Clinical and laboratory characteristics of study participants
A total of 566 patients were included in the AIS group and 197 age‐matched non-AIS patients were set as control group. The baseline characteristics of both study groups are shown in Table 1. The BMI, the proportion of patients with hypertension, diabetes mellitus, metabolic syndrome, and cardiovascular disease, SBP, DBP, Scr, LDL-C, TG, Apolipoprotein A (ApoA), and sd-LDL were significantly higher than those in the control group (P < 0.05). Smoking, and alcohol consumption, and GLU, TC, and Apolipoprotein B (ApoB) did not differ between the two groups, while the HDL-C concentration was significantly higher in the control group.
Distribution of LDL subtractions in AIS patients and controls
As shown in Table 2, levels of LDL-3, LDL-4 and LDL-5 subclasses were significantly higher in the AIS group compared to the control group (p<0.05). In contrast, LDL-1 was significantly lower in the AIS group, while there was no difference between the AIS and control groups in terms of LDL-2, LDL-6 and LDL-7 (p>0.05).
Correlations of clinical characteristics and AIS risk
Spearman correlation coefficient was used to assess the correlation between clinical characteristics and AIS risk, and heatmap was drawn accordingly (Figure 1). It was found that HDL‐C was negatively correlated with presence of AIS in the study cohorts (r=-0.17, P < 0.001). A significant positive correlation was found between AIS and several clinical and laboratory features including BMI, hypertension, diabetes mellitus, cardiovascular disease, SBP, DBP, LDL-C, TG, ApoA, sd-LDL, LDL-3 and LDL-4 (r>0.1, P < 0.001).
Correlations of LDL subtractions and LDL-C
We next performed Spearman correlation coefficient to investigate the correlations between LDL-C and its subclasses (Table 3). As a result, there was a significant positive correlation between LDL-C and LDL-1 (r=0.540, p<0.05), LDL-2 (r=0.547, p<0.05), LDL-3 (r=0.343, p<0.05) and LDL-4 (r=0.227, p<0.05). With regard to the correlation among LDL subtypes, LDL-1 was found positively correlated with LDL-2 (r=0.485, p<0.05), but negatively correlated with the rest LDL subtypes, whereas LDL-2 was positively correlated with LDL-3 (r=0.611, p<0.05) and LDL-4 (r=0.230, p<0.05) besides LDL-1, but negatively correlated with LDL-5 to 7. For the LDL-3 to 7, so called sd-LDL, there is a positive correlation between each other in most cases except for LDL-3 and LDL-7 (r=-0.019, p=0.599).
Correlation analysis between sd-LDL and serum lipids
Spearman correlation coefficient showed a significant positive correlation between sd-LDL levels and serum lipids including TC, LDL‐C, and TG, as shown in Table 4 and Figure 2. Sd-LDL was negatively correlated with HDL-C in control group and all subjects, but not in AIS group.
Clinical and laboratory characteristics of study participants with normal LDL‐C levels
In AIS group, the sd-LDL level was significantly higher in the high LDL‐C group of patients than in the normal LDL‐C group (P < 0.001) (Table 5). Additionally, although there was no significant difference in GLU and LDL-C levels between the AIS group with normal LDL‐C levels and the control group (P > 0.05), the other laboratory characteristics, including blood pressure, Scr, serum lipids, and LDL subclasses, differed significantly between the two groups, as shown in Table 6.
Prediction of AIS risk by laboratory characteristics with an established model
Considering the correlations of laboratory characteristics with AIS, a risk prediction model was established accordingly to identify people most likely to develop AIS (Figure 3). Several typical and recent artificial intelligence (AI) algorithms were assessed in the context of AIS, and the XGBoost stood out as our base model due to its high performance (AUC=0.82±0.04) (Figure 3A). As a frequently used resampling method, k-fold cross-validation was employed to estimate the performance of the XGBoost classifier (Figure 3B). The new prediction model contained 12 variables (ie, Scr, SBP, DBP, TG, TC, LDL-C, HDL-C, LDL-1, LDL-2, LDL-3, LDL-4, LDL-5) with a cut-off value of 0.5.