BRITISH MEDICAL JOURNAL
VOLUME 293
documented as an objective record of follow up.
These photographs are always taken through
dilated pupils since the failure rate is otherwise
unacceptably high.
We would deprecate strongly the suggestion that
a camera of this nature can replace the proper
examination of the diabetic eye, through dilated
pupils, by direct ophthalmoscopy. The only safe
alternative to clinical examination is to use the
camera to photograph the peripheral retina as well
as the macular areas: this is technically possible but
in our experience is difficult and time consuming,
requires at least four photographs taken by a
competent photographer, is less accurate than
ophthalmoscopy, and gives unsatisfactory films in
some patients. Using the camera in this fashion will
effectively negate any advantages in time and cost
effectiveness which are claimed for the machine.
Thus the advent of the camera does not, and
should not, remove the onus placed on the diabetologist to justify his title as a doctor with special skill
in diabetes. The title implies a competence to
examine the eye clinically and detect and assess the
importance of retinopathy. For physicians who
feel that they lack this competence time is better
spent in becoming adept with the ophthalmoscope
rather than poring over photographs-or sending
them to the ophthalmologists to interpet.
T BARRIE
A C MACCUISH
Diabetic Unit, Glasgow Royal Infirmary, Glasgow G3 OSF
1 Scobie IN, MacCuish AC, Barrie T, Green FD, Foulds WS.
Serious retinopathy in a diabetic clinic: prevalence and therapeutic implications. Lancet 1981;ii:520-1.
2 Foulds WS, MacCuish AC, Barrie T, at al. Diabetic retinopathy
in the west of Scoland: its detection and prevalence, and the
cost-effectiveness of a propoaed sreening programme. Heakh
BuL (Edinb) 1983;41:318-26.
3 Deckert T, Poulsen JE, Larsen M. Prognosis of diabetics with
diabetes onset before the age of 31. Diabewtoa 1978;14:
363-70.
Informed consent and the newborn
SIR,-Since Dr Richard Nicholson (25 October,
p 1099) avoids the main issue raised in my letter
(30 August, p 562) 1 must challenge him to answer
it. Randomised trials of different treatments are
fundamental to our efforts to reduce mortality and
handicap in babies. When it is impossible to obtain
prior informed consent from parents of sick babies
should such babies be excluded from trials or not?
I and many other paediatricians would welcome
wide public debate about this, since if trials are not
done or such babies are excluded from them
advances in treatment will inevitably be slowed.
When prior discussion with parents cannot take
place because they are absent, unconscious, or in a
state of shock a choice must be made between two
priorities: the need to prevent death and disability
or the rule of prior informed consent. Faced with
this dilemma I think that society and parents in
general will want trials to go ahead, provided they
have the considered, independent approval of a
properly constituted ethics committee (not composed of doctors alone).
Some comment is needed on tne headlines about
"experimenting on babies" for which Dr Nicholson
acknowledges responsibility. For many, far from
the rigorous comparison of treatments entailed in a
randomised trial, that phrase implies the infliction
of unacceptable procedures to which few parents
would ever consent. Perhaps Dr Nicholson did not
intend that the public should be misled by this
ambiguity and the distress which it has inevitably
evoked. If he did not he should certainly say so.
Those headlines have clouded a particularly sen-
sitive issue.
Oxford OX4 2NL
15 NOVEMBER 1986
Computer aided diagnosis of acute
abdominal pain
SIR,-The findings of Dr I D Adams and others
(27 September, p 797) seem to have promoted the
opening of many an old wound (18 October, p
1025). We believe that the abdominal pain diagnosis trial should be assessed for its positive
contribution to medical practice, without prejudice. Although it may be difficult for many
doctors with traditional training to accept that the
computer has any contribution to make in medicine,' other than perhaps in administration, it
would be a mistake to discard the potential advantage of its use in diagnostic support without reference to the evidence.
The role of the computer in this study should not
be underestimated. It was used, appropriately, to
examine and remember the fine detail of several
thousand hospital case records. It was then able
to define precisely, for each age and sex category,
the relation between signs, symptoms, and the
diseases of acute abdominal pain. As the human
information processor is not capable of handling
such quantities,2 it could perhaps be argued that
each doctor uses a best estimate of this information
when tackling diagnosis. It could be concluded
that the main benefit of the use of the computer is
to provide interactive exposure to its precise diagnostic relationships. Ihis exposure has the benefits
of focusing the mind of the non-specialist on the
important points of the history and emination
and providing a solid foundation on which individual diagnostic stratagems can be based. In
this way the computer provides a condensed
supplement to the small number of cases which a
non-specialist can be expected to see during training.
In clinical practice, the implications are that the
casualty officer or general practitioner could use a
computer such as described in the article to help
with management decision making through its
highlighting important features that may have
been overlooked.
In a modified and "safer" form this type of
diagnostic program is being used by paramedics at
sea for medical decision support. The applications
and their implications need to be explored, discussed,3 and developed, where appropriate, to our
advantage.
G J BROOKS
D G CRAMP
1305
system that we are developing for acute chest pain
probabilities were very polarised when we applied
Bayes's theorem to 25 loosely related indicants. We
have now reduced this number to 12 without loss of
diagnostic performance, with less violation of the
independence assumption, and with better calibration of the resulting probabilities.
Secondly, the results of the trial in mode B
hospitals reveal some important points. Most
strikingly, there was no learning effect at all in the
six month baseline period, despite the fact that the
junior doctors in this group were all in "training"
grades, and most were in "teaching" hospitals: this
raises serious questions about the postgraduate
education of junior staff. There also appeared to be
no learning effect in the forms alone group,
although their diagnostic accuracy was some 100/o
higher than during the baseline period, presumably because of the "checklist effect" combined
with more diligent recording of diagnosis during
the test period. Surprisingly, the graph also
showed no evidence of learning in the computer
group, whose performance started at a high level
but remained nearly static over the six months.
However, during the six month "forms and feedback" trial period diagnostic accuracy did rise
from the forms alone level to the same as that
during the computer period, implying that the
computer itself acted as an aid to performance in
the early months but later on was effective only by
enforcing the discipline of form filling and by
providing feedback.
While help to the new doctor in his first few
months is undoubtedly useful to patients, it may be
that by expropriating his diagnostic capabilities the
computer allows the doctor to perform better while
learning less. Were any data collected about the
doctors' performance after the end of the study
that could confirm this theory?
Finally, especially in view of Dr Spiegelhalter's
postscript on trial design in such studies, it would
be interesting to hear comments on whether
consent should be sought for trials of computer
decision aids and, if so, whose? If we are allocating
the doctors to control or intervention groups
should they be informed of the likely performance
of the computer and asked for their consent, or is it
the patients (whose treatment may be altered if
their doctor is in a trial) who should be consulted?
JEREMY WYArr
PETER EMERSON
NICOLA CRICHTON
Department of Chemical Pathology and
Human Metabolism,
Westminster Hospital,
Royal Free Hospital and School of Medicine,
London SWIP 2AP
London NW3 2QG
1 AndersonJG, Jay SJ, SchweerHM, AndersonMM. Whydoctors
don't use computers: some empirical findings. J R Soc Mcd
1986;79:142-4.
SIR,-Dr I D Adams and colleagues showed that
2 Young DW. What makes doctors use computers?: discussion initial diagnostic accuracy improved from 45-6%
paper.J RSocMed 1984;77:663-7.
3 Young DW. A survey of decision aids for clinicians. Br Med J during the baseline period to 65-3% during the test
1982;258:1332-6.
SIR,-We are also working on computer aided
medical decision making and were pleased to see
the encouraging results of the multicentre trial of
the Leeds abdominal pain system (Dr I D Adams
and others, 27 September, p 800). We would,
however, like to make some comments about the
methods used and the results obtained.
Firstly, we are a little concerned about the use of
the independent Bayes technique, with its assumption that all indicants are unrelated, for probability
calculation with large sets of data. From the full
report of the study (DHSS research report, 1986) it
appears that over 40 symptoms and signs were
used in the probability calculation, including
WILLIAM TARNow-MoRDI several that are closely associated such as "previous
surgery" and "abdominal scar." In the case of a
period. The authors accept that not all of the
improvement observed was due to the computer
feedback but that the computer "created a climate
in which the inexperienced doctor was stimulated
and motivated towards doing the work correctly."
Fig 3 of the study shows that about 500/o of the
improvement in initial diagnostic accuracy using
the computer based system related to the act of
filling in forms. If a full clinical history is taken the
mere transcription of these facts on to a computer
form would not be expected to increase initial
diagnostic accuracy. It therefore follows that the
initial collection of data before the introduction
of the computer system was inadequate and incomplete.
At the completion ofmedical trainiing all medical
students should be able to take a full history in
relation to acute abdominal pain. In my experience
this is not done by the average house surgeon.
Whether this is due to ignorance or the heavy