Introduction

The discovery and widespread use of antimicrobial drugs marked a transformative milestone in modern medicine, significantly improving the treatment of infectious diseases. However, their misuse has accelerated the development of antimicrobial resistance (AMR), now one of the most critical challenges to global public health1,2. AMR not only complicates the management of previously treatable infections but also increases healthcare costs and threatens global health security3,4. In response, the World Health Organization (WHO) has launched a global action plan emphasizing the need for stricter regulation of antimicrobial drug use, increased awareness among healthcare professionals and the public, and advocacy for judicious prescribing practices1. At the national level, many countries have implemented policies such as precise clinical guidelines, healthcare professional training, drug usage monitoring, and targeted public health campaigns5,6.

Despite these efforts, healthcare professionals, especially in resource-limited settings, face persistent barriers to rational antimicrobial use. Challenges include limited access to up-to-date clinical guidelines, patient pressures for immediate treatment, insufficient awareness of AMR, and systemic weaknesses in healthcare infrastructure9,10. Addressing these issues requires not only enhancing healthcare professionals’ knowledge and skills but also understanding the psychosocial factors underlying their prescribing behavior.

This study integrates Social Support Theory (SST)11, Rational Action Theory (RAT)12, Theory of Planned Behavior (TPB)13,14,15, and other psychosocial models to examine the psychological motivations influencing healthcare professionals’ rational antimicrobial use. Using Structural Equation Modeling (SEM)16,17,18, this study quantitatively explores the relationships among these theoretical constructs and their influence on healthcare professionals’ decision-making in antimicrobial drug use. By offering a holistic perspective on the decision-making processes of healthcare professionals, this study provides robust theoretical and empirical support for designing interventions to promote rational antimicrobial use.

Methodology

Scale development

This study employed a multidimensional approach to understand the psychological attitudes and behavioral motivations of healthcare professionals regarding the rational use of antimicrobial drugs. Drawing upon the foundations of Social Support Theory, Rational Action Theory, Theory of Planned Behavior, Cognitive Processing Theory, Health Belief Model, and the constructs of Knowledge and Skills, we designed a comprehensive Likert scale. The scale comprises eight dimensions with four items each, totaling 32 items aimed at evaluating the multifaceted factors influencing healthcare professionals’ behaviors. The scale development was a collaborative effort by an interdisciplinary team of experts, including clinical medicine specialists, epidemiologists, health educators, and psychologists. The content validity of the scale was ensured through a rigorous expert review process, evaluating the relevance, representativeness, and clarity of the items. Following necessary revisions, the scale received validation from the expert panel, affirming its effectiveness in reflecting the study constructs and accurately measuring the target population’s psychological motivations and behavioral tendencies.

Data collection

A cross-sectional survey was conducted from January 8 to February 7, 2024, targeting healthcare professionals at a tertiary hospital in Beijing. Questionnaires were distributed through the hospital’s internal survey system, adhering to strict anonymity protocols to protect participants’ privacy. Participation was voluntary, and no identifiable information was collected.

Target Population Criteria:

Age: Participants were required to be over 18 years old.

Nationality: Participants were required to hold Chinese citizenship.

Employment: Participants were required to be frontline healthcare professionals (e.g., physicians, nurses, pharmacists) directly involved in patient care. Administrative and non-clinical staff were excluded to ensure the study focused on clinical decision-making processes.

Language: Proficiency in oral, written, and reading comprehension skills in Chinese was required.

Sampling Method:

Cluster sampling was employed, selecting entire departments or units of healthcare professionals involved in direct patient care as clusters. This method ensured a comprehensive and representative sample of the hospital’s clinical workforce while maintaining feasibility for large-scale data collection.

Exclusion Criteria:

Cognitive Impairment: Individuals with cognitive impairments were excluded.

Communication Barriers: Individuals with significant communication difficulties (e.g., blindness or deafness) were excluded.

Selection Bias Management: To minimize selection bias, all eligible healthcare professionals within the selected clusters were invited to participate. Demographic data from survey respondents were compared with the hospital’s overall workforce to confirm the sample’s representativeness. Anonymity and voluntary participation further mitigated potential response or selection biases.

Sample size calculation

Utilizing G*Power 3.1 software, the sample size was pre-calculated to ensure sufficient statistical power to detect the expected effects. Assuming a medium effect size (f² = 0.15), a Type I error rate of 0.05, and a power (1-β error probability) of 0.95[19], the minimum required sample size was determined to be 267 participants. Accounting for a potential 20% dropout rate, the adjusted minimum sample size is 290 participants. Ultimately, 720 healthcare professionals completed the survey, far exceeding the required sample size, ensuring robust statistical power for the analysis.

Limitations of the cross-sectional design

While the cross-sectional design allows for efficient data collection and analysis of relationships between variables, it inherently limits causal inferences. The findings of this study should be interpreted as associations rather than evidence of causality. Future research using longitudinal or experimental designs is recommended to validate the causal pathways suggested in this study.

Statistical analysis

Structural Equation Modeling (SEM) using AMOS 23.0 software was employed to test the hypothesized models. The measurement model was first validated through Confirmatory Factor Analysis (CFA) to ensure items accurately reflected their corresponding latent variables18. SEM analysis then assessed the mediating models proposed in the hypotheses, examining path coefficients among independent, mediating, and dependent variables. Model fit was evaluated using various fit indices, including the chi-square (χ2) statistic, degrees of freedom (df), χ2/df ratio, Comparative Fit Index (CFI), Normed Fit Index (NFI), Goodness of Fit Index (GFI), Adjusted Goodness of Fit Index (AGFI), and Root Mean Square Error of Approximation (RMSEA).

Research Hypotheses.

Model 1:

Hypothesis 1 (H1)

Social support (SS) positively influences healthcare professionals’ self-efficacy (SET).

Hypothesis 2 (H2)

Social support (SS) indirectly influences healthcare professionals’ behavioral intentions (BI) to rationally use antimicrobial drugs through self-efficacy (SET).

Hypothesis 3 (H3)

Social support (SS) positively influences healthcare professionals’ rational action (RAT).

Hypothesis 4 (H4)

Social support (SS) indirectly influences healthcare professionals’ behavioral intentions (BI) to rationally use antimicrobial drugs through rational action (RAT).

Model 2:

Hypothesis 5 (H1)

Social support (SS) positively influences healthcare professionals’ self-efficacy (SET).

Hypothesis 6 (H2)

Social support (SS) indirectly influences healthcare professionals’ behavioral intentions (BI) to rationally use antimicrobial drugs through self-efficacy (SET).

Hypothesis 7 (H3)

Social support (SS) positively influences healthcare professionals’ cognitive processing (CP).

Hypothesis 8 (H4)

Social support (SS) indirectly influences healthcare professionals’ behavioral intentions (BI) to rationally use antimicrobial drugs through cognitive processing (CP).

Model 3:

Hypothesis 9 (H1)

Social support (SS) positively influences healthcare professionals’ knowledge and skills (KS).

Hypothesis 10 (H2)

Social support (SS) indirectly influences healthcare professionals’ behavioral intentions (BI) to rationally use antimicrobial drugs through knowledge and skills (KS).

Hypothesis 11 (H3)

Social support (SS) positively influences healthcare professionals’ health beliefs (HBM).

Hypothesis 12 (H4)

Social support (SS) indirectly influences healthcare professionals’ behavioral intentions (BI) to rationally use antimicrobial drugs through health beliefs (HBM).

Model 4:

Hypothesis 13 (H1)

Social support (SS) positively influences healthcare professionals’ knowledge and skills (KS).

Hypothesis 14 (H2)

Social support (SS) indirectly influences healthcare professionals’ behavioral intentions (BI) to rationally use antimicrobial drugs through knowledge and skills (KS).

Hypothesis 15 (H3)

Social support (SS) positively influences healthcare professionals’ planned behavior (TPB).

Hypothesis 16 (H4)

Social support (SS) indirectly influences healthcare professionals’ behavioral intentions (BI) to rationally use antimicrobial drugs through planned behavior (TPB).

Results

Demographics

The survey successfully collected 720 responses from healthcare professionals across various demographics, including gender, age, work experience, education, and departmental affiliation, as illustrated in Table 1. The gender distribution showed a significant female majority, with women constituting 81.1% and men 18.9%, indicating a higher participation rate among female healthcare professionals in this study. The age group of 31–40 years old was the most represented at 37.4%, followed by the 41–50 age group at 30.8%. The distribution suggests a concentration of participants in the mid-career stage. In terms of work experience, individuals with 11–20 years in the field accounted for the highest proportion at 37.4%, indicating a significant number of participants with considerable professional experience. Those with over 20 years of experience represented 26.9%, while newcomers with less than 5 years accounted for 16.9%. Regarding educational levels, a majority held a bachelor’s degree (57.2%), followed by master’s degree holders (17.1%), indicating a generally high educational standard among respondents. Participation varied significantly across departments, with internal medicine leading at 32.5%, suggesting higher engagement in the survey from this specialty. Surgery and otorhinolaryngology followed with 21.1% and 16.4%, respectively, while emergency medicine had a participation rate of 13.9%. Departments like ophthalmology (9.7%), pediatrics (6.0%), and obstetrics and gynecology (0.4%) showed lower engagement, possibly reflecting the distribution of manpower resources within the hospital.

Table 1 Demographic information of respondents in the survey.
Full size table

Confirmatory factor analysis (CFA)

The proposed hypothetical model was validated using Confirmatory Factor Analysis (CFA) on the 32 items across eight dimensions, as shown in Table 2. The results demonstrated that all item factor loadings exceeded the recommended threshold of 0.5 set by Awang20, with t-values and corresponding p-values indicating statistical significance. Internal consistency was confirmed, with Composite Reliability (CR) values surpassing the 0.7 benchmark and Average Variance Extracted (AVE) for all constructs exceeding 0.5, indicating strong convergent validity21. The Squared Multiple Correlations (SMC) for all items also surpassed the acceptable threshold of 0.3022. Discriminant validity was assessed using Fornell and Larcker’s (1981) criterion, comparing the square root of the AVE (diagonal values) with the inter-construct correlations (off-diagonal values)23. The data in Table 3 confirmed discriminant validity among the model constructs, ensuring that each construct is distinct and independent. The CFA results substantiate the statistical reliability and validity of the developed scale, affirming its suitability for subsequent Structural Equation Modeling analysis.

Table 2 Results of confirmatory factor analysis for study constructs.
Full size table
Table 3 Discriminant validity results for study constructs.
Full size table
Table 4 Mediation effect analysis results across different models.
Full size table

Structural equation modeling analysis

Mediation effect analysis

In this study, we employed Structural Equation Modeling (SEM) via AMOS software, supplemented by a Bootstrap method with 5000 samples, to examine and test the research hypotheses related to four two-factor mediation models(see in Table 4). Our objective was to investigate the roles of two specific mediator variables in the relationship between independent and dependent variables and to ascertain which mediator demonstrates a more significant role in mediating the effect of the independent variable on the dependent variable.

Model 1 Analysis results

The SEM analysis revealed that social support (SS) positively influences behavioral intention (BI) through self-efficacy (SET), with a standardized path coefficient of 0.601 (SE = 0.102, Z = 5.892), indicating a statistically significant mediation effect see. This finding is consistent with Bandura’s social cognitive theory, which posits that self-efficacy is crucial in influencing individual behavior24. The Bootstrap bias-corrected 95% confidence interval ranged from 0.453 to 0.914, reinforcing the mediation’s significance, aligning with contemporary SEM approaches that recommend bootstrapping for more accurate confidence intervals25. In contrast, the mediation path through rational action (RAT) also demonstrated a significant positive effect, which is supported by the theory that rational decision-making processes are integral to behavioral intention26. A comparative analysis underscored the stronger mediating role of SET, highlighting the variable’s significant influence on behavior, a finding that aligns with previous research emphasizing the pivotal role of self-efficacy in mediating social influences on behavior26,27.

Model 2 Analysis results

The analysis demonstrated a negative mediation effect of social support (SS) on behavioral intention (BI) via self-efficacy (SET), with a point estimate of -0.168 (SE = 0.053, Z = -3.170). This unexpected inverse relationship suggests that under certain conditions, increased social support might paradoxically decrease self-efficacy, potentially due to over-reliance or diminished personal agency28.On the other hand, cognitive processing (CP) showed a strong positive mediation effect (point estimate = 1.112, SE = 0.059, Z = 18.847), resonating with models that emphasize the role of cognitive factors in shaping behavioral intentions29.The significant difference in the mediation effects of SET and CP underscores the complexity of the mechanisms through which social support influences behavioral intentions, suggesting that the cognitive interpretation of social support plays a crucial role30.

Model 3 Analysis results

The analysis of Model 3 underscored the roles of knowledge and skills (KS) and the Health Belief Model (HBM) as significant mediators in the relationship between social support (SS) and behavioral intention (BI). The mediation effect of KS was substantial (point estimate = 0.395, SE = 0.118, Z = 3.347), aligning with theories that highlight the pivotal role of knowledge and skills in behavior change [30]. Similarly, HBM demonstrated a robust mediation effect (point estimate = 0.436, SE = 0.124, Z = 3.516), consistent with its established role in predicting health-related behaviors15. The lack of significant difference in their mediating effects suggests that both knowledge and individual health beliefs are crucial yet comparable determinants of behavioral intentions, echoing findings from previous research31.

Model 4 Analysis results

The analysis of Model 4 highlighted the differential mediation effects of the Theory of Planned Behavior (TPB) and knowledge and skills (KS) on the relationship between social support (SS) and behavioral intention (BI). While TPB did not exhibit a significant mediation effect (point estimate = 0.155, SE = 0.115, Z = 1.348), KS demonstrated a notable positive mediation effect (point estimate = 0.65, SE = 0.131, Z = 4.962). This significant difference underscores the paramount influence of KS in mediating the impact of social support on behavioral intentions, resonating with the literature that emphasizes the critical role of skills and knowledge in behavior change24. The finding also aligns with Ajzen’s (1991) TPB, suggesting that while attitudes, subjective norms, and perceived behavioral control are important, the practical aspects of knowledge and skills can be more directly influential in shaping intentions and behaviors13,26.

Model fit

In this study, we employed a suite of relative or incremental fit indices along with absolute fit goodness indices to comprehensively evaluate and compare the fit of different structural equation models (SEM). Relative or incremental fit indices, such as the ratio of Chi-square to degrees of freedom (Chi-square/df), the Normed Fit Index (NFI), and the Comparative Fit Index (CFI), were utilized to assess the improvement in fit of one model over another alternative model32,33. Absolute fit indices, including the Chi-square value, Goodness of Fit Index (GFI), Adjusted Goodness of Fit Index (AGFI), Root Mean Square Error of Approximation (RMSEA), and Comparative Fit Index (CFI), were used to evaluate the fit of individual models34,35. Through this approach, we aimed to provide a comprehensive and detailed analysis of model fit to support our research hypotheses and model selection. As illustrated in Table 5, Model 1 displayed a Chi-square/df ratio of 6.4, suggesting potential overcomplexity in the model, although NFI and CFI values near 1 indicated good relative improvement compared to the baseline model32,36. GFI and AGFI values suggested a relatively good fit, and an RMSEA value at the upper limit of the acceptable range further qualifies this assessment37. For Models 2, 3, and 4, the Chi-square/df ratios and RMSEA values indicated less favorable fits, corroborated by the GFI and AGFI values falling below the ideal standard. Despite this, NFI and CFI values showed good relative improvement, emphasizing the nuanced interpretation of fit indices in SEM analysis38.

Table 5 Goodness-of-Fit indices for comparative model evaluation.
Full size table

The comparative analysis of the four models’ fit indices suggests that Model 1 demonstrates a better fit relative to the others, indicating its suitability for supporting the study’s hypotheses and model selection. This nuanced understanding of model fit, supported by a robust framework of fit indices, reinforces the methodological rigor of our SEM analysis in the context of psychological research39.

Conclusion and discussion

This study explored how social support (SS) influences healthcare professionals’ behavioral intentions (BI) regarding the rational use of antimicrobial drugs through four two-factor mediation models. By incorporating Rational Action Theory (RAT), Theory of Planned Behavior (TPB), Cognitive Processing (CP), Health Belief Model (HBM), and Knowledge and Skills (KS) as mediating variables, our Structural Equation Modeling (SEM) analysis confirmed that SS positively impacts rational prescribing behaviors, consistent with prior research7,8. The novelty of this study lies in its quantitative SEM approach, which allowed for a detailed examination of distinct psychosocial pathways. Notably, Model 1 demonstrated the best fit, emphasizing the critical roles of Self-Efficacy (SET) and RAT in enhancing prescribing intentions. Although Models 2, 3, and 4 did not achieve ideal fit indices, they provided valuable insights into the complex cognitive and psychological mechanisms underlying SS’s influence on healthcare professionals’ behaviors. An unexpected finding was the negative impact of SS on BI through CP in Model 2, suggesting that SS may not always facilitate positive behavioral outcomes. This counterintuitive result highlights the potential for information overload and cognitive conflict in healthcare environments, where excessive or misaligned support may overwhelm professionals or contradict their existing knowledge and experiences40. Tailored SS strategies that align with the cognitive capacities and situational needs of healthcare professionals are essential to mitigate these adverse effects. Furthermore, the mediating role of KS revealed varying significance between Models 3 and 4, where TPB emerged as a stronger mediator compared to HBM. These differences emphasize the importance of theoretically grounded and context-specific interventions, particularly those leveraging planned behavior frameworks to enhance decision-making processes. The findings of this study have important implications for healthcare policy and practice. Tailored SS initiatives should prioritize quality over quantity of information to reduce cognitive strain and align with professionals’ needs. Establishing robust professional networks can foster shared learning, reduce decision-making pressure, and promote adherence to rational prescribing practices. Intervention strategies should incorporate theoretical frameworks like TPB and SET to enhance intentional behavior and self-efficacy while addressing cognitive barriers such as those observed in the CP pathway. Educational programs must combine KS and HBM to improve knowledge and shift belief systems toward the importance of rational antimicrobial use. Finally, strategic dissemination of concise, relevant, and actionable information can maximize the impact of interventions. By addressing the broader spectrum of psychological and environmental factors, this study provides a robust foundation for evidence-based antimicrobial stewardship strategies, offering actionable insights to advance healthcare policy and practice globally.