Customize your JAMA Network experience by selecting one or more topics from the list below.
While there have always been those who have cast discredit on the scientific value of statistics, it remains a fact that some medical knowledge must be derived from statistical investigation. Statistics may, of course, be juggled, and it is also fair to assume that statistics are often prepared by persons not skilled in the fundamental principles underlying their preparation. It is probably true that the bulk of medical statistics of the past has been prepared by medical men not trained as expert statisticians.
Recently Mr. Raymond Pearl,1 professor of biometry and vital statistics in the Johns Hopkins School of Public Health, has analyzed, from the point of view of a trained statistician, certain figures in a paper published by Head2 concerning the efficiency of various methods of treatment in pneumonia. In this paper Dr. Head himself suggested that the lowered mortality shown in favor of closed ward treatment might be merely a coincidence. In his analysis of the figures, Pearl shows that while Head’s conclusions are qualitatively correct, they are quantitatively out of the way on account of the neglect to take into account the factor of random sampling. Another neglected factor, frequently overlooked by medical writers, is the natural history of the disease under investigation. It has been asserted by numerous observers that the mortality from influenzal pneumonia at the end of an epidemic is usually much lower than it is at the beginning of the outbreak. Pearl shows that this is the case, and that Head, although recognizing the possibility, did not take it into account in evaluating his figures.
If the statistical method, first extensively introduced into clinical medicine by Louis and the French school, is of value, it goes without saying that the statistics which are used must be based on the well-recognized principles utilized by professional statisticians. So far as mortality statistics are concerned, it may be assumed that the correct methods are usually employed; but it is certain that this is not the case when ordinary clinical statistics are concerned. In any textbook on medicine or surgery, one may find numerous statements covering statistically such matters as the age at which certain diseases occur, the relative proportion of the sexes involved, the frequency of complications, and the relative frequency of different diseases in a given organ or system. It is quite certain from the figures presented that these statistics would be regarded as valueless by a professional statistician, and that while they are perhaps not valueless to the clinician, they are not nearly as valuable or as correct as properly prepared statistics would be. In differential diagnosis, as Southard has pointed out, it is desirable that the physician should know the possibilities. In 10,000 patients with convulsions, what proportion is likely to be due to epilepsy; what proportion to uremia; what proportion to general paresis, etc.? With correct knowledge on such a point, the physician knows when he encounters a case of convulsions that there are certain chances in favor of a given disease, and he can make what Southard calls a diagnosis by orderly exclusion, which is more satisfactory than the old-fashioned diagnosis by exclusion in which the probabilities were ignored. In the matter of treatment, too, the application of correct statistical principles would prevent the flooding of medical periodicals with the views of therapeutic optimists based on uncontrolled observations. The checking of medical statistics by trained statisticians will doubtless serve as a stimulus to more accurate statistical methods.
Physicians and Statistics. JAMA. 2019;322(18):1832. doi:https://doi.org/10.1001/jama.2018.15594
Create a personal account or sign in to: