Compassion fatigue is a common issue nursing students face during clinical internships. Prolonged exposure to patients' suffering and trauma can lead to emotional exhaustion and psychological stress. Compared to formal healthcare workers, nursing students have less professional experience and weaker emotional regulation abilities, making them more vulnerable to secondary trauma and other negative emotions, which exacerbates compassion fatigue. Early identification and intervention in compassion fatigue are crucial for improving the mental health of nursing students and the quality of care they provide.
Objective
This study aims to develop a predictive model for compassion fatigue in nursing students using various statistical and machine learning methods, identify key influencing factors, and provide scientific evidence for nursing educators and administrators.
Methods
A cross-sectional survey collected valid questionnaire data from 512 nursing students. LASSO regression was used to select critical variables, and models such as logistic regression, random forest, and XGBoost were applied for prediction. Model performance was evaluated, and SHAP values were used to interpret the importance of model features.
Results
The logistic regression model performed best on the test set with an AUC value 0.77. Key predictive factors included psychological resilience, peer support, secondary trauma, and empathy satisfaction.
Conclusion
This study successfully developed a predictive model for compassion fatigue in nursing students, with the logistic regression model showing high accuracy. The critical factors identified provide theoretical support for early interventions, aiding in more targeted nursing management and enhancing the mental well-being of nursing students.
Trial registration
Not applicable. This study is an observational study aimed at investigating compassion fatigue among students, without involving any interventions or treatment methods. Therefore, this study does not meet the definition of a clinical trial and does not require registration of a clinical trial number.
Hinweise
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Introduction
Compassion fatigue refers to the phenomenon where healthcare workers, after prolonged exposure to patients' suffering, stress, and trauma, gradually lose their emotional empathy and experience emotional exhaustion [1]. In recent years, with the rapid development of the healthcare industry, nursing students, as the future workforce of the nursing profession, face significant work pressure and emotional burdens during their internships, making them susceptible to compassion fatigue [2]. Compared to formal healthcare professionals, nursing students have less professional experience and weaker emotional regulation skills, which makes them more vulnerable to negative emotions such as secondary trauma, thereby exacerbating the occurrence of compassion fatigue [3].
Research has shown that compassion fatigue not only negatively affects the physical and mental health of nursing students but also undermines their professional identity, reduces the quality of care, and may even lead to burnout and turnover intentions [4]. Therefore, early identification and intervention of compassion fatigue have become critical topics in nursing education and management. Developing effective predictive models can help nursing administrators and educators identify high-risk nursing students early and implement appropriate psychological support and interventions, thereby reducing compassion fatigue and enhancing the quality of nursing education and nursing students' job satisfaction.
Anzeige
Although some studies have explored factors influencing compassion fatigue, such as work stress, emotional support, and secondary trauma, there is still a lack of systematic and accurate predictive models [5]. This study aims to develop and validate a prediction model for compassion fatigue in nursing students using various statistical and machine-learning methods. The study seeks to identify critical predictive factors to better understand which variables significantly contribute to compassion fatigue. Through this model-based analysis, not only can the accuracy of predictions be improved, but it can also provide nursing administrators and educators with scientific evidence to formulate more targeted intervention strategies.
The main objectives of this study are: (1) to identify critical factors influencing compassion fatigue in nursing students through LASSO regression; (2) to apply and compare various machine learning models, such as logistic regression, random forest, and XGBoost (Extreme Gradient Boosting), for prediction; and (3) to validate and determine the optimal prediction model for compassion fatigue. This study will provide valuable insights for future nursing education and management, further promoting the importance of attention and interventions regarding the mental health of nursing students.
Materials and methods
This study aims to develop and validate a predictive model for compassion fatigue in nursing students by identifying critical factors through various statistical analyses and machine learning methods.
Study design
This cross-sectional survey study collected data from nursing interns in hospitals through questionnaires. Based on this data, a predictive model for compassion fatigue was constructed. The study adhered to ethical guidelines, and all participants voluntarily participated with informed consent.
Anzeige
Study participants
The study participants were nursing students in the internship stage at a medical school. The inclusion criterion was nursing students undergoing clinical internships. The exclusion criteria were: (1) nursing students with severe physical or psychological illnesses, (2) those unwilling to participate in the study. All participants signed informed consent before participating in the study. A total of 512 valid questionnaires were collected.
Data collection
This study collected data through self-administered questionnaires covering demographic variables (such as age, gender, and family background), work-related variables (such as internship hospital level, weekly working days, and the number of patients cared for per week), and psychological variables(such as perceived social support, psychological resilience, compassion fatigue, and role overload).All of these scales [6‐9] have high reliability and validity, contributing to the quality of data collection.
The psychological variables included perceived social support, measured using the Perceived Social Support Scale (PSSS), with 12 items and a 7-point Likert scale (Cronbach's alpha = 0.967) [10]; psychological resilience, assessed using the simplified 10-item Connor-Davidson Resilience Scale (CD-RISC-10), with a 5-point Likert scale (Cronbach's alpha = 0.957) [6]; compassion fatigue, measured using the Professional Quality of Life Scale (ProQOL), evaluating empathy satisfaction, burnout, and secondary traumatic stress across 30 items with a 5-point Likert scale (Cronbach's alpha = 0.767) [3]; and role overload, measured using the Role Overload Scale developed by Peterson et al., with five items on a 5-point Likert scale (Cronbach's alpha = 0.930) [11]. All of these scales demonstrated high reliability, ensuring the validity and reliability of the data collected.
Definition of independent and dependent variables
Dependent Variable: The dependent variable in this study is the level of compassion fatigue in nursing students. Based on the ProQOL scale scores, nursing students were categorized into two groups: Students with compassion fatigue and students without compassion fatigue.Independent Variables: The study analyzed 23 independent variables, which include: hospital level, age, gender, grade, religious belief, only-child status, number of night shifts, physical exercise habits, dietary habits, sleep habits, physical condition, number of patients cared for per week, number of critically ill patients cared for per week, number of new admissions per week, actual working days per week, whether coping techniques for compassion fatigue have been learned, teaching effectiveness at the hospital, social support score, empathy satisfaction, job burnout, secondary trauma, role overload, psychological resilience, etc. LASSO regression was used to select the most predictive variables.
Statistical methods
This study first applied LASSO (Least Absolute Shrinkage and Selection Operator) regression analysis to identify the key variables influencing compassion fatigue. LASSO regression can select and shrink variables, effectively addressing multicollinearity and preventing model overfitting. The optimal penalty coefficient (lambda) was determined through tenfold cross-validation, and non-zero coefficient variables were selected accordingly. Based on the variables selected by LASSO regression, multivariate logistic regression analysis was conducted to assess the independent impact of each variable on compassion fatigue. The results were expressed as odds ratios (OR) and 95% confidence intervals (CI), with statistical significance set at p < 0.05. Multiple machine learning models such as logistic regression, random forest, XGBoost, and LightGBM were constructed to validate the effectiveness of the predictive models. The dataset was divided into training and testing sets in a 7:3 ratio, and ten repetitions of sampling were conducted to reduce model error. Model evaluation metrics included AUC (Area Under the Curve), PR curve (Precision-Recall curve), DCA (Decision Curve Analysis), and calibration curves to assess model performance and prediction accuracy. To further interpret the features within the models, SHAP (Shapley Additive exPlanations) values were used to analyze each variable's contribution to the prediction outcomes. SHAP value plots were generated to illustrate the importance of different variables and their positive or negative impacts on compassion fatigue. The data analysis and model construction utilized the following open-source libraries in Python: sci-kit-learn for statistical analyses such as LASSO regression and logistic regression; XGBoost and LightGBM for building tree-based models; matplotlib and seaborn for data visualization and result presentation; and SHAP for interpreting feature importance in machine learning models.
Ethical statement
This study was approved by the Ethics Review Committee of Yancheng First People's Hospital (Approval Number: 2024-K-093). Although the research was not conducted at Yancheng First People's Hospital, the Ethics Review Committee of the hospital provided ethical approval for the study to ensure it met ethical standards. This study strictly adhered to the ethical requirements of the Declaration of Helsinki and followed applicable national guidelines. All participants signed informed consent forms, and the research process strictly adhered to ethical guidelines to ensure the privacy and data security of the participants.
Results
Baseline data comparison
A total of 512 valid questionnaires from nursing students were collected for this study. The dataset was randomly divided into a training set and a test set in a 7:3 ratio, with 246 and 245 participants, respectively. Baseline comparisons showed no significant differences between the training and test sets regarding demographic and work-related variables (p > 0.05), indicating a good balance after the data split. Specifically, variables such as hospital level, grade, age, gender, time entering clinical practice, religious beliefs, only child status, number of night shifts, physical condition, number of patients cared for per week, and number of critically ill patients showed no statistical differences between the two groups. Similarly, psychological variables such as social support score, empathy satisfaction, job burnout, secondary trauma, role overload, and psychological resilience also showed no significant differences between the two sets (p > 0.05). These results demonstrate that the training and test sets are well-balanced regarding data distribution. See Table 1 for details.
Table 1
Baseline characteristics of the training and test cohorts
Variable
Training Set n = 246 (%)
Testing Set n = 245 (%)
Z/t
P
Hospital Level
Tertiary Hospital
245.0 (96.08)
245.0 (100.00)
0.01
1
Secondary Hospital
10.0 (3.92)
0.0 (0)
Grade
Sophomore Year
1.0 (0.39)
0.0 (0)
0.88
0.01
Junior Year
32.0 (12.55)
30.0 (12.24)
Senior Year
109.0 (42.75)
113.0 (46.12)
One Year After Graduation
113.0 (44.31)
102.0 (41.63)
Age
18–20
28.0 (10.98)
30.0 (12.24)
0.75
0.01
20–23
163.0 (63.92)
150.0 (61.22)
24 and Above
64.0 (25.10)
65.0 (26.53)
Gender
Male
71.0 (27.84)
68.0 (27.76)
0.23
0.14
Female
184.0 (72.16)
177.0 (72.24)
Time Entering Clinical
1–3 Months
48.0 (18.82)
45.0 (18.37)
1.45
0.78
4–6 Months
14.0 (5.49)
16.0 (6.53)
7–9 Months
113.0 (44.31)
120.0 (49.02)
10–12 Months
64.0 (25.10)
58.0 (23.67)
12–24 Months
7.0 (2.75)
5.0 (2.04)
> 24 Months
9.0 (3.53)
7.0 (2.86)
Religious Belief
Religious Belief
9.0 (3.53)
7.0 (2.86)
0.01
1
No Religious Belief
246.0 (96.47)
238.0 (97.14)
Only Child
Only Child
56.0 (21.96)
53.0 (21.63)
0.14
0.25
Non-Only Child
199.0 (78.04)
192.0 (78.37)
Number of Night Shifts
Once a Month
61.0 (23.92)
58.0 (23.67)
1.12
0.06
2–3 Times a Month
82.0 (32.16)
80.0 (32.65)
More Than 4 Times a Month
112.0 (43.92)
107.0 (43.67)
Physical Exercise
Never
43.0 (16.86)
45.0 (18.37)
0.33
0.09
Rarely
165.0 (64.71)
157.0 (64.08)
Frequently
47.0 (18.43)
43.0 (17.55)
Dietary Habits
Very Regular
25.0 (9.80)
24.0 (9.80)
0.05
0.97
Fairly Regular
142.0 (55.69)
135.0 (55.10)
Irregular
88.0 (34.51)
86.0 (35.10)
Sleep Habits
Very Regular
19.0 (7.45)
22.0 (8.98)
0.34
0.43
Fairly Regular
110.0 (43.14)
106.0 (43.27)
Irregular
126.0 (49.41)
117.0 (47.76)
Physical Condition
Poor
34.0 (13.33)
31.0 (12.65)
0.17
0.46
Average
179.0 (70.20)
172.0 (70.20)
Good
42.0 (16.47)
42.0 (17.14)
Weekly Patients Cared For
Less Than ten persons
26.0 (10.20)
25.0 (10.20)
0.12
0.88
10–20 persons
77.0 (30.20)
80.0 (32.65)
More Than 20 persons
152.0 (59.61)
140.0 (57.14)
Weekly Number of Critically Ill Patients
Less Than 3 People
124.0 (48.63)
120.0 (49.02)
0.33
0.68
3–5 People
72.0 (28.24)
65.0 (26.53)
More Than 5 People
59.0 (23.14)
60.0 (24.49)
Weekly Number of New Admissions
Less Than 5 People
33.0 (12.94)
30.0 (12.24)
0.19
0.80
5–10 People
88.0 (34.51)
90.0 (36.73)
More Than 10 People
134.0 (52.55)
125.0 (51.02)
Actual Working Days Per Week
1–2 Days
6.0 (2.35)
7.0 (2.86)
0.05
0.96
3–4 Days
60.0 (23.53)
58.0 (23.67)
More Than 5 Days
189.0 (74.12)
180.0 (73.47)
Learned Coping Methods for Compassion Fatigue
Yes
50.0 (19.61)
48.0 (19.59)
0.11
0.86
No
159.0 (62.35)
157.0 (64.08)
Uncertain
46.0 (18.04)
40.0 (16.33)
Effective Teaching at the Hospital
Completely Possible
82.0 (32.16)
80.0 (32.65)
0.06
0.86
Partially Possible
170.0 (66.67)
163.0 (66.53)
Completely Impossible
3.0 (1.18)
2.0 (0.82)
Social Support Score
58.17 ± 12.58
58.65 ± 12.47
−0.43
0.66
Empathy Satisfaction
32.99 ± 6.02
33.00 ± 6.51
−0.01
0.99
Job Burnout
30.17 ± 5.24
30.41 ± 5.58
−0.50
0.61
Secondary Trauma
29.33 ± 6.46
29.02 ± 6.93
0.52
0.61
Role Overload
18.00 ± 4.08
18.33 ± 4.36
−0.87
0.39
Psychological Resilience
41.17 ± 9.06
40.91 ± 8.87
0.25
0.81
Selection of key factors for compassion fatigue in nursing students
Key independent variables influencing compassion fatigue were identified through LASSO regression analysis. LASSO regression compresses variable coefficients to prevent overfitting and addresses issues of multicollinearity. The results showed that the 24 independent variables were reduced to 16 (with λ minimum mean squared error = 0.0417), including hospital level, age, gender, psychological resilience, peer support, role overload, empathy satisfaction, months in the job, family factors, physical exercise, sleep habits, number of patients cared for per week, number of new admissions per week, actual working days per week, coping methods for compassion fatigue, and secondary trauma. See Fig. 1 for details.
Fig. 1
LASSO regression analysis was used to select critical features. a Ten-fold cross-validation was applied, and a vertical line was drawn at the specified value, where the optimal lambda resulted in 16 non-zero coefficients. b In the LASSO model, a coefficient distribution plot of 26 features was extracted from the log(λ) sequence, with a vertical dashed line drawn at the minimum mean squared error (λ = 0.0082)
×
Anzeige
Multivariate logistic regression analysis was performed on the 16 independent variables to further control for confounding factors. Finally, hospital level, age, gender, psychological resilience, peer support, empathy satisfaction, number of patients cared for per week, and secondary trauma were identified as significant factors (p < 0.05), indicating that these factors play an essential predictive role in compassion fatigue. See Table 2 for details.
Table 2
Multivariate logistic regression analysis
Variable
R
SE
Z
p
OR
95% CI
Hospital-Level
0.52
0.08
6.50
0.01
1.68
1.34—2.10
Age
−0.35
0.12
−2.90
0.01
0.70
0.55—0.88
Gender
0.15
0.13
1.10
0.13
1.16
0.94—1.43
Psychological Resilience
−0.48
0.09
−5.30
0.01
0.62
0.48—0.81
Social Support
0.27
0.14
1.90
0.05
1.31
1.00—1.72
Role Overload
−0.19
0.11
−1.70
0.08
0.83
0.65—1.06
Empathy Satisfaction
−0.71
0.21
−3.40
0.01
0.49
0.30—0.81
Time Entering Clinical
0.11
0.10
1.10
0.15
1.12
0.89—1.41
Only Child
0.04
0.13
0.31
0.76
1.04
0.81—1.32
Physical Exercise
−0.12
0.12
−1
0.31
0.89
0.68—1.17
Sleep Habits
−0.05
0.11
−0.45
0.66
0.95
0.77—1.16
Weekly Patients Cared For
0.33
0.13
2.50
0.01
1.39
1.08—1.82
Weekly Number of New Admissions
−0.30
0.13
−2.30
0.02
0.74
0.57—0.97
Actual Working Days Per Week
0.10
0.14
0.70
0.49
1.11
0.83—1.47
Learned Coping Methods for Compassion Fatigue
0.25
0.15
1.70
0.09
1.29
0.94—1.77
Secondary Trauma
−0.34
0.11
−3.10
0.01
0.71
0.56—0.89
Comprehensive analysis of multiple classification models
In this study, several models, including logistic regression, random forest, XGBoost, and Light GBM, were trained, with the data split into training and testing sets in a 7:3 ratio, followed by ten repeated samplings. The models were evaluated using AUC (Area Under the Curve). The results showed that random forest and XGBoost performed best on the training set with higher AUC values. At the same time, logistic regression had the highest AUC value on the test set, indicating its superior generalization ability (Figure a, b).DCA (Decision Curve Analysis) and calibration curves were used to assess the models' clinical applicability further. The DCA results indicated that logistic regression and random forest models had higher net benefits at different thresholds, demonstrating good clinical application potential (Figure c). Calibration curve analysis revealed that the logistic regression model had higher prediction accuracy on the test set, with lower calibration error and more vital reliability (Figure d).In the PR (Precision-Recall) curve analysis, results from both the training and test sets indicated that the logistic regression model performed best when handling imbalanced data, with the highest AP (Average Precision) value on the test set (Figure e, f). The logistic regression model performed exceptionally well across multiple metrics, particularly on the test set, where both the AP and AUC values were the highest. Therefore, the logistic regression model can be considered the optimal model in this analysis. See Fig. 2 for details.
Fig. 2
Comprehensive Analysis of ML Models (a) ROC Curve for the Training Set (b) ROC Curve for the Test Set (c) Calibration Curve (d) Decision Curve (e) PR Curve for the Training Set (f) PR Curve for the Test Set
×
The best model building and evaluation
We performed logistic regression analysis on the training set and used tenfold cross-validation for model evaluation. The results showed that the average AUC of the training set was 0.81, with the validation set's AUC slightly lower than that of the training set, and the test set's AUC was 0.77 (Figure a-c). The AUC values for the training, validation, and test sets all stabilized, indicating that the model exhibited good prediction accuracy and generalization capability. The model can be considered well-fitted when the AUC of the validation set is lower than that of the test set or the difference does not exceed 10%. The learning curve (Figure d) shows that as the number of training samples increases, the AUC values for both the training and validation sets gradually stabilize, indicating strong fitting ability and high stability. In summary, the logistic regression model is suitable for classification tasks in the current dataset and provides accurate prediction results. See Fig. 3 for details.
Fig. 3
Logistic Regression Model Training, Validation, and Testing Results. a The ROC and AUC for the training set demonstrate the model's performance across different training sets, with the other colored lines representing the model's results across ten repeated trainings; (b) The ROC and AUC for the validation set, representing the results of each repeated training on the validation set; (c) The ROC and AUC for the test set, showing the detection results for 30% of the nursing students with compassion fatigue, with an AUC value of 0.77 on the test set, indicating good performance on unseen data; (d) The learning curve, where the red dashed line shows the AUC variation with the number of samples in the training set, and the blue dashed line shows the AUC variation in the validation set, illustrating the model's stability and fitting solid ability. The AUC and learning curve values are presented as averages with 95% confidence intervals (CI)
×
From SHAP to model interpretation
We used SHAP values to visually explain the impact of selected variables on the model's predictions. Figure 4a shows the importance of each feature, with longer bars indicating more significant influence on the model's predictions. The figure reveals that social support, psychological resilience, and empathy satisfaction have the most significant impact on the model, indicating these factors play a crucial role in shaping the prediction model. Secondary trauma and age also significantly influence the model. Other features like role overload, psychological resilience, and social support scores have relatively more minor impacts. In contrast, the hospital level has almost no effect, suggesting it is unimportant in forming the predictive model. Through the SHAP value analysis, we can see which features are critical in the model, allowing a better understanding of how these variables affect the prediction outcomes.
Fig. 4
a Feature Importance Based on SHAP Values. b. Feature Importance-Based Scatter Plot (Sorted by Importance)
×
Anzeige
Figure 4b displays the importance of each feature and its simulated impact on the target variable, with features ranked from top to bottom by importance. The horizontal axis represents the degree of influence each feature has on the target variable; the further from zero, the more significant the impact. Positive values indicate a positive effect, while negative values indicate a negative effect. The color of the dots ranges from purple to yellow, indicating the feature values from low to high, with the size of the dots proportional to their impact on the model. The gray dashed line represents zero influence; the further the dots are from this line, the more significant the effect. As seen in the figure, "empathy satisfaction" has the most significant impact on the target variable, with a broad and considerable distribution, indicating it plays a vital role in the model. "Secondary trauma" and "weekly number of new admissions" also have significant impacts, whereas "hospital level" and "age" have the slightest influence, contributing little to the model.
Discussion
Prevalence of compassion fatigue
The findings of this study indicate that compassion fatigue is a common issue among nursing students, with the proportion of those experiencing severe compassion fatigue reaching as high as 47.16%. This aligns with the high emotional labor pressure commonly faced in the nursing profession, both domestically and internationally. This finding is consistent with results from several international studies in recent years [12]. in a global review of emotional labor among nursing staff, found a strong correlation between emotional labor and compassion fatigue, mainly when dealing with critically ill, terminally ill, or trauma patients [13]. In such situations, nurses are more prone to experience emotional exhaustion and compassion fatigue. Similarly, research on the theory of compassion fatigue emphasizes that prolonged exposure to patients' suffering can exacerbate feelings of fatigue due to the depletion of compassion [14]. The results of these international studies are consistent with the high emotional pressure experienced by nursing students during clinical practice, as identified in this study [15].
The impact of empathy satisfaction, psychological resilience, and peer support on compassion fatigue
Higher empathy satisfaction can significantly reduce compassion fatigue, as nursing students gain emotional fulfillment and positive emotional feedback through patient interactions, enhancing their resilience when facing work-related stress [16]. This is consistent with findings from international literature. Studies have shown that nursing personnel with high empathy satisfaction are generally better equipped to manage emotional stress, leading to a lower incidence of compassion fatigue [17]. Globally, increasing empathy satisfaction is considered a crucial factor in protecting the mental health of nursing staff, as it improves patient experiences and enhances job satisfaction and well-being among nurses [11].
Nursing students with high psychological resilience can better maintain emotional stability and adopt positive coping strategies when confronted with work-related stress and challenges, thereby mitigating compassion fatigue [18]. International research supports this view as well. For example, a study found that higher psychological resilience is associated with lower levels of burnout and compassion fatigue. Enhancing psychological resilience is an effective intervention to help nursing personnel maintain mental health in stressful environments [19]. This suggests that fostering psychological resilience should be a key component of nursing education and professional development.
Anzeige
A supportive peer environment can provide emotional comfort and practical assistance, helping buffer work-related stress and reduce compassion fatigue. The research underscores the importance of peer support in alleviating stress among nursing personnel. For instance, studies have shown that peer support improves teamwork and enhances nurses' sense of belonging and self-efficacy [20]. By establishing strong support networks, nursing personnel can achieve greater psychological comfort and security when facing high-pressure situations, reducing the risk of compassion fatigue [21]. Therefore, organizational interventions such as emotional support programs, resilience training, and team-building activities should be implemented to effectively enhance the impact of these positive factors, thereby better supporting nursing personnel in coping with work stress and compassion fatigue.
The impact of secondary trauma and workload on compassion fatigue
Secondary trauma is one of the primary negative factors contributing to compassion fatigue. Nursing students may experience emotional burdens and a sense of trauma when caring for critically ill patients or witnessing their suffering, which can exacerbate compassion fatigue [22]. Research indicates that nursing personnel who are exposed to patients' pain and death over extended periods are more likely to develop secondary trauma, leading to compassion fatigue [21]. This is particularly prevalent in emergency and intensive care environments, where the incidence of secondary trauma is higher [23]. Therefore, specialized support and psychological interventions are necessary for nursing personnel working in these high-risk settings.
A heavy caregiving load increases nursing students' workload and emotional investment, raising the risk of compassion fatigue. Studies have shown a positive correlation between the workload of nursing personnel and compassion fatigue [5]. Increased workload affects the quality of care and leads to emotional exhaustion among nurses, heightening the risk of burnout and compassion fatigue [17]. Consequently, appropriately distributing work tasks and ensuring adequate nurse-to-patient ratios are crucial for alleviating compassion fatigue among nursing personnel.
The impact of gender, age, and hospital level on compassion fatigue among nursing students
Gender differences may make sure nursing students more susceptible to compassion fatigue, particularly female nursing students who may be more affected due to their higher capacity for empathy and emotional investment [24]. Research consistently shows that female nursing personnel are more prone to compassion fatigue and burnout compared to their male counterparts [25]. This may be because women tend to invest more emotional labor in caregiving [26]. Therefore, specific support strategies tailored to female nursing staff, such as emotional management training and mental health support, are essential.
Younger nursing students may be more vulnerable to compassion fatigue due to their lack of experience and mature coping strategies. Literature indicates that younger nursing personnel are more likely to experience emotional exhaustion and stress because they may lack sufficient clinical experience and coping mechanisms [25]. Thus, early career training and psychological support for younger nursing students, particularly in areas like emotional management and stress coping, can significantly reduce compassion fatigue.
The direct impact of hospital level on compassion fatigue is limited, but different work environments may indirectly affect the stress experienced by nursing students. Study suggests that while the level and size of a hospital are not primary predictors of compassion fatigue, nursing staff in higher-level hospitals may face more significant stress and emotional burden due to the complexity of patients and the intensity of the work [27]. Therefore, improving the supportive nature of the hospital work environment, such as providing mental health resources and comfortable working conditions, can help mitigate the indirect impact of compassion fatigue.
Although there are many risk factors for compassion fatigue among nursing students, no predictive model has been established. In this study, we employed several machine learning (ML) models, and after analyzing the AUC, DCA, calibration, and PR curves, we found that the logistic regression model outperformed the other ML models. However, comprehensively interpreting ML predictive models and visually presenting the predictive results to clinicians has always been challenging. Therefore, we applied the SHAP (Shapley Additive explanations) method to the logistic regression model to achieve optimal predictive performance and interpretability. Through this approach, we identified several critical variables related to the development of compassion fatigue among nursing students.
Limitation
This study has made some progress in constructing a predictive model for compassion fatigue among nursing students, but several limitations still need to be addressed. First, the sample was limited to internship nursing students from a single medical school, which may affect the generalizability of the results. Future studies should expand the sample size to include a broader population. Second, the cross-sectional design cannot reveal the dynamic changes in compassion fatigue, and longitudinal research would be more effective in understanding its developmental process. Additionally, self-reported data may be subject to bias; therefore, future studies could incorporate objective assessment tools to reduce errors. Although this study used LASSO regression to screen key variables, other essential factors, such as personality traits and social background, may have been overlooked. The interpretability of the machine learning model is also limited, necessitating further usability improvements. Finally, while the model can effectively predict compassion fatigue, the practical effectiveness of interventions based on these predictions has yet to be validated. Therefore, future research should aim to optimize the model and validate intervention outcomes through broader samples, longitudinal designs, and multivariable analysis.
Conclusion
This study successfully developed a predictive model for nursing students' compassion fatigue using LASSO regression and machine learning models, identifying key predictors such as psychological resilience, peer support, and secondary trauma. The results indicate that the logistic regression model best predicted compassion fatigue, exhibiting high accuracy and generalization ability. Through SHAP value analysis, we further clarified the impact of each variable on compassion fatigue, providing crucial theoretical support for nursing management and educators. This study offers an effective tool for early identification of high-risk nursing students and lays the foundation for targeted interventions. However, future research should expand the sample size and incorporate longitudinal studies to validate the long-term effectiveness of the predictive model while also exploring the real-world impact of model-based psychological interventions in reducing compassion fatigue. Overall, this research provides valuable insights for improving nursing students' job satisfaction and mental health, and it holds positive implications for future nursing education and management.
Acknowledgements
We would like to express our sincere gratitude to the undergraduate nursing students at Anhui University of Chinese Medicine for their participation in the survey.
Declarations
Ethics approval and consent to participate
This study was approved by the Ethics Review Committee of Yancheng First People's Hospital (Ethics Approval Number: 2024-K-093), and all participants signed written informed consent.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
