Exploring influences on radiation protection compliance: a directed acyclic graph-based cross-sectional study in a non-teaching hospital in western China

Introduction

Radiological examinations have become an important tool in clinical diagnosis due to advancements in medical technology and social progress. While these examinations provide patients with valuable diagnostic information, they concurrently increase their exposure to radiation (Yashima & Chida, 2022). According to the World Health Organization, more than 4.2 billion diagnostic radiological examinations, 40 million nuclear medicine imaging procedures, and 8.5 million radiation treatments are conducted globally each year (World Health Organization, 2016). Consequently, the application of ionizing radiation inspection should be optimized while strengthening radiation protection measures to minimize related health hazards. Given the improvements in inspection technology and protective measures, the radiation dose received during radiological examinations is generally within the safe range (Giammarile et al., 2024).

Despite existing protective measures, prolonged radiation exposure among medical personnel and individual sensitivity can still result in tissue and organ damage, as well as physiological dysfunction (Baudin et al., 2021). Therefore, clinicians and radiologists should utilize personal protective measures during radiological diagnosis and treatment to minimize radiation hazards (Awadghanem et al., 2020).

The detrimental effects of ionizing radiation on the human body are generally categorized into deterministic and stochastic effects. Deterministic effects, such as acute radiation sickness and radiation-induced skin injury, occur when exposure to ionizing radiation exceeds a certain threshold, with the severity of damage increasing at higher doses. In contrast, stochastic effects, such as carcinogenic and genetic outcomes, do not have a threshold dose; instead, their probability of occurrence is related to the dose size (Shafiq & Mehmood, 2024; Talapko et al., 2024). While eliminating stochastic effects is impossible, they can be reduced by implementing appropriate protective measures.

Both acute and chronic exposure to ionizing radiation have been shown to adversely affect human health, with considerable variation in the radiation sensitivity of different human tissues (Britel, Bourguignon & Foray, 2018). Healthcare personnel exposed to ionizing radiation have an increased incidence of developing various tumors and leukemia. Low radiation doses have been linked to an increased risk of cancer. Its prolonged exposure has also been found to affect the motility, normal morphology, and DNA fragmentation index of sperm in Chinese men (Kumar et al., 2014; Zhou et al., 2016).

Considering these findings, the International Commission on Radiological Protection (ICRP) reduced the occupational radiation limit for radiological workers from 50 mSv/year to an average of 20 mSv/year over five consecutive years, with a maximum of 50 mSv allowed in any single year (Frane & Bitterman, 2022). Therefore, current protection levels are based on the principle that, while radiation risk cannot be eliminated, it can be kept at a tolerable level. The ICRP proposes three fundamental principles of radiation protection as follows: “justification”, “optimization”, and “dose limitation” (Yeung, 2019).

Nevertheless, the level of attention paid to radiation protection among radiation workers with varying levels of knowledge differs considerably. Some practice excessive protection—usually driven by an overestimation of radiation risks—while others insufficiently use protective equipment, leading to excessive exposure and potential health risks (Alyousef et al., 2023). Studies consistently show that individuals with greater knowledge of radiation protection are more likely to implement effective protective measures. This underscores the critical role of radiation protection education (Rahimi et al., 2021; Jeyasugiththan et al., 2023; Salih et al., 2023).

An individual’s knowledge of radiation protection significantly influences their attitudes and behaviors, which consequently impact their health outcomes (Partap et al., 2019).

In China, where the number of radiological workers in medical institutions exceeds 450,000 and continues to increase annually (Deng & Sun, 2021), understanding the determinants of radiation protection behavior is essential. However, research in this area—particularly regarding the causal mechanisms underlying these behaviors—remains limited.

While cross-sectional studies have effectively identified associations between knowledge, attitudes, and behaviors (KAB) in radiation protection, they are inherently limited in establishing causal relationships. To address this gap, this study introduces a directed acyclic graph (DAG) framework, which is a robust tool for formalizing causal assumptions based on theoretical models and prior evidence (Akinkugbe et al., 2016; Textor et al., 2016). By constructing a DAG, we aimed to clarify whether radiation protection knowledge directly influences protective behaviors or operates indirectly through attitudinal mediation, while also accounting for potential confounders, such as age, educational level, and training exposure. This approach strengthens the rigor of causal inference and aligns with recent recommendations for integrating advanced causal methods into observational studies in occupational health (Tennant et al., 2021; Zheng et al., 2023). Through this innovative framework, we seek to generate actionable insights to inform and improve radiation protection practices and policies among Chinese healthcare workers.

Materials & Methods

Results

DAG analysis

To address potential confounding and formalize causal assumptions, the DAG was constructed using variables identified as statistically significant in both univariate and multivariate analyses, including age, educational level, radiation protection training, number of children, knowledge about radiation protection, and attitude toward radiation protection. The DAG framework hypothesized causal pathways grounded in the KAB theory and supported by empirical evidence from occupational health studies.

Key pathways demonstrated that higher radiation protection knowledge directly predicted better compliance behaviors (β = 0.551, p < 0.001), meaning each standard deviation increase in knowledge corresponded to a 55% improvement in protective practices. Conversely, older age directly reduced compliance (β = −0.260, p < 0.001), indicating 26% lower adherence independent of knowledge/attitude. Radiation protection training simultaneously enhanced both knowledge (β = 0.394, p < 0.001) and compliance (β = 0.147, p < 0.001). Knowledge indirectly influenced behavior through attitude mediation, with an indirect effect accounting for 18.3% of the total effect (β = 0.024, 95% confidence interval (CI) [0.003–0.045]).

Table 4:

Factors affecting radiation protection behaviors (n = 335).

Variable	B	SE	Beta	t	P
Constant	20.289	2.756	/	7.362	<0.001
Gender	0.398	0.532	0.032	0.748	0.455
Age (years)	−0.246	0.048	−0.390	−5.085	<0.001
Educational level	1.079	0.493	0.101	2.188	0.029
Profession	0.055	0.281	0.009	0.197	0.844
Working unit	0.002	0.133	0.001	0.011	0.991
Marital status	−0.890	0.811	−0.073	−1.098	0.273
Career in present unit (years)	0.078	0.050	0.114	1.554	0.121
Daily exposure time (hours)	−0.598	0.428	−0.058	−1.398	0.163
Radiation protection training	1.953	0.511	0.167	3.824	<0.001
Number of children	1.184	0.549	0.136	2.156	0.032
Knowledge about radiation protection	0.700	0.064	0.555	10.989	<0.001
Attitude towards radiation protection	0.204	0.098	0.086	2.086	0.038

DOI: 10.7717/peerj.20083/table-4

Notes:

R² = 0.533, Adjusted R² = 0.516, F = 30.655, p < 0.001.

Variables such as educational level (β = 0.101, p = 0.029) and number of children (β = 0.136, p = 0.032) were identified as confounders, requiring adjustment in the multivariate models. Comparison with regression analysis confirmed alignment. Age and radiation protection training remained significant predictors (p < 0.05), while the effect of educational level was attenuated post-adjustment, indicating confounding. The DAG’s logical consistency was verified using the DAGitty package, an open-source causal graph modeling tool for constructing, analyzing and validating DAGs, whose functions include identifying confounding variables, generating adjustment sets, and testing conditional independence hypotheses. However, unmeasured confounders (e.g., workplace culture) and the cross-sectional design limited causal certainty (Table 5, Fig. 1).

Table 5:

DAG-guided direct and indirect effects on radiation protection behavior.

Path	Standardized β†	95% CI	p-value
Direct effects
Knowledge → Behavior	0.551	[0.447, 0.655]	<0.001
Age → Behavior	−0.260	[−0.316, −0.204]	<0.001
Training → Knowledge	0.394	[0.302, 0.486]	<0.001
Training → Behavior	0.147	[0.052, 0.242]	<0.001
Indirect effect
Knowledge → Attitude → Behavior	0.024	[0.003, 0.045]	0.026

DOI: 10.7717/peerj.20083/table-5

Notes:

†Standardized β for the indirect effect is the product of the path coefficients (β Knowledge→Attitude × β Attitude→Behavior) and represents the mediated effect size.

Discussion

The occupational health risks associated with prolonged exposure to low-dose ionizing radiation among medical personnel necessitate rigorous scientific inquiry and policy action. While China has established regulatory frameworks to address radiation-related occupational diseases (Jianbiao et al., 2021; Zhu, Hou & Tong, 2021; Baudin et al., 2024), our findings highlight critical gaps in radiation workers’ knowledge and practices, underscoring the urgency of evidence-based interventions.

The KAB theory explains changes in human health-related behaviors by dividing them into three continuous processes as follows: acquiring knowledge, generating attitudes, and forming behavior. According to this theory, knowledge is the foundation, while attitude acts as the driving force behind behavioral change. DAGs provide a structured approach to visualizing these relationships, more intuitively identifying potential confounding variables, and obtaining effect estimates that are closest to the real-world outcomes (Liu et al., 2019). By integrating the KAB framework with causal inference through DAGs, this study advances beyond traditional association-based analyses to elucidate mechanistic pathways and confounding effects, providing actionable insights for policy and practice.

The significant direct effect of radiation protection knowledge on behavior (β = 0.551, p < 0.001) underscores its central role within the KAB framework. Our data reveal an “invisible risk” paradox where workers demonstrated strong behavioral compliance (mean = 28.99/35) despite critical knowledge gaps in biological effects (32% correct), radiation culture (28%), emergency protocols (25%), and dose limits (38%). This paradox operates through dual mechanisms supported by prior literature. For instance, heuristic dread of radiation’s catastrophic potential triggers instinctive caution independent of technical knowledge (Nyathi, 2022; Yurt et al., 2022), while institutionalized safety rituals foster procedural compliance that typically decouples from substantive risk understanding (Partap et al., 2019; Nyathi, 2022). Consequently, 89% reported consistent equipment use, while 63% distrusted dosimeter accuracy—a disconnect reflecting “compliance without comprehension” that creates fragile adherence vulnerable to erosion during periods of high workload. Evidence-based interventions such as risk visualization tools, immersive simulations, and cognitive reframing are recommended to bridge this cognitive-behavioral gap (Fujibuchi, 2021). Our mediation analysis further reinforces this finding: the direct effect of knowledge on behavior far exceeded the indirect contribution of attitude (β = 0.024, 95% CI [0.003–0.045]), confirming that knowledge acquisition alone can drive behavioral improvements without attitudinal shifts. This finding partly challenges conventional KAB models, which position attitude as the primary mediator, implying that training programs should prioritize knowledge dissemination—particularly on underemphasized topics such as radiation culture, biological effects, and emergency protocols—to achieve rapid compliance.

For example, participants demonstrated a limited understanding of dose limits (mean knowledge score = 8.65/14) in radiation protection education, underscoring the need for curriculum standardization. Notably, the weak correlation between knowledge and attitude (r = 0.256, p < 0.01) shows that knowledge alone is insufficient to foster attitudinal change. This disconnect reflects institutional or cultural barriers, such as skepticism toward the accuracy of the personal dosimeter (the lowest-scored attitude item). The lowest attitude score (3.1 ± 1.2) was observed for “trust in personal dosimeter accuracy”, with 63% of participants expressing skepticism. This finding aligns with concerns regarding third-party monitoring transparency (Wrzesień, 2020), as the hospital in our study relies on quarterly thermoluminescent dosimeter reports from external vendors, resulting in a 90-day feedback lag.

While 89% of participants reported the consistent use of protective equipment and patient guidance, 18% admitted non-compliance with dosimeter protocols. Multivariate analysis identified age (β = −0.390, p < 0.001) and radiation protection training (β = 0.167, p < 0.001) as key modifiers, with older workers demonstrating reduced compliance, regardless of knowledge levels.

Age, educational level, radiation protection education, number of children, knowledge of radiation protection, and attitude toward radiation protection were the primary factors influencing radiation protection behaviors among radiation workers.

Specifically, age independently predicted reduced protective behavior (β = −0.390, p < 0.001), with workers aged >50 years scoring significantly lower than those aged <30 years. This decline persisted even after adjusting for knowledge and attitude, implicating factors such as occupational burnout or risk habituation (Chen & Gholamrezanezhad, 2023).

Longitudinal studies indicate that prolonged engagement in high-volume clinical environments reduces vigilance among frontline staff due to occupational burnout or risk habituation (Dyrbye et al., 2021), corroborating our finding of an age-related behavioral decline (β = −0.390, p < 0.001). Older workers with longer career durations (≥10 years) exhibited significantly lower protective behavior scores (p < 0.01), likely due to cumulative burnout from repetitive high-risk tasks, which attenuates adherence to safety protocols (Canon et al., 2022).

We recommend age-specific interventions, including ergonomic adjustments, mental health screenings, and shorter shifts for senior personnel, to mitigate this effect. These measures align with broader occupational health strategies advocating for tailored support in aging workforces (Parikh, 2023).

Although educational level significantly correlated with behavior in the univariate analysis (p = 0.001), DAG analysis identified it as a confounder (β = 0.101, p = 0.029). Higher education enhanced knowledge acquisition but did not directly improve compliance. This finding explains why participants with a bachelor’s degree (or higher education level) scored 15% higher in radiation protection behaviors, likely due to greater access to more sophisticated knowledge resources. However, these workers still required contextual training to translate knowledge into action.

Therefore, policy interventions should be stratified by educational level. Workers with lower educational attainment benefit from simplified, hands-on training, while degree holders require advanced modules on emerging risks—including low-dose chronic effects—to bridge the knowledge-behavior gap (Yashima & Chida, 2022; Alkhayal et al., 2023).

Radiation protection training showed a dual efficacy, enhancing both knowledge (β = 0.394, p < 0.001) and behavior (β = 0.147, p < 0.001). However, its limited impact on attitude indicates that current programs overlook attitudinal components, such as fostering a radiation culture or peer accountability. Therefore, integrating scenario-based simulations (e.g., mock emergencies) or mentorship systems may address this gap (Partap et al., 2019; Yashima & Chida, 2022; Salih et al., 2023; Shafiq & Mehmood, 2024). While the model explained 51.6% of protective behavior variance (F = 30.655, p < 0.001), unmeasured confounders such as family planning significantly influenced outcomes. Workers without children scored 12% higher in protective behaviors than those with two or more children (p = 0.032), possibly due to divergent risk perceptions. Specifically, workers without children may prioritize self-protection due to fertility concerns, as ionizing radiation exposure is linked to impaired spermatogenesis (Kesari, Agarwal & Henkel, 2018; Wdowiak et al., 2019), whereas those with multiple children may perceive reduced reproductive risks, leading to complacency. This inverse relationship (β = 0.136, p = 0.032) underscores the need for tailored messaging emphasizing familial health risks beyond individual concerns. Training emerged as a critical modifiable factor, with participants who received regular training scoring 17% higher than those who did not attend training (p < 0.001). This aligns with previous research demonstrating the dual role of training in enhancing knowledge and compliance (Yashima & Chida, 2022; Baudin et al., 2024).

However, the knowledge-behavior gap persisted. Workers with two or more children scored 8.65/14 on knowledge assessments yet exhibited lower compliance, indicating motivational barriers. Future studies should adopt mixed-method approaches to explore unmeasured determinants such as workplace culture and risk communication efficacy. Furthermore, expanding the sample to include multilevel hospitals and a broader range of healthcare professionals would enhance generalizability.

Longitudinal study designs could further clarify temporal relationships between training exposure, attitude formation, and sustained behavioral change. Future studies should prioritize implementing multi-site studies across diverse hospital settings (e.g., teaching vs. non-teaching, urban vs. rural), integrating electronic health records with real-time dosimetry to objectively track compliance behaviors while overcoming self-report biases. These trials should embed mixed-methods components—combining longitudinal dose measurements with periodic focus groups—to examine contextual mediators like organizational culture. Addressing these gaps will allow interventions to move beyond simple knowledge transfer to foster risk-perception alignment and lasting safety practices.

Conclusions

This cross-sectional study of 335 healthcare workers across western China systematically evaluated radiation protection behaviors using a DAG to clarify causal pathways. Key findings demonstrate that radiation protection knowledge directly drives protective behavior, with age and radiation protection training as critical modifiers. DAG analysis identified educational level and the number of children as confounders, highlighting the need for tailored interventions.

Notably, workers aged ≤30 years, with bachelor’s degrees, and without children exhibited higher compliance rates than the others. This study validates the utility of DAGs in disentangling complex relationships, revealing that knowledge acquisition alone explains 51.6% of behavioral variance, thereby challenging traditional KAB model assumptions.

Multivariate analysis confirmed significant associations between protective behavior and age, radiation protection training, and fertility status; however, the DAG highlighted unmeasured confounders such as workplace culture. We suggest that future research could adopt a mixed-method design to measure or account for such complex factors. For instance, combining longitudinal surveys with ethnographic observations and conducting structured interviews to explore team responsibility norms, thereby enabling clear adjustments to these latent factors in the causal model. These findings underscore the need for standardized training curricula emphasizing emergency protocols, dose limits, and age-specific ergonomic adjustments.

Furthermore, this research provides causal evidence supporting the prioritization of knowledge dissemination and system transparency in radiation protection. By integrating DAG analysis and KAB theory, our findings advance occupational health methodology and inform policy strategies to improve healthcare workers’ safety.

Supplemental Information