A cross-sectional design was employed for this study. Between March and May 2025, clinical nurses were recruited from eight tertiary grade-A hospitals in Sichuan Province using convenience sampling. Inclusion criteria were as follows: (1) age ≥18 years with a valid nursing license; (2) ≥1 year of clinical work experience; and (3) informed consent and voluntary participation. Exclusion criteria included: (1) nurses in internship, advanced training, or administrative roles; (2) those on long-term sick leave or maternity leave; and (3) individuals experiencing major psychological stress during the survey period. This study was approved by the institutional ethics committee (Approval No. 2023-04-083-K01).

Based on conventional sample size estimation, a minimum of 5–10 times the number of independent variables is generally recommended for multivariate analysis (13). This study included 20 key variables, indicating a required sample size between 100 and 200 participants. To enhance statistical power and model stability, and considering an expected 20% rate of invalid responses, the final estimated sample size ranged from 125 to 250.

A total of 600 questionnaires were distributed, and 549 valid responses were obtained after excluding 37 questionnaires with patterned responses or logical inconsistencies, yielding a response rate of 91.5%.

Online data collection was conducted using the “Questionnaire Star” platform. The nursing departments and head nurses at the participating hospitals facilitated the survey process. Trained coordinators were responsible for questionnaire distribution and respondent guidance. Nurses completed the electronic questionnaire voluntarily during work breaks after being fully informed about the study purpose. The survey strictly adhered to principles of informed consent and anonymity.

To ensure data quality, several control measures were implemented. First, required-response settings and logical skip patterns were embedded in the questionnaire to minimize missing data and contradictory answers. Second, a minimum completion time of five minutes was enforced, and questionnaires with abnormally short durations were automatically flagged for exclusion. IP and device ID tracking were used to prevent duplicate submissions. All submitted questionnaires were manually reviewed, and those with highly patterned responses, logical errors, or duplicated content were excluded.

All collected data were independently verified and double-entered into a database. Statistical analyses were conducted using SPSS version 26.0 and Mplus version 7.4.

First, Latent Profile Analysis (LPA) was performed using Mplus 7.4 based on the five items of the WVS to identify latent subgroups of WPV experiences. Starting from a one-class model, additional classes were progressively added. Model fit was evaluated using multiple indices: Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and sample-size adjusted BIC (aBIC), with smaller values indicating better fit. Classification accuracy was assessed using entropy (range 0–1), with higher values indicating clearer class separation. Model comparisons were further validated using the Lo-Mendell-Rubin Test (LMRT) and the Bootstrap Likelihood Ratio Test (BLRT), with P < 0.05 indicating that the model fit was significantly improved compared to the previous class solution.

Second, based on the identified latent classes, further statistical analyses were conducted using SPSS 26.0. Categorical variables were described using frequencies and percentages, and group comparisons were made using Chi-square or rank-sum tests. Continuous variables were analyzed using t-tests or ANOVA, depending on distribution normality. Finally, latent profile membership was used as the dependent variable in a multivariate logistic regression model, incorporating statistically significant predictors from univariate analyses. A two-tailed P-value < 0.05 was considered statistically significant.

A total of 549 clinical nurses participated in this study. The majority were female (86.7%, n = 476). Age distribution was mainly between 36–45 years (32.8%), followed by 26–35 years (27.3%). Married nurses accounted for 69.2%, and unmarried nurses 20.2%. In terms of education, 76.3% held a bachelor’s degree and 12.9% had a master’s degree or higher. Regarding professional title, 46.1% were at the junior level and 43.5% at the intermediate level. Nearly half (47.4%) had over 10 years of clinical experience. Shift patterns showed that 44.8% worked 4–8 night shifts per month. Satisfaction with salary was rated “average” by 48.5% and “satisfied” by 37.5%. Contract-employed nurses comprised 72.9%, while 27.1% were on permanent positions (see Table 1 ).

The prevalence of workplace violence among nurses was 74.13%. The scores of Maslburnout Burnout Inventory, Connor-Davidson Resilience Scale, Emotional Labor Scale and Perceived Organizational Support Scale were (55.45 ± 21.31), (63.36 ± 15.24), (49.46 ± 9.26) and (42.88 ± 10.27), respectively.

Latent profile models with 1–5 classes were fitted based on nurses’ scores on the five WPV items from the past year (see Table 2 ). As the number of classes increased, fit indices such as AIC, BIC, and aBIC declined, indicating improved model fit. With a two-class model, the entropy was 0.915, and both LMRT and BLRT were statistically significant (P <0.001), suggesting excellent classification accuracy and discrimination. Although a three-class model achieved perfect entropy (1.000), LMRT was non-significant (P > 0.05), providing no support for its superiority over the two-class solution. Four- and five-class models showed further decreases in fit indices but generated classes under 5% of the sample and non-significant LMRT tests, indicating potential overfitting and instability.

Considering model fit, interpretability, practical application, and parsimony, the two-class model was deemed optimal, and the score distributions across WPV items were plotted.

As shown in Figure 1 , nurses in Class 1 scored significantly higher across all violence dimensions, especially in “emotional abuse” and “threats/intimidation.” This profile, comprising 27.69% of participants, was labeled the “high-frequency, multi-type violence” group. Class 2 exhibited lower scores across dimensions and was designated the “low-frequency, mild violence” group, accounting for 72.31%.

Univariate analyses of demographic variables by WPV profile are presented in Table 1 . Significant differences (P<0.05) were found between groups in sex, age, marital status, education level, employment type, professional title, years of clinical experience, monthly night shifts, and salary satisfaction.

Further comparison of psychological variables ( Table 3 ) revealed that the “high-frequency, multi-type violence” group scored significantly higher on all dimensions of occupational burnout (emotional exhaustion, depersonalization, reduced personal accomplishment) and total burnout scores, as well as on all aspects of emotional labor (surface acting, deep acting, genuine expression) (P<0.01). Conversely, this group scored significantly lower on psychological resilience (resilience, self-reliance, optimism) and perceived organizational support compared to the “low-frequency, mild violence” group (P < 0.01).

Using WPV profile as the dependent variable (coded as high-frequency, multi-type violence group = 1; low-frequency, mild violence group= 0), variables that were significant in univariate analysis were entered into a binary logistic regression model (coding details in Table 4 ). Results ( Table 5 ) indicated that nurses with a bachelor’s degree (OR = 3.196, P = 0.009) and those with a master’s degree or higher (OR = 5.364, P = 0.001) were more likely to belong to the high-frequency, multi-type violence group. Compared with nurses who were dissatisfied with their salary, those who rated it as “average” (OR = 0.311, P < 0.001) or “satisfied” (OR = 0.140, P < 0.001) had a significantly lower likelihood of being in that group. Additionally, higher resilience scores were associated with a reduced probability of high-frequency, multi-type violence (OR = 0.970, P = 0.001).

The results of this study revealed that the overall incidence of workplace violence (WPV) among the 549 clinical nurses surveyed was 74.13%, which is lower than the 86.7% reported by Li et al. (19), but significantly higher than the global average of 44.9% for nurses (20). This indicates that nurses in China still face a relatively high risk of exposure to violence. Such discrepancies may be attributable to differences in departmental risk levels, disparities in healthcare resource allocation, variation in violence identification criteria, and the robustness of organizational support systems.

Further analysis showed that the total WPV score was 3.02 ± 3.20, suggesting a moderate overall level of violence. However, the distribution of item scores varied significantly, indicating that a subset of nurses experienced more severe forms of violence. In this context, mean scores alone fail to capture the internal heterogeneity of risk within the population, highlighting the need for structural modeling approaches to identify latent subtypes.

To address this, we employed Latent Profile Analysis (LPA), which identified two distinct patterns of WPV experiences. The “high-frequency, multi-type violence” group comprised 27.7% of the sample, with notably elevated scores across multiple dimensions such as verbal abuse, sexual harassment, and threats/intimidation. This pattern indicates cumulative and diverse exposure to violence, which may be associated with increased psychological stress and a heightened risk of occupational burnout (21). Without effective interventions and organizational support, this subgroup is more susceptible to negative professional outcomes, including emotional exhaustion, work avoidance, and even voluntary turnover—necessitating targeted identification and focused interventions (22).

In contrast, the “low-frequency, mild violence” group accounted for 72.3% of participants. Although their overall WPV scores were relatively low, this does not imply complete safety. Some nurses in this group may lack sufficient skills in risk recognition and response, making them vulnerable to unrecognized or subtle forms of violence. These individuals may hover at the edge of potential risk and thus require enhanced coping strategies, psychological resilience, and early warning system development (23, 24).

In summary, the use of LPA effectively uncovered structural heterogeneity in WPV experiences among nurses, addressing the limitations of traditional mean-based analyses in identifying high-risk subpopulations. The findings suggest that nurses with high-frequency WPV exposure should be prioritized for targeted interventions. Meanwhile, those in the low-violence group should also receive support aimed at strengthening self-protective capabilities through training, institutional safeguards, and psychological preparedness.

This study further explored the influencing factors associated with latent types of workplace violence (WPV) among nurses. The results indicated that educational attainment, salary satisfaction, and psychological resilience played significant roles in distinguishing between the “high-frequency, multi-type violence” and “low-frequency, mild violence” groups. These findings suggest that WPV exposure is not solely driven by external circumstances; rather, individual characteristics and internal psychological resources also exert substantial influence.