BACKGROUND Centralising acute stroke services is an example of major system change (MSC). ‘Hub and spoke’ systems, consisting of a reduced number of services providing acute stroke care over the first 72 hours following a stroke (hubs), with a larger number of services providing care beyond this phase (spokes), have been proposed to improve care and outcomes. OBJECTIVE To use formative evaluation methods to analyse reconfigurations of acute stroke services in different regions of England and to identify lessons that will help to guide future reconfigurations, by studying the following contrasting cases: (1) London (implemented 2010) – all patients eligible for Hyperacute Stroke Units (HASUs); patients admitted 24 hours a day, 7 days a week; (2) Greater Manchester A (GMA) (2010) – only patients presenting within 4 hours are eligible for HASU treatment; one HASU operated 24/7, two operated from 07.00 to 19.00, Monday to Friday; (3) Greater Manchester B (GMB) (2015) – all patients eligible for HASU treatment (as in London); one HASU operated 24/7, two operated with admission extended to the hours of 07.00–23.00, Monday to Sunday; and (4) Midlands and East of England – planned 2012/13, but not implemented. DESIGN Impact was studied through a controlled before-and-after design, analysing clinical outcomes, clinical interventions and cost-effectiveness. The development, implementation and sustainability of changes were studied through qualitative case studies, documentation analysis (n = 1091), stakeholder interviews (n = 325) and non-participant observations (n = 92; ≈210 hours). Theory-based framework was used to link qualitative findings on process of change with quantitative outcomes. RESULTS Impact – the London centralisation performed significantly better than the rest of England (RoE) in terms of mortality [–1.1%, 95% confidence interval (CI) –2.1% to –0.1%], resulting in an estimated additional 96 lives saved per year beyond reductions observed in the RoE, length of stay (LOS) (–1.4 days, 95% –2.3 to –0.5 days) and delivering effective clinical interventions [e.g. arrival at a Stroke Unit (SU) within 4 hours of ‘clock start’ (when clock start refers to arrival at hospital for strokes occurring outside hospital or the appearance of symptoms for patients who are already in-patients at the time of stroke): London = 66.3% (95% CI 65.6% to 67.1%); comparator = 54.4% (95% CI 53.6% to 55.1%)]. Performance was sustained over 6 years. GMA performed significantly better than the RoE on LOS (–2.0 days, 95% CI –2.8 to –1.2 days) only. GMB (where 86% of patients were treated in HASU) performed significantly better than the RoE on LOS (–1.5 days, 95% CI –2.5 to –0.4 days) and clinical interventions [e.g. SU within 4 hours: GMB = 79.1% (95% CI 77.9% to 80.4%); comparator = 53.4% (95% CI 53.0% to 53.7%)] but not on mortality (–1.3%, 95% CI –2.7% to 0.01%; p = 0.05, accounting for reductions observed in RoE); however, there was a significant effect when examining GMB HASUs only (–1.8%, 95% CI –3.4% to –0.2%), resulting in an estimated additional 68 lives saved per year. All centralisations except GMB were cost-effective at 10 years, with a higher net monetary benefit than the RoE at a willingness to pay for a quality-adjusted life-year (QALY) of £20,000–30,000. Per 1000 patients at 10 years, London resulted in an additional 58 QALYs, GMA resulted in an additional 18 QALYs and GMB resulted in an additional 6 QALYs at costs of £1,014,363, –£470,848 and £719,948, respectively. GMB was cost-effective at 90 days. Despite concerns about the potential impact of increased travel times, patients and carers reported good experiences of centralised services; this relied on clear information at every stage. Planning change – combining top-down authority and bottom-up clinical leadership was important in co-ordinating multiple stakeholders to agree service models and overcome resistance. Implementation – minimising phases of change, use of data, service standards linked to financial incentives and active facilitation of changes by stroke networks was important. The 2013 reforms of the English NHS removed sources of top-down authority and facilitative capacity, preventing centralisation (Midlands and East of England) and delaying implementation (GMB). Greater Manchester’s Operational Delivery Network, developed to provide alternative network facilitation, and London’s continued use of standards suggested important facilitators of centralisation in a post-reform context. LIMITATIONS The main limitation of our quantitative analysis was that we were unable to control for stroke severity. In addition, findings may not apply to non-urban settings. Data on patients’ quality of life were unavailable nationally, clinical interventions measured changed over time and national participation in audits varied. Some qualitative analyses were retrospective, potentially influencing participant views. CONCLUSIONS Centralising acute stroke services can improve clinical outcomes and care provision. Factors related to the service model implemented, how change is implemented and the context in which it is implemented are influential in improvement. We recommend further analysis of how different types of leadership contribute to MSC, patient and carer experience during the implementation of change, the impact of change on further clinical outcomes (disability and QoL) and influence of severity of stroke on clinical outcomes. Finally, our findings should be assessed in relation to MSC implemented in other health-care specialties. FUNDING The National Institute for Health Research Health Services and Delivery Research programme.