This study employed a cross-sectional design. Data collection took place from May to November 2024 at two Grade-A tertiary hospitals in Liaoning Province, China.

A multistage stratified random sampling method was used. Inclusion criteria were: (1) diagnosis of colorectal cancer confirmed by clinical pathology; (2) voluntary provision of informed consent; (3) clear consciousness and adequate communication ability; (4) age ≥18 years. Exclusion criteria were: (1) critically ill patients unable to complete the process; (2) clinically assessed life expectancy of less than 6 months; (3) presence of cognitive impairment or diagnosed psychiatric disorders; (4) history of psychological interventions that could interfere with study outcomes; (5) individuals unaware of their diagnosis.

The sampling procedure consisted of three stages. Stage 1: Within Liaoning Province, eight Grade III Class A hospitals with specialized oncology departments were selected based on national rankings. Using hospital primary specialty as the stratification variable and a random number table, Dalian University of Technology Affiliated Cancer Hospital and Jinzhou Medical University Affiliated Hospital were selected (sampling proportion 25%). Stage 2: Within each hospital, departments admitting colorectal cancer patients were listed and stratified by primary treatment modality. Using random number tables, selected departments were: Dalian University of Technology Affiliated Cancer Hospital: Colorectal Surgery Wards 5 and 6, Department of Integrated Traditional and Western Medicine; Jinzhou Medical University Affiliated Hospital: General Surgery Colorectal Wards 1 and 3, Oncology Department. Stage 3: Researchers visited selected departments weekly. Eligible patients were screened from the daily inpatient lists. Systematic random sampling was applied: eligible patients were listed by bed number, a random starting point was determined, and every other patient (k=2) was invited. This ensured approximately 50% systematic sampling of daily eligible patients, enabling efficient recruitment while maintaining randomness.

Sample size was determined using a dual approach. First, based on the factor analysis guideline of 10–20 participants per scale dimension, and considering the 7 dimensions across the scales used in this study, a preliminary sample size estimate ranged from 70 to 140 (24). Second, an a priori power analysis was conducted using G*Power 3.1.9.4 software. Specifically, we selected the statistical test “F-test—Linear Multiple Regression: Fixed Model, R² Deviates from Zero.” The analysis targeted the hypothesized moderated mediation model, which included 12 predictor variables (8 demographic variables, perceived stress, illness perception, self-efficacy, and their interaction term). Parameters were set as α = 0.05 (two-tailed), statistical power = 0.95, and a medium effect size (f² = 0.15) (25). This calculation yielded a minimum required sample size of 221. Accounting for an estimated 20% rate of invalid questionnaires or attrition, the target sample size was adjusted to 277 [221/(1-0.20)]. In practice, to ensure data quality and representation, the scope of recruitment was expanded. A total of 319 questionnaires were distributed, and 290 valid questionnaires were ultimately retained (159 from Dalian University of Technology Affiliated Cancer Hospital and 131 from Jinzhou Medical University Affiliated Hospital), resulting in a valid response rate of 90.91%.

Before data collection, all research personnel completed a standardized training program (26–28). Standardized training refers to structured, uniform training procedures implemented for all research personnel to ensure the consistency, accuracy, and reliability of research data. The program consisted of two key modules: (1) scoring criteria and tool familiarization to ensure consistent understanding of the measurement tools (Perceived Stress Scale, Self-Efficacy Scale, Illness Perception Scale, and Depression Scale) and their scoring protocols; and (2) patient-centered communication skills, focusing on techniques for building rapport, delivering clear and neutral instructions, and creating a supportive environment to facilitate honest and accurate responses.

This study was approved by the Research Ethics Committee of Jinzhou Medical University (Approval No. JZMULL2023029). All procedures were in accordance with the ethical standards of the institutional committee and the Declaration of Helsinki. Written informed consent was obtained from all participants.

Data analysis was conducted using IBM SPSS Statistics 25.0 and the PROCESS macro version 3.5. Normality of continuous variables was assessed via the Kolmogorov-Smirnov test; all variables were normally distributed (P > 0.05). Descriptive statistics included frequencies/percentages for categorical variables and mean ± standard deviation for continuous variables. Independent samples t-tests and one-way ANOVA were used to examine differences in depression scores across demographics. Pearson correlation analysis evaluated associations among study variables. Moderated mediation analysis was performed using PROCESS macro 3.5 (Model 14 for moderation, Model 4 for mediation). Effects were examined using bootstrap resampling (5,000 samples) to generate 95% bias-corrected confidence intervals (CI). A two-tailed P < 0.05 was considered statistically significant.

Table 1 presents demographic characteristics and univariate analysis. Participants’ mean age was 59.58 ± 13.09 years (range 19-90). The sample comprised 149 males (51.4%) and 141 females (48.6%). Most were married (81.0%). Univariate analysis revealed significant associations between depression and age, gender, education level, marital status, hospitalization count, and pathological stage (P < 0.05). These variables were retained as covariates in subsequent analyses.

As shown in Table 2, perceived stress was negatively correlated with self-efficacy and positively correlated with illness perception and depression (P < 0.01). Self-efficacy was negatively correlated with illness perception and depression (P < 0.01). Illness perception was positively correlated with depression (P < 0.01). Mean scores were: perceived stress 29.56 ± 6.45, self-efficacy 32.20 ± 7.15, illness perception 29.33 ± 4.63, depression 9.19 ± 4.58.

As shown in Table 3, after controlling for covariates, mediation analysis (PROCESS Model 4) indicated a significant positive correlation between perceived stress and depression (β = 0.483, P < 0.001). When illness perception was introduced as a mediator, the direct effect remained significant (β = 0.400, P < 0.001). Perceived stress was positively correlated with illness perception (β = 0.515, P < 0.001), and illness perception was positively correlated with depression (β = 0.161, P < 0.001). The indirect effect was significant (B = 0.083, SE = 0.026, 95% CI = [0.036, 0.137]), indicating partial mediation. The mediating effect accounted for 17.18% of the total effect (Table 4).

Moderated mediation analysis (PROCESS Model 14) was conducted, controlling for covariates. As shown in Table 5, the interaction term between perceived stress and self-efficacy was positively correlated with illness perception (B = 0.024, P < 0.001), indicating that self-efficacy moderated the “perceived stress → illness perception” pathway.

To illustrate this moderation, self-efficacy was categorized into three levels (low: M - 1SD; medium: M; high: M + 1SD). Simple slope analysis (Table 6; Figure 2) showed that among participants with low self-efficacy, perceived stress significantly positively predicted illness perception (B = 0.233, SE = 0.077, t = 3.026, P < 0.01). This positive predictive effect was stronger among participants with high self-efficacy (B = 0.575, SE = 0.066, t = 8.655, P < 0.001).

Overall, perceived stress directly influenced depression and indirectly affected it through illness perception. Self-efficacy moderated the “perceived stress → illness perception” pathway (Figure 3).

This study has several limitations. First, the cross-sectional design can only reveal correlations between variables and cannot infer causal directionality. Future longitudinal or experimental studies may be conducted to further validate the temporal sequence and causal relationships among variables. Second, the sample originates from two hospitals within the same province, potentially limiting the representativeness of the results. Subsequent research may expand the sampling scope to enhance external validity. Additionally, all data were collected through self-reported questionnaires, potentially subject to social desirability and recall bias. Future studies should integrate multi-source data, such as physiological indicators and clinical assessments, to enhance measurement objectivity and strengthen the robustness of conclusions.

Based on the moderated mediation model established in this study, a multidimensional psychological support program for colorectal cancer patients should be designed across three levels: First, integrate the Brief Perceived Stress Scale, the Illness Perception Questionnaire, and the General Self-Efficacy Scale into routine care for systematic screening. Patients can be categorized based on assessment results to inform subsequent support program development (46, 47). Second, for patients with elevated perceived stress levels, structured programs such as mindfulness-based stress reduction training and relaxation technique instruction should be offered. Patients with negative illness perception may benefit from visual disease education and cognitive restructuring dialogues to foster more adaptive illness understanding. Those with low self-efficacy can participate in symptom management skill workshops and peer experience-sharing groups, providing platforms for learning and exchange (48). Third, establish a multidisciplinary collaboration mechanism involving healthcare providers and mental health professionals. Integrate digital health tools (e.g., health education apps, mood diary mini-programs) as extensions of in-hospital support. Implement regular follow-up protocols to tailor support content according to patients’ conditions at different treatment stages (49, 50). These comprehensive measures will help build a more holistic support system, thereby playing a positive role in reducing depression risk among colorectal cancer patients.