Article Text

Download PDFPDF

Original research
MT-DRAGON score for outcome prediction in acute ischemic stroke treated by mechanical thrombectomy within 8 hours
  1. Wagih Ben Hassen1,
  2. Nicolas Raynaud2,
  3. Nicolas Bricout3,
  4. Gregoire Boulouis4,
  5. Laurence Legrand5,
  6. Marc Ferrigno6,7,
  7. Apolline Kazemi8,
  8. Martin Bretzner3,
  9. Sebastien Soize9,
  10. Wassim Farhat10,
  11. Pierre Seners11,
  12. Guillaume Turc11,
  13. Mathieu Zuber12,
  14. Catherine Oppenheim1,
  15. Charlotte Cordonnier7,
  16. Olivier Naggara4,
  17. Hilde Henon13
  1. 1 Neuroradiology, Centre Hospitalier Sainte Anne, Paris, France
  2. 2 Radiology, Centro-hospitalo Universitaire de Poitiers, Poitiers, France
  3. 3 Department of Interventional Neuroradiology, Centre Hospitalier Regional Universitaire de Lille, Lille, France
  4. 4 Neuroradiology, Hospital Saint Anne, Paris, France
  5. 5 Department of Neuroradiology, Centre Hospitalier Sainte Anne, Paris, France
  6. 6 Inserm U1171-Degenerative and Vascular Cognitive Disorders, Lille, France
  7. 7 Neurology-Stroke Unit, Univ Lille, CHU Lille, Lille, France
  8. 8 Interventional Neuroradiology, Univ Lille, CHU Lille, Lille, France
  9. 9 Radiology, University Hospital Reims, Reims, France
  10. 10 Neurology, Centre Hospitalier Saint Joseph, Paris, France
  11. 11 Department of Neurorlogy, Centre Hospitalier Sainte Anne, Paris, France
  12. 12 Neurology, Groupe Hospitalier Paris Saint Joseph, Paris, Île-de-France, France
  13. 13 Department of Vascular Neurology, Centre Hospitalier Regional Universitaire de Lille, Lille, France
  1. Correspondence to Pr. Olivier Naggara; o.naggara{at}


Objectives The MRI-DRAGON score includes clinical and MRI parameters and demonstrates a high specificity in predicting 3 month outcome in patients with acute ischemic stroke (AIS) treated with intravenous tissue plasminogen activator (IV tPA). The aim of this study was to adapt this score to mechanical thrombectomy (MT) in a large multicenter cohort.

Methods Consecutive cases of AIS treated by MT between January 2015 and December 2017 from three stroke centers were reviewed (n=1077). We derived the MT-DRAGON score by keeping all variables of the MRI-DRAGON score (age, initial National Institutes of Health Stroke Scale score, glucose level, pre-stroke modified Rankin Scale (mRS) score, diffusion weighted imaging-Alberta Stroke Program Early CT score ≤5) and considering the following variables: time to groin puncture instead of onset to IV tPA time and occlusion site. Unfavorable 3 month outcome was defined as a mRS score >2. Score performance was evaluated by c statistics and an external validation was performed.

Results Among 679 included patients (derivation and validation cohorts, n=431 and 248, respectively), an unfavorable outcome was similar between the derivation (51.5%) and validation (58.1%, P=0.7) cohorts, and was significantly associated with all MT-DRAGON parameters in the multivariable analysis. The c statistics for unfavorable outcome prediction was 0.83 (95%CI 0.79 to 0.88) in the derivation and 0.8 (95%CI 0.75 to 0.86) in the validation cohort. All patients (n=55) with an MT-DRAGONscore ≥11 had an unfavorable outcome and 60/63 (95%) patients with an MT-DRAGON score ≤2 points had a favorable outcome.

Conclusion The MT-DRAGON score is a simple tool, combining admission clinical and radiological parameters that can reliably predict 3 month outcome after MT.

  • stroke
  • thrombectomy

Statistics from

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.


Mechanical thrombectomy (MT) has become the recommended treatment for patients with acute ischemic stroke (AIS) due to large cerebral vessel occlusion (LVO)1 Despite rapid and successful recanalization, only 45–55% of AIS-LVO patients treated by MT alone or by bridging therapy (intravenous recombinant tissue plasminogen activator (IV tPA)+MT) have a 3 month favorable outcome.2 Early identification of patients who will not clinically respond to recanalization following MT (futile recanalization) could help select and stratify the best candidates for more advanced therapies (ie, neuroprotection, collateral circulation enhancement), and could help streamline clinical care systems, and anticipate patient orientation and discharge destination.

Several prognostic scores of poor outcome prediction were developed for AIS patients treated by bridging therapy (ie, PRE score, SPAN, THRIVE, etc).3–7 However, the majority of these scores were derived from cohorts treated between 2009 and 2013, before MT became the standard of care with modern stentriever and aspiration devices. In previous thrombolysis studies, the DRAGON score, with its CT8 and MRI versions,9 10 has been shown to be a strong predictor of 3 month clinical outcome. The MRI-DRAGON score is a pragmatic composite tool including clinical (age, National Institutes of Health Stroke Scale (NIHSS) score, glucose level before tPA, pre-stroke handicap, onset to treatment time) and radiological (diffusion weighted imaging-Alberta Stroke Program Early CT score (DWI-ASPECT), M1 occlusion) variables with a high specificity to predict outcome, demonstrated in independent cohorts.

Our aim was to adapt the MRI-DRAGON score to patients with AIS-LVO treated with MT according to current guidelines. We derived the new MRI-DRAGON score (ie, the MT-DRAGON score, from a developmental cohort from one comprehensive stroke center and performed an external cross validation on a population from two different comprehensive stroke centers.


Participants and clinical assessments

Data from three prospective registries of patients from comprehensive stroke centers were collected. All consecutive AIS-LVO patients treated by MT within 8 hours after stroke onset at Lille University Hospital (center 1, from January 2015 to December 2017, n=573 patients), Paris Sainte Anne Hospital (center 2, January 2015 to April 2017, n=318), and Reims University Hospital (center 3, January 2015 to January 2017, n=186) were identified. Inclusion criteria were: age ≥18 years, anterior circulation LVO (tandem occlusion, carotid terminus, M1 segment of the middle cerebral artery (MCA)), pretreatment MRI as firstline imaging of sufficient quality for assessing the DWI-ASPECT score, treatment with MT, with or without IV tPA, and pre-stroke modified Rankin Scale (mRS) score ≤2. Patients with MRI contraindications, absence of pretreatment serum glucose level, or occlusion of the posterior circulation were excluded.

In accordance with the current legislation, this study did not need approval by an ethics committee or written informed consent from patients since it only encompassed analysis of anonymized data that had been collected prospectively as part of routine clinical care.

Clinical assessment and outcomes

Patients were assessed at admission by trained vascular neurologists. Stroke severity was measured using the NIHSS score (range 0–42, with higher scores indicating greater stroke severity). The primary endpoint was favorable outcome, defined by a 3 month mRS score of ≤2. mRS was prospectively assessed by senior neurologists at the 90 day routine follow-up consultation. Clinical and biological variables such as age, gender, NIHSS score, blood pressure, dyslipidemia, current smoking status, diabetes mellitus, pre-stroke mRS score, serum glucose level before treatment, and administration of IV tPA were queried.

Initial brain MRI

In the absence of contraindications, brain MRI is systematically implemented in the three centers as a firstline diagnostic work-up in patients suspected of AIS. All admission MR examinations were performed using standardized protocols (center 1: 1.5 T Philips Achieva; center 2: 3 T Verio, Siemens, Erlangen, Germany or Achieva 2.1, Philips Healthcare; center 3: 1.5 T SignaEchoSpeed, GE Healthcare). The MR protocol included at least DWI (single shot, echo planar, spin echo sequence; three directions, b=1000 s/mm2), and intracranial time of flight MR angiography (MRA). The MRI-DRAGON score, designed in the thrombolysis alone era, included two radiological variables: DWI-ASPECTS score and proximal (M1) MCA occlusion on MR angiography.9 10 Clot location was dichotomized to M1 versus complex occlusion (internal carotid artery (ICA), ICA+M1, tandem occlusions). Location item in the original DRAGON score was adapted with score performance comparison using either M1 or carotid/complex occlusion.

Reading sessions were performed on a dedicated workstation and two neuroradiologists with 5 years' experience in stroke imaging, blinded to clinical outcome, identified occlusion location and independently visually scored DWI-ASPECTS on pretreatment MRI. The ASPECTS score was then dichotomized into >5 and ≤5.2 11

Statistical analysis

Continuous variables with a normal distribution are described as mean±SD and non-normally distributed variables as median (IQR). Variables were compared using the t test, Pearson’s χtest, or Fisher’s exact test, as appropriate.

Score development

The MT-DRAGON score was developed based on patient data from center 1. Logistic regression modeling assuming proportional odds was used to test the interaction of age, sex, glucose level (mmol/L), occlusion site, baseline NIHSS score, IV tPA administration before MT, DWI-ASPECT score, and time to arterial puncture, with the primary outcome as the dependent variable. Multivariate models included all clinical variables of the MRI-DRAGON9 score, with the addition of IV tPA administration, carotid or complex occlusion, and onset to groin puncture (cut-off 4.5 hours) replacing onset to IV tPA time. Other clinical variables were categorized using the previously published cut-offs of the MRI-DRAGON score.9 We assigned 1 point to variables according to the multivariate model regression coefficient: 1 point for regression coefficient values >1 and ≤3, 2 points for variables with a relative risk value >3 and ≤4, 3 points for variables with a relative risk value >4 and ≤5, and 4 points for a relative risk >5. Area under the receiver operating characteristic curve (ie, c statistics) was calculated as a measure of the predictive ability of the models. The model with the highest c statistic was considered the best and was used for internal cross validation, based on 1000 bootstrap replicates. We calculated the c statistic of an alternative MT-DRAGON score, using M1 occlusion rather than complex occlusion.

External validation

External validation of our model was performed using a population from centers 2 and 3. Patients were selected with the same inclusion criteria as those used for score development. Discrimination of the model was quantified with c statistic (from 0.5 to 1, 1 being a perfectly discriminating model). Calibration of the MT-DRAGON score for poor outcome prediction was assessed using the Hosmer–Lemeshow test and a graphical plot.


A total of 1077 consecutive AIS-LVO patients treated by MT were reviewed in the three centers, and 679 patients fulfilled the inclusion criteria (see online supplementary figure 1). Baseline characteristics of the derivation cohort (n=431, 63.5%) and validation cohort (n=248, 37.5%) are presented in table 1. Concordance between readers was high for ASPECTS rating (κ=0.86 (0.78–0.96)). When compared with the derivation cohort, patients in the validation cohort had higher rate of hypertension, dyslipidemia, treatment with IV tPA, lower pre-stroke mRS and DWI-ASPECT scores, and shorter onset to groin puncture delay. At 3 months, the rate of unfavorable outcome was similar in the derivation and validation cohorts (51.5% vs 58.5%, respectively, P=0.7) (table 1). No difference was found in mortality rates between the derivation and validation cohorts (17.4% vs 18.5%, respectively, P=0.32).

Table 1

Baseline characteristics of the derivation and validation cohorts

Derivation cohort

In univariate and multivariate analyses (table 2), unfavorable outcome was significantly associated with all of the MT-DRAGON score clinical variables: age, pre-stroke mRS, NIHSS score on admission, occlusion site, DWI-ASPECTS, serum glucose level at arrival, onset to groin puncture, and absence of prior IV tPA administration.

Table 2

Factors associated with 3 month poor clinical outcome in the derivation cohort

Using the MT-MRI DRAGON score (table 3), the c statistic for predicting poor functional outcome was 0.83 (95% CI 0.79 to 0.88) (figure 1A), higher than the c statistic of the MT-DRAGON score using M1 occlusion rather than complex occlusion (0.75 (95% CI 0.68 to 0.80)). C statistic for mortality prediction showed an area under curve of 0.79 (95% CI 0.74 to 0.85).

Figure 1

Receiving operator characteristic curve for the MT-DRAGON score for 3 month poor outcome prediction in the derivation cohort (A) and the validation cohort (B).

Table 3


Validation cohort

In the validation cohort, the c statistic of the MT-DRAGON score was 0.8 (95% CI 0.75 to 0.86) (figure 1B), not significantly different from the c statistic of the derivation cohort (0.84; 95% CI 0.80 to 0.87; P=0.41) (see online supplementary table 1). The Hosmer–Lemeshow test showed no significant difference between the observed and predicted risks of poor outcome (χ2=9.2, P=0.35). The distribution of 3 month outcomes per increasing points of the MT-DRAGON score in the whole population is shown in figure 2. In our cohort, 95% of patients who had an MT-DRAGON score ≤2 points (n=60/63) had a favorable 3 month outcome whereas 100% of patients who had a score ≥11 points (n=55) had an unfavorable outcome. The c statistic for mortality prediction was 0.76 (95% CI 0.70 to 0.82); 100% of patients with an MT-DRAGON score of ≤6 were alive at 6 months and 60% of patients with an MT-DRAGON score >12 were deceased.

Figure 2

Three month outcome according to the MT-DRAGON score (number) in the whole population. mRS, modified Rankin Scale.


Using the original MRI-DRAGON score, we derived and externally validated the MT-DRAGON score, a simple tool to predict 3 month outcome and mortality of AIS-LVO patients treated with MT within 8 hours of onset, with MRI as firstline imaging.

The performance of this score in predicting clinical outcome was strong and reproducible between two large cohorts. The MT-DRAGON retains the different qualities of the MRI-DRAGON score. Indeed, the evaluation is based on four clinical parameters available at the bedside (age, pre-stroke mRS, NIHSS at admission, and IV tPA administration), biological parameters (serum glucose level), two MRI criteria (DWI-ASPECT and occlusion location), and onset to groin puncture delay. The different items of the score can be obtained directly, without addition of any new step or examination, from routine clinical management of the patient.

Predicting the effectiveness of MT is of clinical importance for several reasons. First, the MT-DRAGON score can provide physicians with an accurate estimation of their patient’s 3 month prognosis, and therefore help families with information and in planning long term living arrangements. Second, it could be used to quickly identify patients with limited benefit from MT who could potentially be the best candidates for potential future trials. One can argue that this score may serve to avoid futile transfer and MT, as the MT-DRAGON score showed a high specificity in extreme values; all patients with a score ≥11/20 had an unfavorable outcome. Models that allow prediction of outcome may be used to define futility of care on the basis of empirical data. As outcome prediction models are developed from populations of patients, they can provide an overall probability of the likelihood of a given outcome, but these point estimates have inherent uncertainty and CIs. Concerns have been raised regarding the possibility of self-fulfilling predictions of poor outcome if early prognostication of a poor outcome in an individual patient leads to care withdrawal, and this is particularly concerning if the specificity of the tool used is not perfect or near perfect. We believe that the specific ability of a scale such as the MT-DRAGON score to allow prediction of individual patient outcome is less conspicuous than the increasing probability of unfavorable outcome as the score increases. This can be very useful in stratifying patients’ level of care, decision making for acute interventional procedures, population stratification for new therapeutic strategies, and communication with families, rather than for decisions on acute care withdrawal.

All parameters of the MT-DRAGON score were significantly and independently associated with 3 month outcome in our study. The MT-DRAGON score incorporates the DWI-ASPECTS, which has been shown to be an independent 3 month outcome predictor, and to be assessible with good intra- and inter-rater agreement.12 13 We chose a score threshold of ≤5 DWI-ASPECTS, used in the original MR-DRAGON score, because this cut-off value was associated with the highest reproducibility values14 and a strong predictive value of malignant infarction evolution in anterior occlusion.15 The choice of the DWI-ASPECTS cut-off for endovascular procedure selection is largely debated, with several studies showing a benefit of MT in patients with large ischemic volume or low ASPECTS.16 17 Our results may have limited extrapolation in the population of low ASPECTS (0–2) as this subset was underrepresented in our population (43/679 patients (6.3%)). Ongoing randomized trials are studying the benefits and safety of MT in large ischemic lesions.18

Several blood biomarkers representing different pathophysiologic pathways, such as copeptin plasma levels,19 hyperglycemia,20 or elevated mean platelet volume,21 measured on admission to the emergency room, have been demonstrated to be associated with clinical outcome, mainly in cohorts treated in the pre-MT era. There is a high potential of improving specificity of predictive scores such as MT-DRAGON using blood biomarkers, and further studies are warranted.

Rather than proximal M1 MCA occlusion used in the original MRI-DRAGON score designed on intravenous thrombolysis cohorts, we choose to use complex carotid occlusions, hence improving the c statistic of the MT-DRAGON score, in line with the HERMES meta-analysis results showing better 3 month outcome (mRS ≤2) in M1 (51.8%) versus ICA (33%) occlusions.11

The incorporation of onset to groin puncture delay into the score may be questionable as it is not directly assessible at baseline, even more so in the case of interhospital transfer. There are many factors that contribute to transportation time besides distance. Ambulance response time, availability or scene time, traffic, and weather, among other things, will contribute to transport time. When considering all of these factors, the relationship between time and distance may not be linear. However, despite variations, onset to groin time can be estimated based on the daily experience of transfer between the primary center and its comprehensive stroke center. In addition, we used a dichotomized value of onset to groin time—that is, 4.5 hours—hence facilitating the determination of the score.

Recent results from the HERMES group22 showed that patients with a glucose level ranging between 5 and 5.5 mmol/L at stroke onset had the largest benefit from endovascular treatment. In the present study, the glycemia cut-off level was left at 8 mmol/L, as in the original score, complying with the lowest level of current guidelines pending larger validation of the HERMES group findings with proper clinical testing in a therapeutic trial.

The choice of keeping only easily accessible criteria explains the non-accounting of leptomeningeal collaterality assessment in the MT-DRAGON score. As recently described, the standard classification (American Society of Interventional and Therapeutic Neuroradiology) based on DSA is not reliable,23 and cannot be obtained at the early stage of management, before groin puncture. Current non-invasive methods of assessing collaterality have not yet been recommended by recent guidelines for early management of patients with AIS-LVO.1

One of the strengths of the MT-DRAGON score is that it was developed in a ‘real world’ setting. Hence if baseline NIHSS scores were similar to those described in patients included in randomized controlled trials, elderly population (≥85 years) represented 17.2% of the population (123 patients). Several other scores can be used to estimate long term outcome in AIS-LVO patients treated with MT, but very few were specifically designed for this purpose and recent recommendations1 have made them obsolete or inadequately suited in modern cohorts. Recently, a clinical decision tool for the selection of AIS-LVO patients for MT was developed based on the Multicenter Randomized Clinical trial of Endovascular treatment for Acute ischemic stroke in the Netherlands (MR CLEAN)  cohort and externally validated based on the Interventional Management of Stroke-III (IMS-III) trial cohort. The choice of IMS-III as a validation cohort raises concerns regarding its applicability, as the population differs from modern MT cohorts and used devices that did not achieve modern recanalization rates.24

Our study had several limitations. First, we performed minor modifications of the MRI-DRAGON score to be applicable in patients treated with MT. Hence our approach could be considered as an adaptation rather than the development of a new score, and we cannot rule out the impact of unmeasured confounds. Second, the MT-DRAGON score has been designed for the anterior circulation and cannot be applied to the posterior circulation, with marked differences in natural history and functional outcome, even in the era of endovascular therapies. Third, the present score is not applicable to patients with CT-CT angiography as the firstline imaging modality. An adaptation using the CT-ASPECT score is strongly warranted. Fourth, the MT-DRAGON score included patients treated within 8 hours of symptoms onset, and needs to be validated in a large population of late window treated patients (8–24 hours). Fifth, in common with the original MRI-DRAGON score, the MT-DRAGON score was of little help in predicting outcome in patients with a mid-range score (ie, 6–11). Refinements of the score could be obtained using more detailed clinical (ie, early neurological improvement) and imaging data (ie, DWI/perfusion weighted imaging mismatch, clot burden). However, our aim was to develop a simple tool including only variables easily available at admission.

In conclusion, the MT-DRAGON score is a new pragmatic tool to predict 3 month outcome in AIS-LVO patients with admission brain MRI and candidates for MT. It can provide physicians with an estimation of their patient’s prognosis and be used to quickly identify poor responders to MT, who may be selected for clinical trials testing innovative treatment modalities.



  • ON and HH contributed equally.

  • WBH and NR contributed equally.

  • Contributors All authors drafted and approved the final manuscript.

  • 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.

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

  • Patient consent for publication Not required.