It doesn't make house calls and its bedside manner is pretty impersonal, but it's on call 24 hours a day, seven days a week.
In a small but growing number of hospitals, one of the first experts consulted by the attending physician in a difficult case is a computer.
Use of computers in medicine dates back to the early 1960s, and today many laboratory tests and medical monitors are computer-based. Most blood tests and about 10 percent of electrocardiograms, for example, are read by computers.
But scientists are also working on a new generation of "smart" computers that not only record medical information but also interpret it and help doctors use the results to diagnose and treat illness.
Such computer systems have the potential of improving diagnoses, saving time and reducing costly and dangerous drug interactions. But they also raise troubling questions about quality control, regulation, liability and protection of privacy.
These issues were aired last week at a hearing on computerization of health care, held by the investigations and oversight subcommittee of the House Committee on Science and Technology.
Though few are commercially available, about a dozen "smart" computer systems for use in medicine are in various stages of development and testing around the nation, said William Roy Baker Jr., chairman of Knowledge Research Associates, a small Potomac software marketing firm.
Baker even predicted that the courts would eventually decide "that it's malpractice not to use one of these systems."
Medical computer programs are designed to advise -- not replace -- a doctor, emphasized Dr. Jack Myers, former chairman of medicine at the University of Pittsburgh School of Medicine who helped design a still-experimental system called Caduceus.
"They're not intended in any sense to replace the physician or be used directly by the patient," Myers said. "These are advisory programs."
Caduceus, named for the serpent-and-staff emblem of the medical profession, is designed to help internists diagnose diseases by recognizing their symptoms and the results of laboratory tests. The system currently includes computer profiles of almost 600 diseases, including more than 4,000 symptoms, Myers said. Eventually, it is expected to cover some 800 diseases.
Caduceus consists of two basic parts: an "electronic textbook" of medical information and an "inference engine" that interprets symptoms and test data. Both parts are designed to overcome what Myers called the two leading causes of incorrect medical diagnosis: limited human memory and limited problem-solving ability.
Because of the knowledge explosion in medicine, he said, the amount of basic information required for a practice in internal medicine is now about four times the capacity of the human memory. And while an expert considering a complex problem cannot deal with more than eight or 10 hypotheses at once, a difficult medical case can require consideration of dozens.
Computer systems such as Caduceus are aimed at those two problems -- the "Why did I forget that?" and the "Why didn't I think of that?"
Caduceus has been tested on more than 1,000 difficult cases, in which Myers said it performed "as well as the skilled clinician but not as well as the very best expert." With another 175 diseases to be profiled, its biggest weakness is its incompleteness, and Myers said it won't be ready for general use by doctors for at least "several years."
Asked by Rep. Harold Volkmer (D-Mo.), the subcommittee chairman, whether such a computer program could improve medical diagnosis, Myers said it already had made a difference in the way he approaches a complex case while teaching medical students and residents during hospital rounds.
"I'm much more willing to say, 'Look, we need more information here. Let's go to the terminal and get it before we continue this discussion,' " Myers said.
Computer programs can "not only increase the quality of care, but reduce the cost," said Reed M. Gardner, chairman of the biophysics department at Latter-Day Saints Hospital in Salt Lake City, which for the past 14 years has been developing and testing a software program called HELP (Health Evaluation through Logical Processing). HELP is marketed by Control Data Corp.
HELP not only improved the accuracy of patients' medical charts and other medical records, Gardner said, but it also saved the hospital's respiratory therapy department $250,000 a year by ruling out needless tests and preventing duplication of prescriptions.
Asked by Volkmer if the hospital had ever lost a patient's records in the HELP system, Gardner said: "Ever is a long time, and I'd say yes, we've lost data." But he quickly added that the computer's electronic data is more secure and less likely to be lost than paper charts.
Computers aren't infallible, of course, and they are only as good as the information they are given -- a limitation known in the trade as the GIGO syndrome (garbage in, garbage out).
But Myers said Caduceus doesn't require perfect information to be useful.
"If the doctor submits a little wrong information, he'll still get the right answer," Myers said. "If the doctor submits a lot of inaccurate data, he'll get the wrong computer interpretation."
"Of course, that happens with this computer up here, too," added Myers, pointing to his head, "so I don't have to apologize for that."
Akey unresolved issue is: Who will regulate medical software programs? The Food and Drug Administration, which regulates the safety and effectiveness of medical equipment under the Medical Devices Amendments of 1976, did not testify at last week's hearing but submitted a written statement that the agency is "in the process of developing an overall policy for computer technology used in medical devices."
"It can be expected that the policy will apply equally to computer hardware and software," the statement said.
But another witness, University of Maryland law professor Vincent Brannigan, called FDA's statement "proof by blatant assertion" that the agency has no jurisdiction over medical software, which he said is not a medical product.
"I think their idea is that this is some sort of new bedpan," Brannigan said. " But it doesn't work the way other products work . . . It doesn't exist the way other products exist. Software is not a device." Both regulation and potential liability should be focused on the users of such systems, not the systems themselves, he said. Regulating the system itself, he said, could lead to "the medical equivalent of the Pentagon's $600 hammer."
Volkmer said he hopes to hold further hearings on computerization of health "later this year" to get the FDA's position in detail. John Villforth, director of the Center for Devices and Radiological Health at the FDA, will be invited to testify, an aide said.
Some experts are also concerned that it would be harder to control access to computerized records.
"The bottom line," said attorney Robert Belair, "is that automation of health records simply makes it easier to collect information in the first place, easier to retain it, and easier and cheaper to say yes to disclosure of the information."
Acknowledging that privacy is an issue "that doesn't have any easy answers," Belair said, "Every public opinion poll taken shows that the public cares more about privacy of health records than other kinds of records."
Already, he said, because of increasing access to a patient's medical data by doctors, employers and insurers, privacy of health records "isn't dead but it certainly is ill."
Dr. Clement McDonald, a professor and self-described "computer nut" at University of Indiana School of Medicine, cautioned that "until such systems are very, very good they will be used relatively infrequently . . . Ten years from now it may be quite different." For all their potential in saving time and money, McDonald said, computers have severe limitations as diagnositicians.
"What they can't do is see the patient -- yet," he said. "The physician may not know that many facts, but they know the patient better than a computer will for a long, long time.
"You want a pilot flying that plane even when it's on automatic pilot."