Direct and Indirect Statistical Associations Between Knowledge, Attitudes, and Behaviours Regarding Subsyndromal Delirium in Psychiatric Nurses: A Cross‐Sectional Study Authors: Lihua Liu, Dongmei Xu, Yuqi Li, Wei Zhan, Wanting Li, Mengqian Zhang, Guangling Shi, Na Zhang, Xinmei Wan, Jing Peng, Jing Gao Published: 2026 ABSTRACT ABSTRACTAimTo examine the indirect association between subsyndromal delirium knowledge and clinical behaviour among psychiatric nurses, with attitude as a linking factor.DesignA cross‐sectional survey design.ResultsThe mean scores for SSD self‐perceived knowledge, attitudes, and practices were 17.32 ± 6.66, 32.55 ± 4.54, and 30.11 ± 6.08, respectively. All three variables showed positive correlations (p < 0.01). Attitudes exhibited a significant association within the knowledge‐practice relationship. The direct association between knowledge and behaviour (81.0%) proved to be more robust than the indirect association (19.0%). The moderating role of professional title was not statistically significant (p = 0.058). Given the cross‐sectional design, the direct and indirect comparisons are descriptive only and should not be interpreted causally.ConclusionAttitude showed a relatively small indirect association with SSD knowledge and behaviour among psychiatric nurses, though the direct association was considerably more robust. Despite favourable attitudes, self‐perceived knowledge levels remained notably low. The findings suggest that system‐level behavioural triggers (e.g., electronic health record prompts, standardized handover checklists) and knowledge education both offer potentially promising directions for future intervention research, though their relative effectiveness requires further investigation. However, the convenience sample was predominantly composed of tertiary hospital nurses (73.2%), limiting generalizability to primary‐level settings. Due to the cross‐sectional nature of this study, the findings indicate associations and necessitate longitudinal validation.Implications for the Profession and/or Patient CareThe direct knowledge–behaviour association suggests that system‐level workflow factors may be relevant intervention targets. Low self‐perceived knowledge scores also indicate a continued need for knowledge education as a complementary goal.ImpactThe predominance of the direct knowledge–behaviour association over the attitudinal pathway suggests that system‐level workflow factors may play a key role in sustaining clinical practices, whereas self‐perceived knowledge deficits point to the continued need for knowledge education. These findings provide empirical support for considering both system‐level and educational approaches in SSD nursing care.Reporting MethodThis study adhered to the STROBE guidelines for cross‐sectional studies.Patient or Public ContributionNo patient or public contribution. Introduction Delirium, an acute syndrome marked by impaired consciousness and cognition, is associated with delayed management, which leads to increased morbidity, hospitalization, and mortality rates (Huang et al. 2024). It significantly affects quality of life and long‐term cognitive function in elderly inpatients (Inouye et al. 2013). However, its subtle and fluctuating presentation, which can emerge within hours, complicates recognition, leading to delayed intervention and an increased disease burden (Gao et al. 2022). Subsyndromal delirium (SSD) represents an intermediate and prevalent state compared to full delirium. Positioned between normal cognition and full delirium, SSD is characterized by the presence of at least one delirium symptom, though not enough to fulfil the complete diagnostic criteria (Cole et al. 2012). It affects up to 33.9% of ICU patients and 23% of elderly individuals, with rates reaching two‐thirds among older adults undergoing joint replacement (Cole et al. 2012; Denny and Such 2018; Yamada et al. 2018; Dalli et al. 2025). Crucially, SSD is not benign: it confers a 1.26‐fold higher mortality risk (Diwell et al. 2018). Additionally, it is associated with transient post‐discharge cognitive decline and predicts long‐term functional and cognitive impairment (Paulino et al. 2023). Delirium is also prevalent in psychiatric settings, with an incidence ranging from 18.2% to 33.6% (Huang et al. 2024). Yet, recognition remains a challenge, with nearly 50% of cases undetected (Hercus and Hudaib 2020). Contributing factors include diagnostic overshadowing, the subtle presentation of hypoactive delirium, limited assessment tools and time, and knowledge gaps among clinicians. Critically, research on identifying and managing SSD in psychiatric settings remains scarce. Nurses, given their close patient contact, are crucial for the early identification and intervention of SSD (Huang et al. 2024; Denny and Such 2018). Research on psychiatric nurses' management of SSD is limited. Similar to ICU nurses, who exhibit knowledge and skill deficits in SSD evaluation (Gao et al. 2022), psychiatric nurses demonstrate limited knowledge of delirium and lack systematic screening practices (Huang et al. 2024). Key gaps include inadequate knowledge, low confidence in identification (affecting up to 85% of nurses), and the absence of routine screening, which leads to inconsistent use of assessment tools. Although about 99% express willingness to be trained, structured models to translate knowledge into effective practice are lacking. While Gao et al. (Gao et al. 2022). identified barriers to SSD assessment among ICU nurses, the knowledge, attitudes, and practices of psychiatric nurses—and the statistical pathways among these constructs—remain underexplored. The Knowledge‐Attitude‐Practice (KAP) model is one of the most widely adopted theoretical frameworks in health behaviour research. Grounded in the “cognition–affect–behaviour” linear logic, it posits that knowledge drives attitude, which in turn drives practice (Tandon et al. 2025). However, this model's applicability may be limited in contexts where behaviour is shaped not only by individual cognition but also by external protocols and routines. The Hierarchical Model of Driving Behaviour proposed by Michon and Keskinen (Keskinen 1996; Michon 1985) offers a possible alternative perspective, organizing behaviour into four hierarchical levels—the social/values level, the strategic level, the tactical level, and the operational level. Given the exploratory nature of this study, this model was not pre‐specified, but was introduced as a tentative post hoc interpretive lens after data collection and analysis to offer supplementary perspectives on the observed variable associations and to generate exploratory hypotheses. Based on the above theoretical divergence, this study adopts the KAP model as a reference framework and introduces the Hierarchical Model of Driving Behaviour as a post hoc interpretive perspective. Specifically, we aim to describe the current status of psychiatric nurses' SSD knowledge, attitudes, and practices, compare the strength of direct versus indirect statistical associations, and generate exploratory hypotheses based on the observed patterns. Materials and Methods Participants Using the convenience sampling method, psychiatric nurses from five provinces in China were selected as the research subjects and investigated between January and February 2025. This study was designed as a multi‐center cross‐sectional study. Inclusion criteria were: (1) Engaged in psychiatric nursing work; (2) Held a nursing professional qualification certificate; (3) Had worked in their institution for at least 1 year; (4) Voluntarily participated in the study. Exclusion criteria were: (1) Internships, continuing education, and rotating nurses; (2) Nurses on sick or maternity leave during the investigation period. Measures Demographic Information Questionnaire A self‐designed questionnaire was utilized to collect participants' demographic information, including gender, age, work experience, education level, professional title, and hospital grade. The Psychiatric Nurse SSD Knowledge‐Attitude‐Practice Scale This questionnaire, developed by our research group, has undergone rigorous reliability and validity testing (Li et al. 2026). It includes general information and three subscales on SSD knowledge, attitudes, and practices. Knowledge measures nurses' self‐perceived understanding of SSD, including its definitions, risk factors, clinical manifestations, assessment tools, and interventions. Attitudes reflect nurses' subjective beliefs regarding the importance they place on SSD, their willingness to prevent and treat it, and their self‐efficacy. Practices involve nurses' actual assessments, prevention, and nursing behaviours in clinical settings. Each subscale consists of eight items, using a 5‐point Likert scale (8–40 points). Higher scores indicate greater knowledge, more positive attitudes, and better practices. In this study, the Cronbach's α coefficients for the knowledge, attitude, and practice subscales were 0.977, 0.967, and 0.909, respectively, with a total Cronbach's α coefficient of 0.939 and a split‐half reliability coefficient of 0.697. Content validity was satisfactory (S‐CVI = 0.958; I‐CVI ranged from 0.833 to 1.000). Exploratory factor analysis extracted three factors that accounted for 80.003% of the cumulative variance, and confirmatory factor analysis demonstrated that the model fit was satisfactory (CFI = 0.939, RMSEA = 0.068). The questionnaire used in this study is the 24‐item version validated by Li et al. (Li et al. 2026). The present study was conducted independently using a separate, larger sample from five provinces, and addresses a distinct research question—examining the statistical associations among knowledge, attitudes, and behaviours—which differs from the psychometric validation purpose of the 2026 publication. The two studies therefore do not share the same participant data. Data Collection The questionnaire was uploaded to a platform to generate a QR code. After obtaining consent from the nursing or research department of each hospital, the QR code was sent via WeChat to the appropriate contact person, who then forwarded it to the psychiatric nurse management group. The survey was conducted anonymously, and the purpose and significance of the study were explained to the nurses, along with the requirements and precautions for completing the questionnaire, to ensure informed consent and cooperation. All questions were mandatory, and each WeChat account was allowed to submit only one questionnaire. Respondents were given a time window of 150 to 900 s to complete the survey. This study utilized a cross‐sectional survey design. Following Kendall's principle, which suggests a sample size of 10–20 times the number of variables, the 30 variables in the study indicated a minimum sample size of 300. To ensure sufficient statistical power and account for potential invalid responses, we recruited 1620 participants, significantly exceeding this minimum requirement. After excluding questionnaires with invalid response patterns, 1599 valid responses remained, resulting in a valid response rate of 98.7%. These valid questionnaires were then subjected to statistical analysis. Data Analysis This study is exploratory in nature, testing associations based on the KAP theoretical framework. The mediation model was specified prior to analysis. Data were analysed using SPSS 26.0. Quantitative data were described using Mean ± Standard Deviation (M ± SD) or rank mean (R), while qualitative data were characterized by frequency (percentage). The knowledge scale distribution was skewed toward the lower end, with 76.5% of participants scoring below the midpoint (20/40), limiting the subscale's discriminative ability. To assess correlations between variables, Pearson correlation was employed. Harman's one‐way method was applied to examine potential common method bias in the questionnaire data. Comparisons across demographic variables with equal variances (age, hospital grade) used one‐way ANOVA with Bonferroni post hoc (α' = 0.05/comparisons). The variable with unequal variances (professional title) used Kruskal‐Wallis with Bonferroni‐corrected pairwise comparisons. To test the proposed mediation model with an exploratory moderator, we utilized Model 7 of the SPSS PROCESS (v4.1) developed by Hayes. To enhance the interpretation of relative effect sizes among variables, all continuous independent, dependent and mediator variables were standardized to Z‐scores. The moderator variable (professional title), an ordinal categorical variable, was coded using Helmert contrasts, which compares each level against the combined mean of subsequent levels. A sensitivity analysis using sequential coding was also performed. Given the exploratory nature of this analysis, this coding scheme was adopted without assuming a specific ordinal trend. In this model, SSD knowledge served as the independent variable (X), SSD behaviour as the dependent variable (Y), SSD attitude as the mediator (M), and professional title as the moderator (W) on the knowledge–attitude path. Based on the results of the univariate analysis, age and hospital grade were included as covariates to control for potential confounding effects. Additionally, a bootstrap resampling method with 5000 repetitions was employed to estimate 95% confidence intervals (CI) for the effects. Effects were deemed statistically significant if the 95% CI did not include zero, while the significance level for other tests was set at p < 0.05. This study followed the STROBE reporting guideline for cross‐sectional studies (see Data S1). Results Demographic Characteristics and Univariate Analysis The characteristics of the participants are summarized in Table 1. A univariate analysis was conducted using the SSD behavioural scores of psychiatric nurses as the dependent variable. This analysis revealed significant differences in SSD behavioural scores based on age, professional title, and hospital grade (p < 0.05). However, no significant differences were found for other general information (p > 0.05). TABLE 1Demographic characteristics and univariate analysis of SSD behavioural scores of psychiatric nurses.CharacteristicsCategories N %M ± SD/R t/F/H GenderMale31619.830.15 ± 6.520.135Female128380.230.10 ± 5.97Age (years)≤ 3038424.031.16 ± 6.14^a7.710**31–4076347.729.84 ± 6.27^b> 4045228.329.67 ± 5.59^bWork experience≤ 10 years69143.230.43 ± 6.331.91611–20 years61238.329.77 ± 5.97> 20 years29618.530.07 ± 5.68Education levelBachelor or above124577.930.11 ± 6.030.015Junior college31319.630.08 ± 6.23Secondary school412.630.11 ± 6.67Professional titleRegistered nurse23314.6876.24^a9.544*Senior nurse51332.1810.18^abSupervisor nurse70143.8773.38^b(Associate) Chief Nurse1529.5771.54^abHospital gradeTertiary hospital117073.230.48 ± 5.97^a9.640**Secondary hospital37223.329.26 ± 6.17^bPrimary hospital573.627.93 ± 6.82^b Note: Symbols ^a, ^b and ^ab are used to represent differences in multiple comparisons. Where: identical symbols indicate no significant difference (p > 0.05); no identical symbols indicate a significant difference (p < 0.05).Abbreviations: M, mean; R, mean ranks (presented for variables analysed by Kruskal‐Wallis test); SD, standard deviation.*p < 0.05, **p < 0.01. Demographic characteristics and univariate analysis of SSD behavioural scores of psychiatric nurses. Note: Symbols ^a, ^b and ^ab are used to represent differences in multiple comparisons. Where: identical symbols indicate no significant difference (p > 0.05); no identical symbols indicate a significant difference (p < 0.05). Abbreviations: M, mean; R, mean ranks (presented for variables analysed by Kruskal‐Wallis test); SD, standard deviation. *p < 0.05, **p < 0.01. Common Method Variance Test The common method bias was evaluated using the Harman's one‐way method. The results indicated that three factors had eigenvalues greater than 1, with the first factor explaining 41.0% of the variance. According to Podsakoff and Organ (Podsakoff and Organ 1986), the variance explained by the one‐way test using EFA (unrotated) should not exceed 50% to avoid serious common method bias. Therefore, this study suggests that common method bias is not a significant concern (Tang and Wen 2020). Correlations of the Main Variables Table 2 presents the SSD variable scores and correlation analysis among psychiatric nurses. The mean scores for self‐perceived knowledge, attitude, and behaviour were 17.32 ± 6.66, 32.55 ± 4.54, and 30.11 ± 6.08, respectively. Table S1 displays the item‐level mean scores for the knowledge, attitude, and practice subscales. The knowledge subscale showed a skewed distribution toward the lower end, with a mean item score of 2.16 (below the midpoint of 3). In contrast, no ceiling effects were observed for the attitude or behaviour subscales. Pearson correlation analysis, also shown in Table 2, revealed significant positive correlations: SSD self‐perceived knowledge was positively correlated with attitude (r = 0.265, p < 0.01) and behaviour (r = 0.429, p < 0.01). Additionally, attitude was significantly and positively correlated with behaviour (r = 0.407, p < 0.01). TABLE 2Correlation analysis.VariablesM ± SDKnowledgeAttitudeBehaviourKnowledge17.32 ± 6.661Attitude32.55 ± 4.540.265**1Behaviour30.11 ± 6.080.429**0.407**1**p < 0.01. Correlation analysis. **p < 0.01. Testing the Proposed Statistical Model After standardizing variables, we tested the model using PROCESS Model 7, including age and hospital grade as covariates. Professional title was specified as an exploratory moderator. The results are summarized in Table 3 and Figure 1. The total association of SSD knowledge on behaviour was significant (total association = 0.422, p < 0.01). In the mediation model, SSD knowledge showed a positive association with attitude (β = 0.256, p < 0.01), and attitude, in turn, showed a positive association with behaviour (β = 0.312, p < 0.01). The direct association of knowledge on behaviour remained significant after including the mediator (β = 0.341, p < 0.01). The overall interaction between SSD knowledge and professional title on SSD attitude was not statistically significant (p = 0.058), as confirmed by both Helmert contrast coding and sensitivity analysis using sequential coding (p = 0.058). Although exploratory Helmert contrasts revealed a significant difference between senior nurses and registered nurses in the first contrast (β = −0.174, p < 0.05; confirmed by sequential coding: β = −0.175, p < 0.05), we refrain from overinterpreting this single significant contrast given the non‐significant overall interaction. These exploratory findings suggest that professional title does not show a robust moderating association in this sample, and the observed contrast differences require further validation in future studies. TABLE 3Equation of the mediation effect of SSD attitudes between knowledge and behaviour.Outcome variablePredictor β SE95% CI t R 2 F AttitudeAge−0.1870.043[−0.271, −0.102]−4.351**0.08716.772**Hospital grade−0.0320.024[−0.080, 0.016]−1.320W10.2240.084[0.059, 0.389]2.660**W20.1780.071[0.039, 0.318]2.502*W30.1630.090[−0.014, 0.340]1.801Knowledge0.2560.029[0.199, 0.314]8.728**Knowledge×W1−0.1740.066[−0303, −0.046]−2.655**Knowledge×W20.0010.063[−0.122, 0.124]0.015Knowledge×W3−0.0750.094[−0.260, 0.110]−0.794BehaviourAge−0.0070.030[−0.066, 0.052]−0.2390.283157.238Hospital grade−0.0820.021[−0.124, −0.040]−3.834**Knowledge0.3410.022[0.298, 0.385]15.373**Attitude0.3120.022[0.269, 0.356]14.146** Note: W1, W2, W3 represent the three Helmert‐coded contrasts for professional title (four levels: registered nurse, senior nurse, supervisor nurse, associate chief nurse or above). Helmert coding compares each level with the mean of all higher levels. The three interaction terms as a set were not significant (p = 0.058); therefore, the isolated significant Knowledge×W1 contrast does not indicate an overall moderating effect of professional title.*p < 0.05; **p < 0.01. Equation of the mediation effect of SSD attitudes between knowledge and behaviour. Note: W1, W2, W3 represent the three Helmert‐coded contrasts for professional title (four levels: registered nurse, senior nurse, supervisor nurse, associate chief nurse or above). Helmert coding compares each level with the mean of all higher levels. The three interaction terms as a set were not significant (p = 0.058); therefore, the isolated significant Knowledge×W1 contrast does not indicate an overall moderating effect of professional title. *p < 0.05; **p < 0.01. Indirect association model of SSD knowledge and behaviour among psychiatric nurses. Standardized path coefficients are shown. SSD = Subsyndromal Delirium. **p < 0.01. We employed PROCESS Model 7 to test the proposed model, with professional title specified as an exploratory moderator on the knowledge–attitude path (see Figure 1). The overall moderating association was not significant (p = 0.058). Discussion A key finding of this study is a pattern of associations that is not entirely consistent with the expectations of the classical KAP model: psychiatric nurses' SSD self‐perceived knowledge score was only 17.32/40 (43.3%), while their attitude and practice scores were 32.55/40 (81.4%) and 30.11/40 (75.3%), respectively. This pattern is consistent with the findings of Shu et al. (Shu et al. 2025), who reported similar scoring rates (self‐perceived knowledge 48.35%, attitude 81.29%, practice 68.84%) in primary hospital nurses. The striking similarity across two independent samples confirms that the pattern of favourable attitudes coexisting with low self‐perceived knowledge is not unique to our sample, suggesting that positive attitudes may be a general strength among nursing populations, while self‐perceived knowledge gaps persist. The Hierarchical Model of Driving Behaviour (Keskinen 1996; Michon 1985) offers a possible explanation: behaviours at the operational and tactical levels are largely triggered by environmental cues, standardized protocols, and team routines, rather than by complex cognitive evaluation processes. Therefore, the relatively high SSD practice scores observed in this study could be speculatively attributed to the potential influence of unmeasured system‐level factors (e.g., routine workflows, team norms), rather than being driven solely by individual knowledge reserves. However, as these factors were not directly assessed in the present study, this interpretation remains speculative and should be viewed as hypothesis‐generating rather than confirmatory. Meanwhile, the high attitude scores may reflect a general professional orientation toward patient safety at the “social/values level” among the nursing workforce, rather than a deep understanding of SSD‐specific knowledge. While Gao et al. (Gao et al. 2022). focused on barriers to SSD assessment among ICU nurses, the present study specifically targets psychiatric nurses and examines the statistical pathways among knowledge, attitudes, and behaviours using a validated 24‐item instrument and a large multi‐province sample (N = 1599), rather than merely describing barriers. The results of the statistical path analysis provide additional descriptive information: the direct statistical association between knowledge and behaviour accounted for 81.0% of the total association, while the indirect association via attitudes accounted for 19.0%. The relatively small proportion of the indirect pathway suggests that, in this sample, the attitudinal pathway may play a more limited role than the KAP model's sequential logic might imply, though this finding does not constitute a test of the model's validity. This pattern may be speculatively interpreted as reflecting that behavioural compliance may be more influenced by routine workflows than by the attitudinal pathway. However, this interpretation remains speculative and requires direct empirical testing in future studies. However, these findings should be interpreted in light of a significant measurement constraint: with 76.5% of participants scoring below the midpoint, the knowledge measure lacked sufficient sensitivity at the lower end of the continuum. Consequently, the observed correlations may be attenuated, and the 19.0% indirect proportion may also be influenced by this restricted range. The pathway estimates reported here are therefore likely conservative and should be interpreted as exploratory rather than definitive. This study found that knowledge, professional title, age, and hospital grade collectively accounted for only 8.7% of the variance in attitude (R 2 = 0.087). This low R‐squared value suggests that most of the variability in SSD attitude may be influenced by other factors—such as organizational support, work stress, and self‐efficacy—though these possibilities were not directly measured in the present study and warrant further investigation. Furthermore, the standardized coefficient of self‐perceived knowledge on attitude was only 0.256, indicating that knowledge level is just one of many contributing factors. The moderating role of professional title on the knowledge–attitude relationship was not supported, possibly because knowledge levels were uniformly low across all title groups, which may have limited the ability to detect moderation. Univariate analysis indicated that younger nurses (≤ 30 years) and those employed in tertiary hospitals demonstrated superior SSD behaviours. As SSD has only recently collected significant attention over the past decade (Cole et al. 2012), it is more probable that younger nurses have undergone recent training on this subject. Moreover, tertiary hospitals generally offer a greater abundance of training resources and practical platforms (Li et al. 2017), thereby fostering a more conducive environment for nurses to effectively apply their knowledge. Based on item‐level findings (Table S1), the lowest self‐perceived knowledge ratings were for assessment tools (K4: 7.0%) and nursing priorities (K6: 9.1%); these areas could be prioritized in educational interventions, while system‐level supports (e.g., protocols and decision aids) may facilitate the translation of knowledge into practice. The lowest practice endorsement rates are for timely reporting (P2: 46.1%) and care planning (P3: 58.2%); managers could consider strengthening system‐level supports such as structured reporting protocols and decision aids. Item A8 showed a high endorsement rate (87%), indicating that most nurses perceived their current SSD knowledge as sufficient for clinical needs. Finally, the advantage of younger and tertiary hospital nurses reinforces that ongoing educational opportunities and supportive environments are both important; however, this interpretation remains speculative and requires direct empirical testing in future studies. Implications for Nursing Practice The direct knowledge–behaviour association (81.0%) substantially outweighed the indirect attitudinal pathway (19.0%), suggesting that SSD‐related nursing behaviours may be associated with unmeasured system‐level factors (e.g., routine workflows, electronic health record prompts, team norms) rather than being driven solely by individual cognitive processing. Given that knowledge was self‐perceived rather than objectively assessed, low scores may reflect low confidence or conservative responding alongside factual knowledge gaps. Thus, system‐level behavioural triggers (e.g., embedding SSD assessment into mandatory documentation fields, standardized handover checklists, and team‐based supervision) warrant exploration as potential intervention targets in future research, while knowledge education—particularly to strengthen factual understanding and confidence—should not be overlooked and may serve as a complementary long‐term goal. However, the relative effectiveness of these approaches requires further testing. Limitations First, the cross‐sectional design precludes causal inferences. The indirect association should be interpreted as a statistical association rather than evidence of a causal mechanism. All findings are exploratory and require validation in longitudinal or experimental designs. Second, online convenience sampling may have introduced self‐selection bias. Although the questionnaire was distributed anonymously, tertiary hospital nurses comprised 73.2% of the sample, while primary‐level nurses were underrepresented (3.6%). As tertiary hospital nurses typically have access to more training resources and better clinical support, this study may have overestimated SSD self‐perceived knowledge levels and practice behaviours; therefore, caution is needed when generalizing the findings to primary‐level hospitals. Future studies could employ stratified sampling to ensure proportional representation across hospital tiers and enhance the generalizability of findings. Third, knowledge subscale scores were generally low (76.5% of participants scored < 20/40), which may have attenuated correlations and biased the indirect association estimates. This subscale measured self‐perceived rather than objectively assessed knowledge; therefore, low scores may reflect not only actual knowledge gaps but also low confidence or conservative responding, rather than definitive factual knowledge deficits. This generally low distribution may limit the subscale's discriminatory power and attenuate its correlations with other variables. Future studies could consider supplementing self‐report measures with objective knowledge tests to more comprehensively assess nurses' SSD knowledge. Fourth, common method bias may be a concern, as all data were self‐reported at a single time point. Although Harman's test showed that the first factor explained 41.0% of the variance, this test is not highly sensitive, and inflation effects cannot be ruled out. Thus, observed associations—particularly the indirect association—may be inflated by shared method variance. The indirect effect proportion of 19.0% should therefore be interpreted as an upper‐bound estimate rather than a precise population parameter. Therefore, caution should be exercised when interpreting the findings of this study. Future research could combine objective knowledge assessments with qualitative interviews and incorporate additional variables (e.g., work stress, self‐efficacy) to improve explanatory power, as well as collect data from multiple sources or at multiple time points to obtain more conservative estimates and mitigate common method bias. Conclusion This exploratory study revealed an indirect association between SSD knowledge and behaviour among psychiatric nurses, with attitude as a linking factor; however, the direct association (81.0%) was larger in magnitude than the indirect association (19.0%). Despite favourable attitudes, self‐perceived knowledge levels remained notably low. System‐level behavioural triggers and knowledge education both offer potentially promising directions for intervention development, though their relative effectiveness remains to be tested. Given the cross‐sectional design, all findings indicate statistical associations and require longitudinal or experimental validation. Author Contributions Lihua Liu: investigation, formal analysis, writing – original draft, software. Dongmei Xu: visualization, supervision, resources. Yuqi Li: validation, investigation, data curation. Wei Zhan: investigation, validation, methodology. Wanting Li: investigation, methodology, validation. Mengqian Zhang: investigation, formal analysis. Guangling Shi: methodology, software. Na Zhang: project administration, conceptualization. Xinmei Wan: investigation, conceptualization. Jing Peng: investigation. Jing Gao: supervision, funding acquisition, writing – review and editing. Funding This work was supported by the Beijing Public Health Talent Project (Academic Backbone ‐03‐45). Disclosure Statistics verification statement: The manuscript has been checked prior to submission by an expert statistician. The statistician's name and email address are provided below: Name of statistician: Dr. Jiahua Xu. Email address of statistician: xujiahuapsy@gmail.com. Ethics Statement This study strictly adhered to the principles of confidentiality and anonymity. First, the researchers obtained approval from the head of the nursing department at the participating hospital; subsequently, nursing assistants detailed the study objectives, methodology, and associated risks to potential participants via an online management group, explicitly informing them of their voluntary participation. All data were anonymized, coded, and processed. This study was approved by the Ethics Committee of Beijing Huilongguan Hospital (approval no. 2024–72‐Ke; date: September 30, 2024). Conflicts of Interest The authors declare no conflicts of interest. Supporting information Data S1: nop270835‐sup‐0001‐Supinfo1.doc. Table S1: Item‐level statistics of the SSD knowledge, attitude, and practice subscales.