| After reading this chapter, you should know the answers to these questions: |
|
To practice medicine
effectively, physicians must have rapid access to the contents of a large
and complex medical knowledge base, and they must know how to apply these
facts and heuristics to form diagnostic hypotheses and to plan and evaluate
therapies. Thus, the goals of medical education are to convey a body of
specific medical facts and to instruct students in general problem-solving
strategies. Medical educators are increasingly aware of the need for all
medical students to learn to use information technology for accessing and
managing medical information-both patient-specific clinical data and more
general scientific knowledge. Computers also can play a direct role in
the education process; students may interact with educational computer
programs to acquire factual information and to learn and practice medical
problem-solving techniques. In addition, practicing physicians may use
computers to expand and reinforce their professional skills throughout
their careers. The application of computer technology to education is often
referred to as computer-aided instruction (CAI).
In recent years, clinical medicine has encompassed an enormous growth in the number of diagnostic procedures that physicians can perform, many costly and some dangerous for the patient. Likewise, the therapeutic armamentarium has become vastly larger, more complex, more expensive, and more hazardous to the patient. When choosing from a wide array of alternative drugs, physicians must understand the indications, contraindications, and possible side effects of the drugs, and they must avoid prescribing interacting combinations of drugs. At the same time, growing pressure to control health-care costs creates additional demands on physicians; today, health professionals must consider not only efficacy and risk to the patient, but also cost of the care, when choosing tests for a diagnostic workup and when planning treatments.
The traditional medical-school curriculum emphasizes the rote memorization of medical facts, and students typically are evaluated based on their ability to recall these facts. This practice is prevalent, even though factual recall is only one of the skills necessary for competent medical practice. Students tend to focus their learning on acquiring the information needed to pass the examinations. In the process, they may neglect the development of problem-solving skills that are crucial to medical practice, yet are rarely tested.
A second weakness of the traditional medical curriculum is its heavy reliance on the lecture method of teaching. Students assume the role of passive recipients of knowledge, even though educational theory suggests that people retain information better when they seek it actively, through questioning, reasoning, and experimentation. The increasing heterogeneity of entering medical-school classes has made the lock-step lecture-dominated strategy increasingly anachronistic. The primary reason for this emphasis on lecture-style teaching is logistic; faculty believe that it is simply too expensive to provide individual instruction to every student in a medical-school class.
Medical students learn clinical-practice skills by observing and eventually by participating with teams Of physicians caring for patients on the wards of teaching hospitals. Growing concern about professional liability, however, has decreased the opportunity for students to assume meaningful responsibility for patients. An other factor that makes clinical education more difficult today than it was a generation ago is the disappearance from the teaching hospital of patients with less than critical disease, or with illnesses that illustrate many of the "typical" medical diseases. The teaching hospital has become predominantly a large intensive-care unit. A high proportion of its beds are filled by patients with acute cardiopulmonary disease. Many patients are chronically ill with multiple diseases, and many patients are terminally ill. In an era of cost containment, few patients are admitted to hospitals for a pure diagnostic workup; thus, many valuable teaching opportunities have been eliminated.
Medical school, internship, and residency are the first steps in a lifelong educational commitment. Physicians must prevent atrophy of existing skills and knowledge, and must continue to acquire new knowledge, to keep up with the rapidly changing field. To gain the mandatory continuing medical education (CME) credits, practi tioners incur large expenses for course-enrollment fees, transportation, lodging, and time lost from medical practices. These problems were recognized by many medical educators; however, before the advent of powerful and inexpensive computers (and of communication networks), there uas little opportunity to rectify the weaknesses of the typical course in continuing medical education.
How are students and experienced practitioners to acquire and maintain the knowledge necessary to function effectively in today's complex and rapidly changing medical environment? The response of the typical medical-school faculty has been to add more factual knowledge to the curriculum. Many thoughtful educators now realize that simply giving medical students more to learn is counterproductive; a better educational policy is to encourage students to become self-motivated and enlightened learners--to shift from a system dominated by a concern with remembering information to an environment in which students learn basic strategies for problem solving and techniques for accessing information as they need it. Some educators believe that the computer, with its superior memory and information-processing capabilities, will play a powerful role in this process.
This philosophy was embodied
in the 1983 report by the Panel on the General Professional Education of
the Physician (GPEP) of the Association of American Medical Colleges. The
GPEP repon recommended that medical schools take the lead in ap plying
information science and computer technology to medical learning, and pro
mote the use of computers in medical schools [Association of American Medical
Colleges, 1984). No longer is it sufficient for medical schools to leach
a static body of facts. Instead, students must learn to think, to solve
problems, and to make decisions--they must learn to learn.
The following characteristics of CAI offer advantages over traditional methods of medical instruction and evaluation:
Computer-based simulations allow students to manage a greater number and variety of patient cases than are typically seen in teaching hospitals. In addition, CAI programs can simulate prototypical, as well as unusual or complex, cases. For the student on the wards for the first time, a patient with several major diseases who is receiving multiple medications can be overwhelming. Simulations can be designed to exemplify a single disease; students can tackle more complex cases as they gain experience.
Simulations also allow students to manage cases as diseases evolve over time. A student's ability to follow a patient does not need to be limited by the length of time a student is assigned to a particular rotation or by the patient's leneth of stay in the hospital. This ability to learn about disease evolution is particularly important for teaching the management of chronic disease, which is becoming a major part of medical practice due to the increase in the elderly population.
Finally, standard cases can be used to assess problem-solving competency according to objective criteria. A student can compare her own performance on these cases with that of her peers, as well as with that of experts. Furthermore, faculty can use performance on clinical simulations as a measure of the student's mastery of problem solving skills. The National Board of Medical Examiners has been developing simulation programs with the plan that such simulations might be used in the board certification of physicians (see Section 17.4.1).
CAI, like every new methodology, is often challenged to prove its superiority over more established teaching methods. Lack of firm data to demonstrate such superiority has been one factor preventing the wider acceptance of CAI. Nonetheless, we know that one indication of successis acceptance by users. From its earliest days, CAI has been well accepted by most students and practitioners. In a recent study of 1200 sessions with CAI programs at the Massachusetts General Hospital, users judged 62 percent of the interactions as superior to lectures or textbooks, and rated only 8 percent as inferior. In 15 percent of the cases, the users thought the methods of teaching were too different to be compared [Hoffer et al., 1986].
Rarely is problem solving taught explicitly in the typical medical curriculum. Therefore, setting up experiments to compare the effectiveness of various methods of teaching problem solving is difficult. Several researchers, however, have performed studies to gain
more objective proof of efficacy.
For example, Hoffer and colleagues published a small study that demonstrated
that nurses learned advanced cardiac life support more effectively from
a computer program than they did from lectures [Hoffer et al., 1975]. The
authors also demonstrated that CAI programs can favorably affect physician
behavior. A study of the use of CAI in five community hospital emergency
departments found that the appropriate use of medications emphasized in
the programs rose substantially after the CAI material uas made available
[Hoffer, 1_975]. Not all such studies have had such gratifying results.
For example, researchers at the University of Wisconsin found that performance
on patient simulations correlated only weakly with students' grades on
final examinations and ward rotations, except on the most complex computer
cases [Friedman et al., 1978]. Physicians' patterns of test ordering and
therapy selection in a related study, however, were similar in computer
simulations and in actual patient-care situations [Friedman, 1973].
Pioneering research in
CAI was conducted in the late 1960s at three primary locations: Ohio State
University (OSU), Massachusetts General Hospital (MGH), and the University
of Illinois. Earlier attempts to use computers in medical instruction were
hindered by the difficulty of developing programs using low-level languages
and the inconvenience and expense of running programs on batch-oriented
mainframe computers. With the availability of time-sharing computers, these
institutions were able to develop interactive programs that were accessible
to users from terminals via telephone lines.
CAI research began at OSU in 1967 with the development ofTutorial Evaluation System (TES). TES programs typically posed constructed-choice, true-false, multiple-choice, matching, or ranking questions, then immediately evaluated the student's responses. The programs rewarded correct answers with positive feedback. Incorrect answers triggered corrective feedback, and, in some cases, the student was given another opportunity to respond to the question. If a student was not doing well, the computer would suggest additional study assignments or direct the student to review related materials.
In 1969, TES was incorporated into the evolving Independent Study Program (ISP), an experimental program that covered the entire preclinical curriculum and was designed to teach basic medical-science concepts to medical students [Weinbere, 1973]. Although the ISP did not use CAI in a primary instructional role, students in the program relied heavily on a variety of self-study aids and used the computer intensively for self-evaluation. The use of COURSEWRITER III, a high-level authoring language, facilitated rapid development of programs. By the mid-1970s, TES had a library of over 350 interactive hours' worth of instructional programs.
Beginning in 1970, Barnett and colleagues at the MGH Laboratory of Computer Science developed CPJ programs to simulate clinical encounters [Hoffer et al., 1986]. The most common simulations were case-management programs that allowed students to formulate hypotheses, to decide which information to collect, to interpret data, and to practice problem-solving skills in diagnosis and therapy planning. BY the mid-1970s, MGH had developed more than 30 case-management simulations, including programs for evaluation of comatose patients, for workup of
patients with abdominal pain, and for evaluation and therapy management for problems such as anemia, bleeding disorders, meningitis, dyspnea, secondary hypertension, thyroid disease, joint pain, and pediatnc cough and fever.
The MGH laboratory also developed several programs that used mathematical or qualitative models to simulate underlying physiologic processes, and thus to simulate changes in patient state over time and in response to students' therapeutic decisions. The first simulation modeled the effects of warfarin (an anticoagulant drug) and its effects on blood clotting. The system challenged the user to maintain a therapeutic degree of anticoagulation by prescribing daily doses of warfarin to a patient who has a series of complications and who was taking medications that interacted with warfarin. Subsequently, researchers developed a more complex simulation model to emulate a diabetic patient's reaction to therapeutic interventions.
About the same time, Harless and researchers at the University of Illinois were developing a system called Computer~Aided Simulation of the Clinical Encounter(CASE), which simulated clinical encounters between physician and patient [Harless et al., 1971]. The computer assumed the role of a patient; the student, acting in the role of practicing physician, managed the patient's disease from onset of symptoms through final treatment. Initially, the computer presented a brief description of the patient; then the student interacted with the program using natural-language queries and commands (Figure 17.1). The propram was able to provide logical responses to most student requests. This feature added greatly to the realism of the interaction, and CASE programs were received enthusiasticaIly by students. The TIME system, under development by Harless and researchers at the National Library of Medicine(NLM), extends CASE's approach to incorporate videodisc technology (see Section 17.4.2).
CAI programs proliferated on a variety of hardware, using a Babel of languages. A 1974 survey of the status of medical CAI identified 362 programs written in 23 different computer languages, ranging from BASIC, FOHTRAN, and MUMPS, to COURSEWRITER III and PLATO. Little sharing of programs among institutions was possible because the task oftransferring programs was typically as large as writing the material de novo. Thus, there waslinle opportunity to share the substantial costs of developing new CAI programs. The lack of portability of systems a_nd the extreme expense of system development and testin_g served as barriers to the widespread use of CAI.
The establishment of a NLM-sponsored,
nationwide network in 1972 was a significant event in the development of
CAI in medicine because it allowed users throughout the country to access
computer-based educational programs easily and relatively inexpensively.
Previously, the programs created at OSU, MGH, and the University of Illinois
were available: to users in selected regions, over voice-grade telephone
lines. Poor quality of transmission and high costs, however, combined to
limit access to CAI programs by distant users. Xcting on the recommendation
of a committee of the Association of American Medical Colleges, the Lister
Hill Center for Biomedical Communications of the NLM funded an experimental
CAI network. Beginning in July 1~72, the CAI programs developed at the
MGH, OSU, and the University of Illinois Medical College were made available
from these institutions' host computers over the NLM network using communication
lines of the Smshare network. During the first :! years of operation, 80
institutions used the programs of one of the three hosts. The high demand
for network use prompted the NLM to institute a stepped char~, first 52.50
and then S5 per hour of use, but use continued to rise. Ha\·ing
exhausted the funds set aside for this experiment, the NLM announced that
the experimental phase would be terminated and that it would discontinue
financial support for the network after May 1975.
|
FIGURE 17.1 A student using the CASE system, developed at the University of Illinois, assumes the role of a practicing physician in simulated encounters between physician and patient. The student poses natural-language queries to the system, which interprets the question and responds as best it can. This transcript is an example of such an interactive session. (Source: Reprinted with permission from Harless, W. G., et al. CASE: A computer aided simulation of the clinical encounter. Journal of Medical Education, 46:443. 1971. The Journal is now published under the name Academic Medicine.) |
| COMPUTER: Il's a Tuesday
morning, the middle of a cold !anuary. You have seen three palicnls in
the clinic and now a rounh person is waiting for you in your office. Tht
woman---middle-aged, while, brown hair---is sitting at the side of your
desk as you open the door. Her features seem wom. and she is slightly underweight.
She looks up and smiles at you with a minimum of enthusiasm and effort.
STUDENT: What brings you to see me? COMPUTER: Well. I'm just tired all the Lime. I don't ever feel like I'm rested and all my energy is gone. STUDENT: How long has this been going on? COMPUTER: About six months ago I started reeling tired all the lime. I finnally felt like I had to do something about it. STUDENT: What did you do? COMPUTER: Are you still talking about my being tired? STUDENT: Yes. COMPUTER: Well, I've been taking vitamins. but they don't seem to help. Maybe you can give me something stronger. STUDENT: Is there anything else bothering you? COMPUTER: I guess my only real complaint is that I am tired all the time. Ijust don't have any energy at all anymore. STUDENT: Do you have any trouble sleeping? COMPUTER: I sleep like a log ... eight ornine hours every day. <student presses a key on the Ierminal keyboard that indicates Physical Exarninarion> STUDENT: Vital signs. COMPUTER: Tcmperalurc: 98.6. Pulse: 96. Rcspira~ion: 20. STUDENT: Tongue. COMPUTER: Tongue examination: Red. some atrophy of the papillae; full movement. STUDENT: Neck examination. COMPUTER: Neck: Trachea central: thyroid not palpable. No lymph node cnlargcmcnt; no abnormal pulsation. STUDENT: Heart auscultation. COMPUTER: Heart auscultation: Auscultarion performed in all appropriate positions. Normal sinus rhythm. Rate of 80/ min. S1 and S2 of good quality. No extra sounds, murmurs, rubs. <student presses the key that indicares Hisrory> STUDENT: Have you ever had any trouble with your heart? COMPUTER: I've never had any heart trouble al all that I know of. STUDENT: Has anyone in your family ever had heart trouble? COMPUTER: No one in my family has ever been bothered by heart trouble. <student presses a key ro rerurn to the Physical Examination> STUDENT: Abdomen. COMPUTER: Abdomen inspection: symmetrical, normal contour. No visible dilated veins, scars, or other abnormality. <student presses the key that indicares Loboratory> STUDENT: Complete blood count. COMPUTER: Complete blood count: red blood cell count, 2.5 million; hemoglobin, 8.5 gm; hematocrit, 26; white blood cell count, 5100. Diffrential: neutrophils,58; coslnophils, 2: basophils, 1; monocytes, 4: lymphocytes, 35. STUDENT: Color of urine. COMPUTER: Urine appearance: Yellow, clear. STUDENT: PBI. COMPUTER: Protein bound iodine: 6 mg. STUDENT: Gastric analysis. COMPUTER: Gastric analysis: fasting. O free acid: 10 mEq/l total: 45 minutes after histamine; stimulation: O free acid: 30 mEqfl total. <student presses the key that indicclres Treatment> COMPUTER: Please type your diagnosis. STUDENT: Pernicious anemia. COMPUTER: Please prescribe your treatment. STUDENT: Vitamin B1Z COMPUTER: After three weeks symptoms begin to vanish. |
As a vivid testimony to the value placed on the educational network by its users, MGH and OSU continued to operate the network as an entirely user-supported activity. The MGH programs are currently available nationwide and in foreign countries over the Telenet communications network. This service is available around the clock, 7 days per week; access to a terminal or personal computer that supports modem communications is the only hardware requirement. In addition, since November 1983, the MGH programs have been offered as the CME component of the American Medical Association's Medical Information Network (AMA/NET). AMA/NET provides to subscribing physicians a variety of services in addition to the CME programs, including access to information databases, to the clinical and biomedical literature, to the DXplain diagnostic decision-support tool (see Chapter 15), and to electronic-mail services. By 1986, approximately 100,000 physicians, medical students, nurses, and other people had used the MGH CAI programs over a network, with about L50,000 total contact hours.
During the early 1970s, medical schools around the country began to conduct research in CAI. One of the most interesting programs was the PLATO system (Programmed Logic for Automated Teaching Operations) developed at the University of Illinois. PLATO used a unique plasma-display terminal that allowed presentation of text, graphics, and photographs, singly or in combination. An electrically excitable gas was used to brighten individual points on the screen selectively. The system also included TUTOR, a sophisticated allthoring language, to facilitate program development. By 1981, authors had created 12,000 hours of instruction in 150 subject areas.The programs received heavy use at the University of Illinois; some of them also were used at other institutions that had access to the system. The high cost of PLATO, and the need for specialized terminals and other computer hardware, however, limited the widespread dissemination of the system.
Research on medical applications of artificial intelligence (AI) stimulated the development of systems based on models of the clinical reasoning of experts. The explanations generated by computer-based consultation systems (for example, why a particular diagnosis or course of management is recommended) can be used in computer-based education to guide and evaluate students' performance in running patient simulations. The GUIDON system is one of the most provocative examples of such an intelligent tutoring system. GUIDON used a set of teaching-strategy rules, which interacted with an augmented set of diagnostic rules from the MYCIN expert system (see Chapter 15) to teach students about infectious diseases [Clancey, 1986]. Subsequently, researchers reorganized and extended MYCIN's knowledge base to form the NEO -MYCIN system by adding explicit knowledge about the process of diagnosis. The NEOMYCIN knowledge base was then used by GUIDONZ to teach students about diagnostic strategies (see Section 17.3.5).
Researchers at the University of U'isconsin applied a different approach to the simulation of clinical reasoning. Their system is used to assess the efficiency of a student's workup by estimating the cost of the diagnostic evaluation [Friedman et al., 1978]. In one of the few successful field studies that demonstrated the clinical significance of a simulated diagnosis problem, Friedman found significant levels of agreement between physicians' performance on simulated cases and actual practice patterns [Friedman, 1973]. Of considerable interest is the commitment of the National Board of Medical Examiners (NBME) to proceed with Friedman's original work by establishing a nationwide delivery system of computer-based evaluation and learning centers. Friedman's simulation model was the prototype for the CBX system, which is being developed by the NBME and will be used to test students' diagnostic skills in the National Board Part in examination (see Section 17.4.1).
The development of PCs, authoring systems, and network technology removed some of the barriers to program development and dissemination, and more CAI software became available. PCs provide an affordable and relatively standard environment for the development and use of CAI programs. The use of networks for program distribution has a number of major advantages. As a two-way medium, it permits users and courseware authors to exchange comments. This interchange is invaluable for finding program bugs and for improving the programs. It also facilitates frequent updating and enhancement of programs. Because users access the software stored on the host computer, the programs are easy to update and the newly modified version is immediately available to all users.
Network distribution is not without its limitations, however. A major disadvantage is the cost of network access. For example, the communications costs incurred by users of the MGH Education Network are nearly equal to the costs of hardware and course development incurred by the MGH host institution. Furthermore, the variety of terminals used over a network limits the program's ability to use graphics or any other techniques that depend on specific terminal characteristics. For these reasons, other modes of program distribution are attracting increasing interest.
A number of publishing firms
are presently distributing medical programs (educational and other) via
floppy disks. The MGH laboratory has arranged with the Williams & Wilkins
Company to publish the programs currently distributed via the MGH Education
Network for use on Apple and IBM personal computers as the RxDx series.
Scientific American Medicine has published case-management problems on
floppy disk (DISCOTEST). Likewise, a quarterly journal called Cyberlog
began distributing programs on floppy disk for Apple and IBM-series computers
in the summer of 1985. These programs contain tutorials, case studies,
and a series of tools, or calculation aids, that can be used by physicians
for independent study. The Universities of Washington and Georgia each
distribute case-management problems on floppy disks. In addition, faculty
at a number of different medical schools have developed and distributed
single examples of medical computer-based educational material on floppy
disks. Most of the general-circulation medical journals currently carry
advertisements for CAI material on floppy disk, and some software reviews
have appeared in the New England Journal of Medicine, in the Annals of
Internal Medicine, and in other respected journals. The unaccredited material
is of uneven quality, however, and a peer-review process is needed urgently.
The goals of medical
education are to teach students specific facts and information! and to
teach strategies for applying this knowledge appropriately to the situations
that arise in medical practice. Thus, students must learn about physiological
processes, and must understand the relationships between their observations
and these underlying processes. They must learn to perform medical procedures,
and they must understand the effects of different interventions on health
outcomes. Medical school faculty employ a variety of strategies for teaching,
ranging from the lecture based one-way transmission of information to the
interactive Socratic method of instruction. In general, we can view the
teaching process as the presentation of a situation or a body of facts
that contains the essential knowledge that students should learn; the explanations
of what are the important concepts and relationships, how can they be derived,
and why are they important; and the strategy for guiding the interaction.
The goals of instruction
and the choices among teaching strategies are reflected in the overall
design of CAI programs (fact-oriented drill-and-practice approach versus
process-oriented simulation approach) and in the structure and style of
the interaction between student and program (relative control of student
versus program in structuring the interaction, and degree of guidance provided
by the system).
Simulations are the basis for most CAI programs used today. These programs are designed to help a student learn by doing. They emphasize learning procedures and problem-solving strategies. Simulation programs now are available to help students learn to take medical histories, to diagnose illnesses, t, perform clinical procedures such as advanced cardiac life support), and to manage patient ca'e over time. As we discussed in Section 17.1.2, students can use computer-based simulations to experiment and to learn irom errors in reasoning and judgment without jeopardizing the well-being of real patients. Furthermore, students can gain experience managing patients with rare conditions, thus gaining familiarity with medical problems to which they might not otherwise be exposed.
Simulation programs may be
either static or dynamic. Under the static simulation model, each case
presents a "patient" who has a predefined problem and set of characteristics.
Figures 17.2 through 17.6 illustrate the interaction between a student
and the RxDx sequential-diagnosis program on abdominal pain. The student
interrogates the computer about history, physical findings, and laboratory-test
results to reach a diagnosis. At any point in the interaction, the student
can interrupt data collection to ask the computer "consultant" to display
the differential diagnosis (given the information that has been collected
so far), or to recommend a data-collection strategy. The underlying case,however,
remains static. On the other hand, dynamic simulation programs simulate
changes in patient state over time and in response to students' therapeutic
decisions. Thus, unlike those in static simulations, the cases of a dynamic
simulation evolve as the student works through them. These programs
| FIGURE 17.2 This transcript shows the first part of an interaction between a student and the sequential-diagnosis program on abdominal pain (which is operational at MGH and is available over the MGH Educational Network or as part of the RxDx series of floppy disks published by Williams 8 Wilkins). The student is presented with a brief description of a patient with abdominal pain; he then collects information by Selecting items from a list of almost 100 different signs, symptoms, orlaboratory tests. The computer program provides immediate responses to each request. The text entered by the student is underlined for purposes of illustration. (Source. Courtesy of the Laboratory of Computer Science.Massachusetts General Hospital.) |
| THE PATIENT IS A 56 YEAR-OLD
MALE.
HE CAME TO THE EMERGENC( DEPARTMENT BECAUSE OF MODERATE EPI GASTRIC PAIN THAT STARTED OVER A MONTH AGO. You may now examine your patient. Item Number: 100 CHARACTER OF PAIN-BURNING Item Number: 102 HOW LONG DO THE PAINS LAST?-MOST OF THE TIME. OCCA - SIGNALLY EASES Item Number: 103 HAVE YOU EVER BEFORE HAD THIS KIND OF pAIN?--SEVERAL YEARS AGO Item Number: 107 VOMITING--YES Item Number: 108 FEVER--NONE Item Number: 109 DIARRHEA--NO Item Number: CONSULTANT |
________________________________________________________________
ITEM INFOBEST TESTSCONSIDERDISCRIMINATELEADING DIAGNOSES TEACHING EVIDENCERETURN TO MAIN PROGRAM__________________________________________________________________FIGURE 17.3 This menu shows the choices available to students using the sequential diagnosis program on abdominal pain. At any time, the student can interrupt the computer to request a consultation--see the last entry in Figure 17.2. The computer program can provide information about the characteristics and potential usefulness of each item of infer mation, and can give general guidance about how to work up a patient who has that specific type of problem. In addition, using a statistical model of clinical reasoning, the computer can provide a differential diagnosis (given the information available at that time), and can recommend the "best tests''--those items that have the highest potential value for clarifying the differential diagnosis. (Source: Courtesy of the Laboratory of Computer Science, Massachusetts General Hospital.)
__________________________________________________________________CONSULTANT option: LEADING DIAGNOSESThe current leading diagnoses and their probabilities are as follows:1) 901 PEPTIC ULCER (GASTRIC DUODENAL) 862) 905 ACUTE GASTRITIS 43) 908 CHRONIC PANCREATITIS 24) 904 GASTRIC CANCER 1__________________________________________________________________FIGURE 17.4 The sequential-diagnosis program displays its list of leading diagnoses. This is the computer model's interpretation of the most likely diagnoses given the informa tion collected thus far. The probabilities are crude approximations, because the underlying model is simplistic and is based on conditional probabilities that have been derived from experts, rather than from clinical studies. The text entered by the student is underlined for purposes of illustration. (Source: Courtesy of the Laboratory of Computer Science, Massachusetts General Hospital.)
help students to understand
the relationships between actions (or inactions) and patients' clinical
outcomes. Tosimulate a patient's response to intervention, the programs
explicity model underlying physiologic processes, often using mathematical
The mechanism for communication
between student and CAI program can take one of several basic forms. At
one extreme, students may select from a constrained list of responses that
are valid in the current situation (such as the set of answers to a
__________________________________________________________________Item Number: 304 TENDERNESS TO PALPATION-EPIGASTRICItem Number: 509 STOOL EXAMINATION--BROWN; HEMOCCULT NEGATIVEItem Number: 306 REBOUND TENDERNESS AND INVOLUNTARY GUARDING--NOItem Number: 303 ENLARGED ORGAN(S) OR MASS IN ABDOMEN-- NO MASSES OR ENLARGED ORGANSYour Attending would like to know what you think is the leading diagnosis. Enter the number of your diagnosis: 902 PEPTIC ULCER (GASTRIC/DUODENAL)She is pleased--you are on the right track.The current evidence strongly supports your conclusion.
__________________________________________________________________FIGURE 17.5 The sequential-diagnosis program requests the student's primary diagnostic hypothesis. The student then coileds further information. The computer-based model is tracking the interaction, updating its internal differential diagnosis with each new item of information as the latter is collected by the student. When the probability of the most likely diagnosis crosses a threshold, the computer interrupts the student and asks that the d~dent enter his primary diagnostic hypothesis. The text entered by the student is underlined for purposes of illustration. (Source: Courtesy of the Laboratory of Computer Science, Massachusetts General Hospital.)__________________________________________________________________THE PATIENT IS A 22 YEAR-OLD MALE.HE CAME TO THE EMERGENCY DEPARTMENT BECAUSE OF MODERATE LEFT UPPER QUADRANT PAIN THAT STARTED WITHIN THE LAST 24 HOURS.
You may now examine your patient.Item Number: 100 CHARACTER OF PAIN--BURNINGItem Number: 103 HAVE YOU EVER BEFORE HAD THIS KIND OF PAIN? --NEVER BEFOREItem Number: 107 VOMITING--NO
Item Number: 109 DIARRHEA--YES; CHRONIC
Item Number: CONSULTANT
CONSULTANT option: LEADING DIAGNOSES
The current leading diagnoses and their probabilities are as follows: 1) 934 IRRITABLE BOWEL SYNDROME 232) 921 ACUTE APPENDICITIS 163) 901 RUPTURED SPLEEN 114) 915 REGIONAL ENTERITIS 10
CONSULTANT option: BEST TESTS1) 104 TRAUMA2) 126 HAVE YOU NOTICED ANY STOOL ABNORMALITY3) 366 REBOUND TENDERNESS AND INVOLUNTARY GUARDING4) 111 WEIGHT5) 304 TENDERNESS XO PALPATION__________________________________________________________________
FIGURE 17.6 In this session, the student is using the sequential-diagnosis program as a consultant on abdominal pain. The student has collected several items of information, and has requested a consultation regarding the computer's suggested diagnostic hypothesis and set of "best tests." The text entered by the student is underlined for purposes of iilustration. (Source: Courtesy of the Laboratory of Computer Science, Massachusetts General Hospital.)
multiple-choice question). At the opposite extreme, students are free to query the program and to specify actions using entirely unconstrained natural language. An intermediate approach is to provide a single, comprehensive menu of possible actions, thus constraining choices in a program-specific, but not in a situation-specific, manner.
The use of a predefined, explicit vocabulary has two disadvantages: (1) it cues the user (suggests ideas that otherwise might not have occurred to him) and (2) it detracts from the realism of the simulation. On the other hand, progiams that providestudents with a list Of actions that are allowable and reasonable in a particular situation are easier to write, because the authors do not need to deal with unanticipated responses. Furthermore, the use of a constrained vocabulary will be less frustrating to students who may otherwise have difficulty formulating valid interactions.
The use of free-text input
works best in well-defined domains or in circumscribed contexts in which
a student's responses are easy to anticipate and interpret. Thus, if the
program asks, "What antibiotic would you prescribe for this patient?" it
needs to be prepared to handle only a relatively small number of possible
responses. In general, both laboratory tests and drug therapies can be
handled relatively easily using natural language. On the other hand, the
range of possible medical-history questions is so large that anticipating
even the most logical set of student inquiries is difficult-devising any
scheme that deals reliably with all possible inputs is impossible. Questions
and actions related to the physical examination fall closer to the feasible
range, although a student will almost certainly overtax the capabilities
of the system if she uses an obscure eponym (such as "Sicar's sign"), or
states an inquiry in a no~el way (for example, "Is there puffiness around
the ankles?" rather than, "Is there pedal edema?"), or asks about an unusual
finding (such as webbed fingers) in a patient case in which such a finding
would usually not be considered. Students quickly become discouraged if
the system fails to recognize what seem to be legitimate inquiries, and
they have little tolerance for learning how to state an inquiry in a way
that the computer can understand.
Abdulla at the University
of Georgia [Abdulla et al., 19&lj, Friedman at the University of W~sconsin
(see Section 17.4.1), and Harless at the Lister Hill Center of the NLM
(see Section 17.4.2) are several of the researchers developing imaginative
teaching programs based on videodisc technology One of the major factors
limiting the development of interactive videodisc programs is the significant
expense involved in producing high-quality images.
In the past decade, CAI
programs have proliferated rapidly. Undoubtedly, this increase in activity
reflects the availability of personal computers. Today, virtually all medical
schools use computers in some capacity to facilitate medical education.
Many of the current CAI programs are simulations that allow students to
diagnose and manage patient cases. The newest programs combine a variety
of media--text, graphics, video, and sound. In this section, we shall describe
several examples of current CAI programs. For discussions of additional
systems, you can consult the proceedings of the Symposium on Computer Applications
in Medical Care (SCAMC), which is cited in the Suggested Readings.
CBX places the physician
in the position of primary caretaker for a patient in a medical environment.
Once presented with a clinical scenario, the physician can assess and manage
the patient's condition by collecting medical-history information, con
ducting a physical examination, performing procedures, ordering diagnostic
tests, ordering therapies, and calling in outside consultants. Almost all
such requests are made through free-text entry of orde rj on a hospital
order sheet. Ln addition, thousands of medical images are available on
videodisc for interpretation and review, including those from X-ray films,
ECG recordings, and slides of tissue specimens. The pa tient's condition
evolves dynamically during the simulation, reflecting both the passage
of time and the physician's therapy-management strategy; likewise, the
results of tests and other clinical information reflect the patient's changing
condition. CBX records the timing and sequencing of the physician's actions
for later evaluation according to criteria defined by a group of expert
physicians.
A typical case shows an introductory scene, then freezes the action in a wait stale.The wait state is the students' cue to take action of some kind. Using a specified set of vocal inquiries and commands, the instructor interviews the patient, directs the diagnostic workup, and manages treatment. The videodisc allows students to witness a variety of scenes that depict past events related to the present illness and to current experiences in the hospital. Other scenes include significant portions of the physical examination, X-ray films, and so on. The instructor can interview the patient directly or can use the command word "thoughts" to hear what the patient is thinking--a feature designed to increase students' awareness of patients' emotional and mental states.
Each TIME case contains multiple decision paints, or situations during the simulation that have uncertain outcomes. Thus, cases are not completely predictable and can unfold in a variety of ways.-The program randomly chooses among the possible outcomes using a table of probabilities associated with the decision point. The prob abilities of the various outcomes change dynamically during the case, depending on the students' behavior in making relevant inquiries and in choosing appropriate interventions. The case study ends following an outcome scene that describes the eventual fate of the patient (for example, complete recovery or subsequent return to the hospital). At this point, the program provides feedback on various aspects of the group's performance. For example, it describes the correctness of the diagnosis, the proportion of critical information obtained during the session, and the cost of the patient's hospital stay.
Prior to running the program,
a user must ~ain the system to recognize a vocabulary of U5 control words.
This training session takes about 30 minutes to complete. The computer
saves the voice patterns of the user as he speaks each word. These patterns
are then used by the computer to recognize and interpret spoken commands
when the simulation is run.
The student can run the program
in each of three modes: laboratory, patient-case analysis, and review.
In laboratory mode, the student selects from menus to control a synthesizer
that can produce heart sounds with a variety of rates and rhythms, aug
menting these with gallops, clicks, and murmurs (Figure 17.7); Accompanying
graphics represent the heart sounds visually and indicate the optimal location
on the chest wall for placement of the stethoscope to hear particular sounds.
In patient-case analysis mode, the system generates patient cases of varying
severity. The student uses a mouse pointing device to position a stethoscope
on the chest wall, then attempts to diagnose the cardiac anomaly based
on the results of auscultation. If the student fails to reach the correct
diagnosis, the program suggests appropriate lessons for review. In review
mode, the program presents a variety of topics pertinent to the cardiovascular
physical examination using text and graphics. A similar program, EKGLab,
allows students to practice the interpretation of ECGs.
FIGURE 17.7 When running the Heartlab simulation program in laboratory mode,students can experiment with the heart sounds associated with a variety of cardiac anomalies. In this case. the user is experimenting with the murmur of a patent ductus arteriosis. The program portrays a graphical representation of the heart sound, as well as producing the sound itself on the computer's synthesizer.(Source: Courtesy of Dr. Bryan Bergeron.)
Images are displayed on two screens: Bit-mapped line drawings and text are displayed on a computer monitor, and photographic and radiographic images stored on videodisc are displayed on a television monitor. The image base is accessible by region of the body (head, thorax, abdomen, and so on) and by system (skeleton, cardiovascular system, nervous system, and so on). A collection of interactive illustrations also is available. Students can explore the effects of facial-nerve damage, for example, by selecting a nerve on the drawin_e, and viewing the corresponding image of a real patient with damage to that nerve (Figure 17.8). Likewise, students can use a mouse pointing device to probe an image of the hand of a patient with sensory nerve damage, testing various locations on the hand for pressure sensitivity. In addi tion toproviding an environment for undirected exploration, ElectricCadaver includes a tutorial that teaches fundamental elements of anatomy and provides mechanisms for testing and for self-evaluation.
FIGURE 17.8 Students using
the ElectricCadaver system can explore a database of digitized anatomical
images to learn about human anatomical structure and physiological function.
This figure shows a dynamic diagram that was designed to illustrate the
effects of facial-nerve damage. The student uses a mouse pointing device
to "damage" part of the nerve (near the earlobe). Immediately, the program
displays on the computer monitor a drawing showing the effects of the damage,
and presents on a television monitor a photographic image of a real patient
with damage to that nerve. (Source: Courtesy of Vesalius, Inc.)
The separation of domain knowledge and teaching knowledge allowed GUIDON to be used with any EMYCIN system.This approach represented a significant advance over traditional CAI methodology because GIJIDON's knowledge was reusable and could be adapted to new applications. Thus, the researchers were able to plug in the knowledge bases from the SACON and PUFF systems, and the same GUIDON program was able to tutor students about structural-analysis p'oblems and the interpretation of pulmonary-function tests.
Although it was adequate for the purpose of diagnostic consultations, the MYCM knowledge base was missing knowledge that was necessary diagnosis of infectious diseases. For example, the knowledge base contained no information about diseases that cause symptoms similar to those of meningitis and bacteremia, and contained linle support knowledge to justify the diagnostic rules. Furthermore, MYCIN's reasoning process differed from physicians' diagnostic inference process--MYCIN performed a top-down search through a prescribed set of diseases, while exhaustively collecting information. Physicians, on the other hand, form hypotheses based on partial evidence, then strategically collect information to refine the diagnosis.
In the early 1980s, researchers
reorganized and greatly expanded MYCIN to create NEOMYCIN, another medical
diagnosis program that addresses many of MYCIN's limitations with respect
to teaching [Clancey, 1986]. The NEOMYCIN knowledge base contains knowledge
of competing diseases and rules that embody explicit strategies for hypothesis
formation, causal reasoning, and grouping and discriminating among competing
hypotheses (Figure 17.9). Thus, the revised teaching system, GUIDON?, is
able to access the strategic knowledge contained in NEOMYCIN's knowledge
base to teach students about strategies for diagnostic reasoning IRodolitz
and Clancey, 1989].
Despite the strong arguments
that have been made in support of computer-based medical education, progress
in developing exciting programs and in the national dissemination of the
existing programs has been agonizingly slow. This lagging development and
diffusion are attributable to the difficulty and expense of writing CAI
programs, to the lack of support for program development within institutions,
and to the barriers to sharing programs among institutions.
Courseware development is labor intensive and time consuming. initially, faculty members must devote conside,ble time to understanding and gaining expertise in the art of creating computer-based CXI programs; both the approach and the materials are significantly different from those used in preparing lectures. Even after this hurdle has been overcome, program development is a lengthy process. Often, there is a si,snifi cant lead-time of many months before any results are seen. The time between the formulation of a concept and the delivery of a completed program can easily be 1 full year if the program is of any depth or complexity.
The development of improved tools for coursework authoring that can gain widespread acceptance is a high-priority need. Such tools could reduce the startup time for and speed the process of program development The ideal authoring language would combine simplicity for novice coursework writers and power for experts, and would be widely available to faculty at many institutions. In addition, the production of educational-support materials on videodisc, and the development of an extensive library of visual and graphical material to be shared among CAI authors, would help individual developers to avoid much redundant work.
At schools where CAI has played a major role, there usually has been a highly visible advocate for its use. Such an advocate, however, will not be successful unless other faculty take the time to learn about the computer programs, and to participate in the introduction of CAI in the curriculum. In most medical schools, such support is absent, usually because of lack of interest and commitment from the dean and senior faculty To encourage medical-school faculty to become involved in courseuare development, medical schools must develop an appropriate reuard system. For example, instructors who author high-quality courseware should receive credit for their work when promotion and tenure decisions are made.
The high costs of software development would be morejustifiable if the completed CAI programs were shared widely, thus reducing the cost per hour of instruction; however, relatively little sharing has taken place. One reason is simple lack of Euniliarity of many medical-school faculty with the medium. There has been little opportunity to have a medical C~ program published or, until recently, even reviewed in the standard literature. It is difficult to learn what computer-based medical educa tional programs are available, what hardware and software are required, what the cost is likely to be, what evaluation has been carried out, and who has used the programs and with what success. One of the more promising developments is the decision of a number of medical journals to publish sofhsare reviews.
Another barrier limiting the transfer of programs is the lack of a standard approach to program development. Medical schools cannot afford to acquire the software and hardware necessary to run every program of interest. PCs provide a relatively standard and inexpensive medium for the development and use of CAI systems, and several promising authoring systems are being developed. However, there is still a troublesome divergence of both hardware platforms and software tools that makes universal sharing difficult, if not impossible. To promote courseuare sharing, leading medical schools could form a consortium to establish standards for program development. If such a consortium were to commit to the use of compatible hardware and one (or several) standard authoring systems, then programs written at one institution could be used elsewhere. The authoring systems should be designed to permit faculty at other institutions to modify the programs, a key factor in overcoming the "not invented-here" syndrome that frequently limits courseuare acceptance. The development of local-area networks within institutions and wide-area network communications linking geographically disparate institutions also encourages sharing of programs and facilitates access by physicians practicing in the community.
An important aspect of courseuare development that is often overlooked is the integration of computer-based materials with the curriculum. Currently. most CAI materials are treated as supplementary material; they are placed in libraries, and are used by students or physicians on their own initiative. This is a valid use, and the programs serve as valuable resources for the students who use them; however, an educator
can use CAI materials more effectively by integrating them into the standard curriculum. For example, programs might be assigned as laboratory exercises or used as the basis of a class discussion. One of the barriers to integration is the initial high cost of acquiring sufficient computing resources. The cost of the computer equipment has fallen substantially in recent years; nonetheless, the cost of enough PCs to accommodate the entire school would be a major item in the cuniculum budget of a school. This consideration will grow less important as more students purchase their own PCs.
In summary, computer-based
educational syste have the potential to help students to master subject
matter and to develop problem-solving skills. Properly integrated into
the medical-school curriculum and into the information systems that serve
healthcare institutions and the greater medical community, CAI can become
part of a comprehensive system for lifelong education. The challenge to
researchers in CAI is to develop this potential. The barriers to success
are both technical and pracrjcal. To overcome them, we require both dedication
of support and resources within institutions, and a commitment to cooperation
among institutions.
Association of American
Medical Colleees. Physicians for the twenty-first century (Report of the
Pdnel on the General Professional Education of the Physician and College
Preparation for Medicine). Journal of Medical Education, 59:1, 1984.This
report b~ the Association of American Medical Colleges presents recommendorions
to medical schools about curnculum changes that are necessary to prepare
physicians-in-taining to meet the challenges of the current health-care
enrironment.
Kingsland m, L· C- (ed). Proceedings ofrhe ~hirreenrh Annual S)mposium on Compurer Applicarionr in Medical Core. Washington, D.C.: DEEE Computer Society Press, 1989.The proceedings of an annual conference on applications of computers in health care, this volume tar well as others in the series) contains articles on much ofrhe current research in CAI.
Piemme, T. E. Computer-assisted learning and evaluation in medicine. Journal of the American Miedical Association, 260:367, 1988.This article traces the development of CAI in the United States, and discLcrses reasons for the slow growth and acceptance of this technology.
Starkweather, J- A- The computer
as a tool for learning. Western Journal ofMedicine, 145:861, 1986.This
overview article describes current major projects in CAI and discusses
directions for the further development of the field.
1. You have decided to
write a computer-based simulation about the management of patients who
have abdominal pain.
a. Discuss the relative advantages and disadvantages of the following styles of presentation: (1) a frame-oriented question-and-answer format, (2) a controlled vocabulary program focusing on the differential diagnosis, (3) a simulation in which the patient's condition changes over time and in response to therapy, and (4) a program that allows the student to enter free-text requests for infer mation and that provides responses.
b. Discuss two issues that you would expect to arise during the process of developing and testing the program.
c. For each approach outlined in part (a), discuss how you might develop a model that could be used to evaluate the student's performance in clinical problem solving.
d. What are the virtues and limitations of including visual material in a teaching program? How can you integrate the technology of the computer-based program and the presentation of the visual material?
2. Consider a simulation that is designed to teach a student how to develop the dif ferential diagnosis of a patient who has abdominal pain. Develop a strategy and a vocabulary that will, allow the computerto recognize the variety of inquiries that the student might make regarding what the character of the pain was, and what the results of a detailed physical examination were.
3. Create a simple transcript consisting of 10 exchanges of a hypothetical session between a student and each of the following programs:
a. A branching-questioning programb. A coaching programc. A tutoring program
4. What barriers inhibit the dissemination of computer-based medical education pro grams from one institution to another? Discuss three factors that could encourage sharing among institutions.
5. Discuss the relative merits of the computer being "in control" of the teaching environment, with the student essentially responding to computer inquiries, ver sus those of having the students being "in control" and thus having a much larger range of alternative courses of action.
6. Discuss the issues that the National Board of Medical Examiners will face when it attempts to use computer-based simulations to assess clinical competency for medical licensure.