Background: The risk of stroke in individuals with very low low-density lipoprotein cholesterol (LDL-C) concentrations remains high. We sought to prioritize predictive risk factors for stroke in Chinese participants with LDL-C concentrations < 70 mg/dL using a survival conditional inference tree, a machine learning method.
Methods: The training dataset included 9327 individuals with LDL-C concentrations < 70 mg/dL who were free of cardiovascular diseases and did not use lipid-modifying drugs from the Kailuan I study (N = 101,510). We examined the validity of this algorithm in a second Chinese cohort of 1753 participants with LDL-C concentrations < 70 mg/dL from the Kailuan II study (N = 35,856).
Results: During a mean 8.5-9.0-year follow-up period, we identified 388 ischemic stroke cases and 145 hemorrhagic stroke cases in the training dataset and 20 ischemic stroke cases and 8 hemorrhagic stroke cases in the validation dataset. Of 15 examined predictors, poorly controlled blood pressure and very low LDL-C concentrations (≤ 40 mg/dL) were the top hierarchical predictors of both ischemic stroke risk and hemorrhagic stroke risk. The groups, characterized by the presence of 2-3 of aforementioned risk factors, were associated with a higher risk of ischemic stroke (hazard ratio (HR) 7.03; 95% confidence interval (CI) 5.01-9.85 in the training dataset; HR 4.68, 95%CI 1.58-13.9 in the validation dataset) and hemorrhagic stroke (HR 3.94, 95%CI 2.54-6.11 in the training dataset; HR 4.73, 95%CI 0.81-27.6 in the validation dataset), relative to the lowest risk groups (presence of 0-1 of these factors). There was a linear association between cumulative average LDL-C concentrations and stroke risk. LDL-C concentrations ≤ 40 mg/dL was significantly associated with increased risk of ischemic stroke (HR 2.07, 95%CI 1.53, 2.80) and hemorrhagic stroke (HR 2.70, 95%CI 1.70, 4.30) compared to LDL-C concentrations of 55-70 mg/dL, after adjustment for age, hypertension status, and other covariates.
Conclusion: Individuals with extremely low LDL-C concentrations without previous lipid-modifying treatment could still be at high stroke risk.
Trial registration: Chinese Clinical Trial Register, ChiCTR-TNRC-11001489 . Registered on 24-08-2011.
Keywords: Conditional inference tree; Low-density lipoprotein cholesterol; Machine learning; Metabolic diseases; Stroke.