1. The problem of assessing nutritional state
This paper emphasizes the need for measurement in diagnosis and recognizes two distinct problems, those of:
(a) Diagnosing specific diseases in the sense of distinct pathological conditions.
(b) Assessing general malconditions where there is no sharp dividing line between the normal and the subnormal.
It suggests a programme of research designed to overcome some of the difficulties inherent in (b) in the case of nutritional assessment.
2. Establishing general health standards
The first requirement is the scientific standardization of the elements making up the concept of health and good physique. This is a complex matter requiring all the help which statistical procedures can give. Moreover, we must check the reliability of the assessors and the indicators which they use before we can place any reliance on results purporting to relate to the population studied.
Provisionally we begin by collecting a battery of indicators of general health; each is then given a graded assessment for each child (and the assessment is repeated to obtain its reliability), and by employing the statistical device of factor analysis we can determine what combination of symptoms will give the highest agreement with the criterion of general health abstracted from these individually imperfect measures. We are actually making explicit the stages which are normally telescoped in the method of ‘general impression’ (which is the method usually adopted in assessing general health, state of nutrition, etc., despite evidence of its unreliability). In this method a number of individual characteristics are fused together without ensuring that different observers will attach equal importance to the signs or even that they will take into account the same ones. The emphasis to be given to each symptom in the final mark is determined statistically by its agreement with the aggregate result. Some indicators may give results so unrelated to the others that they must be eliminated. Other tests, whether physiological, functional or anatomical, may later be added to the battery to reinforce it. A point of practical importance is that some readily assessable symptom or index may be found which gives results in close enough agreement with those of the standard battery to enable it to be used as a substitute, thus shortening the task of diagnosis without appreciably lessening its accuracy.
An alternative criterion could, as we have seen, be the weighted combination of doctors' assessments, for by weighting the results according to each doctor's agreement with his colleagues we were able to increase the agreement between the team result and the hypothetical true mark. However, since the ordinary methods of assessment are unreliable, even the pooled results of the team do not give as good a standard as the standardized battery of tests.
A third standard is sometimes possible in the form of an objective accurately measurable physiological laboratory test or anatomical index. If this is inconvenient to apply generally, any single symptom or pool may be tested by correlation with it to see if the agreement is close enough for practical purposes.
3. Study of more specific nutritional defects
Factor analysis can be used to discover and assess more limited similarities than those covering the whole range of observers or test symptoms. For example, the doctors tended to show group affinities (depending possibly on their different emphasis in diagnosis) in addition to their somewhat low general conformity. Applying similar analytic methods to the study of symptoms, we might isolate specific nutritional deficiencies or it is possible that group factors may be revealed corresponding to the effects of recent or early adverse nutritional conditions respectively.
Whilst the examples of analytic techniques have here been applied to illustrative cases in nutritional and general health assessment, they are equally applicable to other fields of medicine, in particular to endocrinology and to the study of predisposing conditions of disease.
My thanks are due to Prof. Burt and Dr E. H. Wilkins for their suggestions for modifying my original draft. However, this in no way commits them to agreement with the views expressed here.
[Note added in Proof.] Since writing this article I have had an opportunity of correlating and analysing some results of nutritional surveys involving clinical and biochemical signs of malnutrition. The evidence so far obtained indicates a rather weak general factor for clinical signs identifiable with general nutritional state. I hope to be able to give the full results when the material becomes available for publication.
