Development of advanced lung cancer inflammation index-based machine learning models for predicting stroke and...

Area of Science:

• Cardiovascular Research
• Biostatistics
• Computational Biology

Background:

• The Advanced Lung Cancer Inflammation Index (ALI) offers a comprehensive assessment of inflammation and nutritional status.
• Existing research has not extensively explored the role of ALI in stroke patients.
• This study investigates ALI's association with stroke risk and mortality, alongside developing predictive machine learning (ML) models.

Purpose of the Study:

• To examine the relationship between the Advanced Lung Cancer Inflammation Index (ALI) and stroke risk.
• To assess the association between ALI and all-cause mortality in stroke patients.
• To develop and interpret ML models for predicting stroke and patient prognosis.

Main Methods:

• Utilized data from the National Health and Nutrition Examination Survey (NHANES) (1999-2018).
• Employed logistic regression, Cox regression, and restricted cubic splines to analyze ALI associations with stroke and mortality.
• Developed and evaluated six ML models (LR, XGBoost, RF, KNN, SVM, DT) for prediction, using AUCROC and accuracy metrics, enhanced by SHAP and Gini importance.

Main Results:

• Higher ALI correlated with reduced stroke risk.
• Mortality decreased with increasing ALI up to an inflection point (ALI = 40.91).
• Random Forest (RF) models demonstrated strong performance in stroke identification (AUCROC: 0.9657) and mortality prediction (AUCROC: 0.7771). Cardiovascular diseases and age were key predictors for stroke and mortality, respectively.

Conclusions:

• ALI exhibits a reverse dose-response relationship with stroke risk and an "L-shaped" association with all-cause mortality in stroke patients.
• Dual-interpretable RF models incorporating ALI show potential for stroke identification and prognosis prediction.
• This study highlights ALI as a relevant biomarker in cardiovascular health and stroke outcomes.