Principal Investigator: Professor Julia Hippisley-Cox
Institution: University of Nottingham
Lead Collaborators: Dr Juliet Usher-Smith – University of Cambridge, Cambridge, UKTags: 40628, cancer, prediction, prevention, Risk, Screening, validation
Screening for cancers may offer a way to diagnose tumours at an early stage that are treatable, and therefore reduce death rates. Screening methods exist for breast and prostate cancers for example, but these attract some controversy as they may identify tumours that would never progress and cause symptoms, or impact on life. Therefore, a significant number of people every year could undergo tests for cancers that may be invasive and uncomfortable, and cause them to receive intensive treatment for no benefit.
For example, rather than subjecting every woman aged 50 -70 to a test for breast cancer, risk-adapted strategies could identify those people at higher risk of developing breast cancer, and therefore those who may benefit most from undergoing testing. This can be applied to different cancers as well.
Previous research in a sample of almost 5 million individuals has shown that the risk for an individual person developing 11 specific cancers may be predicted by mathematical equations that consider multiple factors about them, such as their age, family history, whether or not they smoke, or how much alcohol they drink. Whilst the results were encouraging, further study using other databases is needed to identify whether or not these tools may be useful to improve screening, or help develop screening for other cancers that do not have a test available yet.
Our project will use the UK Biobank database to test if these developed ‘QCancer’ equations are accurate in predicting the risk of 11 different cancers. By applying the equations to another large database (i.e. the UK Biobank), we will use the equations to try and predict which patients developed breast cancer over a 5-year time period. Then, we will be able to compare this with the observed rates of cancers, and statistical tests can analyse if these are effective tools.
This project will last for 12 months. Should any of the prediction equations prove to be effective in calculating who may be at high risk for specific common cancers, then we may be able to use these to make existing screening tests much more effective, help develop tests for other cancers with no screening programme yet, and target therapy to prevent certain cancers developing in people at high risk.
To externally validate previously developed risk-prediction algorithms for 11 common malignancies derived from the QResearch general practice databases (see Hippisley-Cox & Coupland, BMJ Open 2015, 5;e007825; blood, breast, bowel, gastro-oesophageal, lung, oral, ovarian, pancreas, prostate, renal tract and uterine cancers). For some cancers, especially breast, several risk prediction algorithms have been developed (see Meads, et al. Breast Cancer Res Treat 2012, 132:365-377) however very few have undergone external validation. Such validation in a large cohort is essential for further assessment of their possible utility in clinical practice, specifically whether or not they can be effective tools in targeting screening methods or prevention advice/therapy to individuals at the highest risk. In 2018, Usher-Smith, et al. reported a validation of the QCancer colorectal cancer algorithm in the Biobank cohort and found it to have favourable performance compared to the 13 other algorithms analysed (Usher-Smith, et al. Br J Cancer 2018 doi: 10.1038/bjc.2017.463). We aim to analyse the power, i.e. discrimination and calibration of the QCancer risk prediction tools using the UK Biobank database to ascertain whether or not they may be implemented in efforts to improve the efficacy of screening methodologies for several cancers.
To refine and externally validate previously developed risk-prediction algorithms for a range of outcomes including cardiovascular disease, stroke, thrombosis, fracture, frailty, mortality, kidney disease, diabetes and common malignancies derived from the QResearch general practice databases linked to HES, mortality and cancer registration. The majority of these are in clinical use in the NHS and other parts of the world. For some but not all outcomes, the risk prediction algorithms have been validated on an external database of GP records (CPRD). External validations in a large cohort are essential for further assessment of their possible utility in clinical practice, specifically whether or not they can be effective tools in targeting screening methods or prevention advice/therapy to individuals at the highest risk. We aim to analyse the power, i.e. discrimination and calibration of all of the QPrediction risk prediction tools using the UK Biobank database to ascertain whether or not they may be implemented in efforts to improve the efficacy of identifying those at high risk of a range of different diseases. In addition, we will supplement the existing risk prediction tools with novel variables (eg biomarkers and lifestyle variables) which are on UK Biobank to determine whether or not these improve performance.
Last updated Sep 12, 2019