By Stephen Senn
Everybody knows you need a medic to run a clinical trial but a statistician? What’s the point of one of those? The gruesome truth is that to measure what prolongs life we have to count the dead and those who have made a study of counting the dead are biostatisticians. They are crucial contributors not only to the analysis of clinical trials but also to their design, and this contribution, of course, should be noted in the International Year of Statistics.
But how can it be so difficult? In the long run we are all dead, as John Maynard Keynes famously remarked but he was only an economist. Any biostatistician could have told him that’s not the point. We can classify humanity into the dead and the not yet dead. Since, thank goodness, some patients entered onto a clinical trial will survive beyond the point at which the trial ends, we don’t know how long they will have lived when they have died. For example, we may have data from patients who have lived at least two years since they were entered onto the trial mixed together with those whose exact survival, say 13 months, is known. We don’t have time to observe Keynes’s long run if we want to help future patients soon. Thus we have to have a means of analysis that deals with what are known as censored observations: lifetimes of the form ‘at least two years’.
This topic is known as survival analysis and is only one of many to which statisticians have made extensive and influential contributions. Paul Meier (1924-2011), a famous American biostatistician, co-authored a paper on calculating survival curves with Edward Kaplan in 1958 that has received more than 40,000 citations. David Cox (born in the same year as Meier), a British statistician, introduced in 1972, a means of modelling survival that could take account of many complex influences, that has received more than 34,000 citations. Biostatisticians continue to improve and develop these techniques so that their medical colleagues can sharpen their inferences when it comes to the effect of treatments.
But survival analysis is not the only field where statistics is important in clinical trials. Indeed the very practice of randomly allocating patients to either receive a new treatment or an existing one was introduced by British statistician Austin Bradford Hill (1897-1991) in 1946 to study the effects of streptomycin on tuberculosis. As the 1948 publication of the trial in the British Medical Journal showed, the treatment was a great success but so was the design, which has been copied, developed and adapted in hundreds of thousands of clinical trials since. Hill had originally wanted to study medicine but the fact that he had himself contracted tuberculosis made this impossible. The streptomycin trial was perhaps his revenge on the disease!
Nowadays, wherever the effects of new treatments are being studied you will find clinical trials being run. Such trials, of course would be unthinkable without the participation of expert physicians. However, in the end, whatever the biological theory that led to the drug being tested, it’s the numbers that count and counting the numbers will be a statistician.
Senn is a Swiss statistician working in Luxembourg where he heads the Competence Center for Methodology and Statistics at the CRP Santé, a publicly funded research centre. He is the author of Dicing with Death, published by Cambridge University Press, 2003