Donald Mainland

NORMAL VALUES IN MEDICINE

1. The problem is not something new that came in with computers. It was present even before many medical people realized that there was a problemfor example, in my student days in the early 1920’s, when I was taught that the normal red cell count for men was 5 million per cubic mm, and 4.5 million for women. We do better than that nowadays, but how much better? 2. The problem concerns not only measurement data, but data of the yes-no type and graded values, such as 0, 1+, 2+, and 3+. 3. The problem of how and where to make cutoff points in a series of “normal” or “control” values is difficult enough, but it is, in my opinion, a much more difficult thing to decide what subjects shall provide the standard series, and then, having decided, to find suitable subjects. 4. Even apart from gross laboratory errors, which are still prevalent, good laboratories, using the same technique on the same material, can obtain different results. In addition, the subjects that meet the criterion of “normal” can differ in different places. Thus arises the question: Is it better to use “local norms,” derived from fewer subjects and therefore less precise, or to pool the results from several centers and risk having a range of variation that may be inapplicable to any one center?

The risk of biased selection in forward-going surveys with nonprofessional interviewers

Abstract A questionnaire, given to a class of 129 first-year medical students, revealed that more than one-half of them held opinions which, if allowed to influence selection of subjects in a forward-going etiologic survey, would bias the results. This indicates a risk of bias in the selection by nonprofessional interviewers, but there appears to be no possibility of eliminating bias by training. The only safe method is automatic selection.

AN EXPERIMENTAL STATISTICIAN LOOKS AT ANTHROPOMETRY

The purpose of this paper is to record some of the reactions that occur when the content and methods of anthropometry are looked a t by one who is now an experimental statistician in medicine, but was formerly an anatomist sufficiently concerned with human morphology to become a member of the American Association of Physical Anthropologists. The remarks can be somewhat systematized by examining each of the key words of the title.

Statistical ward rounds--1.

Dr. Mainlands statistical “Notes” have heretofore been sent from time to time to a select group who have enjoyed and benefited from them. In this issue we welcome Dr. Mainlands “Statistical Ward Rounds” to CLINICAL PHARMACOLOGY AND THERAPEUTICS, for now the select group is being expanded to include the readership of this JOURNAL. The JOURNAL will be all the more useful for this additional feature which will appear in each issue. We hope that the “Rounds” will stimulate correspondence. While Dr. Mainland is not offering a consultation service, some letters will be reproduced and answered in these columns, some directly. Please write to Dr. Mainlands unit.

Mortality of bereavement

This is the title of a report12 by a medical practitioner and a statistician on a study of the mortality among persons who had recently lost a close relative by death. The summary ends thus: “People who die following a bereavement are on average slightly younger than the relatives who predeceased them, and they die at an earlier age than is usual for the community in which they live.” It was this sentence that prompted the Editor of this JOURNAL to send me a copy of the report with the following comment: “The last sentence floored me. However I look at it, it means that either (1) an overwhelming number of persons die without anyone caring at all, or (2) there is a progressive and inevitable downward trend in the lifespan of communities sharing this phenomenon. What is the flaw?”

Statistical ward rounds—6

In most of the preceding rounds we have, somewhat to my surprise, hardly moved around the ward at all, because there seemed to be so much to say about individual cases. In this round I propose to mention briefly many cases, hoping to return to them later, if readers will help.

Statistical ward rounds--15.

At a recent symposium offered to specialists in the rheumatic diseases I heard a paper on juvenile rheumatoid arthritis ( Stills disease) which is defined as a form of the disease that occurs before the age of puberty. The speaker was concerned chiefly with diagnosis, signs, symptoms, and therapy, but he remarked that the disease commonly disappears and seldom returns in later life. This statement was questioned by a member of the audience, and the speaker replied that figures indicating recurrence often came from chronic disease units, not from centers that had followed their patients for a long time, such as his own unit, in which the follow-up had averaged nine years. As I listened I found myself applying the kind of excessively simplified and exaggerated arithmetic that is useful in numerical problems that seem simple but often contain traps for the unwary. I visualized one patient who had been followed for 1 year after the disease had disappeared and another patient who had been followed for 17 years. The average (arithmetic mean) follow-up would be 9 years. Then I enlarged the picture to include many patients, with various durations of follow-up, with increasing loss of contact year by year, and

Not-quite-random assignments

A problem of random assignment is discussed.

Statistical ward rounds‐l0

There is probably no disease in which symptoms, and even laboratory values, do not fluctuate to some extent, and in some diseases, such as rheumatoid arthritis, wide and unpredictable variations are a well‐known characteristic. Let us visualize a drug trial in which the patient but not the ph~sician is blindfold. When a patient is domg well on the drug he is transferred to the placebo, or when he is doing poorly on the placebo he is transferred to the drug, or both maneuvers are practiced. A statistical test is applied to the data and if the patients, reckoned either by a majority frequency or by an average of measured differences, did “significantly” better on the drug than on the placebo, the verdict is proclaimed in favor of the drug.

Averages and individuals

A draft report of a small trial. After the completion of a blindfold two‐group trial, with random assignment of a drug and a placebo, I submitted the draft of areport to the clinical investigators. The numbers in the trial had been small (21 on placebo and 22 on the drug), largely because the drug had been widely used for a number of years and patients who had ever received it were disqualified from the trial. Of the very few previous controlled trials of the drug the largest, very carefully conducted, had revealed, in most of the criteria of assessment, a small average superiority of the drug over the placebo, “significant at the 5 per cent level.” The much sm aller trial, on which I reported, had shown similar average differences, but in none of the clinical criteria were the differences “significant.”