Using machine learning to predict patient outcomes after lower limb vascular surgery: preliminary descriptive analyses of the National Vascular Registry
Peripheral arterial disease (PAD) affects 20% of the UK population aged 55-75, equating to a total of approximately 850,000 people. The most severe manifestation, critical limb ischaemia, has an estimated annual cost to the NHS of more than £200 million. People with this condition are now often older and have more comorbidities, and the primary contributory factor is shifting from cigarette smoking to diabetes mellitus. Coupled with this are changes in the way PAD is diagnosed and treated, with modern treatments becoming less invasive. Existing risk prediction models for such procedures were developed in the 1990s and as highlighted there have been significant changes in both the medical care of these patients and the patients themselves. As such it is becoming more difficult to accurately inform patients of their individual perioperative risk.
The UK National Vascular Registry (NVR) records data on patients with PAD undergoing lower limb operations including stenting, bypass and amputation. A total of 40,890 such procedures were recorded between 01/01/2014 and 31/12/2016. Little is known about the outcomes of this patient population in the UK, and there are no commonly used risk prediction models for these specific procedures. By contrast, patients undergoing abdominal aortic aneurysm repair have been extensively studied using NVR data. Previous work using other data sources, primarily from the United States, has used statistical methods to estimate risk of mortality and complications after amputation, stenting and bypass operations, but there is no comparable work in UK patients.
As such there is a need to explore this patient group, define their outcomes, and to develop more accurate and specific risk prediction models for these procedures. This will ultimately have significant patient benefit as we will be able to improve the care these patients receive.
The first stage will be to examine the data from the NVR and to describe what it tells us about the patients undergoing these operations. We will also look at data from the Office for National Statistics and the NHS Hospital Episode Statistics to find out more about what happens to these patients after they leave hospital. We will then use statistical and artificial intelligence techniques to try and develop models to predict the risk for patients having these procedures.
The study team has significant experience and expertise in Vascular surgery and anaesthesia (Professor Rob Hinchliffe and Dr Ronelle Mouton), as well as the statistical methods required to undertake further analyses (Mr Graeme Ambler). The ultimate goal of this research project is to produce a risk prediction tool and/or decision aid for use in clinical practice. The funding requested will enable the preliminary work to be undertaken, namely the data extraction and data linkage from the NVR and NHS Digital datasets. This will be followed by the first stages of the statistical analyses required, in collaboration with colleagues at University of Bristol. We will then apply for a larger research grant from a national funding body to further develop the work towards clinical use.