$\bullet$ Hypertension, chronic obstructive pulmonary disease (COPD), stroke, sleep questionnaire (SLQ), smoking status, PIR, and educational attainment can significantly predict the risk of asthma accompanied by depression. The prevalence rate of depression among asthma patients is 14.45%.

$\bullet$ The eX-treme Gradient Boosting (XGBoost) prediction model showed the best performance in assessing the depression risk of asthma patients, with an area under the curve (AUC) of 0.750.

$\bullet$ The SHapley Additive exPlanations (SHAP) method showed that SLQ, Poverty Index Ratio (PIR), and educational level are the most crucial predictive factors for predicting depression in asthma patients.

Bronchial asthma is a chronic inflammatory airway disease that impacts 262 million people globally [1]. In the United States, approximately 26 million asthmatics are afflicted by this disease, and is accompanied by more than 3000 deaths and a significant health burden of around $81 billion annually [2, 3, 4]. Asthma not only severely affects patients’ quality of life but also triggers psychiatric symptoms. Up to 71.4% of patients experience poorer quality of life and are more prone to psychiatric manifestations such as anxiety and depression [5, 6, 7]. Among patients with asthma and depression, 40% will exhibit symptoms such as poor sleep and insomnia [8, 9]. These issues not only affect the efficacy of asthma control but also lead to refractory asthma, increase the healthcare burden, significantly reduce the quality of life, and raise the risk of death [8]. Studies have indicated that asthma-associated depression is closely related to low-income families, smoking, chronic obstructive pulmonary disease (COPD), hypertension, and stroke [5, 10, 11]. However, there is currently limited research on these risk factors, and clinical predictive models have not been adequately developed. Therefore, it is recommended that a risk prediction model for depression in asthma patients be developed to enhance the management of these risk factors, to reduce the incidence of these co-morbidities.

Machine learning (ML), as a powerful predictive tool, has been widely used in fields such as medicine and engineering [12]. Its predictive accuracy surpasses that of traditional statistical methods [13]. While previous study have developed simple regression predictive models, there is a notable lack of predictive model platforms, which limits their clinical practicality [14]. ML-based algorithms can analyze extensive datasets and uncover subtle predictive risk factors, thereby enhancing clinical predictive capabilities.

This study utilized the dataset of the National Health and Nutrition Examination Survey (NHANES, 2005–2020) to establish machine learning models for predicting the risk of asthma accompanied by depression. A comprehensive machine learning model construction process, including preprocessing, feature selection, model training, and evaluation, was designed. Eight machine learning models were compared to select the model with the best performance as the final prediction model. The aim was to establish an effective prediction platform based on machine learning to provide a theoretical basis for clinical practice and improve the overall health management of the risk of asthma with depression.

Data analysis was conducted using R version 4.3.3 (R Foundation for Statistical Computing, Vienna, Austria; http://www.R-project.org). All data were assessed for normality. Continuous variables that followed a normal distribution were reported as mean $\pm$ standard deviation and analyzed using t-tests, while those that did not conform to normal distribution were presented as the median interquartile range (IQR). Categorical variables were expressed as counts and percentages of patients, and analyzed using the $\chi$² test, with a significance level set at p 0.05 (two-sided).

In the study, eight different machine learning models were used for training and testing: Decision Tree (DT), k-Nearest Neighbor Algorithm (KNN), Light Gradient Booster Algorithm (LGBM), Logistic Regression (LR), Random Forest (RF), Support Vector Machines (SVM), eXtreme Gradient Boosting (XGBoost), and Multilayer Perceptron (MLP). The dataset was divided in a 7:3 ratio, with 70% allocated to the training set and 30% to the validation set. Internal five-fold cross-validation was used, and the evaluation metrics included area under the curve (AUC) and decision curve analysis (DCA). AUC measures the model’s discriminative ability, while DCA assesses the net benefit threshold of the predictions. Additionally, the SHapley Additive exPlanations (SHAP) method was utilized to determine the importance of each variable, highlighting their relative contributions to the model. To facilitate the integration of machine learning into clinical practice, the research team created a predictive modeling website for assessing the risk of asthma-associated depression, accessible at (https://chunyuzhang28.shinyapps.io/asthma/).

Depression is a significant global mental health issue. Genetic, biological, and environmental factors can contribute to comorbid depression in asthma patients. Early prediction and intervention for depression in these individuals are particularly important. ML-based methods have demonstrated promising applications in the diagnosis and treatment of depression. In this study, we developed a predictive model for asthma-related comorbid depression and created an interactive visual prediction platform. Our goal was to offer personalized counseling and enhance medical decision-making, ultimately leading to more effective management strategies in clinical practice. Compared to traditional prediction models, such as nomogram plots, this platform provides more efficient support for clinical applications.

This study referred to previous studies on common risk factors associated with depression in asthma patients, including hypertension, COPD, stroke, SLQ, smoking, PIR and educational level [20, 23]. In a study of 100 patients with bronchial asthma, Zhang et al. [24] found that age $\ge$60 years, asthma control score 20, junior high school or lower education level, rural residence, asthma grade 4 and clinical stage were the main factors for comorbid anxiety and depression. These findings aligned with the results of the current study, based on multivariate retrospective analyses, which showed that hypertension, COPD, stroke, SLQ, smoking, PIR and lower education level increased the risk of comorbid depression in asthma, confirming previous findings.

Behavioral factors such as sleep disturbance and smoking are strongly associated with the onset of depression, particularly in people with asthma. Studies have shown that asthmatics with sleep disturbances are more likely to experience severe depressive symptoms [25, 26]. Approximately 90% of patients experience coughing or wheezing at night, which is associated with nocturnal bronchial hyperresponsiveness and decreased lung capacity [7, 25]. The prevalence of comorbid anxiety and depression is as high as 20% in patients with asthmatic insomnia [9, 27]. Rhoads et al. [5] showed that insomnia not only exacerbates anxiety symptoms in patients with asthma, but can also lead to depressive symptoms such as trouble falling asleep, difficulty staying asleep, and early awakening, which can further exacerbate depressive symptoms [28, 29]. In turn, chronic insomnia can lead to mood problems by affecting brain function and reducing mood regulation through mechanisms that may be related to an imbalance of neurotransmitters involved in mood regulation (e.g., serotonin, norepinephrine and dopamine) and may lead to abnormalities in cortisol secretion, which can induce chronic stress and depression [30]. Similarly, smoking, a risk factor for acute asthma exacerbations and limited irreversible airflow, not only exacerbates asthma symptoms but may also induce persistent limitation airflow [31]. Although the nicotine in smoke may improve mood in the short term, long-term smoking exacerbates asthma and depressive symptoms [32, 33]. Toyama et al. [34] analyzed 1962 asthma patients and found that smokers had a significantly higher prevalence of asthma than non-smokers and that depressed mood had a significant negative effect on asthma control scores. These findings highlight the importance of smoking cessation in the health management of asthma patients.

Socioeconomic factors such as PIR and education level play a key role in the health status of asthma patients [35]. Low-income families face greater economic pressures, including medical, educational and living expenses, and this chronic economic pressure increases the risk of depression in asthma patients [10]. Moreover, education level has a significant impact on managing depression in asthma patients. Tarraf et al. [36] found that less than 30% of asthma patients had good control and that the level of control was closely related to smoking frequency, lower educational level and inappropriate use of medication. Asthmatics with lower education levels often lack awareness of mental illness and effective coping strategies, making them more susceptible to negative emotions like anxiety and depression. Therefore, attention should be paid to the family income and the education level of asthma patients to reduce the risk of depression through early detection and prevention.

Disease factors such as COPD, hypertension and stroke may negatively affect the mental health of asthma patients and increase the risk of depression mood [11]. These chronic diseases not only lead to symptoms such as dyspnea, but also reduce the quality of life of patients, which increases the risk of depression. Chronic health problems may also increase psychological stress [37, 38]. Notably, these diseases may be interrelated, for example, patients with COPD are often associated with complications such as hypertension, which may further exacerbate depression. Therefore, it is recommended to enhance chronic disease management in patients with asthma to reduce the risk of depression.

XGBoost is an integrated decision tree-based learning algorithm widely used in risk prediction, feature selection and clinical decision support, which has the advantages of fast speed, good results, high flexibility and high predictive accuracy. In this study, it was found that the prediction model of XGBoost performed the best in terms of synthesis and performance compared to algorithms such as DT, KNN, LGBM, LR, RF, SVM, MLP, etc., and was especially optimal in predicting asthma-combined with depression. Li et al. [39] found that the Cat Boost model was suitable for depression screening in older adults using data from the China Health and Retirement Longitudinal Study (CHARLS). Su et al. [40] predicted risk factors for depression in older adults using eight ML models and found that self-rated health, marital status, arthritis, and number of cohabits were key factors. Hatton et al. [41] confirmed that the XGBoost model outperformed the traditional logistic regression model in predicting depression in elderly patients. Xia et al. [42] also confirmed that the prediction of depression by XGBoost was superior to that of DT, SVM, Multivariate Adaptive Regression Splines (MARS), Artificial Neural Network (ANN), BT, RF, KNN and other models, supporting the findings of this study.

In conclusion, prevention of depression in asthma patients should focus on those patients with sleep disturbances, lower family economic income and lower educational levels. XGBoost has significant advantages in predicting depression in asthma patients, and by incorporating seven independent risk factors, it helps in the prevention and management of depression improve their mental health.

This study has several limitations: First, all participants were Americans aged $\ge$20 years and the findings may not be generalizable to other populations. Second, the diagnoses of asthma and depression in the NHANES data were based on participants’ self-reports, which might have retrospective biases and consequently affect the reliability of the results; Third, the lack of an external validation cohort in the NHANES data leads to the possibility that the applicability of the developed XGBoost model in clinical practice may be limited; Fourth, although the NHANES dataset covers a wide range of health and nutrition-related variables, it still does not include all potentially important variables, such as certain socioeconomic factors, environmental factors, or genetic background, which limits a comprehensive understanding of complex health issues. It is recommended that in the future, the XGBoost model be applied to asthma patients in different countries and regions to predict the risk of depression onset, so as to conduct internal and external validation of the model and improve its stability.

The interpretable XGBoost prediction model we developed performs optimally in assessing the risk of depression in asthma patients. This interpretable machine learning approach accurately identifies risk factors for depression in asthma patients and enhances physicians’ confidence in the prediction results, thus helping them to identify asthma patients at high risk of depression and provides them with the best treatment options.

These survey data are free and publicly available, and can be downloaded directly from the NHANES website (http://www.cdc.gov/nchs/nhanes.htm) by users and researchers worldwide.

Concept—QN, YXW, YLX, WL; Design—QN, XD, XC, TWL, WL; Supervision—YTL, XC, QSR, JJL, JYL; Resources—YLX, YXW; Materials—QN, YXW, YLX, WL; Data Collection and/or Processing—QN, XD, XC, TWL, WL, QSR, JJL, YTL, JYL, YXW, YLX; Analysis and/or Interpretation—QN, YXW, YLX, WL, XD, QSR, YTL, JYL, JJL; Literature Search—QN, YXW, YLX, JJL, JYL; Writing—QN, YXW, YLX, WL, TWL, XD; Critical Review—QN, YXW, YLX. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

Not applicable.

This research received no external funding.

The authors declare no conflict of interest.