Impact of witnessed workplace violence on job burnout and turnover intention among healthcare workers: a multigroup analysis based on occupational categories

Abstract

Objective

To examine how witnessed workplace violence (WPV) in hospitals influences healthcare workers’ turnover intention, to test the mediating role of job burnout in the association and to compare this mediation across different occupational groups.

Methods

A cross-sectional survey was conducted in six tertiary public hospitals across three provinces or municipalities in China (Shandong, Tianjin and Gansu). Participants were restricted to healthcare workers who only witnessed WPV (excluding those who directly experienced WPV or had no exposure). The final sample comprised 1049 participants. Descriptive statistics, correlation analyses, mediation regression and multigroup structural equation modelling (SEM) were employed to assess relationships and occupational differences.

Results

92.6% of healthcare workers reported exposure to witnessed WPV in the past year. Spearman correlations revealed significant positive associations between witnessed WPV and job burnout, witnessed WPV and turnover intention, and job burnout and turnover intention (p<0.01). Mediation regression showed job burnout partially mediated the association between witnessed WPV and turnover intention (95% CI 0.098 to 0.229), accounting for 44.9% of the total effect. Multigroup SEM demonstrated significant mediation among nurses (indirect effect=51.3%) and doctors (34.5%), but not among medical technicians (95% CI −0.186 to 0.059).

Conclusions

High prevalence of witnessed WPV exists among Chinese tertiary hospital healthcare workers. Job burnout partially mediates the association between witnessed WPV and turnover intention pathway, with significant occupational variation. These findings support the development of occupation-specific intervention strategies.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Multicentre cross-sectional design with participants from tertiary public hospitals across eastern, central and western China.

• Heterogeneity path analyses were performed across three occupational groups, with measurement invariance verified for cross-group comparisons.

• Bootstrap-corrected structural equation models were used for mediating effect analyses.

• Cross-sectional single-wave self-report design prevents establishment of temporal causality.

• Small sample size of the medical and technical personnel subgroup limits statistical power for subgroup analyses.

Introduction

Workplace violence (WPV) in hospital settings has become a major global public health issue, threatening healthcare workers’ mental health and occupational stability.¹ Its severity and the demand for effective control have drawn wide international concern. Notably, WPV not only causes direct physical and psychological harm but also indirectly undermines healthcare workers’ sense of safety and professional engagement, thereby intensifying workforce loss. Many countries are still facing shortages of medical personnel.^{2 3} High turnover intention among healthcare workers further aggravates this problem, compromising both system stability and patient safety.

Turnover intention—defined as an individual’s conscious willingness or tendency to leave their current position—is a key predictor of actual resignation and an indicator of labour-force instability.⁴ A meta-analysis showed that the global prevalence of turnover intention among healthcare workers is 39%.⁵ In developing countries, the proportion of turnover intention related to witnessed WPV is even higher—49.7% in Ethiopia⁶ and 53% in China.⁷ This underscores the urgency of identifying the determinants of turnover intention. Influencing factors exist at multiple levels—individual, job, organisational and environmental.^8–11 Numerous studies have confirmed that job burnout is a major determinant of turnover intention, with a strong positive association between the two.^12–14 Burnout is typically characterised by emotional exhaustion, depersonalisation and reduced personal accomplishment.^{15 16} Manifestations such as fatigue, indifference and decreased work efficacy strongly predict turnover tendencies. Empirical evidence suggests that exposure to WPV accelerates emotional exhaustion, increases burnout risk and heightens turnover intention.^{17 18}

WPV is one of the most severe occupational hazards affecting healthcare professionals’ intent to stay.¹⁹ It includes physical assaults and verbal abuse perpetrated by patients or related individuals. A systematic review estimated that 62% of healthcare workers had experienced some form of WPV in the past year, with verbal abuse being most prevalent and nearly one-fifth reporting physical assault.²⁰ During the COVID-19 pandemic, meta-analyses confirmed persistently high rates of WPV, with increased risk among frontline workers.^{21 22} However, most pandemic-era studies have inevitably confounded the psychological impact of WPV with additional stressors from public health emergencies, making it difficult to isolate the inherent occupational hazard of WPV itself. Well-characterised prepandemic baseline data with rigorous sampling designs are therefore critical for establishing the fundamental association between WPV exposure and occupational outcomes, yet such evidence—particularly for indirect witnessing experiences—remains scarce. In China, over half of healthcare staff reported exposure to WPV,^{23 24} reflecting that despite numerous national policies designed to curb aggression towards medical workers, WPV remains widespread.

Previous research has primarily focused on direct victims, linking violent incidents with psychological distress, burnout and turnover intention.^25–28 However, increasing evidence indicates that mere witnessed violence can also cause adverse outcomes. Havaei et al reported that nurses who observed patients assaulting colleagues experienced trauma, anxiety, depression and burnout.²⁹ Likewise, Chang et al found a significant positive relationship between the frequency of witnessed violence and burnout among hospital staff.³⁰ Witnesses often report fear, worry and anxiety that are associated with emotional exhaustion and diminished job satisfaction—outcomes comparable to or even more severe than those of direct victims. Yet the witness perspective remains under-represented in research on violence-turnover dynamics.

According to social identity theory,³¹ individuals’ self-concepts and behaviours are shaped by group identification and role expectations. Within healthcare settings, doctors, nurses and medical technicians differ in authority, patient-contact frequency and organisational roles, influencing their sense of belonging and identification.³² These differences may yield distinct emotional responses and coping strategies when confronted with violence, resulting in varying patterns of burnout and turnover intention across occupations.

This study applies the stressor–strain–outcome (SSO) model as the theoretical framework. The SSO model was proposed by Koeske to reconceptualise occupational burnout among human service professionals.³³ Subsequently, Um and Harrison further extended and empirically validated the model, elucidating the hierarchical relationships among stressors, psychological strain and negative occupational outcomes.³⁴ The SSO model has since been widely applied in research on workplace adversity across various professional fields. Large-scale empirical studies in China have confirmed its effectiveness in explaining the interrelations among WPV, occupational burnout and healthcare workers’ turnover intentions, thereby providing a reliable theoretical basis for the present study.³⁵ The model proposes that stressors (external environmental stimuli) affect behavioural outcomes indirectly through emotional and psychological strain. In this context, witnessed WPV is conceptualised as a stressor, burnout as strain and turnover intention as the behavioural outcome. Witnessing violence may not inflict physical harm, but it can trigger emotional exhaustion, depersonalisation and reduced personal accomplishment—the core dimensions of burnout—which in turn heighten the desire to leave one’s job.

Accordingly, this study investigates whether job burnout mediates the relationship between witnessed WPV and turnover intention among healthcare workers and whether this mediation differs by occupation. Hypotheses were as follows:

H1: Witnessed WPV is positively associated with turnover intention.

H2: Witnessed WPV is positively associated with job burnout.

H3: Job burnout is positively associated with turnover intention.

H4: Job burnout mediates the relationship between witnessed hospital WPV and turnover intention.

H5: The mediating mechanism varies across occupational groups (doctor, nurse and medical technician).

Materials and methods

Study design

This cross-sectional study was conducted in December 2018 in six tertiary public hospitals across three provinces or municipalities in eastern (Shandong and Tianjin) and western (Gansu) China.

The data were collected in 2018, prior to the COVID-19 pandemic. This dataset was selected for two primary reasons: first, the core occupational stress mechanism under study, grounded in the SSO model and vicarious trauma theory, is theoretically stable over time; second, as scarce prepandemic baseline data covering healthcare workers with witnessed-only WPV exposure across six tertiary hospitals, it provides a valuable reference for future longitudinal trend research.

Sampling strategy and sample representativeness

A purposive sampling approach was used to select study sites and participants, with multidimensional quality control procedures implemented to optimise sample representativeness and mitigate selection bias. Within hospitals, sampling adopted a single-stage cluster approach, supplemented by occupation-based quotas to ensure adequate representation of each occupational subgroup. Recruitment was also conducted across different work shifts.

For geographic and institutional representativeness, the six participating hospitals were chosen to capture heterogeneity in regional economic development, healthcare resource allocation and doctor–patient relationship contexts, while all sites maintained a uniform tertiary public hospital level to ensure consistent institutional attributes. This design strikes a balance between regional heterogeneity and institutional homogeneity, mitigates confounding attributable to hospital tier and enhances the generalisability of the study findings.

At the department level, all major clinical and medical technical departments across participating hospitals were enrolled in the survey, spanning settings with high WPV exposure (eg, emergency department, paediatrics) and technical units with relatively low exposure risk. This broad departmental coverage captures the main spectrum of WPV exposure scenarios and avoids selection bias that could arise from selectively recruiting departments with distinct exposure patterns.

Within each department, questionnaires were distributed under the coordination of department heads, with an effort to cover on-duty staff across a range of age groups, professional titles and lengths of service. This strategy aimed to capture the diversity of internal workforce composition and reduce the risk of over-representation for any single demographic subgroup.

Finally, a post hoc empirical assessment was conducted to evaluate the demographic structure of the valid initial sample (n=2061) against national health workforce data from the 2018 China Health Statistical Yearbook. The distributions of gender, age, educational level and occupational category in the initial sample were generally aligned with national benchmarks for healthcare workers in tertiary public hospitals, providing empirical support for the demographic representativeness of the initial sample. In the final analytical sample, women accounted for 75.5%, close to the national proportion of 71.8%; the occupational distribution of nurses (55.6%), doctors (35.3%) and medical technicians (9.2%) was also broadly consistent with the national tertiary hospital workforce structure.

Data collection

The final questionnaire was developed and validated based on feedback from hospital management experts and mental health professionals, ensuring adequate content validity for the study context. Trained investigators administered the survey after obtaining informed consent from hospital administrators, medical dispute resolution departments, human resources offices and individual participants. All investigators received standardised training on uniform survey protocols and communication scripts to ensure consistent implementation across all sites. Anonymous survey administration was adopted to reduce social desirability bias and improve the authenticity of responses.

Study participants and sample size justification

The inclusion criteria were as follows: (1) doctors, nurses and medical technicians currently employed in hospitals; (2) at least 1 year of work experience and (3) voluntary participation. Exclusion criteria were (1) unwillingness to participate and (2) trainees or interns.

A total of 2450 questionnaires were distributed, and 2061 valid responses were returned, yielding an effective response rate of 84.1%. To focus exclusively on the psychological and behavioural outcomes of indirect WPV exposure, we excluded participants who had personally experienced WPV or had neither witnessed nor experienced such events. The final analytical sample was obtained through a three-step screening procedure: (1) excluded 153 participants with no witnessed WPV experience in the past 12 months, leaving 1908 participants with witnessed WPV exposure; (2) excluded 853 participants who had personally experienced WPV in the past 12 months, retaining 1055 participants who had only witnessed but not directly experienced violence and (3) excluded six participants with missing values in core study variables. The final analytical sample comprised 1049 healthcare workers, including 370 doctors, 583 registered nurses and 96 medical technicians.

This sample size was statistically justified for all planned analyses. Consistent with established guidelines for structural equation modelling (SEM), a minimum sample of 200 participants and a respondent-to-free-parameter ratio of at least 10:1 are recommended for stable parameter estimation, both of which are well met by the present sample. Specifically, a sample size of approximately 1000 has been demonstrated to provide sufficient statistical power to detect small mediation effects while preserving appropriate type I error control.³⁶ The group-level sample sizes also satisfy widely accepted criteria for robust multigroup confirmatory factor analysis and measurement invariance testing.³⁷

Measures

Witnessed WPV

Witnessed WPV was measured using a scale jointly developed by the International Labour Organization, International Council of Nurses, WHO and Public Services International.³⁸ The instrument covers three dimensions: verbal violence, physical violence and sexual harassment. Respondents were asked: (1) “In the past year, have you witnessed a patient verbally insult or scold healthcare staff?” (2) “Have you witnessed a patient physically assault a healthcare worker with body or objects?” and (3) “Have you witnessed a patient sexually harass or assault a healthcare worker?” Each item was rated on a 4-point Likert scale (0=never, 1=once, 2=two to three times, 3=more than three times). The total score ranged from 0 to 9, with higher scores indicating more frequent witnessed of WPV. In this study, the operational definition of witnessed WPV is as follows: participants who report at least one occurrence of any item from the aforementioned dimensions within the past 12 months are classified as positive for witnessed violence.

Job burnout

Job burnout was assessed using the Maslach Burnout Inventory (MBI)–General Survey developed by Maslach and Jackson and revised by Li and Shi for use in Chinese populations.^{39 40} The MBI is considered the gold standard for measuring burnout and has demonstrated strong reliability and validity in Chinese samples.⁴¹ The scale comprises 15 items across three dimensions: emotional exhaustion (5 items), depersonalisation (4 items) and reduced personal accomplishment (6 items). Each item was scored on a 7-point Likert scale ranging from 0 (never) to 6 (every day). Higher scores indicate greater burnout. Job burnout was calculated using a weighted scoring system: Total score=0.4×emotional exhaustion subscale+0.3×depersonalisation subscale+0.3×(6−reduced personal accomplishment subscale). Scores were graded as: <1.5, no burnout; 1.5–3.4, some burnout symptoms; ≥3.5, serious burnout.⁴² In the present study, Cronbach’s α=0.92, with α values of 0.96, 0.95 and 0.95 for the three subscales, respectively.

Turnover intention

Turnover intention was measured using the Turnover Intention Questionnaire developed by Michaels and Spector⁴³ and later translated and revised by Li and Li.⁴⁴ The Chinese version has demonstrated good reliability in local samples.⁴⁵ The instrument includes six items rated on a 4-point Likert scale (1=never, 4=often). It consists of three subdimensions: intention to resign (items 1 and 6), intention to search for a new job (items 2 and 3) and perceived possibility of obtaining another job (items 4 and 5). Higher scores indicate stronger turnover intention. According to the grading criteria proposed by Li and Li,⁴⁴ the total mean score was categorised into four levels: very low (≤1), low (>1 and ≤2), high (>2 and ≤3) and very high (>3). In this study, Cronbach’s α=0.72.

Data analysis

All questionnaires were double-entered using EpiData V.3.1 to ensure accuracy, and incomplete or invalid responses were excluded. Questionnaires with extensive missing data, contradictory answers and stereotyped responses were deleted case-wise.

Missing data processing: Core study variables had a missing rate of 0.6% (6 participants), well below the 5% threshold for robust complete case analysis. Missing data were assumed to be missing at random, and listwise deletion (complete case analysis) was used for all core analyses, as this method produces minimal bias at low missing rates.

Harman’s single-factor test was performed to assess potential common method bias via unrotated exploratory factor analysis of all scale items; a first-factor variance explanation rate below 40% was considered non-substantial.

Normality was assessed using the Kolmogorov-Smirnov test. Normally distributed variables were presented as mean±SD and analysed using Pearson’s correlation; non-normally distributed variables were presented as median (IQR) and analysed using Spearman’s rank correlation. Univariate analyses were conducted to examine between-group differences: independent samples t-tests or one-way analysis of variance (ANOVA) (F-tests) were used for normally distributed variables, and Mann-Whitney U or Kruskal-Wallis tests for non-normally distributed variables. Post hoc pairwise comparisons used unadjusted least significant difference tests after ANOVA, and Bonferroni-corrected tests after Kruskal-Wallis analyses.

All descriptive and correlation analyses were performed in SPSS V.26.0, and SEM was conducted in AMOS 26.0. Bootstrapping with 5000 resamples was used to estimate two-sided 95% bias-corrected CIs for indirect effects; intervals not including zero were considered statistically significant.

A priori structural equation model was developed to test the hypothesised mediation relationships. The model consisted of several latent variables and their corresponding observed indicators derived from the study questionnaires. Each latent variable was measured by multiple observed indicators. To ensure model identification in AMOS, the factor loading of the first observed indicator for each latent variable was fixed at 1, while all residual terms were freely estimated. The hypothesised pathways included the direct effect of the independent variable on the dependent variable, as well as the indirect effect via the proposed mediating variable. This base model was further extended for subsequent multigroup analysis.

Multigroup SEM was used to examine the moderating effect of occupational category on the mediation pathways. Six structural equation models were established for analysis: M1 (doctor-only model) and M2 (nurse-only model) separately tested model fit among doctors and nurses; M3 (medical technician-only model) assessed model fit for medical technicians. M4 (configural invariance model) included all three occupational groups without cross-group equality constraints and served as the baseline for multigroup comparison. M5 (measurement weights model) constrained factor loadings to be equal across groups based on the configural invariance model to evaluate cross-group measurement equivalence. M6 (structural weights model) further constrained mediation path coefficients across groups to test the moderating effect of occupation. Critical ratio of difference (CRD) testing was used to examine specific path differences across occupations; a CRD value >1.96 was considered statistically significant.

Model fit was evaluated using a set of widely accepted fit indices: χ² with df (df), χ²/df ratio, root mean square error of approximation (RMSEA), goodness-of-fit index (GFI), comparative fit index (CFI) and Tucker-Lewis index (TLI). Measurement invariance was examined by comparing nested models sequentially, with changes in CFI (ΔCFI) and RMSEA (ΔRMSEA) adopted as evaluation criteria. The following cut-off values were applied to determine acceptable model fit: χ²/df<5, RMSEA<0.08, GFI>0.90, CFI>0.90 and TLI>0.90. A change of ΔCFI<0.01 and ΔRMSEA<0.015 was defined as evidence of adequate measurement invariance.

Results

Common method bias test

Harman’s single-factor test was used to assess potential common method bias.⁴⁶ Five factors with eigenvalues greater than 1 were extracted, and the first factor accounted for 29.0% of the total variance, below the 40% threshold.⁴⁷ These results suggest that common method bias was unlikely to be a major concern in this study. However, because Harman’s test is only a preliminary diagnostic approach, its results cannot completely rule out the presence of common method bias.

Participant characteristics and group differences

A total of 1049 healthcare workers who had witnessed WPV were included. Participant characteristics are summarised in table 1. Females accounted for 75.5% of the sample. More than half (53.0%) were under 30 years old, while 4.9% were over 50. Nearly half held a bachelor’s degree (47.6%), and 21.2% had a master’s degree or higher. The majority were married (62.6%). Most respondents worked in inpatient departments (80.4%). Regarding profession, nurses represented the largest group (55.6%), followed by doctors (35.3%). The sample in this study matches the national data on healthcare professionals from the ‘2018 China Health Statistical Yearbook’ in terms of demographic characteristics and occupational categories.⁴⁸

Table 1. Demographic characteristics of the entire sample (N=1049).

Variables	Category	N	%
Gender	Man	257	24.5
	Woman	792	75.5
Age group (years)	≤30	556	53.0
	(30, 51)	442	42.1
	≥51	51	4.9
Education level	Junior college or below	328	31.3
	College	499	47.6
	Master or above	222	21.2
Marital status	Married	657	62.6
	Single	384	36.6
	Other	8	0.8
Profession	Doctor	370	35.3
	Nurse	583	55.6
	Medical technician	96	9.2
Working place	Emergency	56	5.3
	Surgery	150	14.3
	Ward	843	80.4

Univariate analyses showed that levels of witnessed WPV differed significantly by gender, age, education, marital status and occupational category. Job burnout differed significantly across occupational categories and working places, while turnover intention varied significantly by age, education and occupation (table 2). With respect to witnessed WPV, although median scores were comparable across occupational groups, the Kruskal-Wallis test assesses overall rank distributions rather than central tendency alone; the statistical significance was driven by divergence in the upper tails of the exposure distribution, with physicians and medical technicians showing higher proportions of frequent witnessed WPV exposure.

Table 2. Univariate analysis and descriptive statistics of study variables.

Variables	Witnessed WPV	Job burnout	Turnover intention
Gender
Man	3.00 (2.00, 5.00)	33.44±12.54	14.56±4.27
Woman	3.00 (2.00, 4.00)	32.68±12.73	14.37±3.75
t/Z	−3.960*	0.827	0.661
Age groups (years)
≤30	3.00 (2.00, 4.00)	33.07±12.10	14.56±3.65
(30, 51)	3.00 (2.00, 5.00)	33.00±13.22	14.88±3.86
≥51	3.00 (2.00, 4.00)	29.51±13.79	12.37±4.63
F/H	21.607*	1.886	7.576*
Education level
Junior college or below	3.00 (2.00, 4.00)	32.29±12.36	14.00±3.81
College	3.00 (2.00, 4.00)	33.07±12.69	14.46±3.81
Master or above	3.00 (2.00, 5.00)	33.27±13.15	14.94±4.11
F/H	14.757*	0.597	3.989†
Marital status
Married	3.00 (2.00, 5.00)	33.11±13.12	14.39±3.96
Single	3.00 (2.00, 4.00)	32.47±11.57	14.40±3.77
Other	4.00 (2.25, 6.00)	32.13±24.68	17.25±1.75
F/H	20.567*	0.327	2.15
Profession
Doctor	3.00 (2.00, 5.00)	33.81±12.85	14.73±3.97
Nurse	3.00 (2.00, 4.00)	32.85±12.40	14.36±3.72
Medical technician	4.00 (2.00, 5.00)	29.35±13.18	13.53±4.43
F/H	15.766*	4.749‡	3.772†
Working place
Emergency	3.00 (2.00, 5.00)	32.34±13.83	14.36±3.41
Surgery	3.00 (2.00, 4.00)	30.38±12.86	14.01±3.95
Ward	3.00 (2.00, 4.00)	33.35±12.53	14.49±3.90
F/H	2.222	3.550†	1.001

^*p<0.001.

^†p<0.05.

^‡p<0.01.

F, one-way analysis of variance; H, Kruskal-Wallis test; t, independent-samples t-test; WPV, workplace violence.

Post hoc analyses indicated that doctors and technicians reported witnessing WPV more frequently than nurses; compared with medical technicians, both doctors and nurses exhibited higher levels of job burnout and stronger intentions to leave their jobs (table 3).

Table 3. Post hoc pairwise comparisons between occupational groups.

Variables	Pairwise comparison	Statistic	SE	P value
Witnessed WPV	Doctor vs nurse	70.407	19.771	0.001
	Doctor vs technician	−12.865	34.070	1.000
	Nurse vs technician	−83.272	32.762	0.033
Job burnout	Doctor vs nurse	0.977	0.840	0.249
	Doctor vs technician	4.459	1.447	0.002
	Nurse vs technician	3.491	1.392	0.012
Turnover intention	Doctor vs nurse	0.386	0.258	0.154
	Doctor vs technician	1.198	0.444	0.007
	Nurse vs technician	0.831	0.427	0.052

For witnessed WPV: Results from Kruskal-Wallis post hoc tests. Statistic=mean rank difference. P values were adjusted using the Bonferroni method. For job burnout and turnover intention: Results from one-way ANOVA with least significant difference post hoc tests. Statistic=mean difference. P values are unadjusted.

ANOVA, analysis of variance; WPV, workplace violence.

The overall mean score of job burnout was 2.17±0.84. Most participants (72.9%) reported some burnout symptoms, while 5.1% experienced serious burnout. For turnover intention, the overall mean score was 2.40±0.65, with 67.9% of participants having high or very high turnover intention. Notably, medical technicians had the highest proportion of very high turnover intention despite their lower burnout levels. Detailed distributions are provided in online supplemental appendix table 1.

Correlations among variables

The Kolmogorov-Smirnov test indicated that the variable witnessed WPV was not normally distributed (p<0.001). Spearman’s correlations (table 4) showed significant positive associations between witnessed WPV and job burnout (p<0.01), between witnessed WPV and turnover intention (p<0.01), and between job burnout and turnover intention (p<0.01). The same pattern held for doctors and nurses, with stronger correlations observed within these groups. Among medical technicians, witnessed WPV was not significantly correlated with burnout or turnover intention (p>0.01), but burnout remained positively associated with turnover intention (p<0.01).

Table 4. The correlation analysis among research variables.

Variables	M	SD	Witnessed WPV	Job burnout
Witnessed WPV	3.32	1.80	1
Job burnout	32.87	12.68	0.084*	1
			(0.151*, 0.167*, −0.038)
Turnover intention	14.42	3.89	0.146*	0.339*
			(0.199*, 0.168*, −0.196)	(0.362*, 0.314*, 0.403*)

Values outside the parentheses represent the correlation coefficients of the whole sample; values inside the parentheses represent the correlation coefficients of the samples from doctor, nurse and medical technician.

^*p<0.01.

WPV, workplace violence.

Measurement model validation

Confirmatory factor analysis was performed to examine the three-factor measurement model comprising witnessed WPV, job burnout and turnover intention. The model exhibited an acceptable overall fit to the data: χ²=165.837, df=24, p<0.001, χ²/df=6.910, GFI=0.966, CFI=0.940, TLI=0.910, RMSEA=0.075.

Standardised factor loadings for all items ranged from 0.35 to 0.93. The composite reliability values were 0.52 for witnessed WPV, 0.82 for job burnout and 0.78 for turnover intention, with corresponding average variance extracted values of 0.30, 0.70 and 0.56, respectively. The core constructs showed generally acceptable reliability and convergent validity, with the witnessed violence dimension presenting relatively lower psychometric properties due to its brief, adapted measurement design. The three-factor structure thus demonstrated satisfactory psychometric properties for subsequent analyses.

Mediation regression analysis

Based on theoretical assumptions, we examined the mediating role of job burnout in the association between witnessed WPV and turnover intention. A measurement model including demographic covariates (age, education and profession) was constructed. The bootstrap test (5000 resamples) revealed a significant partial mediation effect of burnout.

The standardised total effect of witnessed WPV on turnover intention was 0.167 (p<0.001). The standardised indirect effect via burnout was 0.075 (p<0.001), while the direct effect remained significant (0.091, p=0.004). Model fit indices indicated a good fit: χ²/df=4.132 (p<0.001), RMSEA=0.055, GFI=0.978, CFI=0.965, TLI=0.946. The indirect effect accounted for 44.9% of the total effect (table 5, figure 1).

Table 5. Results of mediation effect test.

Method	SE	95% CI
		Lower bound	Upper bound
Witnessed WPV (X) → job burnout (M)	0.142*	0.076	0.208
Job Burnout (M) → turnover intention (Y)	0.533*	0.468	0.591
X→M→Y	0.075*	0.040	0.116
Direct effect	0.091†	0.030	0.149
Total effect	0.167*	0.098	0.229

^*p<0.001.

^†p<0.01.

WPV, workplace violence.

Figure 1. Mediation model showing the relationship between witnessed violence and turnover intention as mediated by job burnout. Values on the paths represent standardised path coefficients. **p<0.01, ***p<0.001.

Multigroup SEM analysis

Measurement invariance across occupations

To further explore occupational differences, a multigroup SEM was conducted. The sample was divided into doctors, nurses and medical technicians. Model-fit indices for the three groups were satisfactory: doctor (M1): GFI=0.966, CFI=0.955, TLI=0.927, RMSEA=0.068; nurse (M2): GFI=0.979, CFI=0.971, TLI=0.952, RMSEA=0.051; medical technician (M3): GFI=0.927, CFI=0.945, TLI=0.910, RMSEA=0.087. All indices met acceptable thresholds, suggesting minimal differences in measurement fit across groups.

Measurement invariance was evaluated by comparing the unconstrained model (M4), the measurement-weight model (M5) and the structural-weight model (M6). ΔCFI and ΔRMSEA values were used as indicators. The ΔCFI between M4–M5 was −0.001 and between M4–M6 was −0.005; ΔRMSEA values were −0.005 and −0.007, respectively. As all ΔCFI<0.01 and ΔRMSEA<0.015, measurement invariance across groups was supported, allowing for parameter comparisons (table 6). Measurement invariance is a necessary prerequisite for the comparison of multiple sets of structural equation models. The test results of this study support measurement invariance across occupational groups, ensuring statistical validity of subsequent comparisons of parameters between groups.

Table 6. Model in variance test using multigroup nested analysis.

Model	χ2 (df)	P value	GFI	CFI	TLI	RMSEA	∆CFI	∆RMSEA
M1	59.987 (22)	0.001	0.966	0.955	0.927	0.068
M2	55.624 (22)	0.001	0.979	0.971	0.952	0.051
M3	37.685 (22)	0.020	0.927	0.945	0.910	0.087
M4	162.080 (33)	0.001	0.958	0.942	0.950	0.061
M5	167.856 (41)	0.001	0.957	0.943	0.962	0.054	−0.001	−0.007
M6	182.047 (47)	0.001	0.953	0.939	0.965	0.052	−0.005	−0.009

CFI, comparative fit index; GFI, goodness-of-fit index; M1, doctor model; M2, nurse model; M3, medical technician model; M4, unconstrained model; M5, measurement weights model; M6, structural weights model; RMSEA, root mean square error of approximation; TLI, Tucker-Lewis index; ΔCFI, change in CFI; ΔRMSEA, change in RMSEA.

Multigroup path differences

Results from the CRD test (table 7, figure 2) indicated that: (1) Among doctors and nurses, burnout significantly mediated the relationship between witnessed WPV and turnover intention, accounting for 34.5% and 51.3% of the total effect, respectively. (2) Among medical technicians, the mediation effect was not significant (95% CI −0.186 to 0.059). (3) Comparison across paths revealed that the predictive effect of burnout on turnover intention was stronger among medical technicians than among doctors and nurses.

Table 7. Standardised path coefficients in SEM for doctor, nurse and medical technician.

Paths	Path coefficient		CRD
	Doctor	Nurse
X→M	0.164*	0.181†	0.401
M→Y	0.481†	0.533†	0.297
X→Y (Direct effect)	0.150*	0.093*	−0.885
X→M→Y (Indirect effect)	0.079*	0.097†
Total effect	0.229†	0.189†
	Doctor	Medical technician
X→M	0.164*	−0.084	−2.064
M→Y	0.481†	0.652†	−2.949
X→Y (Direct effect)	0.150*	−0.134	1.125
X→M→Y (Indirect effect)	0.079*	−0.055
Total effect	0.229†	−0.189
	Nurse	Medical technician
X→M	0.181†	−0.084	−2.399
M→Y	0.533†	0.652†	−2.471
X→Y (Direct effect)	0.093*	−0.055	0.965
X→M→Y (Indirect effect)	0.097†	−0.134
Total effect	0.189†	−0.189

^*p<0.01.

^†p<0.001.

CRD, critical ratios for differences between parameters; M, job burnout; SEM, structural equation modelling; WPV, workplace violence; X, witnessed WPV; Y, turnover intention.

Figure 2. Mediation models for doctor, nurse and medical technician. The three values from top to bottom on each path represent parameters corresponding to the models for doctor, nurse and medical technician. **p<0.01, ***p<0.001; non-significant paths are presented without asterisks.

Discussion

Variable description and group differences

Our findings show that 92.6% of surveyed healthcare workers had witnessed hospital WPV. This prevalence was higher than rates reported among healthcare workers in Pakistan (83.1%) and nurses in China (86.7%), and slightly lower than those documented among paediatric emergency doctors in the United States (96%) and nurses in Lebanon (100%),^49–52 confirming that witnessed WPV is a pervasive issue across the global healthcare sector. The relatively high prevalence in our study can be attributed to two key factors: study setting and case definition. First, our participants were frontline healthcare workers in Chinese tertiary hospitals, which concentrate a large number of critically ill patients. Open-plan clinic layouts and overcrowding during peak visiting hours further increase the probability of witnessing WPV. Second, we used a relatively low threshold for violence exposure: any witness of at least one violent incident in the past year was classified as positive. This annual cumulative incidence approach may have inflated our detection rate.

Furthermore, our analysis found no statistically significant gender difference in job burnout levels among healthcare workers, contradicting the established literature that consistently reports higher burnout prevalence in female staff. This discrepancy is primarily explained by our study’s inclusion criteria: only healthcare workers who had indirectly witnessed WPV but not personally experienced it were enrolled. Most prior studies have focused on direct victims of WPV, where females are more susceptible to elevated burnout following direct victimisation. However, no significant gender-specific stress responses were observed in the context of indirect witnessing. Existing evidence confirms that females can buffer psychological exhaustion and mitigate vicarious trauma through active emotion regulation when witnessing negative workplace events. In contrast, males are more likely to develop a strong sense of injustice and exhibit a more negative professional mindset.^{53 54} This indicates fundamental differences in the psychological mechanisms and gender-specific stress response patterns between direct WPV victimisation and indirect witnessing. In populations exposed solely to vicarious psychological stress, burnout pathways are similar for male and female healthcare workers, ultimately eliminating significant gender differences in job burnout.

Effects of witnessed WPV on job burnout and turnover intention

Our findings supported hypotheses H1, H2 and H3: witnessed hospital WPV was significantly and positively associated with both job burnout and turnover intention, and job burnout was positively correlated with turnover intention. This implies that the impact of hospital violence extends beyond direct victims; witnesses also experience considerable psychological strain.⁵⁵

According to emotional contagion theory,⁵⁶ individuals can experience emotional resonance through verbal, non-verbal and affective exchanges. When healthcare workers witness colleagues being abused or attacked, they may empathically experience fear, anger or helplessness through this emotional contagion. Such negative emotions hinder recovery and contribute to deteriorating occupational well-being.⁵⁷ With prolonged emotional depletion, healthcare workers may experience reduced job satisfaction, emotional exhaustion and a diminished sense of professional mission,^{58 59} eventually heightening turnover intention.

From a social identity perspective, healthcare workers perceive colleagues as members of a shared professional community; witnessing a colleague being assaulted is thus interpreted as an affront to the entire occupational group. When organisational responses to such violence are delayed or inadequate, transient negative emotions can solidify into long-term perceptions of professional insecurity,⁶⁰ accelerating psychological withdrawal. At the organisational level, this process may affect staff shortages and decreased care quality; at the individual level, resignation may become a primary means of escaping emotional distress.

The mediating role of job burnout

Further mediation analysis (supporting H4) revealed that job burnout partially mediates the relationship between witnessed WPV and turnover intention. This indicates that witnessed WPV not only directly heightens turnover intention but also indirectly does so by increasing burnout levels.

Numerous studies have emphasised job burnout as a key mechanism linking WPV to negative occupational outcomes.⁶¹ For instance, Rachel et al found a significant indirect path between hospital violence and turnover intention via burnout.⁶² Gedik et al demonstrated that emotional exhaustion mediates the effect of WPV on health deterioration among healthcare staff.⁶³ Similarly, Liu et al⁶⁴ reported that exposure to WPV correlates positively with burnout and psychological harm. Although these studies primarily focus on direct victims, they collectively highlight the centrality of burnout in explaining how WPV affects healthcare personnel.

Our study extends these insights to witnesses. Even as observers, repeated exposure to violence-related stress may gradually erode professional identity and motivation, potentially leading to a psychological withdrawal tendency. This finding broadens the conceptual scope of hospital violence research, emphasising that its impact transcends direct victims and extends to secondary psychological trauma among witnesses.^{65 66}

From an intervention perspective, the identified mediation pathway holds important practical value: the partial mediating effect of job burnout indicates that the risk of turnover associated with witnessed WPV can be effectively reduced through targeted interventions addressing burnout, even if WPV itself cannot be fully eliminated. This effect size is consistent with those reported in large systematic reviews of WPV in healthcare settings,⁶⁷ supporting the practical relevance of the findings for hospital management and policy.

Although effect sizes were small at the individual level, they carry important public health significance given the extremely high prevalence of witnessed WPV among Chinese healthcare workers. Methodological research has long demonstrated that small individual-level effects can translate to substantial population-level impacts when exposure is widespread across a large workforce.⁶⁸ With over 14 million healthcare workers in China, even small elevations in turnover intention associated with witnessed violence correspond to hundreds of thousands of staff at elevated risk of leaving their positions, with meaningful implications for health system stability.

The effect sizes observed in this study were in the small-to-moderate range, which is consistent with the nature of the exposure examined. As witnessed WPV represents a distal, vicarious stressor rather than direct personal victimisation, smaller effect sizes relative to direct violence exposure are theoretically expected, and align with findings from existing research on vicarious trauma and indirect workplace stress.²⁹

Witnessing WPV and career differences in mediation paths

Multigroup SEM analyses supported hypothesis H5, indicating that the mediation path ‘witnessed WPV → job burnout → turnover intention’ differs significantly across occupations. The path was significant among doctors and nurses but not among medical technicians. This divergence may stem from differences in professional identity⁶⁹ and work environment.⁷⁰

Doctors and nurses occupy the frontlines of patient care, engaging in frequent and intense interactions that increase their exposure to conflict and emotional contagion. Their strong professional identity may amplify empathy toward colleagues’ suffering, thereby exacerbating burnout and turnover risk. In contrast, medical technicians primarily perform laboratory, imaging and diagnostic duties, typically with limited direct patient contact and lower emotional engagement. Their distinct professional orientation may buffer the emotional and occupational consequences of witnessed WPV.^{71 72} Importantly, this finding has two plausible explanations. First, the occupational characteristics of medical technicians—low patient involvement and limited emotional investment—may attenuate the pathway from indirect WPV exposure to psychological stress and job burnout. Second, the small sample size of the medical technician subgroup in our study resulted in insufficient statistical power, increasing the risk of type II error. A true mediating effect therefore cannot be ruled out.

However, despite the weaker direct effects of witnessed WPV on technicians, burnout itself remained a robust predictor of turnover across all groups. This suggests that technicians may be influenced by other occupational stressors—such as workload, limited career advancement and organisational inefficiencies—that increase turnover through the burnout pathway. Consequently, intervention efforts should not be restricted to high-exposure clinical staff but also target systemic stressors across all professional roles.

Implications for intervention

The findings of this study have important practical implications for healthcare organisations seeking to address the growing problem of WPV and its impact on staff retention. Consistent with our results that witnessed WPV is significantly associated with increased burnout and turnover intention, it is critical that institutions move beyond the traditional focus on direct victims and adopt a more comprehensive approach that includes witnesses in their intervention strategies. This represents an important paradigm shift, as most existing programmes have largely overlooked the psychological needs of staff who experience violence indirectly.

Notably, the relatively modest individual-level effect sizes of witnessed WPV on burnout and turnover intention are methodologically expected and consistent with existing literature. As an indirect form of occupational exposure, witnessed WPV generally exerts weaker psychological and occupational impacts than direct victimisation. Nevertheless, the widespread prevalence of such indirect exposure across the healthcare workforce means that even small effects translate into a substantial population-level burden of occupational health impairment and staff attrition. This further justifies the need to expand prevention and support strategies to cover all exposed staff, rather than limiting resources solely to direct victims.

Our identification of occupational burnout as a key mediator between witnessed WPV and turnover intention highlights two complementary targets for intervention: reducing the incidence of WPV at the source and mitigating the psychological impact of exposure for those who witness such events. To achieve these goals, we recommend a multifaceted intervention framework that addresses both organisational and individual-level factors.

At the organisational level, healthcare institutions should establish comprehensive violence prevention systems that include standardised incident reporting protocols, enhanced security measures in high-risk areas and data-driven early-warning mechanisms to identify potential threats before they escalate.^{73 74} Equally important is the development of transparent and fair incident response procedures that demonstrate the organisation’s commitment to staff safety. When staff perceive that violent incidents will be taken seriously and addressed appropriately, they are more likely to report such events and less likely to experience feelings of helplessness and insecurity.

At the individual level, our results underscore the urgent need to expand mental health services to include witnessed WPV. Traditional support programmes that only provide assistance to direct victims fail to address the significant psychological harm experienced by observers. Institutions should implement routine psychological screening for all staff following violent incidents, and provide timely counselling and debriefing services to help witnesses process their emotions and prevent the development of chronic burnout.^{73 75}

In particular, our finding that the mediation pathway differs significantly across occupational groups suggests that targeted, position-specific interventions are essential for maximising effectiveness. For doctors and nurses, who have frequent direct patient contact and are therefore at higher risk of witnessing violence, interventions should emphasise violence prevention training, de-escalation skills and specialised post-traumatic psychological support.^{76 77} For medical technicians, who experience less direct exposure to violence but still suffer from high levels of burnout, efforts should focus on addressing non-violent stressors such as excessive workload, inefficient workflows and limited career advancement opportunities.^78–80

Limitations

Several limitations of this study should be acknowledged. First, the cross-sectional design precludes definitive causal inferences regarding the associations between witnessed WPV, job burnout and turnover intention, despite greater sample diversity achieved through multiregional sampling. Second, reliance on anonymous self-reported questionnaires carries an inherent risk of common method bias. The measure of witnessed WPV, adapted from the WHO framework rather than a fully validated standardised instrument, may also reduce measurement precision. Third, only basic demographic covariates were adjusted for. Key occupational factors including night shift frequency and staffing levels were not included, and residual confounding of the estimated associations therefore cannot be excluded. Additionally, the lower witnessed WPV and burnout levels among nurses deviate from most existing literature, which may be explained by our inclusion criteria restricting to witnessed-only exposure and unmeasured confounding factors. Fourth, the small medical technician subgroup (n=96) confers limited statistical power for multi-group comparisons, which may increase the risk of type II error in subgroup analyses. Finally, generalisability is constrained. This study was restricted to tertiary hospitals, and no in-depth analysis of heterogeneity across individual hospitals and prefecture-level administrative regions was performed. Data were collected in 2018; subsequent changes to healthcare policy, hospital security measures and clinical practice may have altered the prevalence of WPV and the magnitude of its associations with burnout and turnover intention. Replication with contemporary data is therefore recommended.

Future studies with larger multicentre samples should include hospital- and regional-level institutional variables and apply multilevel modelling to explore the impact of contextual differences on the observed associations.

Conclusions

This study examined the psychological and behavioural impacts of witnessed hospital WPV among healthcare workers in tertiary hospitals in China. Unlike prior research focusing mainly on direct victims, our findings provide a theoretical and empirical extension by demonstrating that witnessed WPV also exerts profound indirect effects through burnout. SEM confirmed that witnessed WPV, job burnout and turnover intention are interrelated, with burnout serving as a partial mediator. Furthermore, occupational differences were evident, with stronger effects observed among doctors and nurses compared with technicians.

These findings enrich our understanding of occupational heterogeneity in the ‘witnessed violence–burnout–turnover’ pathway and inform the development of tailored interventions to mitigate the adverse impacts of hospital violence, promote staff retention and safeguard the sustainability of healthcare systems. Addressing witnessed WPV requires a comprehensive, multitiered framework that integrates organisational-level reforms and individual-focused support. Implementing such integrated strategies can effectively attenuate the negative consequences of WPV, stabilise the health workforce and uphold the long-term sustainability of healthcare delivery.

Data availability statement

Data are available on reasonable request.