150 LI:l TERS

‘I WHO Tuberculosis Research Office (1956). Assessment of’BCG vuccinution irl seven countries of Ask, Copenhagen (Mimeographed document).

d WHO Tuberculosis Research Office (I 957). Assessment of BCG vaccination in India, second report. Bulletin qf’the World Health Organization, 17,203.

h WILLIS, S. & VANDIVIERE, M. (1961). The Heterogeneity of BCG. American Review of Respiratory Diseuse, 84,288. c, BETTAG. 0. L., KALUZNY, A. A.. MORSE, D. & RADNER. D. B. (1964). BCG study at a state school for mentally

retarded. Diseases ofthe Chest, 45,503. i COMSTOCK, G. W. &PALMER, C. E. (1966). Long-Term Results of BCG Vaccination in the Southern United States.

American Review of Respiratory Disease, 93;171. 8 COMSTOCK. G. W. & WEBSTER, R. G. (I 969). Tuberculosis studies in Muscogee County, Georgia. VII. A twenty-

year evaluation of BCG vaccination in a school population. American Review of Respiratory Disease, 100, 839. !) SPRINGETT, V. H. (1971). Tuberculosis control in Britain 1945-1970-1995. Tube&e, 52,136.

rD SUTHERLAND, I. (1971). Future policy for BCG vaccination in Britain. Postgraduate Medical Journal, 47, 756. I’ Medical Research Council (1963). BCG and vole bacillus vaccines in the prevention of tuberculosis in adolescence

and early adult life. British Medical Journal. 1,973. I2 ARONSON, J. D. & PALMER, C. E. (1946). Experience with BCG vaccine in the control of tuberculosis among North

American Indians. Public Health Reports, Washington, 61,802. I5 ARONSON, J. D.. ARONSON, C. F. & TAYLOR, H. C. (1958). A Twenty-Year Appraisal of BCG vaccination in the

control of tuberculosis. 4rchives of Internal Medicine, 101,881. I1 SPRINGETT, V. H. & SUTHERLAND, I. (1970). Comparison of the efficacy of liquid and freeze-dried strains of BCG

vaccine in preventing tuberculosis. &fish Medical Journal, 4, 148.

In your leading article on ‘Fifty Years of BCG’, you refer to a paper of mine1 on the future policy for BCG vaccination in Britain and in particular to the hypothesis advanced there that the degree of protection from vaccination may be diminished in areas where the incidence of tuber- culosis is low. Unfortunately some of the evidence for the hypothesis presented in that paper is affected by a statistical fallacy. I would therefore like to explain how the evidence is affected, and to comment on your leading article in the light of this reappraisal.

In the paper’, it was noted that the percentage efficacy of BCG vaccine was positively associated with the incidence of tuberculosis in previously uninfected and unvaccinated subjects in the same area. This association was found in the results of seven controlled trials of BCG vaccine in different parts of the world, and also within the MRC trial in Britain, when the data were examined for 16 separated sub-areas included in that trial.

However, by relating the percentage efficacy to the incidence of tuberculosis in the unvaccinated group alone, any relationship between efficacy and the amount of tuberculosis will be obscured to a greater or lesser extent by a spurious correlation. This is the intrinsic correlation between any compound variable (in this instance the ratio of the rates in the vaccinated and unvaccinated groups) and one of its parts (the rate in the unvaccinated group). In the present instance a chance fluctuation which increases the unvaccinated rate above its true value will reduce the ratio of the vaccinated to the unvaccinated rate, and vice versa; the effect of the chance variability of the unvaccinated rate is consequently to create a correlation which will be superimposed on, and which will therefore obscure, a genuine relationship between the efficacy of vaccination and the incidence of tuberculosis in the area. (Oldham describes and discusses an analogous example of such a correlation.)

The misleading effects of this correlation can be avoided by relating the percentage efficacy instead to an average (namely the geometric mean) of the incidences in the vaccinated and unvac- cinated groups. When this is done the close concordance between efficacy and the incidence of tuberculosis in the seven controlled trials remains. Regarding the two trials in Georgia as one, the chance of as close a concordance is still only 1 in 60.

A corresponding examination of the data for the 16 different sub-areas of the MRC trial shows that the association in this instance was entirely due to the spurious correlation described above. This, however, does not necessarily mean that there is no relationship between efficacy and tuber- culosis incidence. Further analysis shows that the incidence of tuberculosis did not vary greatly between the 16 sub-areas (apart from fluctuations attributable to chance). Where there are no clear geographical differences in tuberculosis incidence there is obviously little scope for investigating

LETTERS 151

whether the efficacy of vaccination is associated with incidence. It follows that the MRC trial can contribute little firm evidence eitherfor or against the suggested relationship.

Because of other differences between them, the association shown by the trials in different parts of the world cannot by itself be regarded as establishing the existence of a genuine relationship between the efficacy of vaccination and the incidence of tuberculosis. But if there is such a relation- ship, the explanation might be that the superinfection of vaccinated subjects with virulent tubercle bacilli, or with other mycobacteria, is necessary to maintain the protection conferred by BCG vaccine. In this connection it is of interest that Wigley3, in studying a community in New Guinea which was virtually free of tuberculosis and of other mycobacterial infections, found a steady and very substantial loss of tuberculin sensitivity among vaccinated subjects over a three year period, whether those subjects had received Australian, Japanese or British BCG vaccine. This indicates the probable role of superinfection in maintaining the tuberculin aZZergy conferred by BCG vaccine, although it leaves open its role in maintaining protection. Until there is further and more direct evidence, the suggested relationship between superinfection and protection must remain only an intriguing possibility.

In your leading article you quote two estimates 1,4 of the expected incidence of tuberculosis in Britain during the next decade. You then state that if the findings of the current national tuberculin survey confirm the essential accuracy of these estimates ‘mass BCG vaccination should be stopped’ in Britain. This conclusion could be justified if it were known that the suggested relationship was genuine and also - a crucial proviso - that protection was negligible at levels of incidence corres- ponding to these estimates. It is not justified in a situation in which neither point can be regarded as proven.

It would be wiser at present to assume, as SpringetP did, that the protective efficacy of BCG vaccine during the next decade in Britain will be similar to that found in the MRC trial. Even with this high level of efficacy, the estimates of the absolute numbers of cases which will be prevented annually by mass BCG vaccination in this country are small, and decreasing4. The time will soon come when it will be appropriate to abandon mass BCG vaccination, but this may not be as immi- nent as your leading article suggests.

MRC Statistical Research and Services Unit, 115 Gower Street, London WCI E 6AS

IAN SUTHERLAND

REFERENCES 1 SUTHERLAND, I. ( 1971). Future policy for BCG vaccination in Britain. Postgraduate Medical Journal, 47,756. 2 OLDHAM, P. D. (1962). A note on the analysis of repeated measurements of the same subjects. Journal ofChronic

Diseases, 15,969. 3 WIGLEY, S. C. (1970). Post vaccination (BCG) conversion reactions in a tuberculosis-free community. Papua und

New Guinea Medical Journal, 13,33. 4 SPRINGETT, V. H. (1971). Tuberculosis control in Britain 1945-1970-1995. Tubercle, 52,136.