Between March 2019 and September 2020, patients presenting with acute cerebral ischemia were referred to the Stroke Centre of the 108 Military Central Hospital for examination and clinical management. Acute ischemic stroke diagnosis relies on clinical manifestations and imaging findings from magnetic resonance angiography (MRI) or computed tomography (CT) scans of the brain. Meanwhile, the diagnosis of narrowed or blocked blood vessels relies on magnetic resonance angiography (MRA) or digital subtraction angiography (DSA). During study period, we enrolled 406 patients with acute IS who met the following criteria: aged ≥18 years old and presenting with acute IS symptoms within 2 days of onset. Patients were excluded if they exhibited the following characteristics: (1) IS associated with valvular heart disease, cardiac arrhythmias, or complete atrioventricular block as detected on electrocardiography; (2) IS patients with implanted pacemakers; (3) IS resulting from conditions associated with increased blood clotting, such as sickle cell anemia or systemic lupus erythematosus; (4) patients with comorbidities including heart failure, renal failure, liver cirrhosis, or thyroid cancer; (5) patients taking medications known to affect laboratory test results lipid profiling, such as carbamazepine, estrogen, lovastatin, simvastatin, phenobarbital, or pravastatin. In addition, we had 14 IS patients with stenosis in both intracranial and extracranial large arteries. Due to the small number of patients compared to the other groups, we have excluded these cases from the analyses.

Upon admission, comprehensive baseline data encompassing age, gender, medical history, and physical examination findings were collected. Each patient underwent thorough clinical assessment, neurological examination, pertinent laboratory investigations, cardiac assessment, and evaluation of imaging methodologies. Lipid profiling was conducted in accordance with established guidelines, with criteria including low levels of HDL-C (<1.03 mmol/L in men, 1.30 mmol/L in women), elevated total cholesterol (≥5.2 mmol/L), increased triglyceride levels (TG ≥ 1.7 mmol/L), and nonoptimal LDL-C values (≥2.6 mmol/L) (24).

The enrolled patients were stratified into two distinct cohorts using the TOAST classification (6, 25). In brief, group A comprised patients diagnosed with IS resulting from large artery atherosclerosis. Imaging assessments revealed cortical or subcortical infarcts, or hemispheric infarcts exceeding 1.5 cm in diameter, accompanied by evidence of greater than 50% stenosis in the affected arteries. This group was further divided into two subgroups: the first subgroup consisted of patients with intracranial atherosclerosis (ICAS, n = 193), while the second subgroup consisted of patients with extracranial atherosclerosis (ECAS, n = 111). Group B consisted of IS patients attributed to small artery occlusion (SAO, n = 102).

The degree of stenosis was assessed using the methodology outlined in the study conducted by Samuels OB et al. (26). It was calculated using the formula: degree of stenosis (%) = (1 – diameter at the narrowest point of the narrow segment/diameter of the proximal normal vessel) × 100%. Patients in this study were categorized into three groups based on the degree of stenosis: mild stenosis group (stenosis ≤49%), moderate–severe stenosis group (stenosis between 60 and 90%), and an extreme stenosis group (stenosis >90% or occlusion).

Blood samples from each patient were collected at hospital admission. The samples were centrifuged according to the standard protocol to separate plasma from blood cells. Subsequently, conventional lipid profile, including triglycerides, total cholesterol, HDL, and LDL levels, was routinely measured on the AU5800 Beckman Coulter Analyzers (Beckman Coulter), employing a quantitative photometric method based on the principles of light absorbance measurement by chemical substances in solution.

Following the routine measurement of conventional lipids, apoA-I and apoB were quantified based on the immunoturbidimetric method. This process was carried out on the AU5800 analyzer according to the manufacturer’s instructions provided by Beckman Coulter. In this method, the sample or standard was mixed with a reagent containing anti-apoA-I or apoB antibodies. Upon binding with these antibodies, apoA-I or apoB in the sample formed a precipitated complex. The turbidity of this complex was then measured at a wavelength of 340 nm. The level of turbidity observed was directly proportional to the concentration of apoA-I or apoB present in the clinical sample.

Statistical analyses were conducted using R version 4.3.2.¹ Chi-square, Kruskal-Wallis, and Mann–Whitney-Wilcoxon tests were employed to compare group differences in qualitative or quantitative variables as appropriate. Spearman’s rank correlation test was applied to evaluate correlations between apoA-I, apoB, and other laboratory parameters. Random forest models were utilized to assess the potential predictor, including apoA-I, apoB, apoB/apoA-I ratio, and other independent factors for IS subtypes and arterial stenosis severity. Receiver operating characteristic (ROC) curves generated from random forest (RF) models assessed the predictive value of variables of interest in discriminating clinical status by calculating the area under the ROC curve (AUC). We analyzed the variable importance and distribution of minimal depth through tree modeling to determine the most influential variable(s) in predicting the outcome. All statistical significance was defined as a two-sided p-value of <0.05.

The baseline profile of 406 consecutive IS patients who met the inclusion criteria enrolled in this study were described in Table 1 and Figures 1–3. In this cohort, the median age at diagnosis in IS subgroups was 65–66 years [ICAS: 65 (18–90); ECAS: 66 (34–93); SAO: 65 (28–86); p > 0.05]. Most of the patients were male (>75% for each group). SAO patient group had a higher percentile (21.6%) of BMI ≥ 25 compared to IS patients with large artery atherosclerosis (ICAS: 8.8% and ECAS:7.2%). There was no significant difference in the distribution of comorbidities, including hypertension and diabetes mellitus, among the groups. There was no difference in the levels of blood lipid types such as total cholesterol, LDL, triglyceride, and HDL between the groups as seen in Table 1 and Figure 1.

The distribution of apoA-I and apoB levels upon hospital admission among subgroups of IS patients was illustrated in Table 1 and Figure 2. Serum apoA-I levels were significantly lower in patients with ICAS compared to those with ECAS and SAO [mean ± SD: ICAS (1.32 ± 0.27 g/L); ECAS (1.43 ± 0.32); SAO (1.51 ± 0.28 g/L) (p < 0.0001), Figures 2A,B]. Conversely, serum apoB levels were higher in ICAS patients [(mean ± SD: 1.24 ± 0.32 g/L) followed by ECAS (mean ± SD: 1.07 ± 0.26 g/L) and SAO patients (mean ± SD: 1.03 ± 0.27 g/L) (p < 0.0001), Figures 2C,D]. These differences persisted with a consistent trend when comparing apoA-I and apoB levels between the ICAS group and the non-ICAS group (comprising ECAS and SAO) (p < 0.0001).

As presented in Table 1 and Figure 3, the apoB/apoA-I ratio exhibited a remarkably higher value in patients with intracranial arterial stenosis (ICAS) compared to those without intracranial arterial stenosis (NICAS) (p = 1.4e-15). Furthermore, upon comparison across subgroups, it was evident that ICAS patients had a significantly higher apoB/apoA-I ratio compared to patients with extracranial arterial stenosis (ECAS) and small artery occlusion (SAO) [mean ± SD: ICAS (0.98 ± 0.33); ECAS (0.78 ± 0.26); SAO (0.7 ± 0.23) (p < 0.0001)]. Conversely, no significant difference was observed between ECAS and SAO patients (p > 0.05).

Analysis of the correlation between apoA-I, apoB levels, and the apoB/apoA-I ratio with routine lipid profile indices, including total cholesterol, LDL, HDL, and triglycerides, we found that apoA-I was significantly correlated with HDL (rho = 0.41, p < 0.0001, Figure 4A), while there was either no correlation or weak correlation with the other lipid indices (data not presented in Figure 4). As for apoB, there was a correlation with triglyceride (rho = 0.26, p < 0.0001, Figure 4B), LDL (rho = 0.41, p < 0.0001, Figure 4C), total cholesterol (rho = 0.40, p < 0.0001, Figure 4D), and but no correlation with HDL (not presented). Regarding the apoB/apoA-I ratio, the analyses showed a positive correlation with LDL (rho = 0.33, p < 0.0001, Figure 4E), total cholesterol (rho = 0.26, p < 0.0001, Figure 4F), triglycerides (rho = 0.27, p < 0.0001, Figure 4G), and a negative correlation with HDL (rho = −0.28, p < 0.0001, Figure 4H).

In addition, we analyzed the apo-AI, apoB, and apoB/apoA-I ratio profiles in a subset of 25 cases with markedly high triglyceride levels (>6 mmol/L). The results showed that the majority of patients (22 out of 25, 88%) had low levels of apo-AI, defined as less than 1.6 g/L. (27) Moreover, elevated levels of apoB and an increased apoB/apoA-I ratio (normal range: apoB <1.3 g/L; apoB/apoA-I ratio < 0.77), were observed in 48% (12 out of 25) and 68% (17 out of 25) of the patients, respectively.

We further analyzed the correlation between serum levels of triglycerides, LDL, cholesterol and apo-A1, apoB, apoB/apoA-I ratio, as illustrated in Figure 5. Our findings revealed significant correlations between apoA-I levels and triglyceride, LDL, and cholesterol, with rho values of −0.45, 0.46, and −0.31, respectively (although not significant correlation for all IS patients as previously reported). Furthermore, a stronger correlation was observed between apoB levels, the apoB/apoA-I ratio and triglyceride, LDL levels in this sub-analysis.

The mild stenosis group exhibited the highest levels of apoA-I [median and range: 1.38 (0.74–2.1)], followed by the severe stenosis [median and range: 1.35 (0.89–1.93)] group and the occlusion group [median and range: 1.28 (0.83–2.52)], although these differences did not reach statistical significance. Conversely, the lowest concentration of apoB and apoB/apoA-I ratio was observed in the mild stenosis group [ApoB: 0.9 (0.69–1.75); ApoB/ApoA-I: 0.75 (0.33–1.28)], followed by the severe stenosis group [ApoB: 1.14 (0.69–1.84); ApoB/ApoA-I: 0.84 (0.49–1.92)]. The highest levels were recorded in the occlusion group [ApoB: 1.23 (0.49–2.16); ApoB/ApoA-I: 0.97 (0.39–2.35)], with statistical significance as depicted in Figure 6.

Upon detailed analysis in the subset of 25 cases with triglyceride levels exceeding 6 mmol/L, data regarding arterial stenosis were accessible for 17 patients. This analysis revealed that out of the 17 patients, 14 (82%) exhibited extreme stenosis, while two patients had mild stenosis, and one patient had severe stenosis.

Within the ICAS group, the serum levels of apoA-I, apoB and the apoB/apoA-I ratio in patients with single-site stenosis/occlusion were not different from patients with multi-site stenosis/occlusion (Figure 7). We could not generate the analyses for ECAS and SAO group because of missing data (ECAS) or unavailable data (SAO).

We utilized different random forest (RF) models to assess factors associated with the severity of arterial stenosis in IS patients. The model-1 included variables apoB/apoA-I ratio, apoA-I, apoB, LDL, cholesterol, triglyceride, age, and gender. The model-2 included variables as apoB/apoA-I ratio, age, and gender. The model-3 included only apoB/apoA-I ratio. The results were illustrated in Figure 8 and we have shown that that all RF models performed well in predicting the degree or severity of arterial stenosis. For instance, in models predicting extreme stenosis from mild–severe stenosis, the AUC values were 1.0 for model-1 (Figure 8A); 0.94 for model-2 (Figure 8B); 0.85 for model-3 (Figure 8C). In addition, to determine the most influential parameter in the model-1, we constructed plots of variable importance and distribution of minimal depth through tree modeling. The results revealed that the apoB/apoA-I ratio played the most significant role in predicting extreme stenosis from mild–severe stenosis compared to other variables as indicated by their low mean minimal depth and high values in both mean decrease accuracy and mean decrease gini values (Figures 8D,E). Similar pattern was observed in the other RF models (extreme vs. severe; extreme vs. mild; severe vs. mild) shown in Figures 8F–H.

To assess the significant predictors of IS subtypes, we constructed RF models, incorporating multiple variables (models 1 and 2) or single variable (model 3, as previously described). The results of our analysis are depicted in Figure 9, illustrating the prediction of ICAS patients from non-ICAS patients (ECAS+SAO), with AUC values of 0.88 (model 1), 0.84 (model 2), and 0.82 (model 3). Plots of variable importance and distribution of minimal depth revealed that the apoB/apoA-I ratio was the most influential variable compared to others such as apoA-I, apoB, LDL.