In this book we have a brief history of "controversies over the use of statistical and probabilistic reasoning within medicine" and "attempts to determine therapy on the basis of statistical comparisons". The stated purpose is to show how "the use of comparative statistics in a therapeutic context" gradually evolved to the point where a double-blind clinical trial (involving random allocation of patients to case and control groups) became "a standard procedure for determining the efficacy of new drugs". But there is a hidden agenda stemming from the author's belief that "the medical practitioner and the medical researcher both share an antipathy towards methods of quantification or statistical inference"; and "that the clinical trial eventually triumphed less because of internal debate within the medical profession and more because society at large decided that, in an era of highly potent industrially produced drugs, the decisions of the medical profession would have to be regulated".
The subsidiary theme is difficult to reconcile with much that J. Rosser Matthews has unearthed. Thus, in 1937, it was The Lancet which asked Austin Bradford Hill "to write a series of articles on the proper method of applying statistics to medical concerns". And, when these were published as Principles of Medical Statistics they were "a resounding success". It was also at the request of the Medical Research Council that Hill designed a clinical trial of streptomycin in relation to tuberculosis. This too was immediately acclaimed as an "exemplary model of how to conduct an 'objective' trial by using randomisation". Finally, although Matthews is correct in saying that it was public protest against Thalidomide which led to the United States Drug Amendments Act of 1962 requiring proof of the efficacy of new drugs, he fails to mention an equally important fact, namely that recognition (by the Food and Drug Administration) that Thalidomide had not been subjected to a randomised clinical trial was the reason why this drug never reached the US pharmacopoeia.
The late arrival on the scene of medical researchers, who not only had a formal training in statistics but were also in a position to put this knowledge to good therapeutic use, has a simple explanation: it was not until after 1870 that statistics was transformed from an empirical social science into a mathematical applied science; clinicians had the precision tools needed for accurate diagnosis; and mathematicians finally freed themselves from the idea that their subject "could only be applied to natural phenomena under the category of causation".
The first influential mathematician to realise that we have in statistical methodology a universal tool of scientific inference was Karl Pearson. He was a professor at University College London from 1883 to 1922, and his conversion was the result of meeting Francis Galton who studied medicine and invented a method of correlation while studying factors related to heredity genius. Together with a fellow professor of zoology, Pearson created a school of biometry, offered the first advanced course in statistical theory and founded the journal Biometrika. Among Pearson's students were three individuals who subsequently pursued distinguished careers in medical research: Raymond Pearl, who was medically qualified but worked as a statistician at Johns Hopkins Hospital; Major Greenwood who was the first professor of epidemiology and public health at the London School of Hygiene; and Bradford Hill who was the second person to hold this chair.
Today there would be no difficulty in meeting the basic requirements of a randomised clinical trial, or conceding that, without quantification and statistical analysis of clinical and laboratory data, medicine would hardly rate as a science. But this was far from being the case when, in 1837, two French physicians publicly disagreed about the proper use of a crude numerical method. Or when a generation later medical researchers bypassed the mathematical and probabilistic concerns of an engineer-turned-physician and insisted that the way forward lay "in experimental induction and not in the numerical method". As late as 1920 a distinguished bacteriologist was insisting that "the mathematical considerations of statisticians are not of serious concern for the trained bacteriologist". But Greenwood, who was representing the opposite opinion, was able to show that without statistical methods of analysis the experimental method is positively dangerous.
In short, what Matthews sees as a special antipathy of clinicians towards the numerical method could be no more than today's equivalent of the earlier (19th-century) antipathy towards the experimental method.
Alice Stewart is senior research fellow, department of epidemiology, University of Birmingham.
Quantification and the Quest for Medical Certainty
Author - J. Rosser Matthews
ISBN - 0 691 03794 9
Publisher - Princeton University Press
Price - £32.00
Pages - 195