The data used in this study were obtained from the NHANES [15]—a comprehensive national cross-sectional survey conducted by the US Centers for Disease Control and Prevention (CDC) to assess the health and nutritional status of the American population. The NHANES protocol was approved by the Ethics Review Committee of the National Center for Health Statistics (NCHS), and all participants provided written informed consent before their inclusion [16].

A total of 70,190 individuals were initially identified from the NHANES database covering the period from 2005 to 2018. The inclusion criteria for this analysis were (1) adults aged 20–80; (2) individuals diagnosed with CHD; (3) participants with a complete Patient Health Questionnaire-9 (PHQ-9) score; (4) those with complete data for all relevant study indicators. Exclusion criteria were (1) individuals younger than 20 or older than 80 or with missing age information; (2) those without a CHD diagnosis; (3) participants with incomplete or anomalous PHQ-9 scores; (4) those with missing data for other relevant indicators. After applying these criteria, 1098 individuals with CHD met the study requirements and were included. A detailed participant selection process flowchart is presented in Fig. 1.

Fig. 1.

Flowchart of the research participant screening process. NHANES, the National Health and Nutrition Examination Survey; PHQ, Patient Health Questionnaire-9.

The PHQ-9 comprises nine items and was used to screen for depression in the NHANES database and assess the frequency of depressive symptoms over the prior two weeks [17]. Each item offered four response options: “almost every day”, “more than half the days”, “several days”, and “never”, scored as 3, 2, 1, and 0 points, respectively. The total score was calculated by summing the individual scores, with a possible scoring range of 0 to 27. Participants scoring 10 or higher on the PHQ-9 were categorized as having major depression, while those with scores below 10 were not [18, 19]. CHD was defined as self-reported CHD, angina pectoris, or a history of myocardial infarction in the NHANES database [20].

Additionally, various demographic and health-related details were extracted from the NHANES database. Basic demographics included age, gender, education, family poverty–income ratio (PIR), marital status, health insurance coverage in the previous year, race, smoking status, alcohol consumption, recent anemia treatment (within three months), overweight status, sleep disorders, and physical activity levels. Physical measurements and laboratory data included height, weight, waist circumference, and hemoglobin levels. Comorbidities encompass hypertension, asthma, arthritis, congestive heart failure, stroke, emphysema, chronic bronchitis, liver disease, thyroid disorders, and diabetes. Gender was categorized as male or female. Education levels were grouped into high school or below, high school graduate/equivalent, some college/AA degree (Associate of Arts degree, a two-year undergraduate academic degree), and college graduate or higher.

The family PIR, representing the household income ratio to the poverty threshold, was based on guidelines from the US Department of Health and Human Services and categorized into less than 1, between 1 and 3, and greater than 3 [21]. Marital status was classified as married/living with a partner, widowed/divorced/separated, or never married. Racial categories included Mexican American, non-Hispanic White, other Hispanics, non-Hispanic Black, and other/multi-racial individuals. Smoking status was divided into daily, some days, or never, while alcohol consumption was categorized into yes or no. Sleep disorders were defined by self-reported difficulties [22]. The need for assistive walking devices was defined as having a walking impairment [22]. Work restrictions were defined as the inability to work due to chronic health conditions [22]. Body mass index (BMI) was calculated using anthropometric data collected by trained examiners at the Mobile Examination Center (MEC). Weight was measured in kilograms using a digital scale, and height in centimeters was measured using a stadiometer. BMI was then derived using the formula: BMI = weight (kg)/(height (m))² [23]. Hypertension was defined as self-reported hypertension, current or previous use of antihypertensive medication, or an average systolic blood pressure of 130 mmHg or higher and/or an average diastolic blood pressure of 80 mmHg or higher [24]. Diabetes history was identified through self-reported diabetes, current diabetes treatment, or a glycosylated hemoglobin level of 6.5% or higher [25]. A random sampling method was applied to divide the dataset into development and validation groups. Each of the 1098 patient records was initially assigned a unique identifier. A random number of seeds of 123 ensured the randomization process was replicable. A random uniform variable was then generated for each patient, and the dataset was sorted according to this variable. From this, a simple random sample of 70% of the records was selected to form the development group. Specifically, the first 769 patients, as ordered by the random variable, were placed in the development group (coded as 1), while the remaining 329 patients constituted the validation group (coded as 2). The dataset was subsequently re-sorted using the original patient identifier to maintain its initial order. This procedure resulted in a development group comprising 70% of the dataset and a validation group representing the remaining 30%.

Statistical analyses were performed using Stata (version 17.0, Stata Corp, TX, USA). Before the analysis, the data were adjusted using three NHANES weighting indicators: “SDMVPSU”, “SDMVSTRA”, and “WTMEC2YR”, to account for sampling variability and enhance the accuracy of results. Weighted measurement data conforming to a normal distribution were reported as the mean and standard deviation, and independent samples t-tests were used to compare the two groups. Categorical variables were presented as frequencies and percentages, with Pearson’s chi-square test used for group comparisons. Weighted univariate and multivariate logistic regression analyses were conducted to identify independent risk factors for major depression among patients with CHD. A nomogram model was developed to predict the likelihood of major depression in these patients. The discriminative power of the model was evaluated using the area under the receiver operating characteristic (ROC) curve, while its calibration was assessed using a calibration curve and Brier score. Clinical utility was evaluated via decision curve analysis (DCA). All statistical tests were two-tailed, and a p-value of less than 0.05 determined significance.

The weighted mean age of the 1098 patients was 64.89 years, with a standard error of 0.440 years. Regarding gender distribution, 66.24% of the sample were men, and 33.76% were women. The weighted prevalence of major depression among patients with CHD was 13.95%. Additional descriptive statistics of the weighted data are provided in Table 1. No significant differences were observed between the weighted data of patients in the development and validation groups (p $>$ 0.05), suggesting that the two groups were comparable and well-balanced, as outlined in Table 1.

The results of the weighted univariate logistic regression analysis in the development group identified sixteen indicators as significant predictors for major depression in patients with CHD (p 0.05). These factors included age, BMI, waist circumference, sleep duration, gender, educational level, poverty–income ratio, smoking status, asthma, arthritis, congestive heart failure, stroke, chronic bronchitis, sleep disorders, work limitations, and walking impairments. A detailed summary of these findings is presented in Table 2.

Sixteen variables, identified as potential risk factors through weighted univariate logistic regression, were incorporated as independent variables, with the incidence of major depression among patients with CHD as the dependent variable. A weighted multivariate regression model was developed using the stepwise regression method based on the minimum Akaike information criterion (AIC). This iterative process sought to determine the most parsimonious model that best fit the data. The final model, with a minimum AIC of 482.96, included nine variables. Among these, the independent risk factors for major depression in patients with CHD were waist circumference, smoking status, arthritis, sleep disorders, and work restrictions. Detailed results can be found in Table 3.

The predictive model developed using the minimum AIC stepwise regression analysis offers a key advantage in identifying variables with the greatest impact on prediction outcomes. This approach refines the model, enhancing both its accuracy and clinical applicability [26]. Consistent with practices employed in similar studies [27, 28] and drawing on the nine predictive indicators identified from the stepwise multivariate regression analysis based on the minimum AIC, a nomogram model was constructed to predict the risk of major depression within the development group. The model was developed using Stata 17.0, as illustrated in Fig. 2.

Fig. 2.

Nomogram model for predicting the risk of major depression in patients in the development group.

The nomogram interpretation is as follows: Each of the nine variables included in the nomogram was assigned a specific score, which can be determined by referencing the horizontal axis labeled “Score”. For each variable, the corresponding score can be identified by drawing a vertical line downward from the value of the variable. Once the scores for all nine variables are obtained, they are summed to calculate the “Total Score”. To estimate the risk, a vertical line is drawn upward from the horizontal axis labeled “Total Score” to the axis marked “Risk of Depression”. The value intersected on the “Risk of Depression” axis provides the predicted risk of major depression for patients with CHD. To facilitate accurate interpretation of the scores for each variable and to determine the major depression risk associated with the total score, a risk score table has been constructed as a supplementary tool to the nomogram. This table offers a clear and user-friendly reference for interpreting risk levels and is displayed in Table 4.

The DCA curve was employed to evaluate the clinical utility of the nomogram-based predictive model for major depression risk in the development group. The curve demonstrates that the nomogram achieves the highest clinical net benefit when the threshold probability for major depression in patients with CHD ranges from 0.04 to 0.54 in the development group and from 0.08 to 0.52 in the validation group. This performance significantly surpasses the “Treat All” and “Treat None” strategies, underscoring the strong clinical applicability of the nomogram. A detailed representation of these results is provided in Fig. 3.

Fig. 3.

Decision curve analysis (DCA) of the nomogram model. Note: (A) The DCA of the nomogram model in the development group. (B) The DCA of the nomogram model in the validation group.

The areas under the ROC curve for the nomogram model were 0.816 (95% CI: 0.776–0.857) in the development group and 0.765 (95% CI: 0.699–0.832) in the validation group, demonstrating good discriminatory power. For further details, see Fig. 4. The Brier scores for the development and validation groups were 0.107 and 0.127, respectively, which are well below the 0.25 threshold, indicating a strong degree of calibration. This calibration analysis is illustrated in Fig. 5.

Fig. 4.

Receiver operating characteristic (ROC) curve of the nomogram model. Note: (A) ROC curve of the nomogram model in the development group. (B) ROC curve of the nomogram model in the validation group.

Fig. 5.

Calibration curve of the nomogram model. Note: (A) Calibration curve of the nomogram model in the development group. (B) Calibration curve of the nomogram model in the validation group. CI, confidence interval.

To validate the predictive model for varying levels of depression severity, the performance of the model was evaluated using PHQ-9 scores of 5, 15, and 20 as thresholds, in addition to the established cutoff of 10 for major depression.

For a PHQ-9 score of 5, representing mild depression, the model showed an area under the ROC curve (AUC) of 0.810 in the development group and 0.797 in the validation group, demonstrating good accuracy in detecting individuals with mild depression.

At a PHQ-9 score of 15, indicating moderately severe depression, the model produced an AUC of 0.811 in the development group and 0.747 in the validation group, suggesting moderate effectiveness in distinguishing individuals with moderately severe depression.

For a PHQ-9 score of 20, representing severe depression, the model consistently achieved an AUC of 0.793 in both groups, indicating reliable identification of severe depression, a group at high risk for adverse outcomes.

These results highlight the predictive effectiveness of the model in categorizing different depression severity levels based on PHQ-9 scores. Furthermore, the consistent performance of the model across severity thresholds offers valuable insights for clinical decision-making and emphasizes the need for targeted interventions in patients with higher PHQ-9 scores.

Several limitations exist in this study that should be acknowledged: (1) While the NHANES database is nationally representative, the selection and participation of the sampled population may affect the generalizability of the model, particularly limiting its applicability to individuals outside the CHD population. (2) The nomogram model relies on lifestyle, behavioral information, and depression scores derived from self-reported questionnaires, which could introduce recall bias and affect the accuracy of data collection. (3) This study has been validated internally but lacks external validation, a critical limitation. While internal validation assesses model performance within the study sample, it does not provide insights into the model’s generalizability across diverse populations or clinical settings. Thus, the applicability of the model to different populations, such as various ethnicities, genders, or age groups, and clinical contexts remains uncertain without external validation. Incorporating external datasets for validation would greatly enhance the generalizability and utility of this model. (4) The model does not account for other potentially relevant predictors, such as laboratory markers, such as glucose metabolism, or lipid-related indexes, which could influence its predictive accuracy. (5) The nomogram model demonstrates correlations between the included indicators and depression risk in CHD but does not establish causal relationships. Further prospective research involving larger sample sizes and rigorous clinical designs is required to explore the extrapolation and generalizability capabilities of the model. (6) Additional sensitivity analyses, such as bootstrap methods, cross-validation, or penalized regression techniques, such as Lasso or ridge regression, could enhance the reliability and predictive performance of the model. (7) A more detailed subgroup analysis could clarify whether the risk factors for depression vary significantly by gender or ethnicity, offering more personalized insights and improving the model’s relevance to specific populations.