Jacqueline Birks

I studied Natural Sciences at the University of Cambridge  (Part 2 Physics) and Applied Statistics at the University of Oxford. I have an extensive range of experience, covering quantitative genetics, clinical trials, individual patient data meta-analysis, analysis of longitudinal data, systematic reviews, and diagnostic and prognostic tests. My interests range over a number of clinical specialities, including surgery, cardiology, emergency medicine, dementia, neurosurgery, pathology and geriatrics.

I am a senior research  advisor for the NIHR - Research Support Service.

I am a co-applicant and lead statistician for ARTICULATE PRO, a project funded by the Accelerative Access Collaborative and NHSx  through a Phase 4 AI in Health and Care Award and lead by Professor Clare Verrill,  consultant pathologist, Nuffield Department of Surgical Science. The objective is to investigate the deployment of AI technology in the prostate cancer pathway by using Paige Prostate to assist pathologists when reading prostate biopsies.

I am a co-applicant and lead statistician for the Risk Of Aneurysm Rupture Study (ROAR) which is lead by Mr Diederik Bulters, consultant neurosurgeon. Intra-cranial aneurysms are common and found in approximately 3% of the population. There is a chance that these aneurysms rupture and cause a subarachnoid haemorrhage which has a 30% mortality rate.  Unruptured aneurysms can be treated but treatment carries risks. Currently there is a risk score (PHASES) that provides estimates of 5-year rupture risk, however this has never been validated  satisfactorily. The objective of the ROAR study is to measure the accuracy of the PHASES score  and develop a more accurate prediction model for the risk of rupture.

One challenging problem that large electronic patient datasets will help us solve is that of early warning scores. These scores are derived from a patient’s current vital signs and are routinely used in hospitals to identify patients who are deteriorating. However, there is little evidence that they work. I am interested in using statistical methods to develop a risk prediction model for deterioration that takes into account both current and previous vital signs and patient characteristics. 

I am evaluating a risk prediction algorithm for unidentified colorectal cancer, derived using machine learning methods and routine patient data on complete blood count from the Clinical Practice Research Datalink (CPRD) database. I am also using this CPRD dataset to develop a risk prediction model using a statistical model. As risk prediction models need to be tested in practice, I am working on trial designs for this purpose.

Clinical trials that compare a new non-surgical intervention with an existing surgical intervention often provoke strong criticisms. I have been the statistician on such a trial, the International Subarachnoid Aneurysm Trial (ISAT), for many years. This trial compared neurosurgical clipping with endovascular coiling to treat ruptured intracranial aneurysms. The first publication in 2002 showed a clear benefit of coiling compared with surgery, but some have questioned the longer-term benefits. It is therefore important to follow up patients, as we have been doing for nearly 20 years in ISAT. Mortality can be assessed using routinely collected data, but the problem I am interested in is dealing with missing data for the patient-reported outcomes.