Synergistic insights: the integrated role of CT/CTP and clinical parameters in hemorrhagic transformation prediction

Descriptive clinical characteristics of subjects

From January 2021 to August 2023, our department managed a total of 224 patients suffering from large vessel-occlusion acute ischemic stroke. These patients were categorized based on the outcomes following thrombectomy: 175 patients were identified with No HT, while 49 patients experienced HT as observed in post-thrombectomy cranial CT scans within 7 days. There was no significant difference in the average age and gender distribution among these groups.

In analyzing the comorbidity profiles of these patients, we observed distinct patterns between the two groups. In the HT group, 36.7% of patients had atrial fibrillation, compared to only 17.1% in the No HT group, a finding that was statistically significant. However, for other comorbidities such as hypertension, diabetes, coronary artery disease, and smoking history, there were no significant differences observed between the HT and No HT groups. Additionally, we collected data on the systolic blood pressure and serum glucose levels of patients upon admission to our hospital. Analysis of these parameters showed no significant differences between the groups.

Significant differences were noted in the National Institutes of Health Stroke Scale (NIHSS) scores on admission between the two groups. Patients in the No HT group had lower NIHSS scores, with a median of 11 (interquartile range (IQR), 7 to 15), compared to those with HT, who had higher scores, with a median of 13 (IQR, 10 to 16). Additionally, a higher proportion of patients in the HT group received intravenous thrombolysis with recombinant tissue plasminogen activator (rt-PA) before undergoing thrombectomy, at 11.6%, compared to only 3.6% in the No HT group, which was statistically significant. Furthermore, the Door-to-Puncture Time (DPT) in patients with HT (median 82.5 minutes, IQR 74.8 to 101.5) was significantly shorter than that in the No HT group (median 93.0 minutes, IQR 83.0 to 116.5). Despite these differences, the Time from Stroke Onset to Groin Puncture was similar between the two groups, indicating no significant variation in the overall treatment initiation timeframe. The above clinical information was illustrated in Table 1.

CTP findings and perfusion deficits in HT patients

Upon admission, computed tomography (CT) and perfusion CT (CTP) imaging were performed on our patients to assess the extent of ischemic stroke. The Alberta Stroke Program Early CT Score (ASPECTS) was utilized to evaluate initial brain ischemia. Patients with no HT demonstrated significantly higher ASPECTS, with a median score of 10 (IQR, 9 to 10), as opposed to those with HT who had lower scores with a median of 9 (IQR, 7 to 9) (Figure 1A).

For CTP imaging metrics, the volume of regions with reduced cerebral blood flow (rCBF) less than 30%, which reflects the core infarction zone, was considerably larger in the HT group with a median volume of 19.1 mL (IQR, 2.3 to 43.9 mL), compared to the No HT group, which had a median volume of 0.9 mL (IQR, 0.0 to 7.7 mL) (Figure 1B). Moreover, the volume of brain tissue experiencing a time-to-maximum (Tmax) greater than 6 seconds was significantly larger in the HT group (median 152.6 mL, IQR 80.4 to 199.1 mL) versus the No HT group (median 111.4 mL, IQR 47.5 to 159.3 mL), suggesting more extensive brain tissue at risk of progressing to infarction (Figure 1C).

Additionally, the mismatch ratio, indicative of the ischemic penumbra—the area of salvageable tissue surrounding the core infarct—was significantly smaller in the HT group (median 3.7, IQR 2.3 to 7.9) when compared to the No HT group (median 11.7, IQR 5.6 to 21.0). This suggests a more limited therapeutic window and potentially lower benefit from reperfusion therapy in the HT group. These imaging findings highlight the value of early and precise CTP assessment in guiding clinical decisions for acute stroke interventions (Figure 1D).

Patients with HT had longer recanalization time during thrombectomy

In the evaluation of endovascular thrombectomy outcomes, recanalization success was measured using the modified Thrombolysis in Cerebral Infarction (mTICI) scale. The proportion of patients achieving mTICI 2b-3 recanalization, indicative of substantial reperfusion, did not differ significantly between the groups, with 85.6% in the No HT group and 87.8% in the HT group. Similarly, the rate of achieving one-pass recanalization showed no significant difference between the No HT group (47.3%) and the HT group (40.8%) (Table 2).

However, the duration from groin puncture to recanalization was notably longer in the HT group, with a median time of 154 minutes (interquartile range (IQR), 109.5 to 221.0 minutes), compared to the No HT group, which had a median time of 132.0 minutes (IQR, 95.0 to 184.0 minutes) (Figure 2A). Moreover, when considering the total elapsed time from stroke onset to recanalization, patients in the HT group experienced a longer duration with a median time of 557.0 minutes (IQR, 466.5 to 847.5 minutes), as opposed to the No HT group, which had a median time of 470.5 minutes (IQR, 351.5 to 655.5 minutes) (Figure 2B).

Independent risk factors for HT in IS

The results of univariate and multivariate logistic analyses between patients with HT and those without HT are presented in Table 3. The univariate analysis suggests that compared to patients without HT, the patients with HT are more likely to have AF history (OR, 2.863; 95% CI, 1.404–5.785; P = 0.003), to have higher NIHSS scores (OR, 1.038; 95% CI, 0.997–1.081; P = 0.068) and ASPECTS (OR, 0.631; 95% CI, 0.506–0.770; P < 0.001), to have a larger volume of regions with rCBF less than 30% (OR, 1.026; 95% CI, 1.015–1.039; P < 0.001), to have a larger volume of brain tissue experiencing Tmax greater than 6 seconds (OR, 1.006; 95% CI, 1.001–1.010; P = 0.009), and to have a longer time from stroke onset to recanalization of the offending vessel (OR, 1.001; 95% CI, 1.000–1.002; P = 0.025). Factors with a p-value less than 0.1 in the univariate analysis, along with factors previously proven to be potentially related to HT in other studies, were included in the multivariate logistic regression analysis. After optimizing the model fitting through stepwise forward and backward methods and adjusting for other factors, a history of AF (OR, 2.834; 95% CI, 1.220–6.547; P = 0.014) and hypertension (OR, 2.604; 95% CI, 1.133–6.523; P = 0.031), lower preoperative blood pressure (OR, 0.982; 95% CI, 0.966–0.997; P = 0.022), lower ASPECTS (OR, 0.755; 95% CI, 0.571–0.990; P = 0.044), a larger volume of brain tissue with rCBF less than 30% (OR, 1.018; 95% CI, 1.004–1.034; P = 0.018), and a longer time from symptom onset to recanalization (OR, 1.001; 95% CI, 1.000–1.003; P = 0.017) were identified as independent risk factors for HT (Table 3, Figure 3).

Development and evaluation of an HT-predicting nomogram

A nomogram for predicting HT was established based on an equation derived from these independent risk factors (Figure 4). The score of each independent predictor corresponds to the score on the upper scale, and the total score for each subject is the sum of the scores from each independent predictor. The total score corresponds to the risk of HT on the risk axis, with higher total scores indicating a higher risk of HT. The nomogram we constructed is available through a free browser-based online calculator at https://chao-yang-hospital-neurosurgery.shinyapps.io/Nomogram_predicting_HT/ (Figure 5). The calibration curve (Figure 6) suggests that, compared to the model based solely on clinical indicators, the nomogram’s calculation model incorporating CT and CTP indicators shows overall better coherence between the predicted probability of HT and the actual clinical characteristics of HT. Additionally, the Receiver Operating Characteristic (ROC) curve (Figure 7) of the model incorporating CT and CTP indicators also performs better. The AUC for the model built on clinical indicators is 0.706. In contrast, the ROC curve for the nomogram’s calculation model that includes clinical and radiological indicators has a larger AUC of 0.808, with the model’s sensitivity and specificity being 0.755 and 0.779, respectively. Decision curve analysis (DCA) curves demonstrate that the nomogram is more effective in predicting HT, as it provides greater net benefits for identifying HT in IS patients across the threshold probabilities from 0.2 to 0.7, compared to predictions based solely on clinical indicators and with both the intervention for all and the intervention for none (Figure 8).

Patients with HT had higher mRS 90 days after onset of stroke

The prognosis of patients post-thrombectomy was quantitatively assessed using the modified Rankin Scale (mRS) at 90 days following the onset of stroke. Our findings indicated that patients in the HT group had a higher median 90-day mRS score of 4 (IQR: 2 to 6), suggesting greater disability, compared to the No HT group, which had a median 90-day mRS score of 2 (IQR: 0 to 4), (Figure 9A). Furthermore, the percentage of patients achieving a favorable functional outcome, defined as a 90-day mRS score of 0 to 2, was significantly higher in the No HT group (54.3%) compared to the HT group (26.5%) (Figure 9B).

Patients

This retrospective study included a total of 224 patients who underwent thrombectomy at Beijing Chao-Yang Hospital, Capital Medical University, from September 2021 to August 2023. All patients provided informed consent for their inclusion in the study. The research protocol was reviewed and approved by the Ethics Committee of Capital Medical University, ensuring adherence to ethical standards in patient treatment and data handling (2022-ke-268).

We screened potential patients according to strict inclusion and exclusion criteria. The inclusion criteria for this study were: 1) age above 18 years old, 2) a confirmed diagnosis of ischemic stroke via imaging scans (computed tomography, magnetic resonance imaging) or cerebral angiography, and 3) treatment with thrombectomy at our center between September 2021 and August 2023. Exclusion criteria included: 1) failed surgery, 2) death prior to operation, 3) missing pre-operative CTP data, and 4) missing follow-up data. Ultimately, a cohort of 224 patients was included in the present study, as depicted in the study flow diagram (Figure 10).

Diagnosis of hemorrhagic transformation

Our protocol for identifying HT involved a rigorous analysis of Computed Tomography (CT) scans performed within 7 days following thrombectomy. To ensure the reliability of the diagnosis, the imaging data were independently reviewed by three experienced radiologists. The determination of HT was based on distinct radiological evidence of hemorrhage in the brain region corresponding to the area of initial ischemic stroke. The consensus among the radiologists was necessary to confirm the presence of HT, with any discrepancies resolved through a joint review process. This structured approach to diagnosis allowed for a consistent and objective assessment of hemorrhagic transformation across our patient cohort.

Clinical data collection

Clinical data were meticulously collected for each patient. This included demographic information (age and gender), vital signs at admission (blood pressure and blood glucose level), neurological status as assessed by the National Institutes of Health Stroke Scale (NIHSS) on admission, door to puncture time (DPT), and post-operation mTICI score determined via digital subtraction angiography (DSA). Additionally, we recorded the presence of comorbidities such as hypertension, diabetes, atrial fibrillation, coronary artery disease, and smoking history. Information on whether patients received intravenous thrombolysis prior to thrombectomy was also noted. The functional outcome was assessed using the 90-day modified Rankin Scale (mRS 90).

CTP parameters collection

The original CTP images were automatically processed using the SHUKUN AI perfusion software (SK-CTPDoc, StrokePro V2.0, SHUKUN Technology, Beijing, China) and six parameter maps of tMIP, CBF (cerebral blood flow), CBV (cerebral blood volume), MTT (mean transit time), TTP (time to peak), and Tmax (time-to-maximum) were obtained rapidly. The infarct core (relative cerebral blood flow, rCBF<30%), ischemic area volume (Tmax>6 s) and mismatch ratio were calculated automatically. Further, the volumes of Tmax>4 s, Tmax>6 s, Tmax>8 s, Tmax>10 s, and HIR (Tmax>10 s/Tmax>6 s) were also calculated automatically. Artery input and venous outflow and threshold were automatically selected using perfusion analysis software. Time–density curve (TDC) was observed and selection of arterial input and venous outflow was manually adjusted when an abnormal time-density curve was found due to incorrect selection of arteries and veins. The key parameters collected included the volume of Tmax>6s and the volume of rCBF<30%, along with the mismatch ratio (Volume of Tmax>6s/Volume of rCBF<30%). Further details of the CTP protocol will be provided subsequently.

Definition of hemorrhagic transformation

Hemorrhagic transformation was identified and classified based on the criteria set forth by the European Cooperative Acute Stroke Study (ECASS) [19]. This was done through post-thrombectomy CT scans, allowing for a standardized assessment of HT across all patients.

Predictive model

Using univariate logistic regression analysis, we investigated potential predictive factors for HT and non-hemorrhagic transformation (No HT) separately. Subsequently, a multivariate logistic regression model analysis was conducted to identify factors independently associated with HT and No HT. Based on the results of the multivariate logistic regression analysis, nomograms were constructed using the rms package in R software (version 4.2.1). The nomogram was derived by scaling each regression coefficient in the multivariate logistic regression to a scale of 0–100. Using variables independently associated, a total score was calculated and then transformed into predicted probabilities. An interactive web-based dynamic nomogram application was built with Shiny, version 1.8.0. To assess the accuracy of the nomogram, internal validation of the nomogram was conducted using the bootstrap method with 1000 resamples, subsequently obtaining the calibration curve. The predictive ability of the nomogram’s calculation model for HT in IS patients was evaluated using Receiver Operating Characteristic (ROC) curves. The optimal cutoff value was determined as the point on the curve where the difference between sensitivity and 1-specificity was maximal at each horizontal coordinate point [20]. Further confirmation of the clinical effectiveness of the nomogram was carried out through Decision Curve Analysis (DCA), a method for assessing the clinical utility of predictive models [21]. The nomogram was evaluated by calculating the net benefit at different threshold probabilities. DCA, capable of illustrating the changing proportion of false positives and true positives with varying risk thresholds, serves to complement deficiencies in the ROC curve.

Statistical methods

For the statistical analysis, GraphPad Prism version 10.0 (GraphPad Software, MA, US) was utilized. The Chi-Square test was employed for analyzing nonparametric data, with results presented as median and quartile ranges. For parametric data, the Student’s t-test was used, with mean values and standard errors reported. A p-value of less than 0.05 was considered statistically significant in all analyses.