Article Text

Download PDFPDF

Original research
Predicting symptomatic intracranial haemorrhage after mechanical thrombectomy: the TAG score
  1. Mayra Montalvo1,
  2. Eva Mistry2,
  3. Andrew Davey Chang3,
  4. Aleksandra Yakhkind1,
  5. Katarina Dakay1,
  6. Idrees Azher1,
  7. Ashutosh Kaushal1,
  8. Akshitkumar Mistry4,
  9. Rohan Chitale2,
  10. Shawna Cutting3,
  11. Tina Burton3,
  12. Brian Mac Grory3,
  13. Michael Reznik3,
  14. Ali Mahta3,
  15. Bradford B Thompson5,
  16. Koto Ishida6,
  17. Jennifer Frontera6,
  18. Howard A Riina7,
  19. David Gordon7,
  20. David Parella6,
  21. Erica Scher6,
  22. Jeffrey Farkas8,9,
  23. Ryan McTaggart10,
  24. Pooja Khatri11,
  25. Karen L Furie3,
  26. Mahesh Jayaraman5,
  27. Shadi Yaghi12
  1. 1 Neurology, Brown University Warren Alpert Medical School, Providence, Rhode Island, USA
  2. 2 Neurology, Vanderbilt University, Nashville, Tennessee, USA
  3. 3 Neurology, Brown University Warren Alpert Medical School, Providence, Rhode Island, USA
  4. 4 Neurosurgery, Vanderbilt University Medical Center, Nashville, Tennessee, USA
  5. 5 Neurology, Rhode Island Hospital, Providence, Rhode Island, USA
  6. 6 Neurology, NYU, New York City, New York, USA
  7. 7 Department of Neurosurgery, New York University Langone Medical Center, New York City, New York, USA
  8. 8 Neurology & Neurosurgery, NYU Langone Health, Brooklyn, New York, USA
  9. 9 Neurointerventional Surgery, Interventional Neuro Associates, Bergenfield, New Jersey, USA
  10. 10 Radiology, Brown University Warren Alpert Medical School, Providence, Rhode Island, USA
  11. 11 Neurology, University of Cincinnati Medical Center, Cincinnati, Ohio, USA
  12. 12 Neurology, Columbia University Medical Center, New York City, New York, USA
  1. Correspondence to Dr Shadi Yaghi, Neurology, Columbia University Medical Center, New York, NY 11220, USA; shadiyaghi{at}


Background There is limited data on predictors of symptomatic intracranial haemorrhage (sICH) in patients who underwent mechanical thrombectomy. In this study, we aim to determine those predictors with external validation.

Methods We evaluated mechanical thrombectomy in a derivation cohort of patients at a comprehensive stroke centre over a 30-month period. Clinical and radiographic data on these patients were obtained from the prospective quality improvement database. sICH was defined using the European Cooperative Acute Stroke Study III. We compared clinical and radiographic characteristics between patients with and without sICH using χ2 and t tests to identify independent predictors of sICH with p<0.1. Significant variables were then combined in a multivariate logistic regression model to derive an sICH prediction score. This score was then validated using data from the Blood Pressure After Endovascular Treatment multicentre prospective registry.

Results We identified 578 patients with acute ischaemic stroke who received thrombectomy, 19 had sICH (3.3%). Predictive factors of sICH were: thrombolysis in cerebral ischaemia (TICI) score, Alberta stroke program early CT score (ASPECTS), and glucose level, and from these predictors, we derived the weighted TICI-ASPECTS-glucose (TAG) score, which was associated with sICH in the derivation (OR per unit increase 1.98, 95% CI 1.48 to 2.66, p<0.001, area under curve ((AUC)=0.79) and validation (OR per unit increase 1.48, 95% CI 1.22 to 1.79, p<0.001, AUC=0.69) cohorts.

Conclusion High TAG scores are associated with sICH in patients receiving mechanical thrombectomy. Larger studies are needed to validate this scoring system and test strategies to reduce sICH risk and make thrombectomy safer in patients with elevated TAG scores.

View Full Text

Statistics from


Mechanical thrombectomy using stent retrievers is the standard of care for select patients with proximal anterior circulation large vessel occlusion within 24 hours from last known normal.1 The most feared complication is symptomatic intracranial haemorrhage (sICH), which occurs in up to 6% of patients and is associated with increased mortality and morbidity.2 3 Identifying predictors of sICH provides a tool for clinicians to improve monitoring protocols, inform acute treatment decisions and provide data to share with the patient/family regarding risks and benefits of endovascular treatment.

While risk factors of sICH in patients receiving alteplase are well-established,2 4 5 limited data exist on predictors of sICH in patients undergoing mechanical thrombectomy, and studies identifying predictors of sICH after mechanical thrombectomy have yielded mixed results and face several limitations.6–11 These limitations include lack of external validation and the fact that most of these studies had a relatively small sample size, low sICH rates and did not include patients treated beyond 6 hours from last known normal.

In this study, we aim to determine factors associated with sICH after mechanical thrombectomy and derive a predictive model of sICH with external validation.


Derivation cohort

We obtained institutional review board approval to analyse data from the Rhode Island Hospital prospective quality improvement acute ischaemic stroke database obtained over a 30-month period (01 July 2014 to 31 December 2017). In general, patients admitted to Rhode Island Hospital with suspected acute ischaemic stroke and within 24 hours from last known normal undergo a CT angiography (CTA) of the head and neck to look for large vessel occlusion. The CTAs were emergently reviewed by a radiologist and the presence of large vessel occlusion was defined as occlusion of intracranial internal carotid artery, middle cerebral artery (M1 or proximal M2 segments) and basilar artery. In our study, we included all patients with proximal artery occlusion who underwent mechanical thrombectomy within 24 hours. In general, patients eligible for mechanical thrombectomy were those with proximal large artery occlusion whose Alberta stroke program early CT score (ASPECTS) was six or more in the anterior circulation, had a disabling neurological deficit and were evaluated within 6 hours, or within 24 hours from symptom onset with favourable perfusion imaging. Some patients not meeting the above criteria underwent mechanical thrombectomy, where the decision was made by the neurointerventionalist and vascular neurologist and discussed with the patient and family.

Mechanical thrombectomy and post-thrombectomy care protocol

Mechanical thrombectomy was performed by one of three dedicated neurointerventional radiologists. In general, interventionists performed the procedure using conscious sedation without general anaesthesia. A standardised setup using a six French sheath in the carotid artery, a distal aspiration catheter and a microcatheter were used in all cases. The goal was to achieve modified thrombolysis in cerebral infarction (mTICI) 2b/3 reperfusion. Adjunctive intra-arterial thrombolytic agents were not used in any cases.

Following mechanical thrombectomy, patients were monitored neurologically in the stroke unit for any signs of neurological deterioration; if any neurological deterioration occurred, an emergent head CT was performed to rule out ICH. Antiplatelet therapy was initiated on admission in patients who did not receive alteplase and at 24 hours after alteplase administration if 24 hours brain imaging did not show ICH. As a general rule, target systolic blood pressure post-thrombectomy was <160 mm Hg if mTICI 2b/3 was achieved, or otherwise <180 mm Hg.

Study variables

Study variables were collected prospectively at the time of admission and included:

  1. Demographic variables: age and sex.

  2. Clinical variables: history of hypertension, history of diabetes, history of atrial fibrillation and National Institute of Health Stroke Scale (NIHSS) score on admission.

  3. Medications prior to admission: antiplatelet agents (aspirin, clopidogrel, aggrenox or ticagrelor) and anticoagulants (warfarin or direct anticoagulants).

  4. Laboratory variables: blood glucose obtained by a finger stick on arrival, platelet count on admission and international normalized ratio (INR) on admission.

  5. Radiographic variables: ASPECTS on admission CT and site of occlusion on vessel imaging (internal carotid artery, M1 segment of the middle cerebral artery, M2 segment of the middle cerebral artery or basilar artery). These were determined by a radiologist and confirmed by an independent neuroradiologist.

  6. In-hospital treatments: intravenous alteplase.

  7. Mechanical thrombectomy details: time to reperfusion and degree of reperfusion defined using the mTICI score and adjudicated by the neurointerventionalist on the last angiographic run after the procedure was complete. We defined successful reperfusion as mTICI 2b or 3.


The outcome variable was sICH as determined by the treating physician and confirmed by an independent vascular neurologist. We defined sICH as any deterioration in the neurological exam with ICH on follow-up brain imaging and haemorrhage as the cause of the deterioration, that is, European Cooperative Acute Stroke Study (ECASS) III definition.12

Statistical analysis

Patients were divided into two groups: those with sICH and those without sICH. We performed univariate analyses to compare demographic factors (age and sex), clinical characteristics (history of hypertension, history of diabetes, history of atrial fibrillation, admission NIHSS score, medications (antiplatelet or anticoagulant use) and laboratory values (INR, platelet count and glucose)), alteplase treatment, admission ASPECTS, mTICI score (0-2a vs 2b-3) and time to reperfusion between the two groups using Fisher’s exact tests for categorical variables and t tests for continuous variables. We then performed multivariable logistic regression analyses to determine factors associated with sICH including variables with p<0.1 from univariate models. For continuous variables, an additional multivariable logistic regression model was created using standard or previously defined threshold values to yield a more practical scoring system. Using the ORs of predictors and strengths of associations, we derived a prediction score and externally validated the score in an independent cohort of patients who received mechanical thrombectomy and were enrolled in the multicentre prospective cohort study: Blood Pressure After Endovascular Treatment (BEST). Statistical analysis was performed using SPSS V.20.0.


Baseline characteristics and univariate analyses

We identified 578 patients with acute ischaemic stroke who received thrombectomy in the study period; 19 patients had sICH (3.3%). The mean age of the sample was 73 years and 52.2% were women.

In univariate analyses, factors associated with sICH versus no sICH were admission glucose reported as mean±SD (mg/dL) (162.2±74.8 vs 137.2±57.9, p=0.067), low ASPECTS reported as median (IQR) (10 (9–10) vs 9 (7–10), p=0.068) and unsuccessful reperfusion (TICI 0-2a) (47.4% vs 15.3%, p<0.001). Other variables were not statistically significant (table 1).

Table 1

Univariate analyses of demographics based on symptomatic intracranial haemorrhage (sICH)

Predictors of sICH in the derivation cohort on multivariable regression analyses

A multivariable logistic regression model using factors with p<0.1 in univariate analyses included the following predictors of sICH: glucose level (OR per 10 mg/dL increase 1.07, 95% CI 1.01 to 1.13, p=0.018), ASPECTS (OR 1.52 per 1-point decrease, 95% CI 1.15 to 2.00, p=0.003) and TICI score (OR 5.13. per unit decrease, 95% CI 1.84 to 14.29, p=0.002) (table 2). When previously defined cut-offs for glucose (glucose>150 mg/dL),13 ASPECTS (<5, 6–7 and 8–10)14 and TICI (0-2a and b-3)15 were used, the results remained unchanged (table 2).

Table 2

Multivariable analysis of association between symptomatic intracranial haemorrhage and continuous and categorical predictors

The TICI-ASPECTS-glucose score

Using the OR of the predictors in the multivariable model, we derived the TICI-ASPECTS-glucose (TAG) score to help to predict sICH risk. The components of the score are described in table 3.

Table 3

Components of the TAG score

The TAG score was associated with sICH (OR per unit increase 1.98, 95% CI 1.48 to 2.66, p<0.001). There was increasing risk of sICH with increasing TAG scores (figure 1). In addition, the score had very good predictive ability (AUC 0.79; figure 2). The association between the TAG score and sICH was not altered by alteplase treatment (no alteplase: OR per unit increase 1.93, 95% CI 1.26 to 2.98, p=0.003; alteplase: OR per unit increase 2.24, 95% CI 1.44 to 3.48, p<0.001).

Figure 1

X-axis showing the TAG score and Y-axis showing the percentage of patients with sICH and the TAG score categories: low risk (0–2), intermediate risk (3–5) and high risk (6–7). ASPECTS, Alberta stroke program early CT score; sICH, symptomatic intracranial haemorrhage; TAG, TICI-ASPECTS-glucose; TICI, thrombolysis in cerebral infarction.

Figure 2

Receiver operating characteristic analyses of symptomatic intracranial haemorrhage classification based on the individual components and cumulative TAG scores in (A) the derivation cohort TAG score AUC 0.79 and (B) the validation cohort TAG score AUC 0.69. ASPECTS, Alberta stroke program early CT score; TAG, TICI-ASPECTS-glucose; TICI, thrombolysis in cerebral infarction.

Validation cohort

To determine the reliability and generalisability of the TAG score, we applied it to a multicenter cohort of ischaemic stroke patients who underwent mechanical thrombectomy. We defined sICH in the BEST cohort as any worsening in the NIHSS score at 24 hours with haemorrhage on follow-up brain imaging within 36 hours from thrombectomy. Patients with asymptomatic haemorrhage were excluded, as a causal association between haemorrhage and neurological deterioration was not recorded in the BEST cohort.

Out of 485 patients enrolled in BEST, 86 patients had asymptomatic haemorrhage and were excluded; 399 patients met the inclusion criteria, 31 of which had sICH (6.4%). The mean age was 68 years and 49.9% were women. In the BEST cohort, the TAG score was associated with sICH (OR per unit increase 1.48, 95% CI 1.22 to 1.79, p<0.001), and there were increasing rates of sICH with increase in TAG score categories (figure 1), and the score had good predictive ability (AUC 0.69; figure 2).

TAG score categories

When the TAG score was divided into three categories, low risk (0–2), intermediate risk (3-5) and high risk (6-7), there was an increased risk of sICH in the intermediate-risk and high-risk categories in the derivation and validation cohorts (table 4).

Table 4

Risk of symptomatic intracranial haemorrhage among each of the TAG score categories: low risk (0–2), intermediate risk (3–5) and high risk (6–7) in the derivation and validation cohorts


In this study, we show that patients undergoing mechanical thrombectomy who developed sICH tended to have hyperglycaemia, low ASPECTS and unsuccessful reperfusion (low mTICI score). We created a new score based on these three metrics, the ‘TAG’ score, as a predictor for risk of sICH in these patients. The TAG score successfully predicts the occurrence of sICH in two independent cohorts. In addition, we found no association between alteplase pretreatment and sICH, a finding reported in prior studies.16

Mechanisms of associations

The association between hyperglycaemia and sICH is not unexpected and has been demonstrated in previous studies.17–19 Hyperglycaemia may have deleterious effects on blood vessel integrity via several mechanisms, including blood–brain barrier injury, excitatory chemokines, acidosis and oedema,20 all of which may lead to a higher risk of sICH. Furthermore, low ASPECTS has been shown to correlate with infarct burden on diffusion-weighted imaging sequence and it is intuitive that symptomatic haemorrhage is more likely to occur with increasing infarct core size.21 The association between TICI scores and sICH is noteworthy. Previous studies showed an association between unsuccessful reperfusion and sICH. This is likely due to the fact that unsuccessful reperfusion leads to a larger ischaemic area22 that is more prone to haemorrhagic transformation. The risk of sICH is also accentuated by ischaemic injury to the neurovascular unit and blood–brain barrier,23 leading to extravasation of blood products into the infarcted territory.

Therapeutic implications

This study has several therapeutic implications. First, prediction models for sICH after thrombectomy may help to inform patients, families and treating physicians when selecting patients for mechanical thrombectomy, particularly in patients with low ASPECTS and hyperglycaemia in whom the risk of sICH is as high as 25%. Second, it is possible that patients with elevated TAG scores may constitute a group who may benefit from a more aggressive blood pressure reduction strategy post-thrombectomy. This, however, should be weighed against the risk of worsening ischaemia and infarct extension, particularly in patients with unsuccessful recanalisation who may have a salvageable penumbra. Most importantly, this study highlights the fact that the most important determinant of sICH in patients undergoing thrombectomy is the core infarct volume (ASPECTS), and therefore treatments that minimise the core infarct volume are very important in reducing the risk of haemorrhagic complications. These include faster treatment times achieved by protocols aiming for rapid identification and treatment of patients with large vessel occlusion24 and state-wide protocols that allow patients with suspected large vessel occlusion to be diverted to comprehensive stroke centres to allow for faster times to reperfusion. In addition, certain neuroprotective agents25 or treatment strategies, such as induced hypertension,26 may be studied to reduce infarct growth until reperfusion is achieved. Furthermore, improvement in thrombectomy techniques27–29 to increase the likelihood of successful recanalisation may help to reduce infarct size and sICH rates. This is particularly important as recent evidence suggests a low rate of sICH in patients with ASPECTS<5 when successful reperfusion is achieved, whereas the rate was significantly higher with unsuccessful reperfusion (adjusted OR 0.24, 95% CI 0.06 to 0.89).30

Strengths and limitations

This study has several limitations. First is the relatively small number of sICH patients in the derivation cohort, as this may have underestimated the effect of other variables. Second, collateral scores were not calculated in the derivation cohort, since over the first 2 years of the study we used a single-phase CTA that may limit accurate assessment of collaterals. Third, sICH definitions were not the same in both cohorts since the derivation cohort used the ECASS III definition of sICH and the validation cohort defined sICH as any haemorrhage associated with worsening in the NIHSS score, regardless of whether or not the haemorrhage was the cause of deterioration. Fourth, we lack data on pre-thrombectomy and post-thrombectomy blood pressure parameters, which may also have been a contributing factor in the development of sICH.

Our study has several strengths. It has a relatively large sample size and externally validates the sICH prediction score. In addition, the score is based on components that can be easily abstracted and used in the real-world setting.


High TAG scores are associated with sICH in patients receiving mechanical thrombectomy. Larger studies are needed to validate this scoring system and test interventions and strategies to reduce the risk of sICH and make thrombectomy safer in patients with elevated TAG scores.


View Abstract


  • MM and EM are co-first authors.

  • Contributors MM, EM, AY, KD, IA, AK, AM and RC: Data collection and manuscript revision. ADC: Data analysis and manuscript revision. SC, TB, BMG, MR, AM, BBT, RMT, MJ, KI, JF, HAR, DG, DP, ES, JF, KFL, and PK: study concept and design, and manuscript revision. SY: Study concept and design, drafting manuscript and full responsibility of data.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data availability statement Data are available on reasonable request.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles