Developing a decision rule using existing blood test data to select primary care patients with non-specific symptoms for urgent cancer investigation.

Background One in two patients ultimately diagnosed with cancer present to primary care with non-specific symptoms, such as unexpected weight loss (UWL) or fatigue. They have longer times to diagnosis and are more likely to be diagnosed as an emergency or at an advanced stage. NHS Rapid Diagnostic Centres (RDCs) are being rolled out in England to urgently investigate this patient group.

However, many patients with non-specific symptoms have a low risk of cancer and GPs need to select who to refer to RDCs.

Preliminary research suggests routine blood tests may help differentiate those at higher risk from lower, which may prevent unnecessary referrals and improve the diagnostic yield of RDCs Aim • To optimise the selection of patients with non-specific symptoms for urgent cancer investigation using results from routine blood tests.

Primary objective • To incorporate simple blood tests into decision rules to select primary care patients with non-specific symptoms for urgent cancer investigation.

Secondary objectives • To extend existing research using blood tests to select patients with UWL for cancer investigation to other non-specific cancer symptoms. • To understand whether incorporating blood test change over time improves cancer prediction by: o adapting established monitoring methods to routinely collected serial blood test data. o receiving training in advanced statistical techniques (joint modelling) and data science (machine learning). • To develop an intervention to select patients with non-specific symptoms for urgent cancer investigation (in NHS RDCs) for future evaluation.

Methods Routinely collected UK electronic health records data (CPRD AURUM) from primary care linked to cancer registry data (NCRAS) will be used to develop clinical prediction rules.

The performance of models derived using static and serial blood test data, and using traditional statistical methods (logistic regression), advanced statistical methods (joint modelling), and machine learning will be compared using standard discrimination and calibration statistics and decision curve analysis.

How the results of this research will be used The feasibility of using the decision rule developed during this fellowship will be subsequently trialled in the network of NHS RDC pathways.

International collaboration will inform the decision rule’s development so that its feasibility may also be tested in other RDC-type pathways, such as the Welsh Single Cancer Pathway and Nordic Multidisciplinary Diagnostic Centres.

The results will also be relevant to settings where GPs have direct access to invasive testing (such as in the US) or to patient self-referral to community based point-of-care testing.