Statistical methods for randomised surgical trials

Randomised Controlled Trials (RCTs) have played a very important role in advancing health care by allowing accurate assessment of a new treatment by comparison with a standard treatment. The field of surgery has been slower to use RCTs than Medicine in general. There are number of features about surgery which make it more difficult to assess, and influence the planning stage of a trial, the day to day running, and also the analysis of the results. The surgeons undertaking the procedure may vary in both experience and inherent surgical ability. The availability of a surgical theatre and the surgical team are limited and therefore the surgery may be delayed. Surgery also differs is in the way that it is administered and a number of apparently valid variations of a particular procedure may be possible. Individually and collectively these issues have inhibited surgical RCTs. This work would look to address each of these issues from a statistical point of view and, by furthering our understanding, can facilitate the uptake and enhance the quality of RCTs in surgery. By doing so our knowledge of the best way to treat patients where surgery is an option will be enlarged.

The randomised controlled trial (RCT) is now widely accepted as the gold-standard design for the evaluation of health care interventions. Whilst the RCT has been accepted into medical practice for many years, there is continuing resistance to its uptake in surgical practice. The design and conduct of randomised controlled trials in surgery is undoubtedly more complex than drug trials for a number of reasons.

A programme of work on statistical methods in surgical randomised controlled trials is proposed to address key issues related to the design, conduct and analysis of surgical trials. This work will be undertaken in collaboration with experts in the UK, Germany and Canada. The specific aims of this work are:

To develop and evaluate new statistical techniques for quantifying expertise in randomised surgical trials
A Bayesian hierarchical models will be developed to investigate the relationship between (surgical) simulator results and various proxy measures (such as operation time) for predicting performance in a randomised controlled trial. Training on a laparoscopic simulator will be undertaken by a sample of surgeons prior to participation in the UKUFF trial.

To create a database of intracluster correlation coefficients for surgical techniques to inform study designs
This database will include intracluster correlation coefficients (ICCs) for a variety of surgical procedures and would also allow investigation of the relationship between surgeon effects and centre effects by comparing ICCs of surgeons and centres.

To build a Bayesian predictive model of recruitment patterns to take into account waiting list effects
The recruitment details from a set of surgical trials will be used as training data for producing and refining the recruitment model. Issues particularly relevant to surgical trials such as delays due to limited availability of resources, a waiting list effect, will be incorporated.

To develop and evaluate the impact of prior distributions for the treatment effect dependent upon different stages of the learning curve
Methods to statistically adjust the surgical trial treatment effect for the learning curve have been previously developed will be extended to incorporate a prior distribution for the treatment effect to investigate the likely influence of learning on the treatment effect.

To test the usefulness of efficacy estimators in randomised surgical trials
The influence of incorporating efficacy estimates within trial cost-effectiveness analysis and decision analytic models will be investigated. The properties of the complier average causal effect approach to estimating efficacy will be empirically investigated using a number of surgical trials.