Interaction Effect of Baseline Serum Glucose and Early Ischemic Water Uptake on the Risk of Secondary Hemorrhage After Ischemic Stroke

Background and Purpose: Intracerebral hemorrhage (ICH) after mechanical thrombectomy (MT) for acute ischemic stroke (AIS) remains a major complication and its early prediction is of high relevance. Baseline serum glucose (BGL) is a known predictor of ICH, but its interaction with early ischemic changes remains uncertain. We hypothesized that BGL interacts with the effect of tissue water uptake on the occurrence of ICH. Methods: Three hundred and thirty-six patients with acute ischemic stroke treated with MT were retrospectively analyzed. ICH was diagnosed within 24 h on non-enhanced CT (NECT) and classified according to the Heidelberg Bleeding Classification. Early tissue water homeostasis has been assessed using quantitative lesion net water uptake (NWU) on admission CT. Multivariate logistic regression was used to identify predictors of ICH. Results: One hundred and seven patients fulfilled the inclusion criteria of which 37 (34.6%) were diagnosed with ICH. Patients with ICH had a significant higher BGL on admission (median 177 mg/dl, IQR: 127–221.75, P < 0.001). In patients with low BGL (<120 mg/dl), higher NWU was associated with 1.34-fold increased likelihood of ICH, while higher NWU was associated with a 2.08-fold increased likelihood of ICH in patients with a high BGL (>200 mg/dl). In multivariable logistic regression analysis, BGL (OR: 1.02, 95% CI: 1.00–1.04, P = 0.01) and NWU (OR: 2.32, 95% CI: 1.44–3.73, P < 0.001) were significantly and independently associated with ICH, showing a significant interaction (P = 0.04). Conclusion: A higher degree of early tissue water uptake and high admission BGL were both independent predictors of ICH. Higher BGL was significantly associated with accelerated effects of NWU on the likelihood of ICH. Although a clear causal relationship remains speculative, stricter BGL control and monitoring may be tested to reduce the risk of ICH in patients undergoing thrombectomy.


INTRODUCTION
Large randomized controlled trials provide efficacy of mechanical thrombectomy (MT) over medical treatment in patients with acute ischemic stroke (AIS) (1, 2). Irrespective of this success, intracerebral hemorrhage (ICH) remains a common and challenging complication with a negative impact on functional outcome. It is these very trials that have further verified a steady rate of symptomatic intracerebral hemorrhage (sICH) with 4.4% after MT when compared with intravenous thrombolysis (1). In addition, also asymptomatic ICH (aICH) has proven to have a negative impact on long-term functional outcome (3)(4)(5)(6). Therefore, evaluating the risk factors for ICH becomes an important issue for continuously improving the efficacy of MT in patients with AIS. In patients with AIS, hyperglycemia has been a long recognized and frequent finding with up to 50% (7,8) and repeatedly associated with increased bleeding events after thrombolytic therapy (IVT) and poor functional outcome (7)(8)(9)(10). Furthermore, a linear relationship between increasing serum glucose and ischemic brain edema has been elucidated (11) and the latter is an independent risk factor for ICH after successful MT. The current state of literature provides no sufficient data for the mutual impact of serum glucose and early ischemic brain edema on the risk of ICH in patients with AIS (12)(13)(14). Utilizing the widely accepted Heidelberg Bleeding Classification (15), the objectives of the present study were to analyze the mutual impact of serum glucose and ischemic brain edema on the risk of ICH in patients with AIS after successful MT and, in particular, to investigate whether different glucose levels modify the effect of early edema formation on clinical outcome. We hypothesized that baseline serum glucose levels interact with the effect of early ischemic brain edema on functional outcome.

Patients
Data of an anonymized cohort of 107 consecutive AIS patients with occlusion in the anterior circulation were retrospectively evaluated, in whom successful MT (TICI 2b/3) was performed (Figure 1). Patients were admitted between June 2015 and March 2018 in the University Medical Center Hamburg-Eppendorf (n = 336). Data were analyzed after ethical board approval, and informed consent was waived by the institutional review board. The data that support the findings of this study are available from the corresponding author in accordance with the institution's data security regulations upon reasonable request. The patients were screened consecutively based on the following a priori defined inclusion criteria: (1) acute ischemic stroke with occlusion of the middle cerebral artery (MCA) or terminal internal carotid artery (ICA), (2) initially performed multimodal CT protocol with CT angiography (CTA) and perfusion CT (CTP), (3) known time from symptom onset to imaging <6 h, (4) follow-up CT (FU-NECT) within 24 h after admission imaging, (5) NIHSS score above 3, (6) documented NIHSS after 24 h and modified Ranking Scale (mRS) after 90 days, and (7) absence of pre-existing thromboembolic or hemodynamic infarctions in admission non-enhanced CT (NECT) within 24 h of stroke symptom onset. Baseline clinical characteristics including baseline serum glucose (BGL) and demographic information were filtered from the medical records. The followup CT (FCT) was analyzed for secondary ICH.

Image Analysis
For image analysis, anonymized CT imaging scans were evaluated independently by two radiologists with 5 (JN) and 7 (GB) years of dedicated neuroradiology experience, blinded to all clinical and imaging information except stroke side.

Quantification of Ischemic Brain Edema
Anonymized admission CT imaging scans were segmented semimanually using commercially available software (Analyze 11.0, Biomedical Imaging Resource, Mayo Clinic, Rochester, MN, USA) to derive net water uptake (NWU). The edematous proportion of the hypoattenuated ischemic lesion (% water uptake) was quantified using CT densitometry as previously published (17,18). In brief, edematous volumetric changes of ischemic lesions due to water uptake were directly quantified by measurements of relative hypoattenuation (Equation 1). Visually evident edematous hypoattenuation was identified as infarct core lesion for further analysis, and a region of interest (ROI) was placed in this infarct core lesion (Dischemic). A symmetric ROI was mirrored automatically within the normal tissue of the contralateral hemisphere in order to obtain the density of the normal tissue prior to the infarction (Dnormal) (19)(20)(21). CT perfusion was used in addition to aid ROI definition of the early ischemic core by simultaneously presenting cerebral blood volume (CBV) parameter maps at a window between 0 and 6 ml/100 ml (19)(20)(21)(22). ROIs were segmented with semiautomatic edge detection and sampled between 20 and 80 HU as described by Broocks et al. (19)(20)(21)(22). Inaccuracies were corrected in consensus reading, if necessary. Relative NWU was calculated based on D ischemic and Dnormal according to Equation

Statistical Analysis
Data were tested for normality and homogeneity of variance using histogram plots and Kolmogorov-Smirnov-tests. Absolute and relative frequencies are given for categorical data. Median and interquartile range (IQR) are given for univariable distribution of metric variables. Patients with ICH vs. without ICH were compared by Mann-Whitney U-test for metric outcome variables and by chi-square test for categorical outcome variables ( Table 1). Kappa statistic and calculated Cohen's k were used for interrater reliability measurement. ROC analysis was assessed to analyze the diagnostic performance for prediction of ICH after successful MT using increasing discrimination thresholds of independent predictors and cutoffs determined according to the Youden index (24,25). The association between clinical and radiological parameters and ICH in patients after successful MT was assessed by univariate logistic regression analysis. For multivariate logistic regression analysis, a model with forward selection was used to identify significant variables for developing ICH (inclusion criterion: P-value of the score test ≤0.05, exclusion criterion: P-value of the likelihood ratio test >0.1) ( Table 2). Given for selected variables are odds ratio (OR) with 95% CI and P-value of likelihood ratio test. For non-selected variables, P-value of score test is displayed. The impact of BGL on the association of NWU and occcurence of ICH was tested using logistic regression analysis. The OR for NWU increase was assessed for patients with five different levels of blood glucose based on the relative distribution of BGL (Table 3). Finally, to further analyze the relationship of early edema formation and BGL, we trichotomized patients into three groups based on the distribution of NWU within the patient cohort. The interaction term for trichotomized NWU and BGL was calculated and plotted ( Figure 2). A statistically significant difference was accepted at a P-value of <0.05. No adjustment for mutlitple testing was performed as the analyses being explorative in nature. Analyses were performed using MedCalc (version 11.5.1.0; Mariakerke, Belgium) and Stata/SE 13.0 (StataCorp, College Station, TX, USA).

Patients
Out of 336 patients, 107 consecutive patients were included according to the inclusion criteria [median age 76 years (IQR: 65.0-81.0) and 51.4% females] with 63 patients (58.9%) having received intravenous thrombolysis before MT. An example is illustrated in Figure 3.   Figure 4 illustrates the relationship of BGL and NWU on the occurrence of ICH.

Prediction of ICH
Univariate ROC analysis was performed to identify the diagnostic accuracy of independent variables of univariate  Odds ratios (OR) for net water uptake (NWU) given for selected levels of serum blood glucose with 95% confidence interval (CI) and P-value of likelihood ratio test.

DISCUSSION
Higher admission BGL as an independent predictor for developing an ICH after successful ET in AIS paired with the association between ischemic edema is the main finding of our study. Furthermore, the effect of NWU on the occurrence of ICH was increased in patients with higher BGL suggesting a specific interrelation between early edema formation, as a sign of blood-brain barrier injury, and BGL as a potential "accelerator" of blood-brain barrier injury. This finding was accompanied by elevated early ischemic edema and poor collateral score as a second independent predictor for ICH. BGL and NWU showed a significant interaction indicating that the slopes for likelihood of ICH with increasing BGL differ significantly according to the degree of NWU, as illustrated in Figure 2. In patients with very low early NWU, BGL increase did not alter the risk of ICH, while a BGL increase in patients with higher NWU resulted in a significant increase in likelihood for ICH. The specific interaction between early edema formation and serum glucose levels and its impact on the occurrence of ICH has not yet been described and might be a hint of a specific pathophysiological association. It has been observed that elevated levels of BGL are associated with aggravated edema formation (13), as a sign of blood-barrier injury, and that elevated levels of early edema formation increased the risk of secondary hemorrhage (12). Hence, the coexistence of high BGL and high NWU might be a constellation of very high risk for ICH and should be a hint for clinicians to indicate stricter monitoring and consider adjustment of glucose levels. Moreover, the findings could enrich for patients to study experimental treatments with antiedematous drugs, such as glyburide. Glyburide, an antidiabetic drug, is an inhibitor of the sulfonylurea receptor 1 and transient receptor potential melastatin 4 (SUR1-TRPM4) (26). The application of glyburide is safe and feasible and has been described for preventing edematous brain edema (27). In previous pre-clinical studies using rodent models with malignant edema, inhibition of SUR1 resulted in lower ischemic lesion volume, reduced mortality, and better functional outcome (28). Supporting these pre-clinical data, retrospective analyses on patients with diabetes and AIS observed that patients with medication of sulfonylurea drugs had an improved clinical outcome and lower rates of hemorrhagic transformation (HT) (29).
The nature of the association between acute elevated levels of admission glucose and increased risk of ICH has already been investigated in animal models of AIS as well as in both FIGURE 2 | Interaction analysis of BGL and NWU on the occurrence of ICH. Occurrence of secondary intracerebral hemorrhage (ICH, y-axis) according to trichotomized NWU (net water uptake; low NWU: <7%; intermediate NWU: 7-12%; high NWU: >12%) and BGL (baseline glucose level; glucose in mg/dl, x-axis). The interaction term between NWU and BGL was significant. FIGURE 3 | Illustrative example of a patient with baseline glucose and water uptake and secondary ICH after successful thrombectomy. Illustration of a patient with high early baseline blood glucose and high ischemic water uptake, with subsequent secondary intracerebral hemorrhage after thrombectomy. On the left, patient admission images are displayed with admission non-enhanced CT and ROIs for ischemic hypoattenuation (D ischemic ) and on the contralateral side (D normal ) as well as relative CBV (rCBV). The rCBV map (ml per 100 ml) is inferred from a quantitative assessment of the partial volume averaging in each pixel. In the middle, follow-up non-enhanced CT images at 24 h are displayed with secondary intracerebral hemorrhage (parenchymal hematoma; grade 1). On the right, follow-up non-enhanced CT images at 48 h are displayed with secondary intracerebral hemorrhage (parenchymal hematoma; grade 2). rCBV, relative cerebral blood volume; NECT, non-enhanced CT; PH1, parenchymal hematoma grade 1; PH2, parenchymal hematoma grade 2.  retrospective and prospective clinical imaging studies (7,(30)(31)(32)(33). Previous studies evaluating the risk and rates of ICH have been hampered by the lack of consensus definitions for bleeding events (31). The recently introduced Heidelberg Bleeding Classification provides a standardized and reproducible tool and basis for evaluating further treatment strategies (34). Nevertheless, our study is methodically limited to the important fact that a systematic approach in classifying symptomatic ICH is missing which should be addressed in future clinical studies with larger patient cohorts. The quantitative stratification of ischemic brain edema via NWU demonstrates the direct relation of hypoattenuation in CT to the percentage of volume of water uptake and has been validated with excellent sensitivity and reproducibility since (19,20,35). Recent findings of our study group give two major explanatory reasons for poor clinical outcome after MT and the risk of an ICH. Firstly, poor clinical outcome has been associated with increased levels of ischemic edema and further associated with poor collateral score and elevated BGL (12,13,36). Secondly, collaterals and the degree of NWU both mediate tissue vulnerability and the risk for an ICH (12). In line with this, the recently published study of Hao et al. also reported poor collateral circulation as an independent predictor for ICH (37). Yet, the relationship of NWU and hyperglycemia with respect to the occurrence of ICH after successful MT remains unknown. The pathophysiology adds support to our hypothesis as both elevated BGL and ischemic edema share a common pathway of impaired blood-brain barrier (BBB) with the risk of increased tissue vulnerability: ICH after MT occurs due to a reperfusion syndrome from rupture of necrotic vessel walls and increased BBB permeability due to prolonged ischemia (38,39). By mediating both oxidative stress and inflammation response in vessel walls, hyperglycemia is also associated with increased reperfusion injury (40)(41)(42). The aggravated breakdown of the BBB results in further edema formation and increased infarct volume (8,43). Thorén et al. have investigated the impact of hyperglycemia on admission as an independent risk factor for cerebral edema in patients with AIS treated with IVT and obtained the extent of ischemic edema with a visual rating system (11). This study is in concordance with our previous findings that proved that the degree of early edema formation using a quantitative imaging biomarker is a risk factor for secondary ICH, possibly potentiated by increased levels of BGL (12,38,39). Taken all together, these independent variables may provide a closed loop that provides a target for therapeutic intervention to disrupt the cycle. Further clinical studies analyzing the mediated effect of NWU on the risk of ICH by increased levels of BGL are therefore needed. Also, potential beneficial treatment effects from the control of hyperglycemia in patients with AIS have been described numerously (44), yet established treatment methods are still under investigation. Although results from experimental studies support a causal relationship between hyperglycemia and poor functional outcome after stroke, multicenter trial data presented in the SHINE, GIST-UK, or THIS trial do not yet support intervention with insulin (45)(46)(47). In the future, clinical trials might consider combining reperfusion with further adjuvant treatment in patients with AIS and elevated levels of admission glucose. Yet, standardized methods to monitor possible effects on ischemic tissue edema are still missing. In this respect, quantitative NWU might provide a feasible imaging biomarker to monitor the effects of these drugs in prospective clinical trials, subject to the condition of a confirmed mediated effect of NWU by BGL (26,27,48).
Several limitations of our study deserve attention. First, for the assessment of NWU measurements, ROIs were drawn by a quantitative edge detection tool and boundaries adjusted manually if necessary, for example, in cases of anatomical asymmetry which may hamper NWU quantification. However, semiautomatic methods of Hounsfield unit value thresholding have been used in our previous studies with excellent reproducibly and partially mitigate a method bias (17,21). Interrater reliability may further help to augment the accuracy of NWU measurements in the future. Additionally, the lack of information on the longitudinal course of glucose levels upon FU-NECT and the undocumented use of blood glucose-lowering drugs limit the generalizability of our results. Furthermore, the purpose of this study was to predict secondary ICH without further differentiating into symptomatic or asymptomatic ICH.
As it has been observed in both retrospective and prospective studies that asymptomatic ICH is also associated with worse functional outcome, we nevertheless consider our observation to be of importance (4)(5)(6). In light of these findings, it is relevant to investigate the risk factors for any ICH, to finally improve the prevention of ICH in patients undergoing MT. Finally, according to the retrospective nature of the study, no causality assumptions can be inferred from the obtained data.

CONCLUSIONS
Our study confirmed that higher BGL increased the likelihood for ICH, but depends on the degree of early ischemic edema. Although a causal relationship between NWU and higher BGL and ICH remains speculative and more data are needed, specific interactions between BGL and NWU may be tested as a further therapeutic target.

DATA AVAILABILITY STATEMENT
The data analyzed in this study is subject to the following licenses/restrictions: According to the institution's strict data security regulations the entire datasets are not publicly available. Requests to access these datasets should be directed to jawed.nawabi@charite.de.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Ärztekammer Hamburg WF-035/18. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

AUTHOR CONTRIBUTIONS
All authors listed have made substantial, direct and intellectual contribution to the work and approved it for publication.