Notable developments in AI in medicine -- curated signal, not commentary.
Stroke trials, regulatory shifts, adoption data, and research. Each entry links to its original source. Updated as the evidence moves.
15 signalsLast updated Mar 22, 2026
March 2026
Stroke★ Landmark
OPTION Trial: Tenecteplase Improves Outcomes in Non-LVO Stroke at 4.5–24 Hours
The OPTION trial (N=570, 48 centers in China) randomized non-large-vessel-occlusion ischemic stroke patients with CT perfusion mismatch to tenecteplase 0.25 mg/kg vs placebo at 4.5–24 hours. mRS 0–1 at 90 days: 43.6% vs 34.2% (RR 1.28, 95% CI 1.04–1.57, P=0.02). Symptomatic ICH occurred in 2.8% of TNK patients vs 0% of controls.
First positive RCT of extended-window TNK for non-LVO stroke with CTP selection — but 2.8% vs 0% sICH, 5.0% vs 3.2% mortality, and single-country enrollment (China) warrant cautious interpretation before generalizing.
OCEANIC-STROKE is a phase 3 trial of asundexian (factor XIa inhibitor) for secondary prevention in patients with recent non-cardioembolic ischemic stroke. Baseline characteristics of the enrolled population have been published (PMID 41614533). Primary efficacy and safety results have been reported at conferences but are not yet indexed in a peer-reviewed publication on PubMed.
Monitoring watch — primary results publication pending. This entry will be upgraded with outcome data when the peer-reviewed PMID is available.
Loberamisal Phase 3 (LAIS): Positive Neuroprotection Signal Reported at ISC 2026
The LAIS trial (N=998, 32 centers in China) tested IV loberamisal initiated within 48h of acute ischemic stroke. ISC 2026 late-breaking results reported a positive primary endpoint. Peer-reviewed primary publication is pending — outcome numbers withheld until PubMed-indexed.
If the peer-reviewed results confirm the conference report and replicate outside China, loberamisal would be the first neuroprotective agent to show clinical benefit in acute ischemic stroke after decades of failed trials.
MedVersa: Generalist Foundation Model Matches Specialists Across 9 Medical Imaging Tasks
MedVersa, trained on 29 million imaging instances across radiology, pathology, and dermatology, matched or surpassed task-specific models on 9 clinical imaging tasks in a multi-institutional evaluation. In a reader study, radiologists rated 64% of AI-generated chest X-ray reports as clinically acceptable or equivalent to radiologist-written reports.
A single generalist model handling diverse imaging tasks marks a shift from narrow AI — but static benchmark accuracy on curated datasets does not equal dynamic sequential reasoning at the bedside.
Systematic Review: 47% of Healthcare Organizations Using or Assessing AI Agents
A systematic review in Frontiers in AI found that nearly half of healthcare organizations are actively using or evaluating AI agents. The review documents a shift from static clinical alerts to continuously adaptive decision support systems.
Baseline adoption data for AI agents in healthcare — useful for tracking the pace of clinical deployment.
Secondary Prevention after Ischemic Stroke (Abstract)
A NEJM review on secondary prevention after ischemic stroke emphasizes early mechanism-specific prevention, aggressive management of modifiable risk factors, and poststroke monitoring of risk metrics, lifestyle behaviors, and medication adherence. This is a review article rather than a new randomized trial, but it is the newest PubMed-indexed NEJM stroke synthesis identified in the audit.
High-signal review support for the secondary-prevention workflow: prevention should be mechanism-routed and started early rather than treated as a generic discharge checklist.
ASSET-IT: IV Tirofiban After Thrombolysis Improves Outcomes in Noncardioembolic Stroke
The ASSET-IT trial (N=832) showed IV tirofiban within 60 minutes of IVT improved mRS 0-1 at 90 days (65.9% vs 54.9%, P=0.001) in noncardioembolic acute ischemic stroke. Symptomatic ICH occurred in 1.7% of treatment vs 0% of control patients (7 vs 0 events).
First positive trial of an IVT adjunct (tirofiban) in acute stroke — but 7 sICH events vs 0 in controls demands careful risk-benefit in practice.
The American Medical Association reported that physician AI adoption rose from 38% in 2023 to 81% in 2025. Top use cases are research summarization and clinical documentation. Over 75% of physicians surveyed believe AI improves patient care.
AI tools have crossed the adoption majority threshold among US physicians — the question is no longer whether physicians will use AI, but how.
LLM Neurology Benchmarking: High Static Accuracy but Lower Dynamic Clinical Reasoning
A Neurological Sciences study comparing five LLMs on neurological diagnosis and lesion localization reported high accuracy on structured clinical cases. However, a separate NEJM AI study using script concordance testing found LLMs underperformed senior residents and attendings on dynamic clinical reasoning under uncertainty — where diagnoses evolve with new information.
Static pattern recognition ≠ clinical reasoning under uncertainty. LLMs can identify diagnoses from textbook presentations but struggle with the sequential, evolving decision-making that defines bedside neurology.
FDA Final Guidance: Clinical Decision Support Software Enforcement Discretion
The FDA issued final guidance on clinical decision support software under section 520(o)(1)(E) of the FD&C Act. CDS tools that display the underlying evidence basis and support — rather than replace — independent clinician judgment are excluded from device requirements. An enforcement discretion policy covers tools where only one clinically appropriate recommendation exists.
Rule-based CDS tools showing evidence for clinician review now have explicit FDA enforcement discretion — lowers the regulatory barrier for transparent decision support.
Medical Management and Revascularization for Asymptomatic Carotid Stenosis (Abstract)
The parallel CREST-2 trials enrolled patients with at least 70% asymptomatic carotid stenosis across 155 centers in five countries. Over 4 years, carotid stenting plus intensive medical management lowered the primary composite outcome versus medical therapy alone (2.8% vs 6.0%), whereas carotid endarterectomy did not show a significant benefit over medical therapy alone (3.7% vs 5.3%).
Asymptomatic carotid disease may need to be modeled as separate stenting and endarterectomy decisions rather than a single revascularization bucket, especially in secondary-prevention pathways.
TNK Extended-Window Meta-Analysis Supports Tenecteplase at 4.5–24h in Selected Patients
A meta-analysis of randomized trials supports tenecteplase 0.25 mg/kg in the 4.5–24h window for patients with perfusion mismatch on CT or MRI. The Chinese Stroke Association recommends this approach for ICA/M1-M2 occlusion with mismatch imaging. Positive trials include EXTEND (alteplase 4.5–9h), WAKE-UP (DWI-FLAIR mismatch), and OPTION (non-LVO 4.5–24h).
AHA/ASA 2026 (PMID 41582814) subsequently incorporated extended-window thrombolysis: Class IIb for LVO patients without EVT access (citing TRACE-III) and Class IIb for non-LVO with perfusion mismatch (citing OPTION).
Intravenous Tenecteplase before Thrombectomy in Stroke (Abstract)
In the BRIDGE-TNK trial, 550 patients with large-vessel-occlusion stroke within 4.5 hours of onset were randomized to intravenous tenecteplase plus thrombectomy or thrombectomy alone. Functional independence at 90 days was higher with bridging tenecteplase (52.9% vs 44.1%; unadjusted risk ratio 1.20), with more pre-thrombectomy reperfusion (6.1% vs 1.1%), while symptomatic intracranial hemorrhage (8.5% vs 6.7%) and mortality (22.3% vs 19.9%) were numerically higher.
Adds pro-bridging signal for tenecteplase before EVT, but the safety margin remains narrow enough that this should still be framed as evolving evidence rather than a settled universal default.
Endovascular Treatment of Stroke Due to Medium-Vessel Occlusion (Abstract)
ESCAPE-MeVO randomized 530 patients with medium-vessel occlusion within 12 hours to EVT plus usual care or usual care alone. A modified Rankin scale score of 0 or 1 at 90 days occurred in 41.6% versus 43.1%, with higher mortality (13.3% vs 8.4%) and symptomatic intracranial hemorrhage (5.4% vs 2.2%) in the EVT group.
Important negative signal: EVT benefit from large-vessel occlusion should not be assumed to generalize to MeVO without more selective criteria.
Endovascular Treatment for Stroke Due to Occlusion of Medium or Distal Vessels (Abstract)
DISTAL randomized 543 participants with isolated medium or distal vessel occlusion within 24 hours to EVT plus best medical treatment or best medical treatment alone. EVT did not improve 90-day disability (common odds ratio 0.90), and all-cause mortality (15.5% vs 14.0%) and symptomatic intracranial hemorrhage (5.9% vs 2.6%) were similar or numerically worse with intervention.
Paired with ESCAPE-MeVO, this pushes current practice away from routine EVT for distal or medium-vessel occlusions outside clearly selected subgroups or trials.