Hierarchical composite outcomes in acute ischaemic stroke with large infarct: a win ratio analysis of the TENSION trial — survival drives the signal

Clinical Question (PICO)

In adults with anterior-circulation large-vessel occlusion, an established large infarct defined by ASPECTS 3–5, and treatment within 12 hours of last known well, how does endovascular thrombectomy plus best medical treatment compare with medical treatment alone when 12-month outcomes are evaluated in a hierarchy: survival time, disability, serious adverse events, then quality of life?

This is a 2026 methodological secondary analysis of the completed TENSION randomized trial, not a new trial or a new treatment indication. Predefined strata in this analysis were sex, age, prestroke disability, consciousness, ASPECTS, occlusion site, and intravenous thrombolysis. The hierarchy was specified for this secondary analysis after the parent trial; “predefined” here does not establish prospective prespecification before the original trial results were known.

Bottom Line

Thrombectomy plus medical treatment had a win ratio of 1.61 (95% CI 1.19–2.18; P=.002) and a net win difference of 22.8 percentage points (95% CI 8.3–36.3; P=.002) under the survival-first hierarchy. Survival contributed 15.7 percentage points of that difference. The authors support hierarchical composite outcomes alongside, not instead of, conventional mRS analyses; this statistic is not an individual patient's probability of benefit or a relative reduction in mortality.

Design

Trial type: Secondary intention-to-treat analysis of an investigator-initiated, multicenter, randomized, open-label, blinded-endpoint thrombectomy trial.
N: 253 randomized patients, all included in the win-ratio analysis.
- Thrombectomy plus best medical treatment: 125.
- Best medical treatment alone: 128.
Randomization: Parent trial used central web-based 1:1 allocation. The published protocol specified stratification by time from onset/last known well (0–6 versus >6 hours) and NIHSS (≤18 versus >18).
Setting: 40 hospitals in Europe and one in Canada, as reported in the parent publications.
Enrollment: July 17, 2018–February 21, 2023. The parent trial stopped early for efficacy at its first planned interim analysis.
Follow-up: Outcomes targeted 12 months. A mean is not supplied in this reanalysis; the parent 12-month report reported median follow-up of 8.36 months (IQR 0.02–12.00).
Analysis: Every intervention patient was compared with every control patient, producing 16,000 possible cross-arm pairs. Each pair resolved at the first different outcome in the hierarchy. Confidence intervals used 10,000 nonparametric bootstrap iterations; no extra baseline adjustment or matching was performed.
Primary analysis here: Survival time → mRS at 12 months → any serious adverse event → EQ-5D. The parent trial's primary endpoint was the full mRS distribution at 90 days, not this composite.
Sensitivity analyses: mRS before survival, then adverse events and EQ-5D; and mRS, survival, EQ-5D, then adverse events.

Population

Inclusion Criteria

Adults with anterior-circulation large-vessel occlusion and ASPECTS 3–5 on CT or diffusion-weighted MRI.
Thrombectomy feasible within 12 hours of onset/last known well.
The trial registry further specifies randomization within 11 hours, NIHSS <26, prestroke mRS ≤2, and an accessible M1 and/or intracranial distal ICA occlusion.
Consent by the patient or an authorized representative, with locally approved emergency-consent provisions.

Exclusion Criteria

The registry and published protocol describe exclusions including neurological or psychiatric conditions confounding outcome assessment, vascular access disease preventing treatment, contrast contraindication, infective endocarditis, mass effect or intracranial tumor, and infarct extent outside the selected ASPECTS range. These are selected major exclusions, not a substitute for the complete trial protocol or a bedside eligibility checklist.

The original protocol had an upper age limit of 80, but the analyzed cohort clearly included older participants. Apply the reported enrolled population, not that obsolete protocol age cap, when interpreting this analysis.

Baseline Characteristics (thrombectomy versus medical treatment)

Women: 56/125 (45%) versus 67/128 (52%).
Age ≤70: 41% versus 41%; age 71–80: 34% versus 34%; age >80: 25% versus 24%. A median age is not supplied in this analysis's baseline table.
Admission NIHSS, median (IQR): 19 (16–22) versus 18 (15–22).
Prestroke mRS, median (IQR): 0 (0–1) in both groups.
Fully awake on arrival: 61% versus 62%.
ASPECTS 3/4/5: 29%/36%/35% versus 38%/30%/32%.
Intravenous alteplase: 39% versus 34%.

Interventions

Thrombectomy plus medical treatment: Endovascular thrombectomy using approved devices within the trial window, plus best medical care. This is a procedural strategy, not a drug-dose comparison.
Medical treatment alone: Standard medical care according to local practice. Intravenous thrombolysis was permitted when appropriate in either arm; randomization did not assign alteplase versus no alteplase.

Outcomes

Primary analysis: survival-first hierarchical composite

The win ratio was 1.61 (95% CI 1.19–2.18; P=.002). Reported pairwise proportions were 60.4% favoring thrombectomy, 37.6% favoring medical treatment, and approximately 2.0% tied. The net difference was 22.8 percentage points (95% CI 8.3–36.3; P=.002), with contributions of:

Survival time: +15.7 percentage points.
mRS at 12 months, among pairs not already resolved: +6.6 percentage points.
Serious adverse events: −0.1 percentage point.
EQ-5D: +0.6 percentage point.

These are pairwise win-difference contributions, not patient-level absolute risk reductions. Longer survival can win even when both compared patients eventually die. Lower-priority outcomes matter only for pairs not resolved earlier.

Predefined subgroup analyses within this secondary analysis

All estimates below are win ratios, not odds ratios for functional independence. These within-subgroup P values do not establish treatment-effect interaction; the paper does not report formal interaction tests in these results.

Stratum	Win ratio (95% CI)	P value
Men	1.62 (1.07–2.47)	.023
Women	1.61 (1.03–2.51)	.035
Age ≤70	1.83 (1.14–2.94)	.012
Age 71–80	1.97 (1.14–3.40)	.015
Age >80	1.00 (0.55–1.83)	.991
Prestroke mRS 0–1	1.68 (1.21–2.34)	.002
Prestroke mRS ≥2	1.89 (0.87–4.10)	.106
Fully conscious	1.96 (1.32–2.91)	.001
Impaired consciousness	1.22 (0.75–1.98)	.426
ASPECTS 3	1.31 (0.78–2.19)	.304
ASPECTS 4	1.54 (0.90–2.65)	.115
ASPECTS 5	2.03 (1.19–3.48)	.010
ICA occlusion	1.18 (0.70–1.97)	.535
M1 occlusion	1.87 (1.32–2.63)	<.001
Intravenous thrombolysis	1.16 (0.71–1.90)	.542
No intravenous thrombolysis	2.08 (1.40–3.08)	<.001

Other efficacy outcomes and sensitivity analyses

Twelve-month mRS, median (IQR): 5 (3–6) versus 6 (4–6).
Functional independence, mRS ≤2: 27/125 (22%) versus 7/128 (5%). The difference calculated from these counts is approximately 16.1 percentage points, descriptive rather than an adjusted effect estimate.
Death or dependency, mRS 4–6: 84/125 (67%) versus 98/128 (77%).
EQ-5D index, median (IQR): 0.7 (0.4–0.9) versus 0.4 (0.2–0.7), with substantial informative missingness; 133/253 (52.6%) lacked these data.
Alternative hierarchies yielded WR 1.63 (95% CI 1.21–2.21; P=.001) and 1.65 (95% CI 1.22–2.22; P=.001). Reordering preserved the overall direction but changed the interpretation of component contributions.

Adverse Events / Safety

Any serious adverse event: 68/125 (54%) versus 91/128 (71%) in this intention-to-treat reanalysis.
The text reports 12-month mortality of approximately 45% versus 56%. Table 2 places death counts beside Kaplan–Meier survival percentages under an ambiguous label; those entries should not be treated as matching crude death proportions.
The original 2023 trial report, using its separate treatment-received safety population, reported symptomatic intracranial hemorrhage in 7 patients (6%) versus 6 (5%). These are parent-trial safety results, not newly estimated 12-month bleeding outcomes.

Figures

Hierarchical outcome components and their net win-difference contributions — *Figure 1.* Hierarchical structure of the WR analysis. This schematic illustrates the hierarchical comparison process used in the WR analysis. Each patient treated with EVT + BMT is pairwise compared with every patient that received BMT alone. Outcomes are evaluated sequentially: (1) time to death, (2) mRS at 12 months, (3) any serious adverse event and (4) EQ-5D at 12 months. For each comparison, the pair is classified as an EVT + BMT “win,” “loss” or “tie” based on the first outcome in the hierarchy that differs between the 2 patients. The overall WR is calculated as the total number of EVT + BMT wins divided by total BMT wins, with corresponding 95% CI and P-value. This hierarchical approach summarises efficacy and safety across multiple clinically important endpoints into a single composite measure. Abbreviation: BMT = best medical treatment; EQ-5D = EuroQOL 5-dimension.

Overall and subgroup win ratios with confidence intervals — *Figure 2.* Forest plot showing the results of the unstratified and stratified WR analysis for EVT + BMT vs BMT alone. A WR > 1 indicates a positive treatment effect for EVT + BMT compared to BMT alone. The vertical dashed line represents the line of no difference (WR = 1). Abbreviations: BMT = best medical treatment; WR = win ratio.

Criticisms

A secondary hierarchy is not a prospective primary endpoint. The authors explicitly acknowledge that the investigator-selected hierarchy was developed for this analysis of a completed trial. It needs prospective co-design with patients and families before outcomes are known.
Survival-first incorporates a value judgment. The thrombectomy-arm median mRS was still 5. Some patients would not value longer survival with profound dependency above all other outcomes; a favorable composite cannot settle their preferences.
16,000 pairs are not 16,000 independent observations. They reuse 253 patients. Neither the point estimate nor the nominal number of pairs justifies treating this as a much larger trial.
Subgroup significance is not interaction. An estimate crossing 1 in older patients or alteplase recipients does not prove no benefit there, and it cannot justify age-based exclusion or withholding thrombolysis.
Quality-of-life missingness remains important. Resolving pairs on death first avoids requiring EQ-5D in deceased patients, but does not reconstruct the missing quality-of-life experience of survivors. With EQ-5D last, it contributes little to this composite.
Parent-trial early stopping and limited generalizability persist. Reanalysis does not remove uncertainty from a trial stopped at 253 participants or extend its population to ASPECTS 0–2, posterior circulation, distal occlusion, or treatment beyond 12 hours.
Small internal reporting inconsistencies deserve acknowledgment. The text's pair counts sum to 15,999 rather than the stated 16,000; Table 2 mixes survival labeling and death counts. The headline WR and CI are reported as published, without silently “repairing” the underlying data.

Funding

The analysis reports funding from the European Union Horizon 2020 research and innovation programme, grant 754640. The registry identifies University Hospital Heidelberg as lead sponsor. The parent trial was investigator-initiated; this secondary article's funding statement does not separately describe the funder's role in design, analysis, interpretation, or writing, so absence of involvement should not be assumed. Several investigators report device-industry grants, consulting, or honoraria in the article's disclosures. Trial registration: NCT03094715.

The paper

Authors. Olga Ciobanu-Caraus et al., for the TENSION investigators.
Title. Hierarchical composite outcomes in acute ischaemic stroke with large infarct: a win ratio analysis of the TENSION trial.
Journal. European Stroke Journal.
Year. 2026; published online June 18, 2026, per the JATS publication metadata.
DOI. 10.1093/esj/aakag063
PMCID. PMC13278760
License. Creative Commons Attribution 4.0. Figure captions below the images are reproduced with attribution; surrounding discussion is this review's interpretation.

Deep Dive — click to expand

What this is

This is a look-back at a June 2026 analysis of data from TENSION, whose original randomized results appeared in 2023 and 12-month follow-up in 2024. It asks a methodological question with a clinical consequence: when thrombectomy helps a person with a large infarct, how much of the measured advantage reflects living longer, how much reflects less disability, and how much reflects other outcomes? Under the investigators' survival-first hierarchy, the WR was 1.61, and survival supplied most of the 22.8-point net pairwise advantage. The result clarifies an existing treatment signal; it does not create a new indication or replace current eligibility guidance.

1. Shadow Audit

The positive headline hides a difficult prognosis: the median 12-month mRS in the intervention group was 5, and 67% had mRS 4–6. Functional independence was more common with thrombectomy, but not the typical outcome. A family hearing only “favorable win ratio” could reasonably imagine a degree of recovery the cohort does not promise.

The component breakdown is the more informative result. Survival contributed 15.7 of the 22.8 percentage points—approximately 69% of the net win difference, calculated from the reported components. That is not the proportion of individual patients saved, nor a causal mediation estimate. Because death sits first, it absorbs comparisons before disability, adverse events, and quality of life get a vote. The almost neutral adverse-event contribution (−0.1 point) does not contradict the descriptive 54% versus 71% SAE frequencies: they answer different questions in different sets of comparisons.

2. Inversion Engine

For the opposite overall conclusion under this exact hierarchy, losses would need to outnumber wins: WR <1 and net win difference <0, rather than WR 1.61 and +22.8 points. The reported CI of 1.19–2.18 does not include 1, and switching the first two priorities did not reverse direction: WR 1.63 when mRS preceded survival.

The clinically consequential inversion is not necessarily a negative trial. A patient may prefer avoiding severe dependency to prolonging life. That preference would require a prospectively justified, patient-informed outcome ordering or utility assessment, not merely relabeling the same survival-first composite. The current study cannot calculate an individual's preferred trade-off. Nor can the age >80 estimate of 1.00 prove futility: its CI spans 0.55–1.83.

3. Second-Order Catalyst

The first practical adopters should be stroke trialists and teams conducting family discussions, not services expanding thrombectomy eligibility on the strength of a new statistic. A trial steering committee can prospectively add a patient-designed hierarchy, a win-difference breakdown, and conventional ordinal mRS results to the same analysis plan. That is a reporting and design improvement; it does not require changing devices or adding a new treatment.

For a comprehensive stroke center reviewing its large-infarct counseling materials, the immediate educational change is to present survival and disability separately. For a trial-eligible patient whose prior function was good, the relevant question is not just whether the procedure improves outcomes on average, but which outcomes the patient considers worth pursuing. That discussion should be integrated into, not allowed to delay, time-sensitive assessment.

4. Asymmetric Leverage

The low-cost leverage is interpretive: adding the component table prevents a single favorable number from becoming an exaggerated promise of independence. The descriptive independence counts were 27/125 versus 7/128, an approximately 16.1-point absolute difference, while the median intervention-arm patient still had severe disability. Both facts belong in the same conversation.

For research design, the authors illustrate that a WR-based study might need 193 participants under simplified assumptions matching these data. That is a hypothesis-generating calculation, not a validated sample-size recipe: it uses observed outcomes and omits loss to follow-up, censoring, and covariate adjustment. The asymmetric opportunity is a better prespecified analysis at low marginal data-collection cost—not permission to shrink trials until a convenient endpoint turns positive.

5. Paradigm Destroyer

The reflex this paper should retire is “a better composite means the patient is more likely to return to independence.” A composite is only as clinically meaningful as its ordering, and even an mRS-first hierarchy still includes death within mRS 6.

A two-sentence educational amendment to a large-infarct counseling protocol would be: “Discuss expected survival, the distribution of disability, and the patient's goals separately rather than translating a win ratio into a recovery probability. Do not create new exclusions based on non-significant subgroup estimates or expand eligibility from this secondary analysis.” This is an interpretation aid, not a replacement for treatment guidelines or individualized clinical judgment.

MVP — Minimum Viable Proof

The minimum defensible statement is: in 253 randomized TENSION participants, a survival-first hierarchical reanalysis favored thrombectomy (WR 1.61, 95% CI 1.19–2.18), with survival contributing 15.7 points of a 22.8-point net pairwise advantage. That is sufficient to improve how this trial is explained and to motivate prospective endpoint development. It is not sufficient to estimate an individual's chance of benefit, establish a new age threshold, or claim a 61% relative reduction in death.

Best Combination

Read the three reports as different views of the same cohort, not independent replications. The 2023 parent trial supplies the randomized 90-day functional estimate (adjusted common OR 2.58, 95% CI 1.60–4.15) and procedural safety context. The 2024 follow-up supplies the 12-month ordinal functional estimate (adjusted common OR 2.39, 95% CI 1.47–3.90), quality-of-life reporting, and longer-term perspective. This 2026 reanalysis shows how a hierarchical composite assigns the pairwise advantage among outcomes.

The combination supports looking beyond a single independence cutoff while retaining the full mRS distribution and explicit component results. It does not prove that a win-ratio reanalysis would overturn a neutral trial in a different occlusion population, or that an investigator's ranking matches every patient's values.

Overvalue Warning

Do not convert WR 1.61 into “61% more patients benefit,” “61% lower mortality,” or an NNT from the 22.8-point win difference. These are comparisons between pairs, not a patient-level risk model, and the latter difference is not an absolute event-risk reduction.
Do not use the subgroup forest plot as a treatment selector. The thrombolysis/no-thrombolysis and older/younger contrasts are not randomized comparisons between those characteristics; within-stratum significance is not an interaction test. These findings cannot justify withholding alteplase or categorically denying thrombectomy to an older person.

Educational use notice. The Time Is Brain Review is dedicated to medical education. Figures from published articles are embedded for teaching purposes, with attribution to the original publication. We do not claim copyright over any embedded figure; all rights remain with the original publisher and authors under the article's published license. If you are a rights holder and would like a figure removed, please contact the editorial team. See our editorial notice for the full policy.