Stress exposure on compassion fatigue in internship nursing students: mediating roles of empathy, psychological capital, and perceived stress

Stress exposure triggers physiological and psychological reactions that may impair well-being [9], particularly under prolonged or intense conditions. For nursing interns, repeated encounters with patient suffering [10, 11], combined with occupational stressors such as heavy workloads, insufficient mentor support, and long working hours, are associated with a significantly increased risk of compassion fatigue [12]. These factors often interact dynamically, as highlighted by Jiang’s Psychological Stress System, cumulatively amplifying stress burden and mental health vulnerability [13].

According to the 2023 China Health Statistics Yearbook, nearly 48% of registered nurses in China hold an associate degree, with approximately 40% practice in rural areas [14]. This highlights the critical role of diploma-level nursing students within primary healthcare, alongside heightened concerns regarding their psychological readiness. Unlike baccalaureate students who undergo four years of structured theoretical training and participate in diverse clinical placements, diploma students typically enter clinical practice after only two years of academically condensed, practice-oriented education. This accelerated pathway may limit their development of coping strategies and resilience, potentially increasing vulnerability to compassion fatigue [15]. Nevertheless, current studies often fail to distinguish between educational backgrounds when examining compassion fatigue among nursing interns [15]. It is therefore essential to investigate how internship stressors are associated with compassion fatigue in this specific subgroup, particularly through potential mechanisms such as empathy, psychological capital, and perceived stress.

The ABC-X model, originally proposed by Reuben Hill in 1949, explains stress outcomes (Factor X) as a function of stressors (Factor A), coping resources (Factor B), and cognitive appraisal (Factor C) [50]. This study applies the model to examine how internship stress contributes to compassion fatigue (Factor X). Work-related stress and negative life events (Factor A) are posited to be associated with compassion fatigue both directly and indirectly by depleting or engaging personal resources (Factor B) and shaping cognitive appraisals (Factor C) [15, 51]. In this framework, perceived stress is positioned as Factor C, reflecting the subjective appraisal of stressors as threatening or overwhelming [52, 53]. Psychological capital is conceptualized as a coping resource (Factor B), representing a state-like positive psychological resource that can be drawn upon to manage adversity [54, 55]. Empathy represents a more complex theoretical case. While often considered a stable trait, we theorize it, for this specific context, as a capacity that can function as a resource (Factor B) by influencing how individuals perceive and engage with emotional demands in clinical settings; its classification as a “resource” is therefore proposed as a testable assumption within the model, contingent upon its regulated, adaptive mobilization. We explicitly acknowledge the theoretical tension herein, as empathy’s dual nature means cognitive aspects may aid in regulated engagement (a potential resource), while unregulated affective aspects may increase vulnerability and drain resources [56, 57]. The conceptual framework based on the ABC-X model is presented in Fig. 1.

Based on the ABC-X model and empirical evidence, we hypothesize that stress exposure is associated with compassion fatigue through both direct and indirect theoretical pathways. Specifically, perceived stress, psychological capital, and empathy are examined as potential mediator in this relationship (see Fig. 2). Figure 2 visualizes these hypothesized relationships within the ABC-X framework: Factor A represents stress exposure (including negative life events and work-related stress), Factor B denotes coping resources (cognitive empathy, affective empathy, and psychological capital), Factor C reflects perceived stress as a form of cognitive appraisal, and Factor X refers to the outcome variable, compassion fatigue. The differentiated functions of cognitive and affective empathy, together with the potential mediating roles of perceived stress and psychological capital, form the theoretical foundation for the mediation hypotheses proposed in this study. Accordingly, the following hypotheses are proposed:

H1: Stress exposure (including work-related stress events and negative life events) is positively associated with compassion fatigue.

H2: Perceived stress mediates the relationship between stress exposure and compassion fatigue.

H3: Empathy (cognitive and affective) mediates the relationship between stress exposure and compassion fatigue.

H4: Psychological capital mediates the relationship between stress exposure and compassion fatigue.

H5: Perceived stress, empathy, and psychological capital sequentially or collectively mediate the relationship between stress exposure and compassion fatigue.

Participants were recruited via convenience sampling from 8 junior colleges in Hunan province, China. These colleges, comprising one private and seven public colleges, were selected to capture the primary institutional type (public vs. private) and broad geographical spread across the province. While not an exhaustive enumeration of all junior colleges in Hunan, this sampling strategy was designed to enhance sample diversity and improve the representativeness of internship nursing students within the regional context. The formal survey was conducted January 7 to 13, 2023. Inclusion criteria were: (a) enrollment in a three-year associate degree nursing program, (b) active participation in clinical internships at secondary or tertiary hospitals for at least eight months, and (c) willingness to participate and provide informed consent. Exclusion criteria included students not involved in direct patient care during clinical practicums. Additionally, to ensure valid response rate, a small gift was provided for each participant upon survey completion.

We employed structural equation modeling (SEM) with maximum likelihood estimation to analyze the associations between variables. The N: q rule (10:1) was applied to determine the minimum required sample size, where N represents the required number of cases and q denotes the number of parameters to be estimated [58]. By using this rule, we calculated a minimum sample size of 220 for 22 estimated parameters. Accounting for a 20% rate of invalid questionnaires, the sample size was adjusted to 264. To ensure generalizability of the results and minimize sampling bias, 640 nursing interns from eight public junior colleges in Hunan Province were surveyed.

Questionnaires were distributed via WeChat and DingTalk, two widely used social media platforms in China. Participants accessed the online survey by scanning a Quick Response code. Before proceeding, they were presented with an electronic information sheet explaining the study’s purpose, procedures, data confidentiality, and voluntary nature. Informed consent was obtained electronically—only those who clicked the “Agree to Participate Voluntarily” button were permitted to access the questionnaire. To ensure anonymity, the survey platform (Wenjuanxing) was configured not to collect any personally identifiable information (e.g., names, contact details, or IP-linked data). All data were securely stored and accessible only to the research team. To maximize the response rate, we (1) required responses to all questions; (2) restricted submissions to one per account, device, and IP address; and (3) offered a financial incentive. Questionnaires exhibiting “straight-lining” (identical answers across questions), “gibberish” (logically inconsistent answers), or excessive speed (completion time under four minutes) were considered invalid and excluded from analysis. Due to the anonymous and aggregated nature of data collection via the online platform, response rates for individual colleges were not tracked; however, of 671 surveys initiated, 640 met the criteria, yielding an effective response rate of 95.4%.

Data were analyzed using SPSS 26.0 (IBM, Armonk, NY, USA) and AMOS 23.0 (SPSS, Inc., Chicago, IL, United States). Descriptive statistics were used to summarize the general characteristics of the subjects and the average scores for each research variable. We applied the Kolmogorov-Smirnov test to assess the distribution of total scores across the CFSS-13, PPQ-22, and PSS-14 scales, except for the BES-20, which presented dimension scores. The results indicated that the data did not meet the normality assumption. However, we opted for parametric tests (independent t-tests and one-way ANOVA) to compare compassion fatigue differences among participants in different categories for two main reasons. First, normality tests such as the Kolmogorov–Smirnov are known to be overly sensitive in large samples (n = 640), often flagging trivial deviations as statistically significant (76). As noted by Ghasemi and Zahediasl (2012) [75], parametric tests remain robust under moderate violations of normality when the sample size is sufficiently large, especially in the range of several hundred observations. Second, the central limit theorem posits that the sampling distribution of the mean approximates normality as sample size increases, regardless of the underlying data distribution [76]. Taken together, these considerations justify the use of parametric tests in our analysis. Concurrently, Pearson’s correlation analyses were employed to examine the relationships between compassion fatigue, burnout, secondary traumatic stress, psychological capital, cognitive empathy, affective empathy, perceived stress, and the number of negative life events and work-related stress events. To assess the robustness of the findings, non-parametric sensitivity analyses (i.e., Mann–Whitney U tests, and Kruskal–Wallis H tests, Spearman’s rank correlations) were conducted for group comparisons and correlation analyses.

We evaluated the overall model using established fit index criteria. According to Schreiber’s statistical guidelines [77, 78], a model is considered well-fitted if the Chi-square (χ2) test does not reach statistical significance (P > 0.05), the χ2/degrees of freedom (df) ratio is between 1 and 3, the root-mean-square error of approximation (RMSEA) is below 0.08, and indices such as the goodness of fit index (GFI), adjusted GFI (AGFI), comparative fit index (CFI), incremental fit index (IFI), and normed fit index (NFI) exceed 0.900. A two-tailed P-value of 0.05 was set as the significance threshold. To determine effect size of standardized direct effects, cutoffs from Cohen (1988) were used: 0.1 = small, 0.3 = medium, and 0.5 = large [79].

Additionally, we conducted tests for indirect effects using a bootstrap approach, calculating 2,000 resamples with replacement to generate 95% bias-corrected confidence intervals. This method was employed to explore the potential mediating roles of psychological capital, cognitive empathy, affective empathy, and perceived stress in the relationship between compassion fatigue and stress exposure during the internship. The significance of the mediating effect is indicated by a confidence interval that excludes zero. To determine effect size of standardized indirect effects, cutoffs from Kenny (2011) were used: 0.01 = small, 0.09 = medium, and 0.25 = large [80].

Participants (N = 640) were predominantly female, aged between 20 and 21 years, and most were nursing majors interning at tertiary hospitals. Further demographic details are summarized in Table 1. Notably, compassion fatigue scores varied substantially by key factors: students averaging < 7 h of sleep per night reported a mean score of (57.65 ± 23.37), compared to (45.31 ± 23.36) for those with ≥ 7 h. Similarly, major preference showed a clear gradient, from (35.03 ± 22.87) to (83.18 ± 19.68). Non-parametric sensitivity analyses (Mann-Whitney U and Kruskal-Wallis H test) confirmed significant group differences in major preference (H = 83.782, p < 0.001), career intention (H = 37.295, p < 0.001), sleep duration (Z = -6.690, p < 0.001), and self-rated health (H = 77.854, p < 0.001), but not in other listed demographics (all p > 0.05), aligning with parametric analyses (see Supplementary Table S1).

Survey data revealed that participants commonly experienced multiple negative life events (Mean = 2.3) and work-related stress events (Mean = 4.0). The most frequent specific stressors were concerns about academic or career pressures and mistakes or accidents at work were, as detailed in Table 2. Mean scores for compassion fatigue, psychological capital, perceived stress, affective empathy and cognitive empathy were (50.2 ± 24.1), (73.1 ± 11.0), (42.4 ± 3.5), (35.1 ± 2.8) and (31.4 ± 2.7), respectively. Table 3 revealed the expected significant intercorrelations among all variables (all p < 0.05). Crucially, both types of stress exposure showed positive correlations with compassion fatigue, perceived stress, and affective empathy, but a negative correlation with psychological capital. Spearman correlation results were consistent (Supplementary Table 2).

Based on the correlation matrix, an initial model with all hypothesized paths were tested. The model fit was suboptimal, and four direct paths were found to be statistically non-significant: from stress exposures to cognitive empathy, from cognitive empathy to compassion fatigue, from affective empathy to compassion fatigue, and from affective empathy to psychological capital. These paths were removed in a post-hoc, primarily data-driven modification to enhance model parsimony, a decision subsequently reviewed for theoretical plausibility. The revised model demonstrated a strong fit (χ2 = 26.531, χ2/df = 2.211, AGFI = 0.970, IFI = 0.988, TLI = 0.973, CFI = 0.988, NFI = 0.979, RMSEA = 0.044). The final path model is depicted in Fig. 3.

It is noteworthy that while the bivariate correlation between affective empathy and compassion fatigue was significant (Table 3), this relationship was not a significant direct or indirect path in the final multivariate model, suggesting its shared variance with compassion fatigue is explained through other model variables like perceived stress. As shown in Table 4; Fig. 3, work-related stress events were associated with nursing interns’ psychological capital (β = 0.121, P = 0.001; between small and medium effect size), affective empathy (β = 0.113, P = 0.007; effect size between small and medium), perceived stress (β = 0.143, P < 0.001; between small and medium effect size), and compassion fatigue (β = 0.093, P = 0.002; very small effect size). Similarly, negative life events were also associated with psychological capital (β = -0.125, P < 0.001; between small and medium effect size), affective empathy (β = 0.102, P = 0.016; effect size between small and medium), perceived stress (β = 0.225, P < 0.001; between small and medium effect size), and compassion fatigue (β = 0.098, P < 0.001; very small effect size). Moreover, psychological capital was found to have a significant negative direct association with perceived stress (β = -0.342, P < 0.001; between medium and large effect size) and compassion fatigue (β = -0.315, P < 0.001; between medium and large effect size). Additionally, the pathways from negative life events to work-related stress events (β = 0.380, P < 0.001; between medium and large effect size), from affective empathy to cognitive empathy (β = -0.100, P = 0.011; small effect size), from cognitive empathy to psychological capital (β = 0.444, P < 0.001; between medium and large effect size), and from perceived stress to compassion fatigue (β = 0.248, P < 0.001; between small and medium effect size) were all found to be statistically significant.

Further analysis utilizing the bias-corrected bootstrap method has revealed indirect association from work-related stress events and negative life events to compassion fatigue. Specifically, work-related stress events are indirectly related to compassion fatigue through the independent mediating effects of psychological capital (β = 0.038, P = 0.005; between small and medium effect size) and perceived stress (β = 0.035, P = 0.001; between small and medium effect size), as well as through their combined mediation (β = 0.010, P = 0.003; small effect size). Similarly, negative life events are also indirectly related to compassion fatigue through the independent mediating roles of psychological capital (β = 0.039, P = 0.004; between small and medium effect size) and perceived stress (β = 0.056, P = 0.001; between small and medium effect size), as well as through their combined mediation (β = 0.011, P = 0.002; small effect size). Notably, our study found that when psychological capital, cognitive empathy, and affective empathy are considered together as mediating variables, they do not exhibit any mediating effects in the pathway from work-related stress events (β = 0.002, P = 0.070) and negative life events (β = 0.001, P = 0.076) to compassion fatigue. The detailed results of these indirect pathways are presented in Table 5.

The average compassion fatigue score in this study was 50.2, which is slightly higher than the 44.99 reported among 2,256 nursing interns in a previous study [81], yet lower than the 60.61 observed among nursing and midwifery students in tertiary grade A hospitals [82]. These differences may be partly attributable to the timing of data collection. Unlike the comparison studies conducted during the COVID-19 pandemic, our study was carried out in the post-pandemic period [81, 82]. Existing studies have shown that compassion fatigue levels among nurses fluctuated throughout different stages of the pandemic [21, 83‐86]. Nursing interns working during the peak of pandemic may have experienced increased infection anxiety and emotional distress due to witnessing patient isolation or death, potentially exacerbating compassion fatigue [82]. Although sociocultural norms can influence how individuals process and express emotions [87], since all cited studies were conducted within similar Chinese contexts, the differences are more likely due to institutional or situational factors rather than cultural differences. Regardless of these variations, the overall level of compassion fatigue remains moderate. Given that compassion fatigue is significantly associated with increased presenteeism and early departure from the profession, continuous monitoring and effective interventions are essential.

The SEM analysis revealed that work-related stress events and negative life events are significantly interrelated and negatively correlated with psychological capital, while positively correlated with affective empathy, perceived stress, and compassion fatigue. According to Conservation of Resources theory [88], nursing interns facing combined work and life stressors expend considerable time, energy, and emotional support, initiating a cycle of resource depletion. This loss of resources reduces psychological capital, weakens coping abilities, and increases perceived stress, ultimately leading to compassion fatigue. Concurrently, sustained stress enhances affective empathy, allowing interns to deeply empathize with patients’ suffering. However, excessive empathy may result in an over-responsibility worry [89], intensifying emotional burden and further depleting psychological resources. These findings highlight the importance of helping nursing students identify potential stressors and implement effective stress management strategies, such as strengthening psychological capital, to prevent and mitigate compassion fatigue.

Using a structural equation model, we examined how stress exposures indirectly associated with compassion fatigue through psychological capital and perceived stress. The results supported these theoretical pathways, indicating that interns with lower psychological capital and/or higher perceived stress are more susceptible to compassion fatigue. These findings align with the Stress and Coping Model, which suggests that individuals implement various coping strategies that focus on stress perception and psychological outcomes such as compassion fatigue [90, 91]. Additionally, Sun et al. found that positive psychological capital enhances nursing students’ ability to manage stress and maintain mental health [70]. Although the mediation effects were statistically significant but small, they still hold practical relevance for educational and preventive strategies. The small effect size suggests that no single pathway in isolation is dominant, which is consistent with the multifactorial nature of compassion fatigue. Practically, this indicates that interventions targeting a single mediator (e.g., only boosting psychological capital) may have limited impact; instead, multifaceted programs that simultaneously address resource building (e.g., psychological capital) and cognitive appraisal (e.g., perceived stress) are likely to be more effective, as small effects can accumulate. For example, identifying at-risk interns, such as those with low psychological capital and high perceived stress, can enable targeted interventions like resilience workshops or mindfulness programs. Small but significant mediated pathways also suggest that multifaceted interventions addressing both psychological resources and stress appraisal may be more effective than single-component approaches.

Contrary to our initial exploratory hypothesis, emotional and cognitive empathy did not mediate the relationship between stress exposure and compassion fatigue in the final model. This non-significant finding suggests that, within our sample and model, empathy did not function as a mediator in the association from stress exposure to compassion fatigue. It does not invalidate theoretical frameworks linking empathy to compassion fatigue, such as Figley’s model of empathic stress and fatigue which posits empathy as a contributor to the “cost of caring” [4]. Instead, it suggests that the role of empathy in this specific stress-outcome pathway for nursing interns may be more complex than a simple mediating one [92]. The non-significant paths may be attributed to several factors Nursing interns’ relatively limited clinical exposure and ongoing development in emotional regulation might be associated with the depth and consequences of their empathetic engagement [93]. Moreover, while the Basic Empathy Scale has demonstrated robust psychometric properties in college student populations and has been applied in studies involving nursing students and interns [66, 94], it is designed to measure general empathy tendency and may not be sufficiently sensitive to capture nuanced empathy-related changes specifically tied to compassion fatigue in transitional contexts such as clinical internship. Therefore, the non-significant mediation effect may reflect not only sample-specific developmental characteristics but also potential limitations in empathy measurement specificity within this population. Consequently, this finding suggests empathy more as an exploratory factor in our model, and an alternative theoretical explanation is that empathy may function not as a mediator but as a moderator in this stress-outcome pathway. For instance, the level of empathy might be associated with the strength of the relationship between stress exposure and compassion fatigue, a possibility that should be explicitly tested in future research. Nevertheless, the observed weak negative correlation between affective and cognitive empathy (r = -0.100) suggests that an imbalance between these components could be associated with future challenges in emotional regulation [95]. Therefore, training programs that help interns balance affective and cognitive empathy may still hold value in fostering long-term emotional health and reducing compassion fatigue.

Despite the meaningful contributions of our study, it is not without limitations. Firstly, the cross-sectional design fundamentally limits the ability to infer causality, even when theoretical relationships are proposed. The mediation pathways identified should be interpreted as testing theoretical pathways of association, not as establishing causal effects. Future research employing a longitudinal design is warranted to elucidate the mechanisms through which adverse life events and occupational events drive the onset and progress of compassion fatigue. Secondly, we assessed stress exposure during the internship by documenting the occurrence of common work-related and negative life events. While this approach helped identify salient stressors and offered initial evidence regarding their cumulative effects, it was limited by its reliance on dichotomous (yes/no) event counts without accounting for variations in frequency or perceived intensity. Consequently, this method may not fully reflect the actual burden of each stressor and could be potentially related to an underestimation of the strength of associations between stress exposure and outcome variables, such as compassion fatigue. Future research would benefit from incorporating validated weighted stress inventories to better quantify the severity and recurrence of stress exposures. Thirdly, the sample consisted exclusively diploma-level nursing interns from a single province in Hunan, recruited via convenience sampling. This approach may have introduced selection bias and considerably limited the generalizability of the findings. Given potential differences in clinical preparedness, psychological resources, and education systems across regions and academic levels, we strongly caution against extrapolating these results to bachelor’s degree nursing students or other cultural and educational settings. Further research should employ stratified or multiregional sampling strategies encompassing diverse educational tiers to enhance external validity. Fourthly, all constructs were assessed via self-report questionnaires collected in a single online survey, which may introduce common method bias. This could inflate the observed correlations among variables, such as those between perceived stress, empathy, psychological capital, and compassion fatigue, potentially leading to overestimated mediation effects. Although procedural remedies were applied (e.g., psychological separation of scales), the cross-sectional self-report design remains as inherent limitation. Additionally, the absence of qualitative data restricts interpretive depth regarding how nursing interns subjectively experience and emotionally process stress. Future studies should incorporate mixed-methods approaches, such as qualitative interviews or external assessments (e.g., supervisor evaluation), to triangulate findings and enhance contextual validity. Fifthly, our study primarily focused on the association of individual resources with compassion fatigue. However, in the ABC-X model, factor B also includes contextual resources such as institutional and societal support. For example, structured mentorship or pre-rotation guidance may buffer stress responses and facilitate emotional adjustment during clinical internships. These factors were not assessed in this study, limiting a more comprehensive understanding of buffering mechanisms. Future research could incorporate such institutional variables—such as mentorship continuity or supervisor accessibility—to explore their potential moderating roles. Finally, the model modification process, while improving statistical fit and parsimony, involved the post-hoc removal of non-significant paths based on statistical criteria. Although theoretically reviewed, this data-driven approach remains a limitation, as it capitalizes on the specific characteristics of our sample.

This study, employing the ABC-X model as the conceptual framework, enhances our understanding of the interactions between stress exposure during internships, empathy (affective/cognitive), psychological capital, perceived stress, and compassion fatigue among internship nursing students. From a practical perspective, the key to preventing and mitigating compassion fatigue lies in accurately identifying and managing stress exposures during this period, reducing levels of perceived stress, and strengthening psychological capital. Specifically, stress management education and optimization of clinical assignments may help interns reframe their cognitive and emotional responses to stressful events, thereby alleviating their subjective stress experience [96]. In parallel, resilience-building content could be embedded into nursing humanities or clinical skills modules — for example, through brief mindfulness-based stress reduction sessions and problem-solving skills training [97, 98]. These curriculum-integrated strategies may help nursing interns regulate stress, enhance coping skills, and build psychological resilience.

Future research should address several key avenues informed by our findings and study limitations. First, longitudinal or prospective cohort designs are recommended to establish temporal precedence and clarify the causal pathways between stress exposure, mediating resources, and compassion fatigue development over time. Second, employing more nuanced stress exposure assessments, such as validated weighted inventories that capture event frequency, perceived severity, and cumulative burden, would provide a more precise understanding of their associations. Third, expanding recruitment to include bachelor’s degree nursing students across multiple provinces or diverse cultural contexts, using stratified random sampling, would enhance the generalizability and comparative potential of findings. Fourth, integrating mixed methods—such as combining longitudinal surveys with qualitative interviews—could illuminate the subjective experiences and emotional processing of interns, offering deeper contextual insights. Fifth, future models should incorporate institutional or contextual resources (e.g., mentorship quality, supervisor support) as potential moderators, aligning with the ABC-X model’s emphasis on external resources, to identify systemic protective factors. Finally, exploring alternative roles for empathy—such as testing its function as a moderator rather than a mediator—using compassion fatigue-specific empathy measures may help clarify its complex relationship with stress and outcomes.