Scholarlink

Background <p>Asthma is a common chronic inflammatory airway disease. Accumulating evidence highlights the roles of demographic, lifestyle, and comorbidity factors in the risk of asthma. This study aimed to identify predictor factors of asthma using machine learning approaches.</p> Methods <p>Data were obtained from the 10th wave (2021–2023) of the English Longitudinal Study of Ageing (ELSA). Participants aged ≥ 50 years with complete information on asthma status and relevant variables were included. Baseline characteristics were compared between asthma and control groups. Subsequently, Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to identify candidate variables. Eight machine learning algorithms were developed and compared to evaluate diagnostic performance. The optimal model was selected and used to determine key variables. Additionally, SHapley Additive exPlanations (SHAP) analysis was applied to interpret variable contributions. Finally, a nomogram was constructed based on the key variables.</p> Results <p>A total of 3429 participants (535 asthma cases) were analyzed. Asthma was significantly associated with 19 baseline variables. LASSO regression retained 14 candidate variables. Among eight machine learning models, the Bagging Tree (BT) model achieved the highest diagnostic performance (micro-averaged area under the curve (AUC) = 0.856; macro-averaged AUC = 0.881). SHAP analysis identified alcohol consumption, marital status, and disease lung as the most influential variables. A total of 11 key variables were identified by the BT model, including marital status, vigorous physical activity, moderate physical activity, alcohol consumption, frequency of feeling isolated, depression, headache, activity limitations, disease lung, arthritis, and psychiatric disease. The nomogram showed good calibration (Hosmer-Lemeshow test <i>p</i> = 0.0857), but its discriminatory ability was moderate (AUC = 0.662).</p> Conclusions <p>This study demonstrated that socio-behavioral factors, psychological distress, and respiratory comorbidities played important roles in asthma risk stratification. Machine learning with multidimensional variables offers a useful exploratory framework for identifying potential predictor factors and generating hypotheses for asthma prevention, although its predictive accuracy remains moderate.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Identifying predictor factors for asthma using machine learning: evidence from the English Longitudinal Study of Ageing

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月3日

查看全文
  • Zhao Chen Yang,
  • Zhou Xiao

摘要

Background

Asthma is a common chronic inflammatory airway disease. Accumulating evidence highlights the roles of demographic, lifestyle, and comorbidity factors in the risk of asthma. This study aimed to identify predictor factors of asthma using machine learning approaches.

Methods

Data were obtained from the 10th wave (2021–2023) of the English Longitudinal Study of Ageing (ELSA). Participants aged ≥ 50 years with complete information on asthma status and relevant variables were included. Baseline characteristics were compared between asthma and control groups. Subsequently, Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to identify candidate variables. Eight machine learning algorithms were developed and compared to evaluate diagnostic performance. The optimal model was selected and used to determine key variables. Additionally, SHapley Additive exPlanations (SHAP) analysis was applied to interpret variable contributions. Finally, a nomogram was constructed based on the key variables.

Results

A total of 3429 participants (535 asthma cases) were analyzed. Asthma was significantly associated with 19 baseline variables. LASSO regression retained 14 candidate variables. Among eight machine learning models, the Bagging Tree (BT) model achieved the highest diagnostic performance (micro-averaged area under the curve (AUC) = 0.856; macro-averaged AUC = 0.881). SHAP analysis identified alcohol consumption, marital status, and disease lung as the most influential variables. A total of 11 key variables were identified by the BT model, including marital status, vigorous physical activity, moderate physical activity, alcohol consumption, frequency of feeling isolated, depression, headache, activity limitations, disease lung, arthritis, and psychiatric disease. The nomogram showed good calibration (Hosmer-Lemeshow test p = 0.0857), but its discriminatory ability was moderate (AUC = 0.662).

Conclusions

This study demonstrated that socio-behavioral factors, psychological distress, and respiratory comorbidities played important roles in asthma risk stratification. Machine learning with multidimensional variables offers a useful exploratory framework for identifying potential predictor factors and generating hypotheses for asthma prevention, although its predictive accuracy remains moderate.