A study on lipid and homocysteine imbalance in Tibetan patients with acute ischemic stroke in Xizang and a machine learning-based case-control classification model

Abstract Objective This retrospective case-control study examined lipid and homocysteine (Hcy) patterns in Tibetan patients with acute ischemic stroke (AIS) in Xizang and developed an exploratory model to distinguish AIS cases from controls. Methods We enrolled 549 Tibetan AIS patients and 559 age- and sex-matched Tibetan controls. Clinical and biochemical indicators were compared between groups. XGBoost feature importance was used only for exploratory feature selection, and a random forest model was then constructed to evaluate the case-control discriminative ability of the selected indicators. Model performance was internally evaluated using AUC, recall rate, and F1 score. Results Compared with controls, AIS patients had lower high-density lipoprotein cholesterol (HDL), lower total cholesterol (TC), and higher Hcy levels. The combined lipid-Hcy imbalance pattern was defined as the co-occurrence of lower HDL and TC levels and higher Hcy levels in AIS patients than in controls. HDL, TC, and Hcy were consistently selected as AIS-associated indicators. The optimized random forest model based on these three indicators achieved an AUC of 0.924 in cross-validation and 0.887 in the test set. Conclusion HDL, TC, and Hcy were associated with AIS case-control status in Tibetan participants from Xizang and should be interpreted as associative markers rather than causal factors or diagnostic biomarkers. Because this was a retrospective case-control study and biomarkers were measured after AIS onset in cases, the model demonstrates only internal case-control discrimination and does not establish pre-onset predictive value. Prospective, multi-center external validation is needed before clinical use.