The MRC Biostatistics Unit (BSU) is one of the largest groups of biostatisticians in Europe, and a major centre for research, training and knowledge transfer, with the mission “to advance biomedical science by maintaining an international leading centre for the development, application and dissemination of statistical methods”.
“Statistics is applicable in all aspects of medicine, epidemiology and public health,”
“Statistics should be the base for designing clinical trials as well as evaluating the effectiveness of public policies.”
– Prof Sylvia Richardson, Director of the MRC Biostatistics Unit, University of Cambridge since 2012.
-Current and recent research on evidence synthesis, policy
evaluation, new models and trial designs, and association
Current and recent research on evidence synthesis, policy evaluation, new models and trial designs, and association between genetic information in health, has had direct impact and influence on clinical practice and public health.
Policy breakthrough in drug-related deaths research
Professor Sheila Bird’s research team at BSU was first to quantify the high risk of drug-related death soon after prison-release, now internationally recognised; and more recently in the 4 weeks after hospital-discharge. Naloxone, the opiate agonist, is used by paramedics to reverse heroin overdoses. Could take-home naloxone be a solution? In 2012, randomization began in the prison-based N-ALIVE pilot trial to find out if naloxone-on-release could reduce overdose deaths soon after release from prison. The N-ALIVE trial did not randomize in Scottish prisons because take-home naloxone was made a funded public health policy in Scotland from 2011. Professor Bird’s advice was accepted that the primary outcome for Scotland’s National Naloxone Policy should be comparison of the proportion of opioid-related deaths that had a 4-week antecedent of prison-release. The proportion fell dramatically from 9.8% in 2006-2010 (193/1970 opioid-related deaths) to 6.3%
(76/1212) during 2011-2013. In the light of these results from Scotland and the N-ALIVE’s own data, the N-ALIVE pilot Trial ceased randomizing in 15 English prisons on 8th December 2014 and offered naloxone-on-release to its already randomized but not-yet-released prisoners.
A model free way of escalating doses in a dual agent phase I trial
The future of drug development in oncology is increasingly to use multiple drugs in combination. This necessitates research in how to design phase I trials to find the maximum tolerated dose combination. There is an increasing need to find a trial design that has solid
statistical operating characteristics, but yet is also easy to understand and use. Dr Adrian Mander’s research programme has developed
a new design that seeks to be simple. The design has the acronym PIPE which stands for the Product of Independent beta Probabilities dose Escalation, and uses statistical distributions in order to identify different toxicity contours over the possible drug combinations. The calculations are relatively simple and can be carried out using Excel, making it more accessible to trialists. The method was published early this year (Statistics in Medicine, 2015) and has already attracted interest from various clinical trials units and the pharmaceutical industry.
Statisticians in Roche Pharmaceuticals have now translated the R-code for the design into a public web application, available soon.
Relaxing the independent censoring assumption in the Cox proportional hazards model using multiple imputation
The Cox proportional hazards regression model is widely used to model time to event data; in 2014, Nature reported that “Cox 1972” is the 24th most cited research paper of all time. However the standard method for fitting this model requires the “independent censoring assumption”, that individuals lost to follow-up have similar survival to those remaining. Dr Ian White’s research programme has developed a
new way to relax this assumption, using easily interpreted sensitivity parameters, which describe the departure from independent censoring.
The new method was applied to data from a HIV trial in Tanzania, where the population of interest was nonpregnant women, so women who became pregnant during the trial were censored at their pregnancy times. Here, the independent censoring assumption is implausible, because the event of interest (HIV infection) and censoring (where due to pregnancy) share a common cause of unprotected sexual intercourse. This study showed that of three apparent predictors of HIV infection (age, alcohol consumption, mobility), only age could robustly be inferred to be a useful predictor. This method is widely applicable for sensitivity analyses.
BSU provides statistical expertise in stratified medicine, longitudinal modelling, clinical trials design, statistical genomics and subject-matter knowledge. June 2015 saw the official start of the MRC-funded MASTERPLANS Consortium in Systemic Lupus Erythematosus (SLE), involving BSU scientists Professor Vern Farewell, Dr Brian Tom and Dr Li Su. Its vision is to significantly improve clinical outcomes in SLE by increasing remission/low disease activity response rates to therapy through a stratified approach that relies on better understanding of key pathways and prognostic biomarkers. This adds to the existing substantial portfolio of public/private partnerships in stratified medicine, which includes rheumatoid arthritis (RA-MAP Consortium), one of the most common auto-immune diseases in the world.