PDF
Citation
Print
INTRODUCTION
Acute ischemic stroke (AIS) is a major cause of morbidity and mortality globally [1]. Advances in acute stroke care, particularly the use of intravenous thrombolysis (IVT) with recombinant tissue plasminogen activator (rtPA), have significantly improved clinical outcomes for eligible patients [2,3]. Nevertheless, its use carries inherent risks, one of the most serious being hemorrhagic transformation (HT), a complication that can negate the benefits of reperfusion and result in worse neurological outcomes or even death.
HT occurs in approximately 2%–7% of AIS patients receiving rtPA [4]. It can present in various forms, ranging from minor petechial hemorrhages, with little clinical impact on large parenchymal hematomas that cause significant neurological deterioration. It is associated with prolonged hospital stay, higher healthcare costs, and elevated mortality rates [5]. Predicting which patients are at an increased risk of HT remains an essential challenge in stroke care, as it can influence therapeutic decisions, guide monitoring strategies, and informed discussions with patients and families.
Historically, risk assessments for HT have focused on static clinical variables available at baseline. These include demographic factors, stroke severity indices, comorbid conditions, and imaging features [5,6]. Although these parameters are informative, they do not account for the dynamic physiological processes that occur in the hours after stroke onset and thrombolytic administration.
Heart rate (HR) is a simple yet powerful biomarker that reflects autonomic nervous system activity and cardiovascular stress. Elevated baseline HR has been linked to an increased risk of cardiovascular events and mortality in various clinical settings [7]. Several studies have demonstrated an association between baseline HR, mean HR, and HR variability, and outcomes such as functional recovery and mortality [8-10]. However, most of these investigations rely on single-point HR measurements, such as HR at admission or the mean HR over a specific period, which may not capture the dynamic nature of physiological changes during the acute phase of stroke.
To address this limitation, heart rate trajectories (HRTs) have been introduced. HRT refers to the pattern of HR change over time and is typically modeled using serial measurements. By capturing temporal trends and variations in the HR, trajectory analysis offers a more nuanced view of autonomic regulation and physiological stress. Recent evidence suggests that persistently high HRT patterns may be associated with worse outcomes in patients with AIS who do not receive IVT [11] or in those who have undergone mechanical thrombectomy [12]. This study aimed to evaluate the association between 24-hour HRT patterns and the risk of HT in patients treated with IVT for AIS.
METHODS
Study population and setting
This retrospective cohort study was conducted at a tertiary care hospital between January 2015 and February 2025. The inclusion criteria were adults aged ≥18 years, with confirmed AIS, who received IV rtPA, and had complete hourly HR data over 24 hours and follow-up imaging. Patients who underwent mechanical thrombectomy, were referred from outside hospitals, or had missing HR data were excluded.
All the patients treated with rtPA received care in accordance with the 2019 Clinical Practice Guidelines for Ischemic Stroke. These are based on protocols established by the National Institute of Neurological Disorders and Stroke [13] and the European Cooperative Acute Stroke Study III [3] for patients presenting within 3–4.5 hours of symptom onset. rtPA was administered at a dose of 0.9 mg/kg (maximum dose, 90 mg), with 10% administered as an initial intravenous bolus and the remaining 90% infused continuously over 60 minutes. If the baseline blood pressure exceeded 185/110 mm Hg, intravenous nicardipine was administered before thrombolysis was initiated. Patients were admitted to the stroke unit or intensive care unit when the stroke unit beds were unavailable for continuous monitoring. Follow-up brain computed tomography (CT) was performed 24 hours after IVT with rtPA or earlier if clinical deterioration occurred to detect the presence of HT.
Data collection and HR measurement
Demographic data, comorbidities, National Institutes of Health Stroke Scale (NIHSS) scores, imaging findings, and treatments were extracted from electronic medical records. Cerebral infarctions were measured qualitatively using CT scans as normal, lacunar, less than half lobe, up to one lobe, or several lobes, as described by Brott et al. [14]. Stroke subtypes were classified according to the Trial of Org 10172 in Acute Stroke Treatment (TOAST) criteria: large artery atherosclerosis, cardioembolism, small vessel occlusion, other determined etiologies, and undetermined etiology [15]. Brachial blood pressure and HR were recorded hourly using a Philips IntelliVue patient monitoring system (MX800, software version M.00.03; Philips) via bedside monitoring. Serial HR measurements were obtained at baseline (before the initiation of intravenous rtPA) and every hour for 24 hours following admission. A total of 24 HR data points per patient were used to model the HRT.
Outcomes
The primary outcome was HT (symptomatic or asymptomatic) detected on follow-up imaging. Symptomatic HT was defined as clinical deterioration or adverse events indicating clinical worsening, such as drowsiness, increased hemiparesis, and increased NIHSS score by >4 points [16]. Secondary outcomes included in-hospital mortality and Modified Rankin Scale (mRS) scores at discharge. Physicians evaluated the mRS (range, 0–6), which assessed the degree of major disability. A score of 0 represented no symptoms, whereas a score of 5 indicated severe disability such as being bedridden, incontinent, or requiring continuous nursing care. A score of 6 corresponded to death [17]. Mortality was verified using hospital-issued death certificates regardless of the cause.
Statistical analysis
Group-based trajectory models (GBTM) are well suited for identifying HRT patterns across 24 hourly time points within the first 24 hours of observation in a cohort. A Monte Carlo simulation study indicated that a minimum sample size of 150 participants is required to achieve a statistical power of 0.81 when the population regression coefficient is fixed at 0.2 [18]. The model selection was guided by several statistical criteria and fit indices. The average posterior probability was examined to evaluate the likelihood of individuals belonging to each trajectory group, with values exceeding 0.70 considered indicative of acceptable classification. Additionally, each trajectory group was required to comprise at least 5% of the sample, based on posterior probabilities. Model comparisons were primarily based on the Bayesian Information Criterion (BIC) with higher BIC values indicating superior model fit [19]. The GBTM handles missing data using the maximum likelihood estimation as part of the model-fitting process.
Categorical variables were summarized as frequencies and percentages and analyzed using either the chi-square test or Fisher’s exact test, as appropriate. Continuous variables are expressed as medians with interquartile ranges and were compared between groups using the Mann-Whitney U-test. Baseline characteristics across trajectory groups were assessed using the chi-square test for categorical variables and the Kruskal–Wallis test for continuous variables, with post hoc pairwise comparisons adjusted by Bonferroni correction. To examine the association between HRT and HT, multivariable logistic regression analysis was performed. Univariate analysis was initially performed, and variables with a P-value <0.10 were considered for inclusion in the multivariable model after assessing for multicollinearity. To account for potential bias arising from the relatively small sample size, Firth’s penalized logistic regression was applied [20]. Odds ratios and 95% CIs were calculated using logistic regression. GBTM was conducted using STATA version 16 (StataCorp.), incorporating the user-written "Traj" command, created by Jones and Nagin [21], specifically for trajectory analysis.
RESULTS
Study population
In total, 164 patients with AIS who underwent IVT with rtPA between January 2015 and February 2025 were included in the final analysis (Fig. 1). The median age was 72 years (interquartile range, 60–81 years), and 60.4% of patients were male. HT occurred in 38 patients (23.2%), of whom 20 (12.2%) had symptomatic HT.
Comparison of clinical characteristics by HT
Patients with HT had significantly higher initial NIHSS scores (median, 13 vs. 9; P=0.002) and were more likely to have atrial fibrillation (31.6% vs. 11.1%, P=0.002), coronary artery disease (26.3% vs. 10.3%, P=0.013), or chronic kidney disease (28.9% vs. 14.3%, P=0.038). HT was also associated with larger infarct sizes, particularly those involving several lobes (21.1% vs. 4.8%, P=0.002), and greater use of intravenous antihypertensive agents (47.4% vs. 27.0%, P=0.018). Functional outcomes were worse among those with HT, as reflected by higher discharge mRS scores (median, 5 vs. 3; P<0.001), longer hospital stay (median, 12 vs. 5 days; P<0.001), and greater in-hospital mortality (21.1% vs. 5.6%, P=0.008) (Table 1).
HRTs
Latent class trajectory modeling identified four distinct patterns of HR changes during the first 24 hours after thrombolysis (Fig. 2). These included a low-early decline HRT group (T1, 7.3%), a moderate-gradual decline HRT group (T2, 37.8%), a moderate-stable HRT group (T3, 40.9%), and a high-gradual increase HRT group (T4, 14.0%). The assessment of the model adequacy is presented in Supplementary Table 1. The average posterior probabilities for all trajectory groups exceeded the recommended cutoff value of 0.7.
Baseline characteristics and clinical outcomes across trajectory groups
Hypertension was more frequently observed in group T3 (80.6%) compared with T2 (56.5%, P=0.024). Initial NIHSS scores were significantly higher in T3 than T1 (median, 11 vs. 7; P=0.017). The proportion of patients with normal infarct imaging was significantly lower in T3 than in T2 (13.4% vs. 41.9%, p < 0.01), and even lower in T4 (8.7%, p=0.024 compared to T2). Regarding stroke subtypes, small vessel occlusion was less common in T4 than in T1 (0% vs. 41.7%, P=0.012), whereas no other pairwise comparisons remained significant after correction. No significant differences in age, sex, diabetes, or stroke location were observed between the groups.
The clinical outcomes also differed according to the HRT. The discharge mRS scores were worse in the higher HRT groups, particularly at T3 vs. T2 (median, 4 vs. 3; P=0.036) and T4 vs. T2 (median, 4 vs. 3; P=0.027). The median length of hospital stay increased progressively from T1 to T4 (4, 5, 6, and 8 days, respectively; overall P=0.007), with significant pairwise differences between T1 and T4 (P=0.015), and between T2 and T4 (P=0.008). The in-hospital mortality rates were not significantly different between the groups (P=0.132) (Table 2).
Association between HRT and HT
After adjusting for coronary artery disease, atrial fibrillation, chronic kidney disease, initial NIHSS score, infarcts involving several lobes, and the use of intravenous antihypertensive agents, patients in the high-gradual rise HRT group (T4) had a significantly higher risk of HT compared with the low-early decline group (T1), with an adjusted odds ratio of 11.1 (95% CI, 1.0–122.6; P=0.049). In Firth’s logistic regression, the association was attenuated, with an adjusted odds ratio of 7.1 (95% CI, 0.9–54.9; P=0.059) (Table 3).
DISCUSSION
This study provided novel insights into the prognostic value of HRT in patients with AIS treated with IVT. Our findings indicate that a persistently elevated HR during the first 24 hours after thrombolysis is significantly associated with HT development. These associations were observed even after controlling for known clinical and radiological predictors of HT, suggesting that the HRT may serve as an independent and dynamic biomarker of neurovascular instability.
The effects of HR and HR-derived values on functional outcomes in AIS have been studied. Previous studies have evaluated the single-point HR on admission [22] and simple mean HR levels [8,23] as predictors of outcomes in AIS. Several focused on HR variability, which is associated with adverse outcomes [24,25]. However, one study failed to demonstrate an association [8].
The trajectory-based analysis used in this study captured the dynamic evolution of HR, allowing for a more nuanced identification of high-risk patients. Our results align with those of Feng et al. [11] and Wang et al. [12], who reported associations between high HRT and poor functional outcomes in patients with AIS who did not receive IVT and in those who underwent mechanical thrombectomy.
The pathophysiological explanation for the effect of the HR on the HT lies in the interplay between autonomic dysfunction and post-stroke cerebral autoregulation. Following an ischemic insult, impaired baroreceptor sensitivity and sympathetic overactivity can lead to elevated HR and blood pressure variability. These fluctuations may exacerbate reperfusion injury, promote disruption of the blood–brain barrier, and facilitate hemorrhagic conversion of ischemic tissue. Liang et al. [26] demonstrated that a higher HR and blood–brain barrier leakage were associated with HT in patients with AIS. Interestingly, we observed that none of the patients in the high-gradual rise trajectory group (T4) had small-vessel occlusion, compared with 41.7% in the low-early decline group (T1). However, more than half of T4 patients were diagnosed with stroke of undetermined etiology, reflecting an incomplete diagnostic workup in this retrospective cohort. Therefore, the findings should be interpreted with caution. This suggests that patients with lacunar infarcts, which are typically associated with milder clinical severity, are less likely to present with a persistently increasing HR. Previous studies have shown that patients with large-artery atherosclerosis exhibit more severely impaired parasympathetic and sympathetic functions than those with small-vessel occlusion, supporting the idea that autonomic dysfunction varies according to the stroke subtype [27].
Various clinical factors, such as NIHSS score, hyperglycemia, elevated blood pressure, advanced age, low platelet count, use of antithrombotic agents, and reperfusion therapy, have been associated with an increased risk of HT [6]. Cardioembolic stroke and atrial fibrillation are known risk factors of HT [28]. In our study, atrial fibrillation was more common in patients with HT, but cardioembolic stroke was not, likely because many cases were classified as having an undetermined etiology based on incomplete evaluations. Several predictive scores were developed based on these factors, showing fair performance with C-statistics ranging from 0.50 to 0.86 [5]. Systolic blood pressure is a variable commonly included in these models. However, none of the existing scores have incorporated HR as a predictor.
To our knowledge, this is the first study to explore the association between HRT and HT in patients with AIS undergoing IVT. Our findings support the integration of real-time HR monitoring and trajectory analysis into stroke unit protocols. Early identification of patients with high-risk HR trajectories may allow for tailored interventions, such as aggressive blood pressure and HR control or intensified monitoring, to reduce the risk of hemorrhagic complications.
This study had some limitations. First, its retrospective, single-center design may introduce residual confounding factors and is inherently limited by the reliance on chart reviews. Secondly, patients who underwent thrombectomy were excluded, which may limit the generalizability of our findings to other stroke protocols involving endovascular treatments. Third, we collected only hourly HR data for the first 24 hours. Continuous long-term HR monitoring may provide additional insights and warrants further investigation. Finally, with 38 HT events, the events-per-variable (EPV) was approximately 6, which is below the conventional recommendation of ≥10 EPV [29]. This limited EPV might have increased the risk of overfitting and contributed to the wide CIs. To address this, we applied Firth’s penalized logistic regression to reduce small-sample bias. Multicenter, large-scale cohort studies are required to validate our findings.
In conclusion, a distinct 24-hour HRTs was associated with HT and functional outcomes in patients with AIS treated with IVT. A persistently high HRT may serve as a useful marker for early risk stratification and as a potential target for intervention.
XML Download