Bryan Shepherd's Ideas

I understand "translational science" to be going from the lab to what is actually practiced in clinics. A piece of this could be translating results from randomized controlled trials (RCT) to what will happen in practice. We perform RCT because they are the best thing at helping us infer causation -- however, the study populations may not always be realistic or the clinical techniques may be impractical to perform in a real setting. Observational studies obviously make causation difficult to assess, but they tend to look at more realistic populations and practice. We should bridge this gap. One way would be focusing on designing RCT in such a way that techniques and study populations are as realistic as possible. From a statistical perspective, some areas of causal inference research could be helpful: As an example, in preventative HIV vaccine trials, one often wants to see whether or not the vaccine helps participants who still become infected during the trial (are they are better off than they would have been had they taken the placebo). The ideal trial to assess the effect of vaccination on post-infection outcomes would be to randomly assign vaccine and placebo and then to infect all participants thru HIV challenges. Of course, this is not ethical. So in order to find the causal effect of a vaccine on a post-infection outcome, one must make assumptions and then perform sensitivity analyses to these assumptions. I use this example because I am familiar with it, and also because it is a situation where the ideal trial cannot be performed but by using modern statistical techniques we can still obtain some information. Using sensitivity analyses, we may be able to take information from RCT and determine benefits in practice.

I'm rambling, and I'm not sure I've made my point very well (no am I sure I really have a point), but I believe the use of modern causal inference techniques with regards to compliance and dynamic treatment regimes / optimal treatment strategies, could be helpful for translational science.

Other thoughts:

• Focus on getting biostatisticians involved in all areas of the process -- particularly sample size and design, to avoid waste. Have a few full-time biostat faculty in the center and I would highlight the use of biostat clinics.
• Keep medical records in-house.
• More rapid/effective IRB process.