Low triglyceride to high-density lipoprotein cholesterol ratio predicts hemorrhagic transformation in large atherosclerotic infarction of acute ischemic stroke

General characteristics

The general characteristics of the training and test cohorts are detailed in Tables 1, 2 and 3. A total of 950 (66.76%) men and 473 (33.24%) women were enrolled in the training cohort. The median National Institute of Health stroke scale (NIHSS) score was 4. The mean±SD of triglyceride (TG), TC, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and TG/HDL-C values were as follows: 1.56±1.16 mmol/L for TG, 4.41±1.09 mmol/L for TC, 1.09±0.41 mmol/L for HDL-C, 2.66±0.83 mmol/L for LDL-C, and 1.60±1.44 for TG/HDL-C (Table 1). Baseline characteristics of the training cohort according to stroke etiology are detailed in Table 2. Data from an additional 558 patients attributable to LAA were retrospectively recruited into the test cohort. The median NIHSS score was 4. The baseline characteristics and mean concentrations of serum lipids are shown in Table 3.

Correlation of lipid level with HT

For the training cohort, HT after AIS was observed in 155 (10.89%) patients. Comparisons of baseline characteristics in patients with or without HT are shown in Table 1. Compared with patients without HT, patients with HT had higher proportions of no smoking, diabetes mellitus, coronary artery disease, atrial fibrillation and prior use of antiplatelets and anticoagulants; elder age; higher levels of SBP and fasting glucose; lower levels of platelets, TG, TC, LDL-C and TG/HDL-C; and higher scores of NIHSS (all P<0.05). The distributions of Trial of Org 10172 in Acute Stroke Treatment (TOAST) classifications in patients in the absence and presence of HT were significantly different (P<0.001) (Table 1). Furthermore, HT was more frequent in cardioembolism (38.06%) than in other TOAST classifications (Table 1), as previously depicted [16,17]. In patients attributable to LAA, HT was more likely to be present if the patients had diabetes mellitus (P=0.003) and prior use of anticoagulants (P=0.049), higher scores on the NIHSS (P=0.006), or lower levels of TG (P=0.016) and TG/HDL-C (P=0.006) (Table 2). In patients with cardioembolism, HT correlated significantly with NIHSS score on admission (P=0.001) and with fasting glucose (P=0.001) as well as systolic blood pressure (SBP) (P=0.045), but not with lipid level (Table 2). However, baseline characteristics showed no correlation with the incidence of HT in patients with small-vessel occlusion (Table 2). For the test cohort, HT occurred in 50 (8.96%) patients. Similar results were observed in patients attributable to LAA (Table 3).

A novel predictive model

The univariate logistic regression analyses for baseline characteristics in the training cohort are shown in Tables 4 and Table S1. The results showed that age, smoking, SBP, NIHSS scores, diabetes mellitus, coronary artery disease, atrial fibrillation, stroke etiology, prior use of antiplatelets, fasting glucose, platelets, TG, TC, LDL-C and TG/HDL-C were associated with HT (all P<0.05) (Table 4). The multivariable logistic regression analyses showed that patients attributable to cardioembolism had higher risk of HT than those attributable to LAA (OR=3.03, 95%CI=1.91-4.80, P<0.001); however, TG, TC, LDL-C and TG/HDL-C were not independently associated with HT in patients attributable to cardioembolism (P>0.05). To further investigate the association of stroke etiology with HT, we performed logistic regression analyses in patients attributable to LAA, cardioembolism and small-vessel occlusion. The findings indicated that lower levels of TG (OR= 0.61, 95%CI=0.27-0.98, P=0.042) and TG/HDL-C (OR= 0.53, 95%CI=0.20-0.93, P=0.032) were significantly associated with a higher risk of HT in patients with LAA (Table 4), but not in patients with cardioembolism or small-vessel occlusion (Table S1). These results were further confirmed in the test cohort of patients with LAA (Table 4).

To investigate the predictive power of lipid level for HT in patients with LAA, we established a nomogram according to the results from univariable analyses to predict the incidence of HT. The univariable analyses showed that NIHSS scores on admission, diabetes mellitus, TG and TG/HDL-C were associated with HT in the training and test cohort patients with LAA. Furthermore, TG/HDL-C had the greatest AUC among lipid classifications, and NIHSS scores on admission, diabetes mellitus, and TG/HDL-C were included in the nomogram for both the training (Figure 1A) and test (Figure 1B) cohorts by stepwise logistic regression analyses. The novel models for both cohorts indicated that higher NIHSS scores, diabetes mellitus and lower TG/HDL-C were indicators of HT in patients with LAA. These findings were similar to those obtained previously in the multivariate logistic models.

Figure 1. Nomograms of patients attributable to large artery atherosclerosis to predict hemorrhagic transformation after acute ischemic stroke. Locate the TG/HDL-C on the respective axis; draw a straight line up to the Points axis to determine how many points toward hemorrhagic transformation the patient receives for the TG/HDL-C; repeat this process for diabetes mellitus and NIHSS; add the points and locate this number on the Total points axis; and draw a straight line down to find the patient’s estimated risk of hemorrhagic transformation. The c-indexes for the training and test cohorts of patients attributable to large artery atherosclerosis are 0.734 (A) and 0.698 (B), respectively. TG, triglyceride; HDL-C, high-density lipoprotein cholesterol; NIHSS, National Institute of Health Stroke Scale.

Predictive values of TG, TG/HDL-C for HT

To study the predictive power of TG and TG/HDL-C, we performed ROC curves and AUC analyses regarding HT in patients attributable to LAA from the training cohort. We also analyzed the combined predictive value of TG and HDL-C for HT. The results showed that the predictive value of TG/HDL-C (AUC=0.77, 95%CI=0.72-0.81) was significantly superior to TG (AUC=0.68, 95%CI=0.63-0.72) and their combined effect (AUC=0.67, 95%CI=0.62-0.71) (Figure 2A). In addition, similar findings were observed when their predictive power was further analyzed by ROC curves and AUC analyses in patients attributable to LAA of the test cohort (Figure 2B).

Figure 2. Predictive power of TG, HDL-C, TG+HDL-C, and TG/HDL-C for hemorrhagic transformation in patients attributable to large artery atherosclerosis of the training (A) and test (B) cohorts.

Study population and study protocol

The present study was approved by the Nanjing Medical University Ethics Committee and complied with the Declaration of Helsinki. All of the patient data were acquired from Nanjing First Hospital, Nanjing Medical University between January 2012 and June 2018 and from the Affiliated ZhongDa Hospital of Southeast University between January 2011 and January 2017. The patients were divided into two cohorts, including training and test cohorts. The clinical characteristics of the training and test cohorts are summarized in Tables 1 and 3. A total of 1423 AIS patients who were retrospectively collected from a prospective consecutive hospital-based stroke registry in Nanjing First Hospital were enrolled in the training cohort. An additional 558 AIS patients attributable to LAA who were admitted to the Affiliated ZhongDa Hospital of Southeast University were retrospectively recruited for the test cohort. Individual patients were enrolled in this study if they met the following conditions: (1) first ictus of stroke; (2) admission within 7 days of stroke onset; (3) blood sample extraction within 24 hours on admission (all enrolled patients had fasting lipid levels drawn the morning after hospital admission); (4) ischemic stroke; (5) had not received thrombolytic therapy with tPA and (6) standard in-house procedures were followed [15]. Several patients were excluded as described in our previous study [15].

A multiset and hospital-based study was conducted to explore the precise effect of serum lipid levels on HT after ischemic stroke. Briefly, the association between serum lipid levels and HT was initially investigated in the training cohort. We found a significant association of serum lipid levels and HT in patients with AIS attributable to LAA by using a logistic regression model. Furthermore, we also constructed a prognostic model to predict the incidence of HT in AIS patients attributable to LAA. For the test cohort, we further confirmed the influence of serum lipid levels on HT in a total of 558 patients with AIS attributable to LAA, and we validated the prognostic model in another independent cohort.

Clinical and laboratory assessments

All patients underwent initial brain computed tomography (CT) or magnetic resonance imaging (MRI) before the initiation of antithrombotic therapy within 24 hours after admission. Subsequently, these patients underwent CT or MRI within 7 days after hospital admission or with any worsening. All CT or MRI scans were performed by a neuroradiologist blinded to the baseline characteristics and progress of AIS patients. The protocol of the MRI scan consisted of axial diffusion-weighted imaging (DWI), fluid-attenuated inversion recovery (FLAIR), and axial gradient echo (GRE) sequences. HT was defined as a high-density shadow within the area of low attenuation on the follow-up CT scan or a low-signal area within the acute ischemic lesion on the follow-up GRE sequences of the MRI scan. HT was confirmed according to the European Cooperative Acute Stroke Study I (ECASS I) [31]. As symptomatic HT and asymptomatic HT are part of the same continuum and their discriminatory criteria are somewhat inaccurate, all HTs were considered independently of their clinical symptoms. Stroke etiology and stroke severity were determined as described in our prior study [32].

The baseline characteristics extracted on admission included (1) demographics (sex, age, weight and height), (2) clinical characteristics (smoking, admission blood pressure and NIHSS score), (3) medical history (hypertension, diabetes mellitus and coronary artery disease, atrial fibrillation and previous TIA), (4) stroke etiology, (5) prior use of antiplatelets (aspirin, clopidogrel and cilostazol) and anticoagulants (warfarin, rivaroxiban and dabigatran), and (6) laboratory characteristics.

Statistical analysis

All statistical analyses were performed using SPSS software (Version 20, Chicago, USA) and R 3.0.3 software (Institute for Statistics and Mathematics, Vienna, Austria). The TG/HDL-C was obtained by dividing TG by HDL-C. Categorical variables were expressed as frequencies and percentages and were compared by the Chi-squared test. Continuous variables were expressed as the means ± standard deviations (S.D.) or medians (interquartile ranges, IQR), which were compared by the Student’s t test, one-way ANOVA or the Mann–Whitney U-test if necessary. The discriminatory power of serum lipid levels was assessed by receiver operating characteristic (ROC) curves and corresponding area under the ROC curve (AUC). Comparisons of the AUCs of serum lipid levels were performed by nonparametric statistics [33]. Univariable logistic regression analyses were applied with HT as the dependent variable to obtain odds ratios (ORs) and corresponding 95% confidence intervals (CIs) of HT by serum lipid levels. Significant variables in the univariate analyses were selected into the multivariable logistic regression model to acquire independent predictors. Furthermore, a nomogram based on the independent predictors was constructed by R software with the package rms. The predictive capacity of the nomogram was determined by Harrell’s c-index. P<0.05 was regarded as statistically significant.