HDI has been awarded a Cancer Research UK Pioneer Award for research to improve the early diagnosis of cancer using machine learning and computer-based inference algorithms. We will analyse very large linked and unlinked datasets to ask:
1) Is it possible to identify patterns in medication given prior to the diagnosis of cancer and other data to derive an “index of suspicion” that a patient is at increased likelihood of developing subsequent cancer?
- for different types of cancer; our suspicion is that the index may be most valuable for patients with cancers that present with vague symptoms – e.g. pancreatic, ovarian, stomach or brain cancer.
- for different stages of presentation of the same cancer: for example, can we use the index to help identify common cancers at an earlier stage when they would have a better prognosis?
2) Using the index of suspicion derived in 1), can we risk-stratify patients in the unlinked prescription data to identify those who might be most at risk of developing a particular cancer?
By detecting and clustering patterns in prescriptions and other data such as combinations of medication and comorbidities, we expect to be able to identify typical patients or coincident treatments. This will allow us to build human-readable predictive models for outcomes such as future cancer diagnosis grouped by cancer site and stage of presentation of the potential cancer. Odds’ ratios, flowcharts that users can understand, numerical predictions of risk scores and the likely error rates will also be produced from these models.
When applied to the whole population this approach may allow the risk-stratification of patients with an increased likelihood of presenting with cancer and help GPs to prioritise which patients to refer for further investigation or place on active follow-up.