Article Text

Download PDFPDF

Impact of infarct location on functional outcome following endovascular therapy for stroke
  1. Charlotte Rosso1,2,
  2. Raphael Blanc3,
  3. Julien Ly1,2,
  4. Yves Samson1,2,
  5. Stéphane Lehéricy1,4,
  6. Benjamin Gory5,
  7. Gautier Marnat6,
  8. Mikael Mazighi3,
  9. Arturo Consoli7,
  10. Julien Labreuche8,
  11. Suzana Saleme9,
  12. Vincent Costalat10,
  13. Serge Bracard5,
  14. Hubert Desal11,
  15. Michel Piotin3,
  16. Bertrand Lapergue7
  17. on behalf of the ASTER Trial and Pitié-Salpêtrière Investigators
    1. 1 Inserm U 1127, CNRS UMR7225, Sorbonne Université, UPMC Univ Paris 06 UMR S 1127, Institut du Cerveau et de la Moelle épinière, ICM, Paris, France
    2. 2 APHP-Urgences Cérébro-Vasculaires, Hôpital Pitié-Salpêtrière, Paris, France
    3. 3 Department of Diagnostic and Interventional Neuroradiology, Rothschild Foundation, Paris, France
    4. 4 APHP-Neuroradiology Department, Hôpital Pitié-Salpêtrière, Paris, France
    5. 5 Department of Diagnostic and Interventional Neuroradiology, University Hospital of Nancy, Nancy, France
    6. 6 Department of Diagnostic and Interventional Neuroradiology, University Hospital of Bordeaux, Bordeaux, France
    7. 7 Department of Stroke Center and Diagnostic and Interventional Neuroradiology, University of Versailles and Saint Quentin en Yvelines, Foch Hospital, Suresnes, France
    8. 8 Department of Biostatistics, University Lille, CHU Lille, Lille, France
    9. 9 Department of Diagnostic and Interventional Neuroradiology, University Hospital of Limoges, Limoges, France
    10. 10 Department of Diagnostic and Interventional Neuroradiology, Hôpital Gui de Chauliac, Montpellier, France
    11. 11 Department of Diagnostic and Interventional Neuroradiology, Guillaume et René Laennec University Hospital, Nantes, France
    1. Correspondence to Charlotte Rosso, Institut du Cerveau et de la Moelle épinière, Paris 75013, France; charlotte.rosso{at}


    Objectives The relationship between stroke topography (ie, the regions damaged by the infarct) and functional outcome can aid clinicians in their decision-making at the acute and later stages. However, the side (left or right) of the stroke may also influence the identification of clinically relevant regions. We sought to determine which brain regions are associated with good functional outcome at 3 months in patients with left-sided and right-sided stroke treated by endovascular treatment using the diffusion-weighted imaging-Alberta Stroke Program Early CT Score (DWI-ASPECTS).

    Methods Patients with ischaemic stroke (n = 405) were included from the ASTER trial and Pitié-Salpêtrière registry. Blinded readers rated ASPECTS on day 1 DWI. Stepwise logistic regression analyses were performed to identify the regions related to 3-month outcome in left (n = 190) and right (n = 215) sided strokes with the modified Rankin scale (0–2) as a binary independent variable and with the 10 regions-of-interest of the DWI-ASPECTS as independent variables.

    Results Median National Institute of Health Stroke Scale (NIHSS) at baseline was 17 (IQR: 12–20), median age was 70 years (IQR: 58–80) and median day-one NIHSS 9 (IQR: 4–18). Not all brain regions have the same weight in predicting good outcome at 3 months; moreover, these regions depend on the affected hemisphere. In left-sided strokes, the multivariate analysis revealed that preservation of the caudate nucleus, the internal capsule and the cortical M5 region were independent predictors of good outcome. In right-sided strokes, the cortical M3 and M6 regions were found to be clinically relevant.

    Conclusion Cortical non-motors areas related to outcome differed between left-sided and right-sided strokes. This difference might reflect the specialisation of the dominant and non-dominant hemispheres for language and attention, respectively. These results may influence decision-making at the acute and later stages.

    Trial registration number NCT02523261.

    • acute treatment
    • magnetic resonance imaging
    • prognosis

    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.


    The Alberta Stroke Program Early CT Score (ASPECTS) is a simple scoring system to estimate the extent of acute stroke lesions on diffusion-weighted imaging (DWI).1 2 The total score has been shown to help predict treatment response to thrombectomy as well as outcome.3 The extent-based or volume-based approach guides our decisions as clinicians. When there is a proximal occlusion of the middle cerebral artery (MCA), the more regions spared by the infarct there are, the more potential penumbral tissue has to be salvaged by recanalisation. However, topography (ie, location of infarcted tissue) rather than volume should be taken into account when considering functional outcome.4 5 Some areas are more eloquent and thus more clinically relevant to functional outcome than others due to their functional anatomy. The pathophysiological status of these specific areas (eg, infarcted, penumbral or normal) may be important for therapeutic decisions and prognosis.4 For instance, if motor areas are inside the penumbral tissue but not yet infarcted, outcomes may be optimised by treating such patients, even if the National Institute of Health Stroke Scale (NIHSS) is low or the infarct volume small.

    Another issue is that eloquent structures may differ depending on the side of the infarcted hemisphere.5 6 Obviously, functional outcome assessed by the modified Rankin Scale (mRS) depends on motor components and captures disability for cognitive deficits such as aphasia for the left and neglect for the right hemisphere.7 A group from Atlanta demonstrated, using DWI-ASPECTS, that only right parieto-occipital (M6) and left superior frontal (M4) involvement independently predicted poor outcome (defined as a mRS from 3 to 6).6 To identify these regions, the template of the DWI-ASPECTS regions has proven useful, as it is increasingly used in clinical practice and shows good reliability.8 Furthermore, the template is based on anatomical structures and thus the individual regions covering different parts of the deep and cortical territories.

    The aim of this study was to determine which regions captured by DWI-ASPECTS at day 1 was related to 3-month good outcome and whether there was a difference in these critical regions according to the side of the stroke lesion.

    To this purpose, we merged two large acute stroke datasets (ASTER trial9 and Pitié-Salpêtrière registry10) of patients treated by endovascular therapy (EVT) for a large vessel occlusion in the anterior circulation and who had MRI at day 1, when the final extent of the infarct was reached. Our hypothesis was that critical regions associated with 3-month good outcome would differ between the left and right hemisphere, especially at the cortical level where the specificity of the dominant and non-dominant hemispheres are more marked.

    Material and methods


    The dataset included patients both from the ASTER trial and the Pitié-Salpêtrière registry treated by EVT for an anterior circulation ischaemic stroke in the first 6 hours of stroke onset. The ASTER trial was a randomised controlled trial, which showed no statistical difference between aspiration or stent retriever as a frontline thrombectomy approach.9 The Pitié-Salpêtrière registry is a registry of patients treated with EVT, intravenous thrombolysis or both.10 For this study, we examined the same study period (2015–2017) as the ASTER trial. In both cohorts, patients underwent EVT within a 6 hour time window with or without intravenous thrombolysis (in a 4.5-hour time window from symptom onset) if they had imaging evidence of an occlusion of the intracranial internal carotid artery or the origin (M1) or branches (M2) of the MCA (online supplementary methods).

    Supplemental material

    All patients were examined with the NIHSS at baseline and day 1. The 3-month mRS was rated at follow-up. The mRS is a commonly used scale for measuring the degree of disability or dependence in the daily activities of people who have suffered a stroke.11

    To conduct this study, we included patients from both databases who had MRI performed 1 day after the stroke onset, when the final extent of the infarct was reached. In total, 319 patients from the ASTER trial and 86 patients from the Pitié-Salpêtrière registry were merged into a single dataset (n=405). The baseline characteristics of the eight centres from the ASTER trial were compared with those of the Pitié-Salpêtrière registry. No difference between the baseline characteristics was detected (baseline (pretreatment) NIHSS, p: 0.18; day 1 NIHSS, p:0.56; baseline DWI-ASPECTS, p: 0.30; day 1 DWI-ASPECTS, p:0.05) except for age (p: 0.04). (online supplementary file 1).

    ASTER was a randomised controlled trial led with ethics committee approval and written informed consent for each patient. In accordance with the French legislation, the Pitié-Salpêtrière registry did not need approval by an ethics committee or written informed consent from patients, as it is a retrospective database implying only analysis of anonymised data collected prospectively as part of routine clinical care.

    DWI-ASPECTS scoring and frequencies of the affected DWI-ASPECTS regions

    DWI-ASPECT scores were rated on the MRI performed at day 1 after stroke onset by investigators blinded to the clinical data for each database. The total score as well as damage to each region-of-interest (ROI) was recorded. The ASPECTS template (figure 1) is described at two levels: at the level of the basal ganglia for the following regions (M1 to M3, caudate-C, lenticular nucleus-L, internal capsule-IC, insula-I) and those superior to this level for M4 to M6.1–3 Three of these regions belong to the deep territory (C, IC and L) and seven to the cortical superficial territory (I, M1 to M6). One point is deducted for partial or total involvement by acute infarction in each of the 10 regions. A DWI-ASPECTS rating of 10 represents no visible infarction, and a score of 0 represents extensive infarction throughout the MCA territory.

    Figure 1

    ASPECTS Template. Panel A is the DWI slice at the level of the basal ganglia, and B at the supraganglionic level. C indicates caudate nucleus; I, insular ribbon; IC, internal capsule; L, lenticular nucleus; M1, anterior MCA cortex; M2, MCA cortex lateral to the insular ribbon; M3, posterior MCA cortex and M4, M5, M6, anterior, lateral and posterior MCA territory immediately superior to M1, M2 and M3. This patient has a total ASPECT score of 5. Image is presented in the neurological convention (left is left). ASPECTS, Alberta Stroke Program Early CT Score; DWI, diffusion-weighted imaging; MCA, middle cerebral artery.

    As a first step, we determined the frequency of infarct damage in each region and whether certain regions were likely to be damaged together. As infarcts usually involve several regions at the same time and as the pattern of lesion topography depends on the territory vascularised by each artery, it seemed obvious that some regions would be more likely infarcted together and others less. We also computed correlation matrices between these regions to better identify which ones covaried with each other. Multiple comparisons were handled with Bonferroni corrections.

    Impact of hemisphere on the association between infarct topology and 3-month outcome

    First, all left and right DWI-ASPECTS ROIs were evaluated using univariate analyses in order to determine whether they were clinically relevant to good outcome at 3 months poststroke. ORs and 95% CIs were calculated. Then, we ran separate stepwise logistic regressions for left-sided and right-sided strokes with the 3 month mRS binarised for a good outcome as the dependent variable. The independent variables were the DWI-ASPECTS ROIs found to be significant in the univariate analysis. A potential problem for this type of analysis is multicollinearity of the independent variables, which might affect the precision and choice of the variables. We therefore verified the absence/presence of multicollinearity with two factors. First, the variance inflation factor (VIF) was computed using JASP (V.0.8.6, Amsterdam, The Netherlands). By convention, multicollinearity is considered present if the VIF of one variable was at least 10.12 Second, we also calculated the Condition Index (CIx) for the 10 ROIs. The CIx is a measure derived from principal component analysis. A variable is considered collinear if the CIx was >30.13

    Statistical analysis

    Descriptive statistics are given as the median and IQR. Non-parametric tests were used for the comparison of continuous data. Comparisons of proportions were done with χ² tests. Correlations between ROI damage were analysed using the Spearman correlation coefficient. 95% CI are given with the correlations. Statistical analyses were performed with MedCalc (V.12.5.0, Belgium, 2013) for the univariate and stepwise logistic regression analyses and JASP, an open source R based program for the multicollinearity and principal component analyses.



    Four hundred and five patients (median age: 70 years, IQR: 58–80, 55% of males) with available DWI-ASPECTS at day 1 and 3-month mRS were included in the study. There were 190 left-sided strokes (47%) and 215 right-sided stroke (53%). Baseline clinical characteristics as well as DWI-ASPECTS ROI involvement are summarised in table 1. NIHSS scores at baseline and at day 1 were statistically different between patients with left and right stroke but not the 3-month mRS (whether considered as a dichotomised or continuous variable). The difference in the NIHSS could be expected as it measures the severity of strokes in the right and left hemisphere differently since 7 points are dedicated to aphasia and only 2 for neglect.14 15

    Table 1

    Demographics and radiological features of patients with left-sided and right-sided stroke

    Characterisation of infarct topography according to the DWI-ASPECTS

    Stroke lesions involved the cortical and deep territory simultaneously in 251 patients (62%) were purely cortical in 110 patients (27%) and purely deep in 44 (11%).

    Table 2 indicates the frequency of multiple regions of damage for cortical and deep structures. A total of 355 patients (84 %) had at least one cortical ROI infarcted, with an equal distribution between each number of cortical ROIs involved (1–7). A total of 295 patients (737%) had damage in at least one deep ROI and most of them in two (43%, n=128).

    Table 2

    Proportion of patients with damage for each ROI

    The comparison between left-sided and right-sided infarcts revealed an asymmetry for the infarcted territories. The deep territory was less frequently involved in left-sided vs right-sided strokes (p: 0.02), driven by damage to the lenticular nucleus (table 1, p: 0.02). The cortical M4 region was more frequently infarcted in left-sided than right-sided strokes (table 1, p: 0.03). The other comparisons did not show significant differences.

    In order to determine which regions were more frequently affected, we computed the proportion of patients for whom a given ROI was infarcted alone or combined with others (figure 2). The insula was the most frequent ROI damaged (72%), followed by M2 (63%), which corresponds to the same blood supply. For the deep ROIs, the lenticular nucleus was more often infarcted (69%) than the other deep regions (p<0.0001). Individual regions were rarely infarcted in isolation but in combination with other regions (figure 1). To identify which regions were more likely damaged together, we calculated the correlation matrix presented in figure 3. We also plotted in pie charts the distribution of combination of regions (only two by two as the possibilities are too numerous) for each region (see online supplementary figure 1). As an example, M1 damage (n=178) was correlated to damage in all others ROIs of the deep and cortical territory. This is consistent with the fact that M1 was never infarcted alone (ie, in isolation), but always with other ROIs such as the insula (95% of damage in M1 occurred simultaneously in the insula) and M2 (90% of the damage co-occurred in the M1 and M2 regions).

    Figure 2

    Distribution of damage in the cortical and deep territories of the DWI-ASPECTS ROIs. *P<0.005. Blue represents the proportion of patients for whom the ROI is damaged with another and red for a ROI damaged in isolation. ASPECTS, Alberta Stroke Program Early CT Score; DWI, diffusion-weighted imaging; ROI, region-of-interest.

    Figure 3

    Correlation matrix between damage to DWI-ASPECTS ROIs in the entire cohort Correlation coefficients are given with the 95% CI. Colour code is based on the strength of the correlations. **P<0.0001, *p<0.001, § for p>0.005 (does not survive multiple comparisons), NS=not significant. ASPECTS, Alberta Stroke Program Early CT Score; DWI, diffusion-weighted imaging; ROI, region-of-interest.

    Identification of regions related to 3-month outcome in left-sided and right-sided strokes

    As a first step, we performed univariate logistic regression analyses to determine the weight of each ROI separately for 3-month good outcome (table 3). For left-sided strokes, all DWI-ASPECTS regions were significant, with the IC having the highest OR for explaining good outcome. For right-sided strokes, the caudate and lenticular nuclei were not significant, and the highest OR to explain good outcome was attributed to the right M6.

    Table 3

    Univariate logistic regression: OR of each ROI and its 95% CI for 3-month good outcome in left-sided and right-sided strokes

    As none of the regions were detected as collinear (online supplementary material) material, ‘multicollinearity analysis’, and online supplementary figure 2), we entered all of the 10 DWI-ASPECTS regions for the left-sided strokes. For left-sided strokes, the stepwise logistic regression (p<0.0001) retained the caudate nucleus (C) (OR: 2.07, 95% CI 1.06 to 4.06), the IC (OR: 4.55, 95% CI 1.82 to 11.34) and M5 (OR: 3.39, 95% CI 1.77 to 6.48). The model yielded 66% of patients properly classified. As for right-sided strokes, we did not include the caudate and the lenticular nuclei as these regions were not significant in the univariate analysis. The model for right-sided strokes retained M3 (OR: 2.89, 95% CI 1.39 to 5.98) and M6 (OR: 2.76, 95% CI 1.28 to 5.98). The model yielded 67% of patients properly classified. Adding rtPA administration as an independent variable did not change the results or the accuracy of the models. For left-sided strokes, the logistic regression retained the same ROIs in addition to rtPA administration (OR: 2.21, 95% CI 1.11 to 4.40, p: 0.02) with 67% of patients properly classified. For right-sided strokes, the model remained unchanged.


    In this study, we showed that not all DWI-ASPECTS regions have the same weight in predicting 3-month good outcome after EVT. Moreover, these regions differed with regard to the side of the infarct. The left caudate, IC and M5 region, when spared, were independently associated with good outcome. The right M3 and M6 were independently associated with good outcome. The other DWI-ASPECTS regions were not retained, implying that they were not independently predictive of good functional outcome.

    Identification of clinically relevant regions for functional outcome in left-sided and right-sided strokes

    In left-sided strokes, three independent regions were associated with good outcome when spared: two regions in the deep MCA territory, the caudate nucleus and the IC, and one cortical region, the M5 frontoparietal region. Proximity of the deep MCA regions and their blood supplies anatomy explained why both regions were often damaged together (43% in our cohort). The most frequent associations were either the caudate and lenticular nuclei or the lenticular nucleus and IC together.5 6 For the caudate nucleus, its role in daily living activities and independency is somewhat surprising. Two explanations can be proposed. The first concerns its structural connectivity and its functional role. The caudate is an associative structure, which has been involved in motor, behaviour and dysexecutive disorders due to its important connections with the frontal and the parietal lobes.16–18 Second, the caudate’s association with outcome could also be the reflection of an efficient deep artery collaterality. If the caudate nucleus is spared by an efficient collaterality network, this collaterality may also decrease the depth and extent of ischaemia and therefore be a factor of good outcome in and of itself. On the contrary, the association between damage of the IC and functional outcome was expected. Previous studies have found that corticospinal tract regions (corona radiata, posterior limb of the IC) are crucial for motor outcome19 20 and consequently for daily living activities (and thus, the mRS). Preservation of the left M5 was also associated with good outcome. The left M5 is a frontoparietal region that contains important tracts for language (such as the arcuate fasciculus, which travels nearby) or for higher cognitive functions (such as the frontoparietal modules connected by U fibres or some segments of the superior longitudinal fasciculus) (figure 4).21–23 Injury to this cortical region and the underlying white matter may therefore reduce possibilities of language recovery and contribute to more severe disability. In contrast to our results, Rangaraju et al 6 found that left M4 and not M5 was predictive of poor outcome when infarcted. Some differences between the two studies may explain this discrepancy, including the selection criteria, the distribution of infarction, the endpoint (poor outcome), the timing of the MRI scan that was performed in a wide range of hours (12–72 hours) as in Rangaraju et al,6 but more probably, the imprecision in the localisation of lesions between raters for M4 and the adjacent M5.6 It could be surprising that the left M1 or M3 is not retained in the logistic regression model as they are also important areas for language processing. We strongly believe that our findings and those of Rangaraju’s are related to the underlying damage of the language dorsal white matter pathway.6

    Figure 4

    Anatomical relationships between the M3 and M5 regions with white matter fasciculi on axial (A,C) and coronal slices (B,D) on DWI. Top row: The M5 region (red) overlapping with the arcuate fasciculus in yellow. Bottom row: The M3 region (red) overlapping with the IFOF (blue) and ILF (green). DWI, diffusion-weighted imaging; IFOF, inferior fronto-occipital fasciculus; ILF, inferior longitudinal fasciculus.

    In right-sided strokes, good outcome was observed when right M3 and M6 (both are located in the right temporoparietal junction-TPJ) were spared from infarction. These results are in line with a Voxel Lesion Symptom Mapping study which reported asymmetrical patterns between hemispheres, especially involving the right inferior parietal lobe and superior temporal gyrus.5 Indeed, damage to the right TPJ in the non-dominant hemispheres may result in neglect,24 which in turn disturbs certain activities of daily living measured by the mRS. The ventral pathway including the inferior fronto-occipital fasciculus (IFOF) is located just beneath M3 as a direct connection from the occipital cortex to the prefrontal lobe25 as well as the inferior longitudinal fasciculus which connects the occipital cortex to the temporal lobe and runs laterally to the IFOF at the level of the temporo-occipital junction (figure 4).26 The role of these fasciculi in the right hemisphere is still controversial but has been shown to be involved in non-verbal semantic processes or non-verbal consolidation learning.26 To summarise, the right TPJ and its structural connectivity are crucial for attention27—as pointed out by a recent meta-analysis—spatial orientation and exploration,28 and its injury can lead to greater disability. It could be surprising that the IC was found predictive in the left but not the right hemisphere. These results were also found in another cohort of 490 patients imaged at 48 hours poststroke.29 There was an association between injury to left motor areas and poor outcome that was not found in the right hemisphere. This could be because the major determinant of good outcome is the absence of neglect and that large strokes often injure areas responsible for motor function and neglect in a combined way. Indeed, other ROIs were not retained in the model, suggesting that they were either not associated with outcome at all or had no independent contribution to outcome, once other regions were accounted for. For example, a right basal ganglia infarct does not lower 3-month outcome.

    Characterisation of infarct topography according to the DWI-ASPECTS

    We described the frequencies of damage to the DWI-ASPECTS regions in a large cohort of patients (n=405). The insula, the lenticular nucleus and M2 were the three most frequently damaged regions in our cohort. Although most affected, these regions were not the most important in explaining functional outcome at 3 months, in agreement with a previous study.6 This discrepancy may be explained by two factors: (1) these regions were rarely infarcted in isolation and (2) the weight of these regions on the functional outcome was lower when other regions were taken into account.

    We highlighted a strong relationship between the damage in some DWI-ASPECTS regions. Correlation coefficients above 0.5 (considered as strong correlations in Rangaraju et al 6) were observed between regions supplied by the same artery such as the insula and M2 (the rolandic operculum nearby the insula), M1 and M4 (the same area as M1 but at the supraganglionic slice) and M3 and M6 (two close regions at two different slices). Fortunately, we did not identify multiple collinearity with two independent measures in our cohort, which allowed us to enter all the variables in the logistic regression. In our study as well as in Phan et al in 200630 and Rangaraju et al,6 the DWI-ASPECTS ROIs were not collinear, yet collinearity was found in one publication in 2013.31

    Third, we found an asymmetry in the frequency of injury between left-sided and right-sided strokes in the deep (lenticular nucleus) and the superficial territory (M2). However, neither of the two regions was identified as a clinically relevant area.

    Our study has several limitations. First, DWI-ASPECTS presents limitations for precisely evaluating infarct volume as there is no clear cut-off defined to count a region as affected or not. Second, its binary nature (0 if damaged, 1 if spared) does not take into account the proportion of the ROI actually damaged by the infarct. For example, an isolated injury of the cortical ribbon will be scored the same as a territorial damage involving 100% of the ROI. Nevertheless, this is one of the most used score nowadays in clinical routine.

    Second, these results originated from cohorts where clinical assessments were performed using simple measures such as NIHSS at the acute and mRS at a mid-term stage. We cannot exclude that other regions may play a role in modalities not explored by the mRS. In addition, the analysis was carried out on ischaemic strokes of the anterior circulation (with exclusion of the posterior circulation and the subsequent areas-brainstem, cerebellum and occipital lobes). Third, one could argue that the multisite recruitment could be a bias. However, we rather view this as an advantage since the cohort was not biased by monocentric practices. The results should therefore be more generalisable to other centres. Finally, a complementary analysis would be to study DWI reversal for the same regions (ie, the reversal of the DWI abnormalities between pretreatment and follow-up MRIs), in order to identify whether the reversal in the left C, IC and M5 or the right M3 and M6 and not in other regions is associated with a dramatic early improvement.32 This association would definitely prove that these specific regions are involved in processes that hamper independency. However, even if the sample size of our cohort was large, DWI reversal occurred only in less than 3.1% for the deep territory and 5.5% for the cortical superficial territory (online supplementary table 2). Unfortunately, these percentages lead to insufficient statistical power to analyse the influence of laterality on DWI reversal of our clinically relevant regions.


    Not all brain regions have the same weight in predicting good functional outcome at 3 months in patients with ischaemic stroke; moreover, these regions depend on the affected hemisphere. The left IC and M5 as well as the right M3/M6 were independently associated, when spared, with 3month good outcome. These results may affect decision-making in acute and later stages. Moreover, this study highlights the important role of these regions, especially the cortical regions, in brain function and processes leading to disability.


    We thank Eric Moulton for proofreading the English of our manuscript.



    • Collaborators ASTER Trial Investigators: Rothschild Foundation: Michel Piotin, Raphael Blanc, Hocine Redjem, Gabriele Ciccio, Stanislas Smajda, Mikael Mazighi, Robert Fahed, Jean-Philippe Desilles; Foch Hospital: Bertrand Lapergue, Georges Rodesch, Arturo Consoli, Oguzhan Coskun, Federico Di Maria, Frédéric Bourdain, Jean Pierre Decroix, Adrien Wang, Maya Tchikviladze, Serge Evrard; Hospices Civils de Lyon: Francis Turjman, Benjamin Gory, Paul Emile Labeyrie, Roberto Riva; Limoges University Hospital: Charbel Mounayer, Suzanna Saleme; Montpellier University Hospital Center: Vincent Costalat, Alain Bonafé, Omer Eker, Grégory Gascou, Cyril Dargazanli; Nancy University Hospital Center: Serge Bracard, Romain Tonnelet,Anne Laure Derelle, René Anxionnat; Nantes University Hospital Center: Hubert Desal, Romain Bourcier, B Daumas-Duport; Bordeaux University Hospital Center: Jérome Berge, Xavier Barreau, Gauthier Margnat; Department of Biostatistics,University of Lille: Lynda Djemmane, Julien Labreuche, Alain Duhamel. Pitié-Salpêtrière registry: Yves Samson, Sophie Crozier, Sandrine Deltour, Anne Leger, Flore Baronnet, Charlotte Rosso, Fredéric Clarencon, Nader Sourour, Ehmad Shotar, Christine Pires.

    • Contribution CR: study design data analysis drafting of manuscript. BL and MP: study design, data collection, manuscript revision. RB, JL, YS, SL, BG, GM, MM, AC, SS, VC, SB and HD: data collection, manuscript revision. JL: data analysis, manuscript revision.

    • Funding The Pitié-Salpêtrière registry was supported by the French Ministry of Health grant EVALUSINV PHRC AOM 03 008. The research leading to these results has received funding from ‘Investissements d’avenir’ ANR-10-IAIHU-06. The ASTER trial was sponsored by the Fondation Ophtalmologique Adolphe de Rothschild. An unrestricted research grant was provided for the ASTER trial by Penumbra, Alameda, California. No grant was provided for this analysis and this study.

    • Competing interests None declared.

    • Patient consent Not required.