From Rothamsted to Northwick Park: designing experiments to avoid bias and reduce variance

In March 2006, a clinical trial at Northwick Park hit the headlines when it went badly wrong. Rosemary Bailey explains what went wrong, and how statistics developed for one area of science dramatically improved the outcomes of clinical trials.

In March 2006 the topic of designed experiments briefly hit the newspaper headlines when a clinical trial went badly wrong. Even when the health of volunteers is not at stake, there are some important factors to consider in the design of any comparative experiment. If you want to find out how much school milk improves children's health, it is no good giving the milk just to the poorest children, no matter how laudable that may be on other grounds. To avoid bias, some children in each income group should be given milk while the others in each group are not.

As well as avoiding bias, we want to reduce variance, because smaller variance indicates that our estimate of treatment differences is more likely to be close to the true value. Statisticians thought that they knew how to combine variance reduction with avoidance of bias, but the explosion of microarray experiments in the early 2000s showed that our received wisdom was wrong.

Experience from designing experiments in one field of science can be applied to other fields, once the necessary vocabulary has been learnt. In particular, the variance in dose-escalation trials can be reduced by a factor of three without reducing safety or using more volunteers.

Register to tell a friend about this lecture.

Comments

If you would like to comment about this lecture, please register here.

©2003-2006 Niminim Limited. Created by

Any ad revenue is entirely reinvested into the Lecture List's operating fund