insights, the three-year study period may not fully capture the long-term effects of burnout. Continuation of the survey would be recommended for long-term results. Finally, cohort bias should be considered, as differences observed between bachelor, master, and residency cohorts may not solely reflect educational influences but could also be attributed to cohort-specific characteristics.