MAARS Multimodal Active Adaptive Risk Stratification for Cancer

Compared to similar economically developed countries, the UK has an appalling record for diagnosing cancers at a late stage, when curative treatment is not possible. For example, the UK has one of the poorest survival rates for colorectal cancer in Europe[1] which is thought to be partly related to late presentation, delays in diagnosis and delays in treatment. The five-year survival for early-stage colorectal cancer is greater than 90% compared with 10% for widespread cancer at diagnosis[2,3]. Evidence suggests that increased awareness of symptoms and earlier diagnosis could help improve treatment options and improve 5 year survival[4]. It has been estimated that such an approach might save 5,000 lives even without any new medical advances[5].

Whilst there have been some advances over the last decade, there is currently no adaptive interpretable decision tool to enable earlier cancer diagnosis available for use at the point of care. There is no risk scoring system which updates and refines cancer risk over time as further clinical information becomes available along the patient pathway.

Existing systems implemented into the NHS, such as the First-of-Type QCancer risk stratification system[6,7], only provide static risk estimates and do not include results of blood tests or scans and are not updated as further information becomes available along the diagnostic pathway.

Using the Qresearch linked database a cohort of adults in primary care will be assembled between 2010 and 2022. This will be used to develop and validate an active adaptive interpretation risk/tool to generate personalised clinical phenotypes linked to appropriate clinical pathways. This will help enable targeting of resources to those at highest risk of having a cancer, as yet undiagnosed, and also most likely to benefit. It would also identify low-risk patients who can be reassured.

As the patient progresses along the clinical pathway, their risks will be updated to incorporate the results of investigations to further refine the class assignment and enable earlier referral or discharge to watchful waiting. The cancer risk estimates will be generated and shared with clinicians (e.g. gastroenterologists) in a coded form so they can be used to triage patients on waiting lists for scans, endoscopies & other required investigations. Such a model can then be adapted to any cancer type.

Comparisons will be made against existing cancer prediction algorithms (such as the QCancer scores) to determine the utility and potential added benefit of this new approach.

demographics - age, sex, ethnicity, deprivation, geographical region
cancer diagnoses
co-morbidities associated with increased risk of cancer
investigations (including blood tests, scans)
medication

The linked hospital episode statistics include information on process variables including outpatient appointments, hospital admissions, diagnostic tests undertaken as well as outcomes (cancer diagnoses) and treatments (surgical procedures, chemotherapy).

The linked mortality data include cause and date of death.

The linked cancer registry data contain data on cancer diagnoses, histological type, grade, stage, route to diagnosis, treatment with radiotherapy or Systemic Anticancer Therapy (SACT)

Professor Julia Hippisley-Cox (University of Oxford)

Professor Carol Coupland (University of Nottingham)

Dr Tingting Zhu (University of Oxford)

Dr Jun Wang (Queen Mary University London)

Professor Konstantin Nickolic (University of West London)

Dr Robert Kerrison (University of Surrey)

Dr Joe Geraghty (Manchester University NHS Foundation Trust)

Dr Jennifer Hirst (University of Oxford)

Dr Diana Withrow (University of Oxford)