Randomized trials have recently demonstrated benefit of endovascular mechanical thrombectomy in late-presenting patients with acute ischemic stroke (AIS) due to large vessel occlusion (LVO) who have favorable tissue profiles (small ischemic core, large rescuable penumbra) on multimodal brain imaging. Perfusion-core mismatch patient selection between 6 and 16 hours after last known well and clinical-core mismatch patient selection between 6 and 24 hours after last known well identified patients who benefitted substantially from reperfusion despite relatively late presentation [1,2]. Accordingly, international guidelines have extended the treatment time-window in AIS due to LVO to include select patients with favorable penumbral profiles on brain imaging between 6 and 24 hours after last known well. As a result, systems of acute stroke care are being reorganized and expanded internationally to facilitate endovascular mechanical thrombectomy in late, imaging-selected LVO-AIS patients [3]. An important component of stroke care systems requiring optimization for late, imaging-selected LVO-AIS patients are processes to inform patients, families, and physicians regarding the benefits and risks of therapy to facilitate treatment decision-making.

Among all late-presenting LVO-AIS patients, multimodal imaging selects those who have had slow or intermediate, rather than fast, infarct progression up to the time of brain scan. As a result, the pace of benefit loss from endovascular thrombectomy (EVT) associated with prolonged care process times is likely to be less swift among late-presenting, imaging-selected LVOAIS patients than among unselected early-presenting LVO-AIS patients. Nonetheless, multicenter cohort studies have shown that, even among AIS-LVO slow and intermediate progressors, ischemic core growth proceeds at substantial rates [4]. As a result, experts have emphasized that faster start of EVT after qualifying imaging is desirable to maximize patient benefit even in late, imaging selected patients [5]. After emergency department arrival, immediate actions leave only a short interval for counselling the patient and/or family about endovascular intervention. In this short period, patients and family without medical knowledge or prior experience must decide whether to pursue a major medical procedure that carries both substantial potential benefits and also substantial potential risks. It is therefore critical to rapidly convey information about the potential beneficial and adverse effects on outcome of mechanical thrombectomy performed in the extended time window. Figural displays are powerful tools to help patients, family, and healthcare providers rapidly understand response patterns to therapy [6,7].

We sought to use the methods for creating rapidly understandable figural displays from ordinal disability outcomes in parallel group trials [6,7] to develop visual guides for decision-making regarding ordinal outcome of EVT among late-presenting, imaging-guided, ischemic stroke patients based on data from the two salient pivotal randomized trials, DAWN (Clinical Mismatch in the Triage of Wake Up and Late Presenting Strokes Undergoing Neurointervention With Trevo) and DEFUSE 3 (Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke 3).

In DAWN there were 206 patients, 107 assigned to the thrombectomy group and 99 to the control group, and in DEFUSE 3 there were 182 patients, 92 assigned to the thrombectomy group and 90 to the control group. The study-level averaged distributions of the 3-month mRS with EVT and with control is shown in Figure 1. The mean mRS score in the EVT group was 2.94 (standard deviation [SD], 1.94) and in the medical group 3.98 (SD, 1.67), yielding a mean difference of 1.04. The study-level averaged rates of SICH were 6.1% with EVT and 3.7% with medical therapy and the rate of INT, available from DAWN, was 3.7% with EVT and 0% with medical therapy. For each of the six possible dichotomizations of the 7-level mRS, net BPH values are shown in Table 1 and ranged from 4 to 32.

The independent expert joint outcome analyses indicated that the biologically most plausible BPH, of patients having a better outcome with thrombectomy by ≥1 levels on the mRS, was 52.1 (interquartile range [IQR], 48.2 to 52.6) and the HPH 2.5 (IQR, 1.8 to 3.5). The likelihood of help to harm ratio was 20.8. The net BPH from the expert JOT (eJOT) analysis was 47.6 (IQR, 45.4 to 49.6), comparable to the automatically derived net BPH of 53.0. The NNTB was 1.9 (IQR, 1.9 to 2.1), and NNTH was 40.0 (IQR, 29.2 to 58.3).

Combining the eJOT full mRS shift and the BPH analysis indicated that among 100 patients treated with thrombectomy 52 have a reduced final level of disability as a result of treatment, including 32 who improve to an independent functional outcome (mRS 0-2) and 20 whose final disability reduction occurs across other transitions on the mRS besides the mRS 3 to 2 transition. The eJOT HPH analysis also indicated that among these 100 treated patients, three would have an increased final level of disability as a result of thrombectomy, including one who worsens to a severely disabled or dead (mRS 5-6) outcome and two whose final disability worsening occurs across other transitions on the mRS besides the mRS 4 to 5 transition. These values were used to generate the lead decision matrix visual decision aid for late-presenting imaging-selected patients treated with thrombectomy (Figure 2). In a similar manner to above analyses, three additional Choice Consequence Matrix type visual decision aids were: (1) 7-level mRS analysis using DAWN data only (Figure 3A); (2) 7-level mRS analysis using DEFUSE 3 data only (Figure 3B); and (3) 6-level mRS analysis using pooled DAWN-DEFUSE 3 (Supplementary Figure 1) and DAWN or DEFUSE 3 individually (Supplementary Figure 2).

The Choice Comparison visual decision aid using seven color shades to depict all 3-month mRS outcomes in both EVT and medically treated patients is shown in the Supplementary Figure 3.

In the present study, visual aids have been produced to convey key decisional information to patients, families, and providers regarding benefits and harms of EVT in late-presenting, imaging-selected patients, encompassing benefits and harms over all levels of 3-month disability, as well as early adverse effects of SICH and INT. Expert-informed derivation of benefits and harms per hundred treated patients considering disability transitions across all seven levels of the mRS and automated, mathematical derivations demonstrated good agreement. In the expert-assisted derivation, among every 100 patients treated with EVT, 52 will have an improved and three a worse 3-month global disability level. Clinicians can use the developed visual decision aids to rapidly inform patients and family members of the benefits and risks associated with EVT for AIS in penumbral-imaging selected patients 6 to 24 hours after last known well.

The BPH and HPH values for shift in 3-month mRS outcomes derived in the current study, and depicted in the resulting visual aids, are consonant with, and extend, prior related findings. The magnitude of EVT shift effects in the late-presenting, imaging-selected EVT patients is comparable to the effects of early EVT in largely under 6-hour patients ineligible for intravenous thrombolytic therapy, and the net benefits substantially exceed that of intravenous thrombolysis alone in under 3-hour patients [6,7]. This accords with the greater reperfusion efficacy of EVT over intravenous thrombolysis and with the capacity of multimodal imaging to identify those late-presenting patients who are similar to the preponderance of early-presenting patients in still harboring substantial salvageable penumbra.

Producing visual decision aids delineating dichotomized end points is relatively straightforward, but figural displays are more challenging to develop for ordinal outcomes that are arrayed along a range of disability. Fortunately, techniques to produce rapidly understandable figural displays from ordinal disability outcomes in parallel group trials have been developed [6,7]. Though, to our knowledge, expert derivation of differentiated BPH and HPH for EVT in imaging-selected, late-presenting patients has not been previously performed, our findings are consistent with an automated derivation of net BPH performed on the DAWN trial dataset using a joint outcome table multiple resampling technique [2]. The net BPH value for DAWN alone of 50 in that study is comparable to the net BPH values for pooled DAWN-DEFUSE 3 data in the current study using the expert joint outcome table population technique (net BPH 48) and the automated, algorithmic min-max, joint outcome table population technique (net BPH 53).

The general comparability in the current study of the net BPH derived by the expert-informed and expert-independent methods strengthens confidence in the results and provides assurance that the expert judgments are reliable. That the expert-derived net BPH was actually mildly lower than the algorithmically derived net BPH further indicates that the experts did not “game” their joint outcome table completions to inflate the BPH. Rather, as a result of their content expertise, they judged that those patients who benefit from EVT benefit substantially, so that the group-level benefit of EVT observed in overall trial results is distributed to fewer patients than indicated by the automated algorithmic technique.

The visual decision aids developed in this study meet key recommendations of the International Patient Decision Aids Collaboration for developing effective decision aids for emergency treatments [6,11]. These include (1) figures are based on input from disease-specific experts and experts in multiple related specialties; (2) figures are generated according to the highest available quality of research data; (3) an appropriate graph type is chosen for the type of information to be conveyed (in this instance, Kuiper-Marshall personographs for betweengroup outcome comparison); (4) information includes both benefits and risks associated with a treatment option; (5) the elements of the graphical display are shown proportional to the data quantiles; (6) absolute indices rather than relative outcomes are displayed; (7) a common denominator is used to compare outcome probabilities in different treatment options; and (8) color and symbol formatting is aligned with the degree to which outcomes are important and meaningful for patients. In the lead visual decision aid display format (Single Personograph Choice Consequence Matrix), large person-icons convey the most important outcome, 3-month level of disability, using color-coding to distinguish five effect levels (improvement to functional independence, improvement though not to functional independence, no alteration, worsening though not to severe disability-death, and worsening to severe disabilitydeath). In addition, the less important (because short-term) outcomes of early SICH and imaged INT are conveyed by smaller bars placed within appropriate person-icons. In the alternative visual decision aid display format (Double Personograph Choice Comparison), person-icons convey the most important outcome, 3-month level of disability, using color-coding to distinguish all seven levels of the mRS, showing the expected outcome distributions with EVT+medical management vs. medical management alone side-by-side. Physicians, nurses, patients, and families vary in their preference regarding whether to use Choice Consequence Matrix or Choice Comparison visual displays for decision-making, so both have been generated. But most end-users prefer to use the Choice Consequence Matrix display as focuses viewer attention solely upon changes in outcome directly attributable to thrombectomy treatment, leaving aside other influences upon outcome that may be more challenging for patients and families to process quickly. The Choice Consequence Matrix display was therefore developed as the lead display in this study.

A caution to note is that, while both the DAWN and DEFUSE 3 trials used advanced imaging techniques for patient selection among late-presenting patients, some technical aspects of the imaging-selection strategies differed between the two trials. Both trials used advanced imaging to identify patients with favorable penumbral profiles, characterized by large volumes of threatened but still salvageable tissue (penumbra) and small volumes of likely irreversibly-infarcted tissue (ischemic core) [1,2,12-14]. DEFUSE 3 used a perfusion—core mismatch imaging strategy to identify favorable penumbral profile patients. DAWN permitted sites to either also use the perfusion—core mismatch strategy or, at magnetic resonance imaging (MRI) sites able to perform diffusion but not perfusion MRI, to use a clinical—core mismatch strategy identify favorable penumbral profile patients. As both trials’ imaging strategies select for similar small core/large penumbra patients, it would be pathophysiologically expected that they would be enrolling very similar patients, and formal studies have empirically confirmed this similarity by showing that patients selected by both imaging strategies respond to EVT similarly [12,15].

The findings of the current study provide further support that the DAWN and DEFUSE 3 imaging strategies select largely similar patients. The BPH values for the DAWN trial alone (52) and the BPH values for the DEFUSE 3 trial alone (55) were very similar, as were the HPH values for the DAWN trial alone (3) and for the DEFUSE 3 trial alone (3). As a result, the Choice Consequence Matrix visual decision aids for each trial alone were very similar in appearance.

An additional caution is that the visual aids generated in this study show outcomes and outcome differences expected for a prototypical patient of the type enrolled in the DAWN and DEFUSE 3 trials, and each individual patient in front of a clinician at time of acute treatment decision-making is likely to differ in several features from the prototypical trial patient. Every individual patient has particular prognostic features, including age, admission National Institutes of Health Stroke Scale (NIHSS) score, time from last known well, and imaged infarct core size, that differentiate them somewhat from the overall trial population and will shape their specific outcomes with EVT and with medical therapy alone [16]. Nonetheless, fixed decision aid images that reflect a prototypical patient enrolled in trials with largely similar populations can be useful in emergency settings where rapid access to the decision aid is a requisite. For this reason, fixed decision aid images for prototypical patients merging data from several acute stroke studies have found wide adoption for intravenous thrombolysis and EVT therapy for AIS [7,17,18]. Personalized, patient-specific visual decision aids able to be generated in real-time after input of individual patient characteristics have been developed for thrombolytic stroke therapy and other cardiovascular treatments, but have thus far been confined to presentation of binary dichotomized outcomes, and so substantially underestimate the full benefits and full harms of therapy [19,20].

This study has additional limitations. First, while data regarding 3-month global disability early SICH outcomes were available from both trials, data regarding infarcts in new territories was available only from DAWN. Second, the generated figures reflect outcomes of EVT when performed by experienced neuro-interventionalists during the time period of conduct of the DAWN and DEFUSE 3 trials. The benefits of EVT may not be as great when performed at centers with less experienced interventionalists and supporting services; conversely, the benefits of EVT may be heightened when performed with technologies that have become available only after the completion of the DAWN and DEFUSE 3 trials [21]. Third, the DAWN and DEFUSE 3 trials each recruited patients with extremely favorable patterns on advanced imaging. The substantial net benefit from EVT in these trials suggests that patients with mildly less favorable imaging profiles (modestly larger ischemic core volume and/or modestly smaller penumbra volume) are still likely to benefit meaningfully from EVT. But the current figures are not applicable to such patients who do not quite formally meet either DAWN or DEFUSE 3 entry criteria. Lastly it is important to emphasize that direct patient-practitioner and/or family-practitioner discussions are pivotal for optimal treatment decision-making; the visual aids generated in the current study are intended to facilitate, not substitute for, those discussions.

Mechanical thrombectomy in late-presenting, imaging-selected patients with ischemic stroke due to LVO confers substantial benefit, reducing long term disability across the entire mRS range in 52% of treated patients, and minimal harm, increasing long term disability in 3% of patients. Figural decision guides are now available that amalgamate in one rapidly assimilable image, long term benefits, long term harms, and short term complications. These visual displays can convey the most salient information necessary for decision making rapidly and efficiently in emergency settings. The figures may also be helpful in educating policy-makers, the general public, and key stakeholders designing regional systems of acute stroke care regarding the magnitude of benefit and harm of EVT in latepresenting, imaging-selected stroke patients.