This retrospective study included adult patients diagnosed with AIS who were hospitalized in three tertiary medical centers in China: (1) the First Affiliated Hospital of Soochow University between March 2018 and January 2021, (2) the First People’s Hospital of Lianyungang between October 2019 and December 2021, and (3) the Second People’s Hospital of Lianyungang between January 2022 and July 2024. All participating hospitals used standardized diagnostic protocols and electronic health record systems to ensure data consistency. AIS diagnoses were confirmed by neurologists using clinical criteria in accordance with the World Health Organization definition and supported by neuroimaging (Computed Tomography or Magnetic Resonance Imaging). After admission, fasting blood samples were collected from all patients within 24 h by trained nursing staff for laboratory analysis, including lipid profiles and metabolic markers. Additional demographic and clinical data were extracted from hospital records by two independent investigators using a structured data collection form.

The inclusion criteria were: (1) Patients aged ≥ 18 years; (2) Patients who met the diagnostic criteria for AIS. On the other hand, the exclusion criteria were: (1) Patients with Intracranial Hemorrhage or mass bleeding; (2) Patients with severe infection or septic shock; (3) Patients with liver or kidney failure; (4) Patients with incomplete laboratory or clinical data; (5) Stroke of undetermined etiology.

This study was approved by the Ethics Committees of the First Affiliated Hospital of Soochow University (Approval No. 2020272, 2,019,057), the Second People’s Hospital of Lianyungang (Approval No. 2020050), and the First People’s Hospital of Lianyungang (Approval No. KY-20210917001-01).

Experienced clinicians collected general patient information, including age, gender, Systolic Blood Pressure (SBP), Diastolic Blood Pressure (DBP), history of hypertension, history of diabetes, history of atrial fibrillation, smoking history, and alcohol consumption history. Within 24 h of admission, professional medical staff collected hematological samples for laboratory testing, including Total Cholesterol (TC), TG, LDL-C, HDL-C, Glucose (Glu), Glycated Hemoglobin (HbA1c), Albumin (Alb), Creatinine (Cr), and Homocysteine (Hcy).

Neurological status was assessed by two experienced neurologists using the National Institutes of Health Stroke Scale (NIHSS) at admission and discharge. Patients were categorized into large-artery atherosclerosis (LAA) or non-LAA subtypes based on the Trial of Org 10,172 in Acute Stroke Treatment (TOAST) criteria (21). Stroke severity was defined as mild (NIHSS < 4) or severe (NIHSS ≥ 4).

Herein, AIP values [log(TG/HDL-c)] were determined and grouped into quartiles (22): Q1:<−0.0904; Q2: −0.0904-0.0814; Q3:0.0814–0.255; and Q4: ≥ 0.255.

All statistical analyses were conducted using R software (version 4.3.2). The Kolmogorov–Smirnov test was applied to assess the normality of continuous variables. Variables with normal distribution were summarized using mean ± standard deviation (SD), and comparisons between groups were made using independent samples t-tests. For non-normally distributed variables, data were presented as medians and interquartile ranges (IQRs), and group comparisons were performed using Mann–Whitney U tests. Categorical variables were expressed as counts (percentages), and intergroup differences were evaluated using the chi-square test or Fisher’s exact test when expected frequencies were <5.

Logistic regression analyses were used to examine the association between AIP and LAA, allowing adjustment for potential confounders. Multiple covariance tests were performed prior to the logistic regression analysis. Restricted cubic spline (RCS) analysis was conducted to explore the potential non-linear association between AIP and LAA risk. ROC curve analysis was performed to evaluate the discriminative power of AIP (23), with Youden’s index used to determine the optimal cutoff point.

Missing data were handled according to the percentage of incompleteness: variables with >10% missing data were excluded from analysis, while variables with <10% missing values were imputed using the Random Forest method via the mice package in R. All tests were two-tailed, and p-values <0.05 were considered statistically significant.

Herein, as shown in Table 1, the baseline characteristics of participants were categorized by AIP quartiles, revealing significant differences across various demographic and clinical parameters. Among the total of 2,382 patients included in the study, 1,483 were male (62.3%) and 899 were female (37.7%). There was a notable trend in age distribution (p < 0.001), with median ages decreasing from Q1 (71 years) to Q4 (64 years). Furthermore, there was a decreasing trend in the proportion of participants aged ≥ 60 years across the quartiles, with Q1 and Q4 having the highest (84.2%) and lowest (63.8%) proportions, respectively. On the other hand, gender distribution was relatively balanced across the quartiles, although each group had more males, but with no significant intergroup difference (p = 0.085). Smoking and drinking habits increased considerably across the quartiles (p < 0.001 for both), with Q4 showing the highest proportions (47.5 and 56.7% for smoking and drinking, respectively). The prevalence of diabetes and atrial fibrillation also increased significantly (p < 0.001), with Q4 showing the highest prevalence (56.7 and 45.3%, respectively). On the other hand, the prevalence of hypertension remained consistent across groups (p = 0.119). Regarding metabolic markers, the quartiles showed significant differences in HbA1c, Alb, Glu, TC, TG, HDL-C, and LDL-C levels (all p < 0.001), with an upward trend in glucose and lipid indices from Q1 to Q4. Conversely, the quartiles showed no significant differences in SBP and Cr levels. Moreover, the median NIHSS scores at admission decreased slightly across quartiles (p = 0.008), with Q4 showing a larger proportion of mild cases (NIHSS < 4). As shown in Figure 1, according to the TOAST classification system, the groups exhibited an increase in LAA prevalence, with Q4 accounting for most cases (50.7%) (p = 0.040).

Figure 2 shows a Restricted Cubic Spline (RCS) curve depicting the relationship between AIP and LAA. Notably, AIP correlated significantly positively with LAA (p = 0.011), with an inflection point observed at an AIP value of 0.10, after which the odds ratio (OR) increased significantly. Furthermore, the relationship between AIP and LAA did not deviate significantly from a linear trend (p for non-linearity = 0.233).

We assessed collinearity before analysis (Supplementary Table 1). We also assessed the association between various characteristics and LAA, as shown in Table 2, where NIHSS score at admission, smoking, HbAlc, Glu, TC, LDL-C, and AIP were independent risk factors for LAA. The risk of LAA occurrence increased by approximately 42% for each 1-unit increase in AIP when using AIP as a continuous variable [OR = 1.42, 95% confidence interval (CI): 1.07–1.89]. Categorical analysis of AIP showed that only Q4 was significantly associated with risk of LAA (OR = 1.39, 95% CI: 1.11–1.75).

This logistic regression analysis assessed the association between various characteristics and the likelihood of LAA, adjusting for different sets of covariates across five models, as shown in Table 3. When the AIP was treated as a continuous variable, each unit increase was associated with a higher risk of LAA, though the OR increased as more covariates were included: from OR = 1.42 in Model 1 to OR = 1.61 in Model 5. On the other hand, when analyzed categorically, Q4 of AIP showed a consistent and significant association with higher LAA odds across all models, with ORs ranging from 1.39 (95% CI:1.11–1.75, p = 0.004) in Model 1 to 1.32 (95% CI: 1.06–1.80, p = 0.005) in Model 5. Trend tests indicated a significant trend association between AIP groups and LAA across all models. These findings collectively suggest that higher AIP levels correlate with an increased risk of LAA.

Herein, an ROC plot was generated depicting the predictive ability of the five models in Figure 3. According to the plot, the area under curve (AUC) values increased gradually as covariates were progressively added from models 1 to 5, indicating an improvement in the models’ predictive value. The AUC value for Model 1 was 0.532, the optimal threshold value for the AIP index was 0.008, and the Yoden index, sensitivity and specificity values were 0.061, 63.9 and 42.2%, respectively. In Model 5, the AUC value was 0.646, the optimal threshold was 0.514, and the Youden’s index, sensitivity, and specificity values were 0.219, 45.5, and 76.4%, respectively.

According to the stratified analysis results, among individuals with consistently high AIP levels, the risk of developing LAA was higher in those aged ≥ 60 years compared to those aged < 60 years, and patients with diabetes had a higher risk of LAA than patients without diabetes, as well as in those with LDL-C < 3.4 mmol/L compared to those with LDL-C ≥ 3.4 mmol/L (both p < 0.05; Table 4). Furthermore, in the ≥ 60 years group, compared to Q1, the risk of LAA was higher in Q3 (OR = 1.31, 95% CI: 1.01–1.70) and Q4 (OR = 1.60, 95% CI: 1.22–2.09). In the LDL-C < 3.4 mmol/L group, only Q4 was significantly associated with LAA compared with Q1 (OR = 1.35, 95% CI: 1.05–1.74). On the other hand, in the LDL-C ≥ 3.4 mmol/L group, the risk of developing LAA in Q2, Q3, and Q4 did not change significantly relative to Q1 (all p > 0.05). In addition, in the group with an NIHSS score of ≥4 on admission, compared to Q1, compared with Q1, the risks of LAA in Q2, Q3, and Q4 increased by 34% (OR = 1.34, 95% CI: 1.01–1.85), 41% (OR = 1.41, 95% CI: 1.02–1.96), and 48% (OR = 1.48, 95% CI: 1.06–2.06). In contrast, only Q4 was associated with LAA in the NIHSS score at admission <4 group (OR = 1.41, 95% CI: 1.02–1.95).