NHS Scales Stroke AI, but Outcomes Still Need Proof

<p>A stroke scan at an English hospital can now trigger an AI-assisted image review and rapid referral workflow; a senior NHS official says Brainomix is deployed across all 100 stroke units.</p> <p>At a 16 June parliamentary hearing, Dr Alec Price-Forbes, NHS England and DHSC’s National Chief Clinical Information Officer, said Brainomix had reached all 100 stroke units. His testimony is not an audit of scan use or recovery.</p> <p>Thrombectomy removes a clot from a brain artery. Patients at hospitals without the procedure may need rapid transfer to a specialist centre; Brainomix 360 aims to help clinicians identify candidates and share scans.</p>

<h2>What the scan assistant does</h2> <p>Brainomix does not decide treatment. e-ASPECTS analyses non-contrast CT for early ischaemic change and bleeding; e-CTA flags possible large-vessel blockages on angiography; e-CTP estimates damaged and salvageable tissue from perfusion scans. The platform also shares images between hospitals.</p> <ul> <li><strong>Scan:</strong> software returns image measurements or flags.</li> <li><strong>Review:</strong> a clinician checks them against the patient’s scan and condition.</li> <li><strong>Referral:</strong> the team shares the case and arranges transfer if appropriate.</li> </ul> <p>The National Institute for Health and Care Excellence says these are fixed algorithms that do not learn live from routine NHS scans; clinicians remain responsible for diagnosis and treatment.</p>

<h2>England’s national data show a treatment signal</h2> <p>A prospective observational study in The Lancet Digital Health used England’s Sentinel Stroke National Audit Programme to examine 452,952 stroke admissions at 107 NHS hospitals from January 2019 to December 2023. Brainomix 360 was evaluated in 26 hospitals across four thrombectomy networks.</p> <p>At the 26 sites, thrombectomy rates rose from 2.3% to 4.6%—a 100% relative increase. At the other 81 hospitals, rates climbed from 1.6% to 2.6%, or 62.5%. The difference in change favoured evaluation sites (interaction odds ratio 1.24).</p> <p>Patient-level data covered 71,017 people with ischaemic stroke at evaluation hospitals. Adjusted analysis found 1.57 times the odds of thrombectomy with AI review (95% CI 1.33–1.86). At primary centres, transferred patients’ median door-in-door-out time was 64 minutes shorter.</p> <ul> <li><strong>2.3% to 4.6%:</strong> evaluation-site thrombectomy rate.</li> <li><strong>1.6% to 2.6%:</strong> comparison-site rate.</li> <li><strong>64 minutes:</strong> shorter median time from arrival to departure at referring centres for transferred patients with AI review.</li> </ul> <p>The paper reports a discharge-function association, but its figures conflict: 7,238 of 13,293 is 54.4%, not the reported 45.6%. Six-month data were too incomplete to assess longer-term disability. Authors disclosed Brainomix employment, share options or consulting ties; NHS England funded the study.</p>

<h2>Why the headline numbers need limits</h2> <p>Hospitals were not randomly assigned, and stroke pathways changed over five years. By 2023, 56 of 81 comparison hospitals used some stroke AI. Adjustment helps, but cannot isolate the software from staffing or referral changes.</p> <p>It does not prove a threefold rise in long-term independence. That claim traces to a 2024 single-centre study of 40 thrombectomy patients: favourable three-month scores rose from 16% to 48%. The authors flagged the small sample and a non-significant adjusted disability-shift result. Three authors were Brainomix employees; the lead author reported company lecture honoraria. The national study’s outcome data also remain uncertain.</p> <p>A Swedish study published in August 2026 found a similar workflow gain but no significant clinical-outcome difference among 970 thrombectomy patients: median CT-to-puncture time fell from 47 to 35 minutes after rollout.</p> <p>NICE’s final 2024 guidance said more cost-effectiveness evidence was needed. It did not recommend routine NHS use or stopping existing use; e-Stroke, RapidAI and Viz could be used during evidence generation only with healthcare-professional review and NHS England Digital Technology Assessment Criteria approval. NICE also says existing scan-reporting protocols should be maintained. A December 2025 migration left recommendations unchanged; the guidance predates the national study.</p>

<h2>Economic projections are not observed savings</h2> <p>A 1 October 2026 economic model used an estimated 81,565 acute ischaemic-stroke admissions in England in 2023. It projected 8% more intravenous thrombolysis, 44% more mechanical thrombectomies, £3.97 million lower costs and 1,721 added quality-adjusted life-years. At £20,000 per QALY, incremental net monetary benefit was £38.39 million, combining health gains and costs rather than cash savings. Its treatment-effect inputs draw on the same five-year, 26-hospital observational evaluation described above, so the model is not independent confirmation.</p> <p>At CBK’s 15 October rate (£1 = KSh171.30), lower costs equal about KSh680 million and modelled net benefit KSh6.58 billion; neither is a recorded saving or Kenyan forecast. Three model authors disclosed Brainomix employment; another worked for OI Pharma Partners, which received funding from Brainomix.</p>

<h2>For Kenya, referral capacity comes first</h2> <p>This is not a Kenyan deployment. A preliminary 2023 Africa–UK Stroke Partnership survey received responses from 45 of 54 countries. Experts in 16% of respondents reported endovascular-thrombectomy access and identified 69 centres; that is not the share of African patients or hospitals with access. Kenya’s 2024 Ministry of Health guidelines call for CT angiography at NIHSS scores above six and emergency transfer for M1, A1 or P1 occlusions to an appropriately trained, qualified and accredited neuro-intervention specialist.</p> <p>That pathway still needs CT access, image exchange, trained clinicians, ambulances and a specialist receiving centre. For Kenya and the region, software coverage cannot substitute for those links.</p> <p>The next proof point is whether routine records show more eligible patients safely reaching treatment, fewer delays across networks and better disability outcomes months later—not just more licences or connected units.</p>