Patients who develop kidney failure require renal replacement therapy (RRT) in order to survive. Renal transplantation is the best treatment but is often not forthcoming. Many patients therefore require haemodialysis treatment, for which a means of vascular access (VA) is necessary. There are substantial morbidity, mortality and cost benefits from using arteriovenous fistulae (AVF) rather than central venous catheters (CVC) for this purpose. Despite audit standards recommending that most patients should dialyse using AVF there has been longstanding, marked variation between centres in Scotland and further afield as to the proportions of patients who dialyse using each VA modality. Many studies have documented this variation and its clinical consequences, but little progress has been achieved over more than a decade of registry-documented practice. The present study aims to understand this variation using systems approaches to delineate the structure of VA clinical pathways in Scotland and the manner in which they function; to quantify the clinical workload associated with VA services; to illuminate gaps between ‘work-as-imagined’ and ‘work-as-done’ from the perspectives of those working in the services; and to present the findings in a manner that facilitates quality improvement activities. The study characterises VA as a complex socio-technical system, with reference to the patient safety, quality improvement and systems theory literatures. A novel approach to investigating complex clinical systems was developed, in keeping with the principles of Safety II and healthcare resilience engineering. A mixed-methods approach was used to investigate every Scottish VA service, including detailed semi-structured interviews, a clinical activity census, and linkage with pre-existing registry data. An in-depth, thematic analysis of audio-transcripts was considered in light of clinical activity and registry data. The results were distilled into four major themes: VA creation, VA maintenance, service performance, and development needs. A substantial associated clinical workload was quantified for the first time, despite a shortage of clinical resources dedicated to the service. VA creation procedures were proportional to the size of the local RRT cohort, but maintenance activities did not reflect the local cohort size and varied widely between centres. Recommendations for practice were disseminated using a novel ‘scorecard’ tool, designed with the principles of resilience engineering in mind. Centres were encouraged to report their concordance with recommendations; the resulting data suggested a statistical relationship between published incident and prevalent AVF use, and the degree to which the recommendations were implemented by each centre.