TCS Daily

A Black Eye for Medicine

By Jerry Pournelle - June 12, 2000 12:00 AM

Computers change the way professionals work -- at least most professionals. Despite the mounting costs of heath care, computer assistance is sorely lagging in the huge industry.

Twenty years ago when small computers first came out, there were several medical diagnosis programs to run on them. All indications were that they worked. I saw one at the Insititut fur Informatik at ETH in Zurich that by all accounts was better in diagnostics than many experienced physicians.

The problem was that the program consisted of 70,000 lines of Borland Turbo Pascal. That is, on the computers available to the ordinary research worker or physician at that time, the programs that were complex enough to be effective took a long time to run and even longer to develop. The input interface was slow and klunky. Storing medical records was no guarantee. Even so, there was still strong support for development of such programs.

In particular, Dr. Lawrence Weed campaigned for development of computer programs that take patient case histories and examine possible matches for rare disorders. In the early 1980s, I spent a few days with Weed at a meeting of computer-using physicians, and even then it was clear that developing computerized clinical diagnostics programs could be valuable, although the computer hardware available to the average physician wasn't yet up to the job. There were centralized programs -- one called Teiresias showed great promise -- but they did not run on small computers, and there was not any easy access to them.

I gave one of the keynote speeches at that conference. I talked about Moore's Law and its effect on small computers. I also repeated my then-startling prediction that by 2000 everyone in Western civilization would be able to get the answer to any question that had a known answer, and that this would include access to computer-enhanced medical databases and diagnostics programs. I maintained that medical diagnosis by computer would be the wave of the future.

Of course, I encountered opposition to such computer programs. A young intern stated flatly that if a computer program ever got to be a better diagnostician than he was, he would give up medicine. He did not think it would be possible, but if it happened, he would feel so demeaned that he could not continue. That sentiment was popular.

So much has changed

Today the computer world has greatly changed. Most of us now have access to machine capabilities that some predicted but didn't quite believe. We also have communications and bandwidth. We have CD-ROMs and DVDs. We have, literally, the capability to store every word every written, every picture ever painted, every photograph ever taken, every movie and video ever made, and keep duplicate copies in many places.

Moreover, all of this is available to nearly anyone in the United States, regardless of location. For $240 a year (less than some medical journals), physicians anywhere can access the Stanford e-Skolar service, for instance, and access the same information available to colleagues at Stanford or Johns Hopkins. There are other services for similar fees. And these fees likely will come down: Internet information wants to be free, and many companies would love to sponsor a service for physicians. The upshot is that any physician can, in potential, know anything that any other physician does. The only requirement is to ask.

When CD-ROMs first came out, a number of medical research companies offered a CD-ROM drive, and quarterly (later monthly) CD-ROMs containing the latest medical information: a precursor to the Internet. At least one of these research services included a diagnostics program similar to Teiresias: The physician entered as much of the patient case history as possible, and the program suggested possible diagnoses. If he accepted one of the diagnoses, the program suggested further tests, and eventually would suggest treatments and dosages (based on patient weight, age and other conditions). The programs were not anything like perfect, but the evidence at the time was that the program plus a physician was considerably more accurate than the program alone -- or a physician alone.

I recall a psychiatrist friend's experience. He was at Walter Reed Hospital, and a patient had alarming mental symptoms. There was nothing in the textbooks about them, at least nothing he could find. There were drugs to control the symptoms, but nothing seemed to effect a cure. Then he recalled a lecture from medical school: A visiting professor had mentioned an obscure and rare tropical parasite that caused odd behavior. The patient had been a Foreign Service officer and had spent several days on the Indonesian island where the parasite was observed. Voila! But, as Ed told me later, it was sheer chance that he had heard that lecture, or that he remembered it. A computer program would not have left that to chance. Computers can know about rare disorders and routinely check for them, even if you never heard of the condition.

Rare disorders become more likely as we travel more. Another friend suffered for years from an undiagnosed parasite she contracted in an Indian restaurant in Edinburgh Scotland; since Kathleen had never been to the Middle East no one thought to test for that. She suffered for five years.

New diseases are discovered annually. No physician can know them all. Faced with a thousand possibilities, with HMOs expecting physicians to see six or more patients an hour, the unaided physician is in trouble. Why not bring in computers to help?

This seems reasonable to the layman. Few airline pilots resent either safety checklists or the computer automation of many system checks formerly done by the flight crew. Today's auto mechanics could not operate without computerized diagnostic equipment. We all use various diagnostics programs for computer upgrades and repairs.

Indeed, as our world becomes more complex, we must rely on computer programs to examine, diagnose, maintain, and prescribe for most of the tools and conveniences we depend on. Our society would collapse without computerized maintenance. So why shouldn't the same techniques help physicians?

So much hasn't

It may seem reasonable to you and me, but there is strong resistance and little grant support for medical computer assistance. I found little when I did an Internet search for the Teiresias program. Searching further, I found that Weed is still crusading for computer-aided diagnosis, and physicians are still resisting. Despite the great advances in computer technology, we do not seem much further along in computer-aided diagnostics than we were in the 1980s, and physicians are still saying that if we do develop computer diagnostics, it will be "a black day for clinical judgment" (the title of one article on the subject, dated January 2000).

This is an unfortunate and doomed attitude. Like it or not, "evidence-based medicine" (the newest name for computer-assisted diagnosis) will expand and thrive, if for no other reason than patients will demand it. The Internet has made it possible for everyone to gain access to medical knowledge. Patients nowadays come armed with nine pounds of Internet printouts suggesting that the patient -- or more likely the patient's child -- has contracted some obscure disorder requiring elaborate and expensive tests and treatments. Such patients often have hungry lawyers as well. Physicians and their employers will ignore that combination at peril.

And there is the rub. As physicians gain access to more information, their liabilities go up as well. What you do not know and do not test for certainly can hurt you, both as a patient and a physician -- and no one can know about everything. Like it or not, we are going to have to develop computerized-diagnostics programs that match symptoms against disorders. The good news is that such tools can save a great deal of money. Getting the right answer in a timely manner is cost effective, whether in electronics, roof maintenance, car repair -- or medicine.

It is even more true in medicine. We all know that prevention is cheaper than cure: Big medical insurance companies learn that every year. Spending money on prevention saves money on cures, thus lowering costs and increasing profits. Good early diagnosis is cheap. It may not always have been so: Left to progress without treatment some diseases led to swift and early death, and funerals were cheaper than prevention. No longer. We now have the means to keep people more or less alive for a long time, and death is often neither swift nor early, and certainly is not cheap.

Even with all the potential benefits of computer-assisted diagnosis, it's no real wonder that it has taken so long. It needs the active cooperation of practicing physicians, and many physicians do not want to think computers are smarter than they are. But the alternative is not the old family practitioner who gets to know his patient and uses intuition to come up with what's best. There is precious little of that kind of medicine now, anyhow. The reality is six patients an hour, with an HMO employer urging in more. The reality is little time to think about that visiting lecturer 20 years ago. The reality is the same as it was in the 1980s: Physician plus computer is a far better combination than computer alone -- or physician alone.

A good day

It clear that there is little effort going into developing expert systems for assisting physicians. When you listen to some of the idiotic computer enthusiasts who talk as if the computers are going to replace doctors, you can see why many physicians are reluctant to get involved in developing those tools.

As computers get faster and databases larger and access to them easier, patients will begin using the tools even if doctors will not. One way or another, we will have better medicine at a lower cost, and the best way is if physicians enthusiastically cooperate with the computer people. Certainly mistakes will be made, but each mistake is a case that can be fed back into the database so that does not happen again. Unlike elephants and physicians, computers do not forget. Building diagnostic programs is not easy, but it can and must be done.

And it won't be "A Black Day for Clinical Judgment" at all.

Jerry Pournelle has written about computers and civilization for 20 years. He is a contributing editor for, where this column originally appeared.

TCS Daily Archives