John Fox is working to make sure that computers make life easier for doctors. Geoff Watts reports.
You probably know the despair of seeing your computer respond to a simple command with an incomprehensible error message. What to do?
You could wade through a manual written in a language you never learned, or phone a technical helpline manned by someone fluent in that language but not in English. Small wonder that some doctors are unenthusiastic about any computerised "enhancement" of their skills that has to be carried out in full view of the patients.
Computer scientist John Fox is well aware of the hazards of what he describes as the "third party in the medical consultation". As director of the Imperial Cancer Research Fund's Advanced Computation Laboratory, he is better placed than most computer experts to hear from doctors that even when computer systems operate faultlessly, their designers have rarely programmed them to reason like humans. This insight has informed Professor Fox's thinking throughout the 20 years he has worked in medical informatics.
Professor Fox has spent most of the past three years on a European Union funded collaboration called Prompt. "The aim was to develop and test a system that could be used at the point of care and would provide clinicians with software to give prompts, reminders and suggestions about diagnosis and treatment."
He has concentrated on the part of the project concerned with computer support in decision-making. Besides a general knowledge of medicine - facts about drugs, diseases, tests and so on - any system also needs information about individual patients. This, Professor Fox says, is the difficult part. "Not only for technical reasons, but because the patient is there with the doctor in the clinic. There's time pressure, the difficulty that some people have working with computers, and so on. For years this has been one of the major blocks on using computers in medicine."
The work of artificial intelligence researchers is fostering radical changes in the software used in systems such as Prompt. As Professor Fox sees it: "We need to understand what medical knowledge and expertise consist of and what doctors are doing when they exercise these skills. We are trying to build systems that are compatible with doctors' intuitions rather than give them a lot of mathematics that is completely alien to them and to the ways they think about patients and disease and treatment.
"That's why we spend a lot of time thinking how, in the jargon of the trade, to 'represent' knowledge, and to make decisions in a safe and sound way. Software has to explain itself, and say why it is making a particular recommendation. So an understanding of natural intelligence influences the design of artificial intelligence."
Tests on Prompt suggest that it can be valuable. "Colleagues at Erasmus University in Rotterdam looked at the results of using it in 60 GPs' surgeries in the Delft area. They compared blood tests ordered with and without computer support. They found that the computer brought about a 20 per cent reduction in the number of tests requested and a better selection. In other words - a lowering of cost and an improvement in quality."
A similar but smaller study on dyspepsia carried out by Peter Wilson and Colin Lyons at North End Medical Centre in London showed that the practice's drug bill had fallen, but that more patients were referred for investigation. "One of the less frequent causes of dyspepsia is cancer," Professor Fox says. "Early detection is vital if you're to get the best treatment results."
One outcome of Prompt is Infermed, a company set up by ICRT, the commercial arm of the Imperial Cancer Research Fund. This will take over the further development of the system.
Another project, RAGS (Risk Assessment in GeneticS) illustrates the kind of unanticipated problem that computers sometimes create. Developed with the support of the Economic and Social Research Council, the Cancer Research Campaign and ICRF's General Practice Research Unit in Oxford, RAGS is designed to help GPs compile a family tree of patients with ovarian and breast cancer, assess their risk of developing these diseases and then suggest what to do.
Professor Fox offers this typical scenario. "A youngish woman whose mother and aunt both had cancer sees a programme on television about breast cancer and becomes anxious. She goes to her GP and says, 'Do I have anything to worry about, and should I be doing anything?' At present, although GPs seem willing to provide this advice, they do not feel they have the expertise to do the genetics, to take the family history, analyse it and decide whether or not to refer patients for specialist advice and testing for cancer genes."
In a soon-to-be-published pilot trial of RAGS, actresses played the part of patients. In line with current thinking about the need to get patients more involved in consultations, the doctors allowed the women to watch the computer screen as they keyed in the information. All went well until the moment came to seek the computer's advice. "As you pressed the button, up would come this sign saying 'High risk'. This is not the way you want to introduce bad news."
The GPs testing the system felt they were losing control of the consultation. They suggested modifying the software so that doctors using it would have time to collect their thoughts on how best to break what might be bad news. The researchers concerned principally with RAGS have learned valuable lessons. With no harm done to real patients, a proper controlled trial is now planned.
Professor Fox knows that many doctors remain to be won over to a thorough going use of computers. They are already deluged with sometimes conflicting guidelines on diagnosing this and treating that. They view the prospect of having another lot on computer as multiplying their problems. And if researchers such as Professor Fox cannot devise systems that make doctors' work easier, quicker and more reliable, computerisation will never go beyond mere record-keeping.
On the other hand, it is the spread of this purely clerical use that gives Professor Fox confidence for the future. "I think that all medical information will eventually be recorded electronically. Once you have overcome that barrier, more and more doctors will use systems such as Prompt. I see it as a progressive process, and I think decision support will be widely used in just a few years' time."
Projects such as Prompt are part of the rapidly developing enthusiasm for evidence-based medicine. The aim is to give doctors a method of making decisions on the basis of established fact rather than habit, prejudice or incomplete memories of now outdated undergraduate lectures. The accumulation of medical knowledge is now such that without computer help there can be no prospect of achieving this ambition.