.png){kind=icon}
.png){kind=icon}
Return
Machine learning-based identification of blood lipid-related factors associated with COPD among non-smoking women in the Yili region of Xinjiang, China
.png){kind=icon}
OA
AI
S
B
C
S
B
L
DOI：10.1186/s40001-026-04929-1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Chronic obstructive pulmonary disease (COPD) in non-smoking women, particularly in rural regions, remains poorly understood. This study employed machine learning to identify key associated factors for COPD among never-smoking women in the Yili region of Xinjiang, China. A cross-sectional study was conducted involving 115 COPD patients and 50 healthy controls. All participants were non-smoking Kazakh or Uyghur women aged ≥ 40 years. Twelve machine learning algorithms were evaluated using entropy-weighted TOPSIS, with the optimal model selected for SHAP interpretability analysis. Traditional logistic regression was performed for comparison. Gradient Boosting Regression Tree (GBRT) achieved the highest TOPSIS score and demonstrated robust performance (AUC = 0.857 on test set). SHAP analysis identified total cholesterol/high-density lipoprotein cholesterol (TC/HDL) ratio, age, high-density lipoprotein cholesterol (HDL-C), and hemoglobin as the most influential predictors. The TC/HDL ratio showed a monotonic inverse association with COPD probability, with an inflection point at approximately 2.95. Age exhibited a nonlinear effect, with probability markedly increasing after 67 years. Hemoglobin demonstrated a U-shaped relationship, with both low and high levels associated with increased probability. A synergistic interaction between low TC/HDL ratio and advanced age was markedly associated with elevated model-predicted COPD probability. In contrast, multivariate logistic regression retained only HDL-C as a significant predictor, failing to capture these nonlinear relationships. Machine learning models effectively identified TC/HDL ratio and age as synergistic core factors of COPD in never-smoking women, revealing complex nonlinear relationships undetected by conventional methods. These findings suggest a potential departure from the traditional smoking-centric paradigm in this specific population and highlight the need for population-specific associated stratification strategies.
Keywords:
Chronic obstructive pulmonary disease
Machine learning
Blood lipids
Never-smoking women
Journal
IF:
3.4
Papers:
2.0K
Citations:
5.9K
Organization
Cited Papers