Network analysis of factors associated with lung cancer screening behavior among high-risk rural adults in Fujian, China

Abstract

Background

Lung cancer is the leading cause of cancer-related death in China. Low-dose computed tomography (LDCT) improves early detection and reduces mortality among high-risk populations, but uptake remains low in rural settings. The interrelationships among psychological, cognitive, and health belief-related factors remain unclear. In this study, used network analysis was used to examine the factors associated with lung cancer screening (LCS) behavior among high-risk rural adults in Fujian Province, China, and to compare networks by socioeconomic status (SES).

Methods

A cross-sectional survey was conducted from February to October 2024 among high-risk rural adults who were recruited from eight township health centers in Fuzhou and Putian. The data covered demographic information, knowledge, health beliefs, anxiety, depression, smoking-related stigma, trust in medical professionals and LCS behavioral stage. A mixed graphical model was estimated. Centrality and bridge indices were calculated, SES-based networks were compared using network comparison tests, and stability was assessed with bootstrapping.

Results

Among the 1,975 individuals who received the questionnaire, 1,905 completed the survey and were included, yielding a completion rate of 96.5%. The strongest edge was between depression and anxiety, followed by the edge between knowledge and the LCS behavioral stage. Excluding the LCS behavioral stage node, depression had the highest strength centrality and expected influence, followed by perceived barriers. Depression and anxiety were the main bridge nodes, whereas knowledge and depression had the highest bridge expected influence. Network comparison revealed no significant difference in global strength between SES groups, but network structure differed significantly. In the medium-low SES group, perceived barriers and depression were most central, and the self-efficacy–LCS behavioral stage edge was relatively strong. In the high SES group, depression and smoking-related stigma were key nodes, and the knowledge–LCS behavioral stage edge was more prominent. Bootstrapping indicated good stability.

Conclusions

Among high-risk rural adults in Fujian, LCS behavior was embedded in a network of psychological, cognitive, health belief-related and social-contextual factors. Depression was a stable central and bridging factor across SES groups. Interventions may need to reduce perceived barriers and strengthen self-efficacy in medium-low SES populations while improving knowledge translation and reducing smoking-related stigma in higher SES populations. Further validation in diverse, probability-based samples is warranted.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13690-026-01990-4.

Text box 1. Contributions to the literature
• This study applies network analysis to clarify how psychological, cognitive, health belief-related, and social-contextual factors are interconnected in lung cancer screening behavior.
• The findings identify depression, perceived barriers, self-efficacy, knowledge, and smoking-related stigma as potentially important targets for improving screening uptake.
• SES-based network comparisons suggest that screening promotion strategies may need to be tailored to different socioeconomic contexts.
• The results provide evidence to support more precise, equity-oriented interventions in rural primary care and community screening programs.

Introduction

Lung cancer is the leading cause of cancer-related death worldwide and the most common malignancy in China, with approximately 1.06 million new cases and 730,000 deaths reported in 2022 [1–3]. Prognosis is strongly stage-dependent: the five-year survival rate for early-stage lung cancer ranges from 68% to 92%, yet fewer than 20% of lung cancer cases in China are detected at stage I [4]. Evidence from the U.S. National Lung Screening Trial revealed that low-dose computed tomography (LDCT) screening reduces lung cancer mortality by approximately 20% among high-risk populations [5], and prospective evidence from China suggested that even a single LDCT scan may reduce lung cancer mortality by approximately 31% [6]. These findings underscore the importance of promoting LDCT screening among high-risk groups to improve early detection and reduce disease burden.

Despite the proven benefits of early detection, the use of lung cancer screening remains suboptimal. Similar challenges have been reported internationally, where screening uptake often remains below program targets [7, 8]. In China, the “Early Diagnosis and Early Treatment of Cancer” program has been progressively implemented across urban and rural areas since 2005. However, participation remains insufficient in many regions, with a pronounced urban–rural disparity: the screening rate among rural residents has been reported to be 26.59%, which is markedly lower than the 43.67% reported among urban residents [9]. Improving lung cancer screening participation among rural high-risk populations is therefore important not only for early detection but also for promoting health equity.

Rural populations face unique challenges that may shape health behaviors, including lung cancer screening (LCS). These challenges include limited access to healthcare facilities, lower health literacy, transportation barriers, financial constraints, and cultural beliefs that influence perceptions of disease risk and preventive care [10–12]. Such structural and contextual factors may directly affect screening participation and may also interact with psychological and cognitive processes, jointly shaping individuals’ screening decisions. Understanding these interactions is essential for developing context-sensitive interventions for high-risk rural populations.

Socioeconomic status (SES) is an important social determinant of preventive health behavior and is typically measured using indicators such as education, income, and occupation [13]. Individuals with lower SES may have less access to health information, fewer opportunities to use preventive healthcare services, and greater practical barriers to screening [14, 15]. Several factors associated with lung cancer screening participation, including lung cancer and screening-related knowledge, health beliefs, trust in healthcare providers, anxiety, fatalism, and smoking-related stigma, have been identified [16–18]. However, most existing studies have examined these factors separately. When interactions among these factors are ignored, the mechanisms underlying screening participation may remain unclear, which may limit the effectiveness of intervention strategies.

Network analysis provides a useful approach for examining complex interrelationships among multiple factors. In a network model, variables are represented as nodes, and conditional associations between variables are represented as edges [19, 20]. This approach can identify central nodes that are strongly connected with other factors and bridge nodes that connect different conceptual domains. In health behavior and symptom research, network analysis has been increasingly used to reveal potential intervention targets and to move beyond traditional single-factor analytical frameworks [21, 22]. For LCS, a network perspective may help clarify whether screening behavior is more closely connected with knowledge, health beliefs, psychological distress, stigma, or trust in healthcare providers and whether these patterns differ across socioeconomic contexts.

In the context of LCS, screening behaviors may be shaped by several interrelated domains. Health belief factors, including perceived barriers, perceived benefits, perceived risk, and self-efficacy, reflect individuals’ appraisal of screening based on the Health Belief Model (HBM) [23]. At the same time, knowledge provides a cognitive basis for understanding screening; psychological distress, such as anxiety and depression, may influence motivation and decision-making, and social-contextual factors, such as trust in medical professionals and smoking-related stigma, may further facilitate or inhibit screening participation [24, 25]. These factors were therefore considered interrelated components of a broader screening decision system rather than isolated predictors.

To our knowledge, limited evidence is available on the network structure of factors associated with lung cancer screening behavior among high-risk rural adults in China. To address this gap, this study constructed a network model of psychological, cognitive, health belief-related, social-contextual, and behavioral factors associated with LCS behavior among high-risk rural adults recruited from township health centers in Fujian Province, China. We further compared network structures across SES groups to explore whether the mechanisms associated with screening behavior differed by socioeconomic context. These findings may provide evidence for the development of more targeted and equity-oriented interventions to improve lung cancer screening uptake in rural primary care settings.

Methods

Study design and setting

This cross-sectional study was conducted in Fujian Province, China, from February to October 2024. Participants were recruited from eight township health centers in Fuzhou and Putian. Township health centers are the main primary healthcare institutions that serve rural residents within their catchment areas and provide both basic medical care and public health services. This study was approved by the Biomedical Research Ethics Review Committee of Fujian Medical University (Approval No. 2023075). All participants provided written informed consent before participation. The study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [26].

Study participants

This study used convenience sampling. The sampling frame consisted of residents registered in the health records and chronic disease management systems of the participating township health centers. Potentially eligible individuals who met the high-risk criteria for lung cancer screening were first identified by local health workers. Trained investigators then contacted eligible individuals during outpatient visits or by telephone, explained the study purpose and procedures, and invited them to participate. A final sample of 1,905 high-risk individuals who met the inclusion criteria was included in the analysis.

The inclusion criteria were as follows: (1) meeting the 2023 Chinese Expert Consensus on Early Lung Cancer Diagnosis criteria for high-risk populations; [27] age 40–80 years and the presence of at least one of the following risk factors: smoking history of at least 20 pack-years; a history of environmental or occupational exposure, such as exposure to radon, silica, cadmium, arsenic, beryllium, chromium, nickel, asbestos, diesel or coal smoke, or radioactive substances; a first-degree family history of lung cancer; chronic obstructive pulmonary disease, diffuse pulmonary fibrosis, or previous pulmonary tuberculosis; previous malignant tumor; long-term passive exposure to secondhand smoke for more than 2 h per day for at least 10 years; or long-term exposure to cooking fumes from frying, sautéing, or deep-frying; (2) local residency for at least six months; and (3) willingness to share personal opinions and experiences. Individuals with brain injuries, neurological disorders, mental illness, or current mental health treatment were excluded.

Previous methodological work suggests that in sparse networks with fewer than 20 nodes, a sample size of 250–350 participants may provide adequate sensitivity, specificity, and stability for edge weight estimation [28]. This range was used as a minimum benchmark for estimating the overall network. The present study included 1,905 participants, substantially exceeding this threshold. A relatively larger sample size may help improve the precision and stability of network estimation and provide a certain degree of statistical support for subgroup analyses, including network comparison tests. Further stability analyses revealed that the correlation stability coefficients (CS-C) for centrality indices were all above 0.5, suggesting good stability of the network structure. Together with the non-parametric bootstrap results, the edge-weight estimates demonstrated relatively good precision. Overall, these findings suggest, to some extent, that the sample size in this study is adequate to support a relatively stable estimation of the network structure and to provide support for the interpretation of the results.

Data collection

Data were collected using either a paper-and-pencil anonymous questionnaire or an online survey through the Questionnaire Star platform (https://www.wjx.cn/vm/OCzUgMw.aspx#). The average completion time was approximately 15–20 min. All survey administrators received standardized training before data collection to ensure consistency in eligibility assessment, questionnaire administration, and data checking. Participants who had difficulty completing the questionnaire independently because of literacy or other barriers received assistance from trained research staff, who read the questions using standardized wording and recorded participants’ responses without interpretation or guidance. Completed questionnaires were checked for completeness and consistency before data entry and analysis.

Environmental and occupational exposure, secondhand smoke exposure, and cooking fumes exposure were assessed using a standardized checklist administered by trained investigators. For each exposure, participants were asked whether they had ever been exposed, the main source or setting of exposure, and the approximate duration and frequency when they could recall this information. Investigators provided standardized examples of common occupational, household, and environmental exposure scenarios to improve comprehension and support more consistent responses.

Measures

Lung cancer and screening-related knowledge

Lung cancer and screening-related knowledge were assessed using a 9-item questionnaire developed on the basis of previous lung cancer screening research and a cancer prevention survey conducted in Fujian Province. The questionnaire assessed knowledge of lung cancer symptoms, prevention, screening recommendations, screening methods, and screening frequency. It included four single-answer items and five multiple-answer items. Single-answer items were scored as 1 for a correct response and 0 for an incorrect response. Multiple-answer items were scored as 2 for a fully correct response and 0 for an incorrect or partially correct response. Higher total scores indicated greater lung cancer and screening-related knowledge [29]. In this study, the questionnaire demonstrated good internal consistency, with a Cronbach’s alpha of 0.89.

Stage of adoption algorithm for LCS

LCS behavior was assessed using a five-question stage of adoption algorithm corresponding to seven stages: unaware, unengaged, undecided, decided not to act, decided to act, action, and maintenance [23]. For descriptive purposes, participants in stages 1–4 were classified as non-participants in LCS, whereas those in stages 5–7 were classified as LCS participants.

Lung cancer screening-related health beliefs

Lung cancer screening-related health beliefs were assessed using the Lung Cancer Screening Health Belief Scales, originally developed by Carter-Harris et al., which are based on the Health Belief Model [30]. The Chinese version was used in this study [31]. The Chinese version was developed through a standardized translation and cultural adaptation procedures, including forward translation, back-translation, expert review, pilot testing, and psychometric validation. No substantive item modifications were made in the present study.

The scale includes 35 items across four domains: perceived barriers to lung cancer screening, perceived benefits of screening, self-efficacy for screening, and perceived risk of lung cancer. Items are rated on a 4-point Likert scale. The perceived barriers subscale was reverse-coded so that higher scores indicated lower perceived barriers. Higher scores on the remaining subscales indicated greater perceived benefits, stronger self-efficacy, and higher perceived risk. The total score ranged from 35 to 140, with higher scores reflecting more favorable health beliefs toward lung cancer screening. In this study, the Cronbach’s alpha coefficients for the subscales ranged from 0.83 to 0.93.

Smoking-related stigma

Smoking-related stigma was measured using the smoking-related subscale of the Cataldo Lung Cancer Stigma Scale [32]. Items were rated on a 4-point scale ranging from 1, strongly disagree, to 4, strongly agree, with higher scores indicating greater smoking-related stigma. One item, “Some people act as though it is my fault that I have lung cancer,” was excluded because it was not applicable to high-risk individuals without a lung cancer diagnosis. The resulting four-item subscale demonstrated good internal consistency in this study, with a Cronbach’s alpha of 0.85.

Anxiety

Anxiety symptoms were assessed using the Generalized Anxiety Disorder Scale-2 (GAD-2), a brief version of the GAD-7 developed by Kroenke et al. [33] The GAD-2 includes two items assessing nervousness and uncontrollable worrying over the past two weeks. Items are rated on a 4-point scale from 0 to 3, yielding a total score ranging from 0 to 6. Higher scores indicate more severe anxiety symptoms. The Chinese version of the GAD-7 has been shown to have good reliability and validity [34]. In this study, the Cronbach’s alpha of the GAD-2 was 0.84.

Depresssion

Depressive symptoms were assessed using the Patient Health Questionnaire-2 (PHQ-2), a brief version of the PHQ-9 [35, 36]. The PHQ-2 assesses depressed mood and anhedonia over the past two weeks. Items are rated on a 4-point scale from 0 to 3, yielding a total score ranging from 0 to 6. Higher scores indicate more severe depressive symptoms. The Chinese version of the PHQ-9 has demonstrated good reliability [37]. In this study, the Cronbach’s alpha of the PHQ-2 was 0.85.

Patient trust in the medical profession

Trust in medical professionals was assessed using a five-item scale [38]. Items were rated on a 5-point scale, and one negatively worded item was reverse-scored. Higher scores indicated greater trust in medical professionals. In this study, the Cronbach’s alpha of the scale was 0.80.

Socioeconomic status (SES)

SES was assessed using a composite score based on education, occupation, and household per capita annual income [39]. Each component was scored from 1 to 6 according to its rank. Education was classified into six levels, from no formal education to higher education. Occupation was grouped into six categories, ranging from unemployed individuals to professionals and business owners. Household per capita annual income was divided into six brackets. The total SES score was calculated as the sum of the three components and ranged from 3 to 18, with higher scores indicating higher SES. The participants were initially categorized into low SES (≤ 8), medium SES (9–11), and high SES (≥ 12) groups [39]. To compare SES-based networks, the low and medium SES groups were combined into a medium-low-SES group to ensure an adequate subgroup sample size and to compare participants with relatively lower socioeconomic resources against those with higher SES.

Sociodemographic characteristics

The participants reported their age, sex, residential location, education, employment status, occupational exposure history, smoking status, family history of cancer, medical insurance status, and other relevant characteristics. Environmental and occupational exposure, secondhand smoke exposure, and cooking fume exposure were recorded as part of the sociodemographic and risk factor questionnaire.

Data analysis

Descriptive analyses were performed using SPSS version 26.0. Categorical variables are summarized as frequencies and percentages, and continuous variables are presented as the means with standard deviations. These analyses were used to describe participants’ sociodemographic and behavioral characteristics, including age, sex, smoking history, and SES.

Network analysis was conducted in R version 4.4.3. The network included 10 nodes: lung cancer and screening-related knowledge, perceived barriers, perceived benefits, perceived risk, self-efficacy, anxiety, depression, smoking-related stigma, trust in medical professionals, and LCS behavioral stage. These variables were conceptualized as interrelated components of a broader screening decision system, including cognitive, psychological, health belief-related, social-contextual, and behavioral domains. Given that the study variables included continuous and ordered variables, a mixed graphical model (MGM) was used for network estimation [40, 41].

In the network, nodes represent observed variables, and edges represent conditional associations between pairs of variables after controlling for all other variables in the network. Network sparsification was performed using the extended Bayesian information criterion (EBIC) to select tuning parameters and reduce spurious edges [40]. Networks were visualized using the qgraph package version 1.9.8, with the Fruchterman–Reingold algorithm placing more strongly connected nodes closer together [42].

Centrality indices, including strength, closeness, and expected influence, were calculated using the igraph package [43]. Strength reflects the sum of the absolute edge weights connected to a node, whereas expected influence preserves the sign of edge weights and therefore reflects the net positive or negative association of a node with the rest of the network. Bridge centrality indices were calculated using the networktools package to identify nodes that connected different conceptual communities [44].

The stability and accuracy of the network were assessed using the bootnet package [40]. Edge weight accuracy was evaluated through non-parametric bootstrapping with 1,000 iterations to generate 95% confidence intervals. The stability of centrality estimates was examined using the correlation stability coefficient (CS-C). A CS-C greater than 0.25 was considered acceptable, and a value greater than 0.50 was considered good.

The network comparison test (NCT) was used to compare network structures between the medium-low SES group and the high SES group [45]. Global strength invariance and network structure invariance were tested, and edge-level differences between the two SES groups were further examined. All the statistical tests were two-sided, and a p value less than 0.05 was considered statistically significant.

Results

A total of 1,975 eligible individuals were invited to complete the questionnaire. Of these, 1,905 completed the survey and were included in the final analysis, corresponding to a questionnaire completion rate of 96.5%. Among the included participants, 1,070 were male (56.2%), and the mean age was 61.51 years (SD = 9.30). Most participants were covered by urban-rural resident basic medical insurance (n = 1,459; 76.6%). A total of 1,357 participants (71.2%) were classified as having medium-low SES. Current smokers accounted for 36.3% of the sample (n = 692), and more than half of the participants reported exposure to secondhand smoke. Occupational exposure was reported by 89.4% of the participants. Additional participant characteristics are presented in Table 1.

Table 1.

Summary of the participants’ background characteristics (n = 1905)

Characteristic	Frequency	Percent (%)
Gender
Male	1070	56.2%
Female	835	43.8%
Educational category
Low literacy	347	18.2%
Less than elementary school education	216	11.3%
Elementary school education	314	16.5%
Junior high school education	566	29.7%
High school or equivalent/Secondary vocational education/Technical school	328	17.2%
College/Associate degree or higher	134	7.0%
Employment
Out of work	407	21.4%
Construction/transportation, farming, fisheries, forestry, or water conservation.	1152	60.5%
Commercial service industry	142	7.5%
Administrative, technical, or legal	134	7.0%
Executive in government, political organization, corporation or higher education	25	1.3%
Other worker	45	2.4%
Annual household income
≤ 2000.00	37	1.9%
2000.01 ~ 3333.33	121	6.4%
3333.34 ~ 5000.00	548	28.8%
5000.01 ~ 8000.00	508	26.7%
8000.01 ~ 12960.00	494	25.9%
≥ 12960.01	197	10.3%
Socioeconomic status
Low (≤ 8)	535	28.1%
Medium (9–11)	822	43.1%
High (≥ 12)	548	28.8%
Medical insurance type
Urban employee basic medical insurance	387	20.3%
Urban-rural basic medical insurance	1459	76.6%
Self-paid (uninsured)	36	1.9%
Other	23	1.2%
Smoking status
Non-smoker	980	51.4%
Current smoker	692	36.3%
Former smoker	233	12.2%
Secondhand smoke exposure
Yes	1124	59.0%
No	781	41.0%
History of COPD
Yes	802	42.1%
No	1103	57.9%
Occupational exposure history
Yes	1703	89.4%
No	202	10.6%
Family history of lung cancer
Yes	235	12.3%
No	1670	87.7%
Previous LCS behavior
Yes	1372	72.0%
No	533	28.0%

Notes: COPD Chronic Obstructive Pulmonary Disease, LCS Lung cancer screening

Estimation of the overall network of factors associated with lung cancer screening behavior

Figure 1 presents the estimated network of factors associated with LCS behavior in the overall sample. The network consisted of 10 nodes. Among all the edges, the strongest association was observed between depression and anxiety (edge weight = 0.79), followed by the association between lung cancer and screening-related knowledge and LCS behavior (edge weight = 0.39) (Fig. 1 and Table S1, Additional File 1).

Fig. 1.

Estimation of the network influencing lung cancer screening behavior. Notes: BS = Perceived barrier; ES = Perceived benefit; RS = Perceived risk; and SS = Self-efficacy

Among all the nodes, LCS behavior had the highest centrality value (1.50). Given that LCS behavior is the outcome node in this study, its high centrality mainly reflects its broad associations with multiple psychological, cognitive, and health belief-related factors. After the LCS behavior node was excluded, depression had the highest strength centrality and expected influence (0.96), followed by perceived barriers (0.54) (see Table 2). Strength centrality reflects the overall connectivity of a node, whereas expected influence captures the direction of associations, indicating that depression is strongly and directionally consistently connected with multiple factors and may play a coordinating role within the screening-related network. In terms of bridge centrality, depression was identified as the most important bridge node (1.13), followed by anxiety (1.00), suggesting that both may link different domains of factors. In terms of the bridge expected influence, lung cancer and screening-related knowledge had the greatest influence (1.58), followed by depression (1.57) (Table 2 and Figure S1 in Additional File 2).

Table 2.

Centrality measures of each node in the lung cancer screening network. (n = 1905)

Index	Strength	Expected influence	Bridge strength	Bridge expected influence
Lung cancer and LCS knowledge	-0.46	-0.46	0.85	1.58
Perceived barriers	0.54	0.54	0.82	1.29
Perceived benefits	0.49	0.49	0.66	1.07
Perceived risk	-0.50	-0.50	0.71	1.10
Self-efficacy	-0.10	-0.10	0.61	1.07
Smoking-related stigma	-0.58	-0.58	0.83	1.47
Depression	0.96	0.96	1.13	1.57
Anxiety	0.27	0.27	1.00	1.43
Medical trust	-2.11	-2.11	0.52	0.93
LCS behavioral adoption stages	1.50	1.50	1.24	2.06

Notes: LCS Lung cancer screening

Bootstrap analysis showed that the 95% confidence intervals of edge weights were relatively narrow (Figure S2, Additional File 2), suggesting acceptable accuracy of edge weight estimation. The network demonstrated good stability, with correlation stability coefficients (CS-C) of 0.52 for strength centrality and 0.75 for expected influence (Figure S3, Additional File 2).

Comparison of LCS behavioral networks between the medium-low and high SES groups

A comparative analysis of LCS behavioral networks was conducted between the medium-low SES group (n = 1,357) and the high SES group (n = 548) (Figs. 2 and 3). The NCT revealed no statistically significant difference in global strength between the two groups (global strength: 3.33 vs. 2.30; S = 1.03; P = 0.08). However, a statistically significant difference in network structure was observed (M = 0.33; P = 0.009).

Fig. 2.

The lung cancer screening behavioral network in the medium-low socioeconomic status group

Notes: BS = Perceived barrier; ES = Perceived benefit; RS = Perceived risk; and SS = Self-efficacy

Fig. 3.

The lung cancer screening behavioral network in the high socioeconomic status group. Notes: BS = Perceived barrier; ES = Perceived benefit; RS = Perceived risk; and SS = Self-efficacy

In the medium-low SES group, excluding the LCS behavioral stage node, the nodes with the highest strength centrality and expected influence were perceived barriers (1.19) and depression (0.84) (Table 3). In terms of edge connectivity, the strongest association was observed between depression and anxiety (edge weight = 0.77), followed by the association between self-efficacy and LCS behavior (edge weight = 0.37) (Table S2, Additional File 1).

Table 3.

Centrality measures of each node in the lung cancer screening network stratified by socioeconomic status (medium-low vs. high socioeconomic status subgroups)

Index	Strength		Expected influence		Bridge strength		Bridge expected influence
	LM-SES	H-SES	LM-SES	H-SES	LM-SES	H-SES	LM-SES	H-SES
Lung cancer and LCS knowledge	-0.51	0.13	-0.51	0.13	0.85	0.76	1.58	1.25
Perceived barriers	1.19	-1.07	1.19	-1.07	0.88	0.51	1.32	0.70
Perceived benefits	0.69	0.45	0.69	0.45	0.62	0.68	1.11	0.93
Perceived risk	-1.19	-0.85	-1.19	-0.85	0.62	0.47	0.89	0.62
Self-efficacy	0.45	-1.33	0.45	-1.33	0.68	0.27	1.16	0.51
Smoking-related stigma	-1.10	0.85	-1.10	0.85	0.74	0.89	1.28	1.26
Depression	0.84	0.94	0.84	0.94	1.09	0.96	1.42	1.01
Anxiety	-0.41	0.32	-0.41	0.32	0.87	0.83	1.17	0.94
Medical trust	-1.23	-1.00	-1.23	-1.00	0.72	0.51	1.33	0.81
LCS behavioral adoption stages	1.27	1.56	1.27	1.56	1.17	1.08	1.99	1.58

Notes: LM-SES Medium-low socioeconomic status, H-SES High socioeconomic status, LCS Lung cancer screening

In the high SES group, excluding the LCS behavioral stage node, the nodes with the highest strength centrality and expected influence were depression (0.94) and smoking-related stigma (0.85) (Table 3). The strongest association was again observed between depression and anxiety (edge weight = 0.83), whereas the second strongest association was observed between lung cancer and screening-related knowledge and LCS behavior (edge weight = 0.41) (Table S3, Additional File 1).

Bridge centrality also differed between the two SES groups. In the medium-low SES group, the main bridge nodes were depression (1.09) and perceived barriers (0.88). In the high SES group, the main bridge nodes were depression (0.96) and smoking-related stigma (0.89) (Table 3, Figures S4 and S5, Additional File 2).

The CS-C values for strength centrality were 0.517 in the medium-low SES group and 0.516 in the high SES group, while the CS-C values for expected influence were both 0.75, indicating good stability of the subgroup networks (Figures S6 and S7, Additional File 2). The bootstrap results revealed relatively narrow 95% confidence intervals for edge weights, supporting the accuracy of network estimation in both SES groups (Figures S8 and S9, Additional File 2).

Edge-level comparison showed that two edges differed significantly between the medium-low SES group and the high SES group (Fig. 4). The greatest difference was observed for the edge between self-efficacy and LCS behavior (E = 0.33; P < 0.05), followed by the edge between lung cancer and screening-related knowledge and perceived barriers (E = 0.23; P < 0.05).

Fig. 4.

Network of edge differences between the middle-low and high socioeconomic status groups. Notes: BS = Perceived barrier; ES = Perceived benefit; RS = Perceived risk; SS = Self-efficacy

Discussion

To our knowledge, this is one of the first studies to apply network analysis to examine the factors associated with LCS behavior among high-risk rural adults in China. Unlike previous studies that examined individual predictors separately, this study conceptualized LCS behavior as part of a broader screening decision system involving psychological distress, health beliefs, knowledge, smoking-related stigma, trust in medical professionals, and socioeconomic context. The findings showed that depression occupied a prominent position in the overall network and served as an important bridge node linking the psychological and behavioral domains. Differences in network structure and key nodes were also observed across SES groups. Because LCS behavioral stage was included as the behavioral outcome node, its centrality should be interpreted descriptively rather than as evidence that it represents an independent intervention target.

Overall, the findings suggest that LCS behavior is associated with a complex pattern of psychological, cognitive, health belief-related, and social-contextual factors. Depression played a relatively stable central and bridging roles across SES groups, whereas SES-specific patterns were observed for perceived barriers, self-efficacy, knowledge, and smoking-related stigma. These results extend traditional single-factor approaches by highlighting the interconnections among multiple factors related to screening behavior. They may provide preliminary evidence for developing more tailored and equity-oriented LCS promotion strategies in rural primary care settings.

Psychological factors associated with LCS behavior: the dual role of anxiety and depression

This study identified anxiety and depression as important psychological factors associated with LCS behavior. Across the overall network and both SES subgroup networks, anxiety and depression consistently showed the strongest edge, indicating a stable co-occurrence pattern of psychological distress among high-risk rural adults. These findings suggest that anxiety and depression may jointly shape individuals’ health risk appraisal, screening-related concerns, and behavioral decision-making processes [23, 46].

Anxiety may increase attention to health risks and strengthen the motivation to seek screening, but excessive anxiety may also lead to avoidance because of fear of screening results or concerns about subsequent diagnosis and treatment [47]. Depression, in contrast, may reduce behavioral initiative, weaken perceived control, and decrease the willingness to engage in preventive health behaviors [48, 49]. In the present study, depression showed high strength centrality, expected influence, and bridge centrality, suggesting that it may be closely connected with multiple domains of the screening decision system.

Notably, depression consistently exhibited relatively high centrality, bridge centrality, and expected influence within the screening behavior network. In particular, its relatively high expected influence suggests that its associations with adjacent nodes are directionally consistent overall, potentially exerting broad interconnected effects on other psychological, cognitive, and behavior-related factors within the network. This pattern may reflect the widespread links between depressive symptoms and reduced behavioral motivation, impaired cognitive processing, and lower engagement in health management [50, 51]. From an intervention perspective, these findings suggest that brief psychological assessment, emotional support, and referral pathways may be useful components of LCS promotion programs for high-risk rural adults. However, given the cross-sectional design, depression should be interpreted as an important factor associated with LCS behavior rather than as a confirmed causal determinant.

Health belief-related factors across SES groups

The present study revealed that health belief-related factors had different network positions across SES groups. In the overall network and the medium-low SES subgroup, perceived barriers showed high centrality. Its central position suggests that perceived barriers were closely connected with more favorable screening-related beliefs and LCS behavioral progression. Conversely, these findings also imply that perceived barriers may remain important obstacles to screening participation, particularly among individuals with medium-low SES.

For medium-low SES participants, perceived barriers may reflect not only subjective concerns but also real-life constraints, such as time costs, financial burden, transportation difficulties, limited access to screening services, and uncertainty about screening procedures [52, 53]. Therefore, interventions for this group should not rely solely on information provision. Practical measures, such as reducing out-of-pocket costs, simplifying appointment procedures, providing transportation support, integrating screening into township health services, and offering clear explanations of low-dose computed tomography procedures, may help reduce perceived barriers and facilitate screening participation.

In contrast, smoking-related stigma emerged as a more prominent node in the high SES subgroup. This finding suggests that when material and access-related constraints are relatively lower, social-evaluative concerns may become more salient. Smoking-related stigma may reduce screening willingness through increased shame, self-blame, fear of judgment, or reluctance to disclose smoking history [54]. Therefore, screening promotion for individuals with higher SES may benefit from stigma-sensitive communication strategies that frame LCS as a preventive health behavior rather than as a consequence of personal blame.

Taken together, these findings suggest that health belief-related and social-contextual factors may operate differently across socioeconomic contexts. For medium-low SES populations, reducing perceived barriers and improving access may be particularly important. For high-SES populations, reducing smoking-related stigma and improving supportive communication may be more relevant.

SES differences in knowledge, self-efficacy, and screening-related pathways

At the network edge level, this study observed different patterns linking knowledge, self-efficacy, perceived barriers, and LCS behavior across SES groups. In the overall network and the high SES subgroup, the edge between lung cancer and screening-related knowledge and LCS behavior was relatively prominent. This pattern may indicate that, among individuals with relatively more socioeconomic resources, knowledge is more directly connected with screening behavior progression [55].

In contrast, in the medium-low SES subgroup, the edge between self-efficacy and LCS behavior was stronger. This finding suggests that, under more resource-constrained conditions, knowledge alone may not be sufficient for behavioral progression; individuals may also need to believe that they are capable of arranging, attending, and completing screening despite practical barriers [56]. Strengthening self-efficacy may therefore be especially relevant for medium-low SES populations.

Edge difference analysis further revealed that the connection between lung cancer and screening-related knowledge and perceived barriers was stronger in the medium-low SES group. This may suggest that in this group, knowledge is more closely linked to how individuals appraise screening-related barriers, such as time demands, costs, accessibility, and uncertainty [57]. In other words, knowledge may be associated with barrier appraisal before it is associated with screening behavior. However, because this study used a cross-sectional design, this interpretation should be considered hypothesis-generating rather than evidence of a temporal or causal pathway.

In the high SES group, the weaker connection between knowledge and perceived barriers may suggest that barrier appraisal is less dependent on knowledge and more closely related to actual resource availability or other contextual factors [58]. Meanwhile, the stronger knowledge-LCS behavior edge may indicate a more direct cognition-to-action pattern in this subgroup. These SES-specific patterns suggest that LCS promotion strategies may need to be tailored according to socioeconomic context. For medium-low SES individuals, interventions may need to combine knowledge provision with barrier reduction and self-efficacy enhancement. For individuals with high SES, improving knowledge translation and reducing stigma-related concerns may be more appropriate.

Overall, these findings suggest that LCS behavior may be associated with multiple pathways rather than a single linear knowledge-attitude-behavior process. Network analysis provides a useful exploratory framework for identifying potential intervention-relevant factors, but longitudinal and experimental studies are needed to confirm the temporal and causal relationships among these factors.

Limitations

This study has several limitations. First, the cross-sectional design limits causal inference. The observed network edges represent conditional associations rather than causal pathways. Future longitudinal or experimental studies are needed to examine how psychological factors, health beliefs, knowledge, stigma, trust in medical professionals, and SES dynamically influence LCS behavior over time.

Second, the sample was recruited from eight township health centers in Fuzhou and Putian in Fujian Province, using convenience sampling. Although this approach facilitated the recruitment of high-risk individuals in real-world rural primary care settings, it may have introduced selection bias and limited the generalizability of the findings to other regions or rural populations in China. Future studies should use more representative probability sampling methods, such as stratified multistage sampling or probability proportional to size sampling and include more geographically diverse populations.

Third, environmental and occupational exposure history, secondhand smoke exposure, and cooking fumes exposure were self-reported and may therefore be subject to recall bias and exposure misclassification. Although standardized checklists, examples, and investigator training were used to improve consistency, objective exposure assessments were not available.

Finally, some potentially relevant variables, such as family support, healthcare access, transportation distance, physician recommendation, local screening availability, and cultural norms, were not included in the network. Future research should incorporate these factors to provide a more comprehensive understanding of LCS behavior among high-risk rural populations.

Conclusions

In this study, network analysis was used to examine the factors associated with LCS behavior among high-risk rural adults who were recruited from township health centers in Fujian Province, China. The findings suggest that LCS behavior is embedded in a network of psychological, cognitive, health belief-related, and social-contextual factors. Depression played a stable central and bridging role across SES groups, indicating that psychological distress should be considered in screening promotion efforts.

SES-specific network patterns were also observed. Among medium-low SES participants, perceived barriers and self-efficacy were closely connected with the LCS behavioral stage, suggesting that interventions may need to focus on reducing practical barriers and strengthening individuals’ confidence in completing screening. Among high SES participants, knowledge and smoking-related stigma appeared more prominent, suggesting that improving knowledge translation and reducing stigma-related concerns may be particularly relevant. These findings provide preliminary evidence the development of tailored and equity-oriented LCS promotion strategies in rural primary care settings. Longitudinal studies using more representative samples are needed to validate these findings and clarify causal pathways.

Abbreviations

LDCT

Low-dose computed tomography

SES

Socioeconomic status

NCT

Network comparison test

CS-C

Correlation stability coefficient

EBIC

Extended bayesian information criterion

GAD-2

Generalized anxiety disorder scale-2

PHQ-2

Patient health questionnaire-2

LCS

Lung cancer screening

STROBE

Strengthening the Reporting of Observational Studies in Epidemiology

HBM

Health belief model

MGM

Mixed graphical model

Authors' contributions

All the authors meet the authorship criteria of the International Committee of Medical Journal Editors and approved the final manuscript. The study concept and design were performed by FFH. Data acquisition was performed by YLL, YZH, and YAL. Data analysis and interpretation were conducted by FFW. Manuscript drafting was led by FFW and YJC. Critical revision for important intellectual content was provided by RA, WTC, and FFH.

Funding

This research was supported by the National Natural Science Foundation of China (Grant No. 72304068).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

All procedures involving human participants were conducted in accordance with the Declaration of Helsinki. All study participants provided written informed consent, and the study design was approved by the ethics committee at Fujian Medical University (No 2023075).

Competing interests

The authors declare no competing interests.