A cross-sectional study was conducted. A convenience sample was recruited of 2,624 mental health nurses from psychiatric hospitals across seven regions in China. The participating hospitals were located in seven major geographic regions of China (e.g., North China, South China, East China, etc.), ensuring a broad geographic coverage that enhances the national representativeness of the sample. In the Chinese healthcare system, hospitals are officially classified into levels (Level-1, Level-2, and Level-3) based on their scale, technological sophistication, and service complexity, which is associated with differential resource allocation and management capacity. The present study deliberately excluded Level-1 hospitals (which are primary care institutions that typically lack specialized psychiatric inpatient units) and focused solely on Level-2 and Level-3 hospitals. This strategy was adopted to ensure homogeneity in the care setting and to facilitate a valid comparison of PSC perceptions between the two higher-level hospitals that form the backbone of China's specialized psychiatric inpatient care. Level-3 hospitals represent tertiary referral centers with greater resources and specialized expertise, while Level-2 hospitals serve as regional centers with more general services. This distinction provides important context for interpreting the observed disparities in PSC perceptions. Inclusion criteria comprised: (1) registered nurses working in psychiatric inpatient settings, (2) at least six months of tenure in their current unit. Participants completed the survey anonymously via the Wenjuanxing platform. After excluding 100 participants due to incomplete data or not meeting the inclusion criteria, the final analytical sample comprised 2,524 nurses.

The survey instrument comprised two main sections: (1) demographic and work-related variables, and (2) the Chinese version of the Hospital Survey on Patient Safety Culture (HSOPSC). The HSOPSC is a widely validated tool containing 42 items that measure 12 composites, encompassing 10 safety culture dimensions and 2 outcome dimensions (15). Items are scored on a five-point Likert scale. Following established methodology, the percentage of positive responses (PPR) for each dimension was calculated (15, 25).

We used a validated Chinese adaptation of the HSOPSC, which has been extensively applied in prior research within the Chinese context (24, 26, 27). This questionnaire was distributed as an anonymous online survey to the defined study sample via the Wenjuanxing platform between October 2019 and October 2023.

Data analysis was performed using SPSS software (version 25.0). Descriptive statistics were used to summarize participant characteristics and the Percent Positive Response (PPR) for each safety culture dimension. Continuous variables with normal distribution are presented as mean ± standard deviation (SD), while those with skewed distribution are presented as median (interquartile range), and categorical variables as frequencies and percentages (%). The PPR was calculated as the percentage of positive responses (e.g., “agree/strongly agree” or “always/most of the time”) for each dimension following established methodology (15).

The choice of specific statistical tests was guided by the nature and distribution of the data. Multiple linear regression analysis was employed to identify predictors of the outcome dimensions (“Overall Perception of Safety” and “Frequency of Events Reported”) while controlling for potential confounders. For subgroup analyses, non-parametric tests were selected as the PPR data were not normally distributed. The Kruskal–Wallis H-test was used to compare PPR across the two hospital levels (Level-2 vs. Level-3) for each of the 12 dimensions, and the Mann–Whitney U-test was used for comparisons between genders.

For the regression models, model fit was assessed using the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), with lower values indicating a better fit (28). Multicollinearity was checked via the variance inflation factor (VIF), and a value of less than 5 was considered acceptable (29). Statistical significance was set at a p-value of less than 0.05 for all tests.

Ethical approval was obtained from the Ethics Committee of the University (Project Identification Code: 20180206). Participation was anonymous and voluntary.

A total of 2,524 participants from Level-2 and Level-3 hospitals were included in final analysis. The sample was predominantly female (87.2%), with a mean age of 32.1 years (SD = 7.6). Over half of the participants held an undergraduate degree (58.9%), and the majority were married (69.5%). Regarding professional experience, 22.1% had 1–5 years of experience, and 24.0% had 6–10 years. A substantial proportion of nurses (55.3%) reported working more than 30 min of overtime per day. The detailed demographic and work-related characteristics of the participants are summarized in Table 1.

The analysis of PPR across the 12 PSC dimensions revealed substantial variation (Table 2). Based on established criteria (25), three dimensions were identified as strengths (PPR≥75%): “Teamwork within units” (83.3%), “Organizational learning/continuous improvement” (82.8%), and “Feedback and communications about error” (75.8%). Conversely, three dimensions, with PPR below 50%, were categorized as requiring improvement: “Staffing” (43.3%), “Nonpunitive response to error” (46.6%), and “Frequency of events reported” (49.3%). The remaining six dimensions had PPR ranging from 50% to 75%.

The Kruskal–Wallis H-test revealed significant differences in PPR between Level-2 and Level-3 hospitals across several patient safety culture dimensions (Table 3). Level-3 hospitals demonstrated significantly higher PPR than Level-2 hospitals in five dimensions, including “Feedback and Communication about Error”, “Communication Openness”, and “Hospital Management Support for Patient Safety”. In contrast, for the dimension “Nonpunitive Response to Error”, Level-2 hospitals had a marginally higher PPR, but this difference was not statistically significant (P = 0.451).

The multiple regression analysis (Table 4) showed that eight dimensions of PSC were significant predictors of “Overall Perceptions of Safety”, explaining 15.8% of the variance. For “Frequency of Events Reported”, gender, years of experience, and three safety culture dimensions were significant predictors, collectively accounting for 41.2% of the variance. Specifically, daily overtime hours were a significant negative predictor of the overall PSC score [β = −0.22, 95% CI (−0.38, −0.06), P = 0.007].

Our results corroborate extant international literature, confirming that “Staffing”, “Nonpunitive Response to Error”, and the correlated “Frequency of Events Reported” represent pervasive and interconnected challenges in psychiatric settings (20, 22, 30).

The subgroup analysis revealed statistically significant disparities in PSC perceptions between Level-2 and Level-3 hospitals. Level-3 hospitals demonstrated superior performance across key dimensions such as “Hospital management support”, “Communication openness”, and “Feedback and communication”, underscoring the influence of underlying structural determinants associated with hospital hierarchy.

This divergence may be attributed to a multifactorial etiology. First, disparities in financial and institutional resources likely enable Level-3 hospitals to implement more robust safety-specific training and infrastructure (26). Second, variations in managerial sophistication and leadership practices may foster greater transparency and supportive supervision in Level-3 settings. Third, differences in human capital, as evidenced by the higher proportion of degree-qualified and experienced nurses in Level-3 hospitals (Table 1), may contribute to a workforce more adept in systems-based safety practices.

These findings carry significant policy implications, indicating that PSC enhancement initiatives must be stratified. Interventions in Level-2 hospitals should concentrate on foundational elements, such as strengthening leadership engagement, establishing basic communication protocols, and implementing nonpunitive reporting systems. In contrast, Level-3 hospitals can leverage their advantages to focus on advanced, iterative quality improvement and sophisticated root-cause analysis.

Finally, while our regression models were statistically significant, the modest explained variance (Adjusted R²= 15.8% for “Overall Perceptions of Safety”) warrants comment. This is consistent with the complex, multifactorial nature of PSC, which is influenced by numerous factors beyond our model's scope (35). In contrast, the model for the “Frequency of Events Reported” accounted for a substantially higher proportion of variance (Adjusted R2 = 41.2%), suggesting that the predictors in our framework are more strongly associated with this behavioral outcome than with global perceptions. Despite this difference, the remaining unexplained variance for both outcomes underscores the influence of other unmeasured contextual or individual factors. Nonetheless, the primary value of our analysis lies in identifying significant and actionable predictors, such as staffing and nonpunitive response to error, which remain crucial for targeted interventions.

This study has several limitations that should be considered when interpreting the findings. First, its cross-sectional design precludes the inference of causality. Second, the deliberate exclusion of Level-1 hospitals, while necessary to ensure sample homogeneity, limits the generalizability of our findings to the entire spectrum of psychiatric hospitals in China. Third, the exclusive reliance on nurses' perspectives omits the viewpoints of other clinical and non-clinical staff (e.g., physicians, aides, administrators), which would provide a more holistic cultural assessment. Finally, although our sample of 2,524 nurses from multiple regions is substantial, the use of a convenience sampling method may limit the generalizability of the findings to all psychiatric nurses in China and could introduce selection bias. Therefore, while the large sample size is a strength, the findings should be generalized to the broader national context of Chinese psychiatric nurses with caution.

Based on these limitations, future research should prioritize several directions. First, and foremost, while the large sample size is a strength, the findings should be generalized with caution. Nationally representative studies employing probability sampling methods are warranted to enhance the external validity of the findings. Second, longitudinal or mixed-methods designs are needed to elucidate the causal mechanisms underlying the inter-hospital disparities identified here. Third, there is a particular need for intervention studies that develop and evaluate tiered improvement strategies tailored to the specific needs and resources of different hospital levels. Finally, future investigations should explore the role of specific leadership behaviors and organizational climates that mediate the relationship between hospital level and PSC.