I recently returned from Phoenix where I summarized a consensus conference on Comparative Effectiveness Research (CER). The conference was sponsored by the American College of Sports Medicine (ACSM), one of my favorite clients. Among other things, the conference focused on the idea that unlike randomized control trials (RCTs), long considered the gold standard in clinical research, factorial designs allow you to construct adaptive interventions. As I understand it, at the heart of these study designs are a form of game theory and bayesian modeling. I know little on either topic so I asked my colleagues from the American Medical Writers Association what they knew.
One of my AMWA colleagues let me know that Bayesian methods have the potential to significantly reduce sample sizes, and therefore research costs. Game theory isn’t used all that much for adaptive clinical trials except in the form of statistical decision theory. A list of references on Bayesian inference and decision theory is provided at the bottom of this blog post (thanks to my AMWA colleague Robert Ryley).
Another AMWA member stated that within the medical device community people turn to Don and Scott Berry for information on adaptive clinical trial design. Check out Berry Consultants website. Another trusted AMWA colleague pointed me toward a recent NPR article, in which Stuart Kauffman states, “…when RCTs work, they do really work, often well. But they often fail in complex biological-medical situations where causality is multifactorial, as it typically is. In place of RCT, our group has found a better alternative in these cases which we call ‘Team Learning.’” Finally an AMWA member said that while she couldn’t comment on adaptive clinical trial design, she did know that game theory, particularly a form of “crowd sourcing”, is being used in the design of diagnostic algorithms.
The Patient-Centered Outcomes Research Institute (PCORI) will be awarding grants in CER through 2019. What are your thoughts on applications of Bayesian principles and decision theory to medical research?
Bayesian Inference and Decision Theory References-
Provided by Robert Ryley (rryley1976@GMAIL.COM)
Here is a list of decision theory references with a numerical ranking of the mathematical competence required to understand the text:
1. High School Level — elementary algebra or geometry at most
2. Undergraduate Level text — Multivariable calculus helpful. Basic calculus a must.
3. Advanced Undergraduate/Graduate Level Text — Complex Analysis and Measure Theory presumed
Title: Elementary Decision Theory
Initially Published: 1959
Authors: Herman Chernoff, Lincoln Moses
Difficulty Rank: 1
Summary: A classic in decision theory that provides a very good introduction to basic frequentist statistics and their application to decision problems. It has been kept in print by Dover publications and is very cheap compared to more modern texts. One thing to keep in mind — the data analysis methods described here were for people who only had pencil and paper as an aid. Start here first if your understanding of basic undergraduate statistics is a bit rusty.
Title: Games and Decisions: Introduction and Critical Survey
Authors: R. Duncan Luce, Howard Raiffa
Initially Published: 1957
Difficulty Rank: 2
Summary: Provides more intuitive justifications for game theoretic reasoning. The authors wrote the text to broaden the knowledge of these methods for social scientists. Apparently, this text was used by John Nash when he taught an intro game theory course. A more modern, and less mathematically oriented approach is also provided in the book Negotiation Analysis, which has Howard Raiffa as one of the co-authors.
Title: Theory of Games and Statistical Decisions
Authors: David Blackwell, M.A. Girshick
Initially Published: 1954
Difficulty Rank: 3
Summary: I believe this text was widely used in graduate-level statistics programs. At the very least, it is widely cited in more modern statistics books. It is probably overkill for most of us, but those who want to understand how statistical procedures are evaluated by experts will likely want to study this. I have a copy lying around somewhere. This has also been kept in print by Dover publications.
Title: Statistical Decision Theory and Bayesian Analysis
Author: James O. Berger
Initially Published: 1985
Difficulty Rank: 3
Summary: A highly recommended graduate level text in statistical decision theory. Although it continues to be used in graduate-level statistics programs as far as I know, it could be followed by someone with college-level algebra and calculus who has persistence to work through the examples and look up what is unfamiliar.
Title: Bayesian Data Analysis: A Tutorial
Authors: D.S. Silva, John Skilling
Initially Published: 2006
Difficulty Rank: 3
Summary: This book was designed for undergraduates in science and engineering. It encourages thinking in probabilistic terms and shows how to apply mathematical methods commonly used in engineering toward statistical problems. Physicist and prominent Bayesian protagonist E.T. Jaynes recommended this book as a complement to his more theoretical book Probability Theory: The Logic of Science.
Title: Introduction to Applied Bayesian Statistics and Estimation for Social Scientists
Author: Scott M Lynch
Initially Published: 2007
Difficulty Rank: 2
Summary: A very good intro for social scientists — psychology, sociology, etc. The author provides some basic methods from calculus and matrix algebra for those who are lacking in this area. It will certainly help you bridge the gap from conventional frequentist methods toward a Bayesian way of thinking about problems.
Title: Bayesian Adaptive Methods for Clinical Trials
Author: Scott M Berry
Initially Published: 2010
Difficulty Rank: N/A
Summary: I have not purchased this book, but it is certainly on my wish list.