BLOod Test Trend for cancEr Detection (BLOTTED): an observational and prediction model development study using English primary care electronic health records data
Simple single blood tests can play an important role in identifying patients for cancer investigation. The current evidence base is limited almost entirely to tests used in isolation. However, combining several blood tests and investigating trends in repeated blood test results over time could be more useful. We aim to explore whether combinations of blood tests and blood test trends are more useful than symptoms and single blood test results in selecting primary care patients for cancer investigation. We aim to develop clinical prediction models incorporating blood test trends for risk of cancer to help general practitioners identify patients who need referral for further investigation.
A recent clinical review confirms that simple blood tests have an important role in identifying patients for cancer investigation. However, analysis of National Cancer Diagnosis Audit in Primary Care data suggests that primary care investigations may delay referral, as it is often unclear which department the patient should be referred to. Smarter use of blood tests is needed to reduce referral delays and therefore increase cancer detection rates and reduce unnecessary referrals.
Our group has recently completed analyses investigating test combinations for cancer and trends in full blood count blood test results specifically for colorectal cancer diagnosis. Our studies found that test combinations and trends among repeated tests may hold greater potential to rule-in and rule-out further cancer investigation than symptoms and single blood test results.
WHAT WE ARE DOING
Using population-level primary care data, we will first describe patterns in blood testing and identify trends in blood tests for cancer overall and by cancer site. We will then use statistical testing to assess the association between blood test trends and cancer presence. Among the relevant blood test combinations, we will use advanced analytic methods, such as joint modelling and machine-learning, to develop prediction models that utilise patient-level trends to identify risk of cancer. We will assess how well these prediction models perform using conventional performance measures for discrimination and calibration.
Blood test trends and combinations may have greater diagnostic ability than single tests, so could facilitate early detection of cancer and therefore improve survival rates.
The models developed in this study will undergo implementation into electronic systems in a single/small group of primary care practices for testing. Patient information, such as trends over repeated blood tests, would be pulled automatically from the patient’s electronic record, and the prediction models executed on the patient’s data every time new blood tests are performed, thereby updating risk of cancer diagnosis. The models will rely on only routinely available data so it is anticipated that little-to-no additional input is needed from GP staff or patients.
External project members
Lara Chammas (Big Data Institute, University of Oxford)
Eva Morris (Centre for Statistics in Medicine, NDORMS, University of Oxford)
Jacqueline Birks (Centre for Statistics in Medicine, NDORMS, University of Oxford)
Tingting Zhu (Department of Engineering Science)
PARTNERS ON THIS PROJECT
Big Data Institute, University of Oxford
Centre for Statistics in Medicine, NDORMS, University of Oxford