37
mortality differences between the series. An alternative would
be to use the log-rank test, as suggested by Dr Haybittle, because this provides a comprehensive test of significance of the difference between the survival curves over their entire duration. Far from being inappropriate, as the authors suggest (June 12, p. 1291), the log-rank test represents the most satisfactory way of assessing fairly the reality of the mortality difference which they claim to exist. Mr McDonald and his colleagues estimate that the probability of occurrence of the observed mortality difference at one year between those lymph-node positive patients given routine postoperative radiotherapy and those not given radiotherapy routinely, if there is really no mortality difference between the series, is less than 0.05. However, because their calculations are inappropriate this estimate is likely to be too low, and perhaps much too low. The evidence the authors have presented for the higher mortality after routine postoperative radiotherapy in breast cancer patients with lymph-node involvement must therefore be regarded as much less strong than they claim, and it seems to be inadequate to justify the categoric statements in the discussion and summary of their paper. There is little advantage in having "direct evidence from a properly controlled clinical trial" unless that evidence is also properly evaluated statistically. M.R.C. Statistical Research and Services Unit, University College Hospital Medical School, London WC1E 6AS
IAN SUTHERLAND
to apply the same objectivity to statements about general practice that they would expect to be applied to therapeutic trials within their own specialty. Paulley4 asserts that "while many of one’s general practitioner colleagues are excellent ... quite frankly there are as
sultants
many who are not". How does he know? What criteria did he apply to end up by finding 50% excellent and 50% not? Elsewhere in his letter he asserts that G.P.s rush to the "consultant’s doorstep when they or their families take ill". Some may, but how many? Does he know? I suspect not. In our five-
partnership each of us chooses a partner with whom he family are registered. The partner who looks after my and does so superbly, uses the same criteria for referral family, to hospital as he does for all his patients; and so do the rest
man
and his
of us.
Paulley says that my figure of 10.6 new outpatients per week for each general physician is "undoubtedly wrong". I checked again and found that inadvertently I had stated that 10.6applied to whole-time equivalents: in fact it is the correct figure for the total number of consultants. The figure for whole-time equivalents is 12. 1. I apologise for the error. But Paulley’s suggestion that I should "check with consultant colleagues" to arrive at the "true" figure is so unscientific as to be ridiculous. The source of the data on which the figures are based is shown in the references at the end of my paper. Paulley’s final suggestion that either I or you (or both of us) deliberately distort the truth for political reasons is too absurd to be insulting.
OUTPATIENT FOLLOW-UP
SIR,-My paper and your editorial2 have provoked a number of letters that should be answered. There has never to my mind been any question of the importance of outpatient follow-up for uncommon serious disease such as renal failure3 or Crohn’s disease4 or certain other longdisease including those with multiple pathology.33 I also realise the educational value of follow-up appointments, especially in a specialty such as neurology.s The follow-up appointments that were considered unnecessary were the mindless routine appointments of no educational value, and of no advantage to the patient. Coggon and Goldacre6 illustrate very well indeed that while there may be a case for outpatient follow-up for serious complicated cases of common illnesses (such as appendicitis), routine follow-up of the rest is not only unnecessary but also ineffective as a guard against complications. I recognise that from the point of view of care after discharge from hospital (and from other points of view) there are bad G.P.S just as there are bad barristers and butchers, dentists and dockers, or members of any other profession or occupation. Olsen’s plan of identifying bad G.P.S in the patient’s notes by a code word seems to me a sensible plan compared to those who advocate routine outpatient follow-up because all or most or 50% of G.P.s are not to be trusted.4 The question of the quality of care provided by general practitioners often underlies statements in defence of the unnecessary follow-ups. Let us suppose that we could all agree on the criteria for measuring the quality of care provided by each G.P. (and criteria based on G.p./hospital contact would be only a part since this forms only a minority of our work). It would then, in theory, be possible to classify G.P.S into good, mediocre, and bad. This has not been done, and it surprises me how often consultants make sweeping statements about general practice (a branch of medicine of which they seldom have any postgraduate experience) based soley on subjective impressions. I would plead with conterm
major
1 Loudon, I. S. L.Lancet, 1976, i, 736. 2. ibid.p.1168. 3 Kerr, D. ibid. p. 1287. 4. Paulley, J. W. ibid p. 1347. 5. Matthews, W. B. ibid. p. 1287. 6. Coggon, D., Goldacre, M. J. ibid. p. 1346. 7. Olsen, N. D. L. ibid. p. 854.
Everything I have written stems from my belief that the future of medicine in Britain is grim unless it is firmly based on general practice of high quality. The unnecessary routine follow-up does four things: (a) it leads, to overcrowding in outpatients; (b) it fails to be an effective guard against serious complications; (c) it removes all incentives for G.P.S to undertake the care of their patient on discharge from hospital; and (d) most of all, it devalues general practice in everyone’s eyes and undermines the mutual respect and confidence on which the care of our patient depends. Health Centre, Garston Lane,
Wantage,
I. S. L. LOUDON
Oxon OX12 7AS
SEROTYPING OF E. COLI
SIR,-We have expressed concern’
at the suggestion that of Escherichia coli is no serotyping longer necessary in the study of infantile enteritis and that the ability to detect enterotoxin may be all that is required. We welcome the letter by Dr Sackin which he supports the value of serotyping in the study of epidemic diarrhoeal disease. It has been shown in numerous studies that outbreaks of infantile enteritis in many countries are frequently caused by E. coli belonging to a restricted range of serotypes.3 Dr Sack doubts the value of serotyping E. coli from sporadic cases of diarrhoea in infants and children, but this argument does not take into account the common epidemiological pattern of hospital outbreaks of infantile enteritis. Outbreaks frequently occur in units housing young babies, and epidemic spread of infection follows the admission of a sporadic case. During a survey in a Dublin hospital4 E. coli 0142.H6 was found in 13 of 70 apparently sporadic cases of diarrhoea on admission to hospital. A cross-infection outbreak subsequently occurred, and 49 of 68 babies who developed diarrhoea after admission were shown to have acquired E. coli 0142.H6 whilst in hospital. Recognition of an enteropathogenic serotype in sporadic cases of diarrhoea on admission to hospital allows the 1. Rowe, B., Gross, R. J., Scotland, S. M. Lancet, 1975, ii, 925. 2. Sack, R. B. ibid. 1976, i, 1132. 3. Taylor, J. J. appl. Bact. 1961, 24, 316. 4. Hone, R., Fitzpatrick, S., Keane, C., Gross, R. J., Rowe, B. J. med. Microbiol. 1973, 6, 505
38
implementation of spread of infection.
measures
which may prevent
epidemic
Dr Sack has criticised
our report5 that strains of E. coli traditional belonging enteropathogenic serotypes isolated during epidemics of infantile enteritis failed to produce enterotoxin detectable in laboratory tests. He suggests that these strains may have lost a plasmid controlling the production of enterotoxin during storage. This is certainly a possibility and is supported by Raska and Raskova.6 However, Falkow reports that in a study of 100 enterotoxigenic strains of E. coli only 21% could be shown to harbour enterotoxin plasmids;7 the possibility therefore remains that in some strains the determinants of enterotoxin production may be chromosomal, as is the case in Vibrio cholerae.8 We have lately tested strains of E. coli 0148.H28 isolated during 1965 from 10 diarrhoea cases in an epidemic among British troops in Aden,9 and the strains from all 10 cases remain enterotoxigenic even after storage on Dorset egg medium for over 10 years. Similarly, the widely used enterotoxigenic control strains B7a and H10407 have remained enterotoxigenic for many years even though enterotoxin production in strain H10407 is known to be controlled by a transferable plasmid.1O These strains may represent a stable, strongly enterotoxigenic group frequently causing diarrhoea of adults in the tropics, and such strains might be expected to give consistently positive results in all the available test systems. Strains associated with infantile enteritis in temperate zones seem to behave differently. The controversy over the relative value of serotyping and enterotoxin testing remains unresolved. Further studies are required of E. coli strains recently isolated from cases of infantile enteritis, especially those from epidemic situations. These strains should be tested for enterotoxin production and subjected to detailed serotyping using the complete range of antisera. We shall be pleased to receive such strains.
to
Salmonella and Shigella Reference Central Public Health Laboratory, London NW9 5HT
Laboratory
B. ROWE R. J. GROSS
SYLVIA M. SCOTLAND
"NORMALISED" VALUES SAVE CLINICIANS’ TIME an addendum to the controversy in your columns SI units and in particular to the suggestions that it would be better to avoid the use of units altogether and use some method of presenting results in terms of their divergence from a "normal" or reference value,I1-14may I offer the results of a small experiment designed to find out how a mixed sample of medical users would cope with data presented to them in one
SIR,-As
over
such way? I made out a series of pairs of cards, each in the format of a cumulative biochemical report representing daily values for seven substances for one patient over eight days. In each pair, one gives the values as though they were "real" values (omitting units for simplicity) and the other the same figures converted to "normalised" value-i.e., the number of standard deviations above or below the mean. On the "real values" card the normal range was given for each substance: on the "normalised" card the mean and the standard deviation. The accompanying examples taken from different pairs will enable any reader who wishes to try the experiment on himself. The experiment was tried on 26 colleagues, mostly medicals but including a few B.sc. research assistants and computer 5. Gross, R. J., Scotland, S. M., Rowe, B. Lancet, 1976, i, 629. 6. Raska, K., Raskova, H. ibid. p. 1300. 7. Falkow, S. Infectious Multiple Drug Resistance. London, 1975. 8. Vasil, M. L., Holmes, R. K., Finkelstein, R. A. Science, 1975, 187, 849. 9. Rowe, B., Taylor, J., Bettelheim, K. A. Lancet, 1970, i, 1. 10. Gyles, C., So, M., Falkow, S. J. infect. Dis. 1974, 130, 40. 11. Lennox, B. Lancet, 1975, ii, 1085. 12. Wright, B. M. ibid. p. 1261. 13. Gräsbeck, R. Lancet, 1976, i, 244. 14. Ransil, B. J. ibid. p. 245.
e
ranged from students to very senior consulphysicians. Each was briefly told of the object of the experiment, and the normalisation process explained (practically all knew at least in general terms what a standard deviation was). They were shown examples of the cards, and told that "normal" for the purpose of this experiment meant "within the given normal range" for one type and "within 2 S.D.S of the mean" for
staff. The medicals tant
the other.
They were then handed a card of one or other type and asked three questions: (a) which substances have remained normal for all eight days?; (b) is there any single value on the card which is so discrepant from the general trend of other values for that substance that it looks like a mistake?; (c) is there a substance on the card for which all values are within the normal range, but either all are high or all are low? Time to produce each answer was measured with a stopwatch. A second card of the other type (but not of the same pair) was then presented, and the process repeated. The results in 26 subjects were:
It takes
on
average rather less than half the time
to
find the
right answer with normalised values, and the error rate is ven greatly reduced. What was particularly striking is the fact that not once did any subject answer a question on real values faster than the corresponding one on normalised values. It made surprisingly little difference which type of card was offered first. Of
course a single experiment by a prejudiced observer ought not to be accepted as complete proof. But this seems to be strong evidence that users would find little difficulty ir adapting to a device of this kind, and that it would save din)
cians’ time and reduce their error rate. It is perhaps worth saying that the word "normalisation" u convenient statistical jargon for this method of presenune data, and has no reference to the largely unprofitable dispute that surround the word "normal". Department of Pathology, Western Infirmary,
Glasgow G11
6NT
BERNARD LENNOX