Soluble ST2 and risk of cognitive impairment after acute ischemic stroke: a prospective observational study

The CATIS trial, a single-blind, blinded end-points randomized clinical trial, was designed to evaluate whether immediate blood pressure (BP) reduction would reduce death and major disability at 14 days or hospital discharge. The study design and main results had been described previously [15]. The enrollment criteria were as follows: (1) first-ever ischemic stroke; (2) age ≥ 22 years; (3) time from onset to admission within 48 h; (4) systolic BP between 140 and 220 mmHg. The exclusion criteria were as follows: (1) BP ≥220/120 mmHg; (2) treated with intravenous thrombolytic therapy; (3) severe heart failure, acute myocardial infarction or unstable angina, atrial fibrillation, aortic dissection, cerebrovascular stenosis, resistant hypertension, or in a deep coma. Finally, 4071 patients with first-ever ischemic stroke within 48 h of onset and elevated systolic BP were recruited. The present prospective observational study was a pre-planned ancillary study of CATIS trial, which was designed to test cognitive function at 3 months. In the ancillary study, acute ischemic stroke patients from 7 randomly selected participating hospitals were consecutively recruited for neurological evaluations. Each of the 7 participating hospitals planned to recruit 80–100 patients. During the period of August 2009 to November 2012, 660 participants were enrolled. After further excluded participants without available cognitive evaluations (n = 22) or sST2 data (n = 19), 619 participants were finally included in the present analysis (Supplementary Figure 1).

This study protocol was approved by the ethical committee at Soochow University in China and Tulane University in the United States, as well as ethical committees at the 7 participating hospitals, in compliance with the Declaration of Helsinki. All participants provided written informed consent.

Fasting blood samples were drawn within 24 h of patients’ hospital admission, and were frozen − 80 °C until testing. The concentrations of plasma sST2 and serum high-sensitive C-reactive protein (hsCRP) were measured by commercially available immunoassays (R&D Systems, Minneapolis, MN). The intra-assay and inter-assay coefficients of variation for sST2 were below 5.5 and 2.4%. Similarly, the intra-assay and inter-assay coefficients of variation for hsCRP were below 8.4 and 1.6%. Laboratory technicians who performed measurements were blinded to clinical features and outcomes of patients.

Baseline data regarding demographic characteristics, clinical features, medical history, and prior use of medications were collected using a standard questionnaire at the time of enrollment. The National Institutes of Health Stroke Scale (NIHSS) and the modified Rankin Scale (mRS) were applied to assess stroke severity by trained neurologists. Trial of Org 10,172 in Acute Stroke Treatment criteria was used to classify the ischemic stroke subtypes as large-artery atherosclerosis(thrombotic), cardiac embolism (embolic) and small-vessel occlusion (lacunar), according to the symptoms and imaging data of the patients by experienced neurologists [16]. BP measurements were performed by trained nurses according to a standard protocol adapted from procedures recommended by the American Heart Association.

Cognitive function at 3 months after stroke was evaluated by trained neurologists using the Montreal Cognitive Assessment (MoCA) and Mini-Mental State Examination (MMSE). Prior studies found that people whose education ≤12 years tended to perform worse on the MoCA test [17]. Thus, 1 point was additionally added for participants with education ≤12 years on their crude MoCA score (if < 30) to correct for education effects. Lower scores indicate worse cognitive function, and PSCI was defined as a MoCA score of less than 23 [18, 19] or MMSE score of less than 27 [20].

Baseline characteristics were summarized according to the quartiles of plasma sST2 levels. Means with standard deviation(SD) or median with interquartile range were used for continuous variables, and the generalized linear regression models were further used to test for trend across the sST2 quartiles. Frequency with percentage was used for categorical variables, and the Cochran-Armitage trend χ² tests were used to test for trend.

First, we performed generalized linear regression models to test the association between plasma sST2 levels and continuous MoCA and MMSE score. Second, both MoCA and MMSE score were categorized into two groups (without versus with PSCI). Logistic regression models were used to calculate odds ratios (ORs) and 95% confidence intervals (CIs) for the association of plasma sST2 levels with the incident of PSCI. According to baseline characteristics, previous studies and professional knowledge, three models were used with an increasing level of adjustment. Model 1 was adjusted for age, sex and education level (illiteracy, primary, high school and college or higher). Model 2 further adjusted for current smoking, alcohol drinking, time from onset to randomization, systolic BP, baseline NIHSS and mRS score, medical history (hypertension, diabetes mellitus, hyperlipidemia, and coronary heart disease), use of antihypertensive medications, randomized treatment, ischemic stroke subtype (thrombotic, embolic and lacunar), anticoagulant treatment and hypoglycemic treatment. Available evidence suggested that elevated sST2 levels were involved in pro-inflammatory response [21, 22], and inflammation may be an intermediary in the relation of sST2 levels to cognitive impairment. In consideration of adjustment for a risk factor that is in the causal pathway from the exposure to outcome will reduce or even remove the effect of interest, we further conducted a separate analysis with hsCRP included in a multivariable-adjusted model based on the model 2. We tested for linear trends across the median of sST2 quartiles.

Moreover, Hosmer Lemeshow χ² statistic was used to assess the calibration of the model with or without sST2. C statistic, net reclassification index (NRI) and integrated discrimination improvement (IDI) were applied to evaluate the predictive ability of plasma sST2 beyond conventional risk factors model, which including age, sex, education level, systolic BP, baseline NIHSS, baseline mRS, hypertension, diabetes mellitus, ischemic stroke subtype, hsCRP, anticoagulant treatment and hypoglycemic treatment. In addition, random forest regression model, one of the most robust ensemble machine learning methods for classification and regression, was used to assess the importance of potential predictors. “Mean Decrease Accuracy” is a type of importance measure and the value is larger suggests that the predictor is more important. Multiple imputation for missing data was performed using the Markov chain Monten Carlo method (Supplementary Table 1). All two-sided P values < 0.05 were considered to be statistically significant. Statistical analysis was conducted using SAS statistical software, version 9.4 (SAS Institute Inc., Cary, NC).

A total of 619 patients (434 men and 185 women) with a mean (SD) age of 60.0 (10.5) years were enrolled in the present study. The median of plasma sST2 level was 163.51 pg/mL (interquartile range, 117.60–238.77 pg/mL). The baseline characteristics across the quartiles of sST2 levels were summarized in Table 1. Compared to the patients in the lowest quartile of sST2 levels, those in the higher quartile were tended to be male and older, to have higher baseline NIHSS and mRS score, to have higher level of hsCRP, and to have shorter time from onset to randomization and lower proportion of antihypertensive medications use.

At 3-month follow-up, the median (interquartile range) score of MoCA and MMSE were 22.0 (18.0–26.0) and 26.0 (22.0–29.0), respectively. There were significantly decreasing trends in MoCA and MMSE scores as sST2 increased from the lowest quartile to the highest quartile (Fig. 1). 325 (52.5%) or 323 (52.2%) participants developed PSCI according to MoCA or MMSE, respectively. The incidences of PSCI increased from the lowest quartile (sST2 < 117.60 pg/mL) to the highest quartile (sST2 ≥ 238.77 pg/mL) (Table 2).

After adjustment for age, sex, education level, current smoking, alcohol drinking, time from onset to randomization, systolic BP, baseline NIHSS and mRS score, medical history (hypertension, diabetes mellitus, hyperlipidemia, and coronary heart disease), use of antihypertensive medications, randomized treatment, ischemic stroke subtype, anticoagulant treatment and hypoglycemic treatment (model 2), the odds of developing PSCI (defined by MoCA score) increased significantly with elevated baseline sST2 levels (p for trend = 0.002). This significant association remained when further adjustment for hsCRP levels (model 3). Compared to patients in the lowest quartile of sST2, those in the highest quartile had 2.38-fold (95% CI 1.42–4.00) risk of PSCI. Similar independent association was also observed when PSCI was defined according to MMSE score, the corresponding adjusted OR (95% CI) of PSCI was 1.82 (1.09–3.03) for patients in the highest quartile of sST2 in comparison with those in the lowest quartile of sST2 (Table 2).

The Hosmer Lemeshow test showed that model calibration was adequate after adding plasma sST2 to the basic model containing conventional risk factors (MoCA p = 0.20; MMSE p = 0.74). Incorporation sST2 into the basic model also significantly improved the risk reclassification performance (continuous NRI 18.5%; IDI 1.6%; both p < 0.01) when PSCI was defined by MoCA score (Table 3). Similar incremental predictive value of sST2 was observed when PSCI was evaluated by MMSE score.

Additionally, to test the contribution of potential predictors in relation to PSCI, mean decrease in accuracy was calculated. Moreover, we found that sST2 was superior to NT-proBNP in predicting PSCI, according to the receiver operating characteristic curves (Supplementary Figure 2). The random forest regression models suggested that baseline plasma sST2 was one of the promising predictors for PSCI (Fig. 2).