Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
APPENDIX THE DEVELOPMENT AND DIFFUSION Â£ OF A MEDICAL TECHNOLOGY: MEDICAL INFORMATION SYSTEMS* Donald A. B. Lindberg THE CONCEPT OF MEDICAL INFORMATION SYSTEMS Systems Rationale There is no official body to authenticate the definition of the term "medical information systems." Nonetheless most workers in the field are likely to agree that medical information sys- tems have certain attributes in common. It is clear that the phrase implies an automated system, generally a digital computer- based system. The information is about persons, at least that information about them that is relevant to their health, their health complaints, the management of their complaints, and the treatment of their illnesses by health professionals. To most of us there is a strong implication that whatever information the system contains is organized relative to the person to whom it pertains. That is, there is an integrated patient record. Similarly, there is the strong implication that the system it- self is meant to -facilitate treatment or health maintenance of the individuals whose information it contains. Most people familiar with medical computing would assume that a medical information system contained information about patients * I am grateful to Bruce Waxman, Octo Barnett, Lawrence Weed, and Polly Ehrenhaft for kindly providing advice and reference mate- rials, and to the staff of the Health Care Technology Center, especially Pat Mullins, John Simpkins, Don Foster, and Jean Gorges, for assistance in preparation of this report. 20l
202 that aroseâor at least entered the systemâfrom multiple sources. This circumstance has historically provided the major rationale for the development of such systems. Some workers might disagree, however, believing that even single source data systems such as those containing sets of observations made in a private physi- cian's office might be called medical information systems. About a number of other common attributes of past and current medical information systems there would also not be agreement among workers in the field. For example, many workers would stipulate that in a medical information system the integrated pa- tient record would be available to and used by a number of health professionals in addition to the person who initially entered the information. An especially critical issue is the property of some medical information systems, that they are designed and implemented in toto, cut from whole cloth, as it were. Hodge argues that this is an essential characteristic:27 "MIS is achievable only through an integrated approach and direct professional use. ... An in- tegrated approach implies a single integrated system serving the institution. 'Generalized) means application independent." An equally strong argument has been made by others that MIS's to be successful must be built up from a number of subsystems. In this view, the integrity of the system arises in the concept- ualization of the interrelationships between the parts. The com- ponent subsystems are implementable individually in various sequences, and in various combinations. Presumably they may come and go over time, and presumably multiple copies of sub- systems might exist within one overall MIS. Barnett has been an advocate of the view that the MIS must be made up of subsystems: "... developing a modular system is far better for the present, than seeking a so-called 'total medi- cal information system.' Although the long-term objectives of the two strategies are identical, the methods of procedure and the intermediate goals are very different, as is the prognosis for their relative utility and near-term success."7 Collen defined a medical information system in operational terms: "A medical information system ... is one that utilizes electronic data processing and communications equipment to pro- vide on-line processing with real time responses for patient data within one or more general medical centers, including both hos- pital and out-patient services."12 In his view, subcomponents of MIS's may include a hospital in- formation system, a laboratory data system, a hospital administra- tive information system, and (presumably) others. He provides further specifications for immediate objectives and general func- tional requirements.
203 Clearly, both the holistic and the subsystem philosophy could produce workable systems, which ultimately might be indis- tinguishable from one another. I shall, for the purposes of this report, reject Hodges' argument and shall accept that MIS's can be either holistic or can be the sum of the subsystem "parts." Definition In order to facilitate our analysis of alternative approaches to the MIS problem, it is tempting to fall back upon the fundamental approach of the system designer. This is to ask one's self: What data elements are being collected, and what is being done with the single elements and the combination of elements? Essential Data Elements Listed below are the barest bones of the content of a medical information system. That is, there are the minimum essential data elements that would need to be collected, disregarding for the moment all considerations of purpose, usage, and setting. They are: patient identification (number, name, address); hos- pital or ambulatory care location; demographic information (including sex and age and sometimes occupation); past hospital- izations; diagnosis/diagnoses; linkage information (this would vary from one installation to another) Something more than name and number is required in order to link new information trans- actions in this kind of file to the preexisting patient record. Some systems would use items such as maiden name, mother's name, check digits on patient numbers, transaction numbers, source codes, or time-data qualifiers. So far as I know there is no MIS in the United States that collects only these minimal elements. The Danderyd System in Sweden, however, limits itself to not much more than these ele- ments.1 One notes that Danderyd is one of the largest operating MIS's in the world, even though limited in its depth of data. Optional Data Elements There is a host of data elements that are included in some systems. These are: billing information; insurance status; physician(s) responsible for the patient's care; invariant phys- iologic information (e.g., blood type, leukocyte antigen type[s], chromosome karyotype); elements of health status provided by the patient (e.g., complaints, history of present and past illnesses, health status as reflected by daily work and personal functions, immunizations, etc.); results of measurements or observations
204 performed upon the patient (e.g., height, weight, laboratory test results, ECG's, radiological studies, results of physical examinations, including the traditional physician's general ex- amination, and special examinations such as range of motion, proctoscopy, pelvic and cystoscopy, etc.); and interpretive in- formation (generally provided by the physician, such as problem list, provisional or working diagnoses, treatment plans, thera- peutic and diagnostic orders, descriptions of surgical procedures. CURRENT STATE OF THE TECHNOLOGY OF MEDICAL INFORMATION SYSTEMS The data elements collected by a particular medical information system identify it in a kind of common sense way. The various circumstances in which many MIS's have been operated, considered with the purposes for which they were operated, may be taken to characterize the state of the art of such systems. Using this kind of a general analytic framework, we can examine the extent to which the entire possible range of MIS settings (i.e., the problem domain) has actually been explored, and to what extent applications have been successful. In this sense, each general aspect of an MIS may be considered as a dimension that bounds the problem domain. Dimensions of a General Analytic Framework The "dimensions" of a general analytic framework are: (l) data elements collected, (2) functions that are being performed, (3) medical service area (usually a hospital or ambulatory care de- partment or division), (4) health care of institutional setting, (5) patient population, (6) uses of the output of the system, and (7) financial basis of the system. Each of the individual "dimensions" have subdivisions or multiple values. The data element content and the function undertaken by a medical information system are interrelated. That is, a func- tion may not be possible unless a data item or group of data items has been collected. However, collecting data elements (with purpose A in mind) does not assure that they will be used (either for purpose A or purpose B). Collection of data elements is a necessary but not sufficient condition for the functions to occur. Careful consideration of the dimensions will convince the reader that the interrelationship between the dimensions is not limited to data content and function. All of the dimensions are
205 in some ways interrelated, and in other ways unrelated. Con- sequently, the dimensions are not independent in a mathematical sense. I make no pretense that it is clear how to assess this technology in any overall sense by numerical techniques. I assert that the potential contribution of medical information systems to progress in health care, and the assessment of the present state of the technology, can only properly be done within the true full problem domain that is characterized by the dimen- sions named. One may define a medical information system by selecting one or more attributes from each of the seven dimensions or aspects. Together they will define a valid form of medical information system. Example: Feasibility of MIS's With Many Kinds of Medical Service Functions When we examine the current and past MIS's along one sample di- mension, i.e., medical service areas, it is clear that a con- siderable range of application systems have been feasible. The medical service areas have included: Admissions Office;14'16,31 Business Office;5,23 Medical Record Department;2,l4,17 Clinical Laboratory;2'35'38 Radiology;5'40,42 Electrocardiog- raphy;10,11,52-54,60 Intensive Care; 30 , 51, 55, 58 Obstetrical Care;46 Mental Health;15,20,24,59 Pharmacy.5,21,29,32,45 Each of these sytems has other dimensions, or secondary character- istics, such as type of patient population, institutional set- ting, etc. Space does not permit presentation of the full characterization of feasible systems. Tasks Thus Far Infeasible Not all applications of MIS's will necessarily prove feasibleâ even in a technical or scientific sense. From outside the medi- cal domain, the best-known example of an infeasible computer function is automatic language translation. There may well be similar functions in medicine that appear to be reasonable but that will remain technically infeasible. The following set of important elements that have thus far proven infeasible, I offer merely as a personal view: (l) cre- ation of a generally usable thesaurus with a standardized medical terminology, (2) processing of free text medical entries, (3) processing of text diagnoses, (4) execution of a general diag- nostic logic, (5) automated certification of clinical medical
206 competence, (6) positive specimen identification in the labora- tory, (7) automatic clustering of symptoms so as to recognize new clusters, (8) automatic analysis of cardiac arrhythmias, (9) recognition of n-tuple drug interactions, (l0) automation of even simple treatment plans, (ll) basic improvement in clinical laboratory control, (l2) a practical means of recording the gen- eral physical examination, (l3) a computer programming or even command language for use by the medical staff as opposed to the expert programmer. DESCRIPTION OF EXAMPLES OF MEDICAL INFORMATION SYSTEMS Systems Designed for a Particular Institution There are medical information systems that have been successful in more than one medical application area simultaneously, and certainly have been successfully used by many kinds of health care professionals for many different purposes simultaneously. By considering a few of these, albeit briefly, one can achieve a kind of gross inspection of the rather considerable extent to which the total problem domain has been explored. The specific instances to be reviewed areâlike any other short listâan arbitrary selection. Examples are: Institute for Living, Hartford, Connecticut. Overall Description: This medical information system serves an entire 400-bed private mental institution. The system grew out of l5 years of experience with computer-based information systems under the direction of Bernard Glueck. The institution has been a pioneer in such developments. The primary purpose of the sys- tem is to facilitate patient care. It utilizes two DEC PDP-l5 computers; the code was written in the MUMPS-l5 language. Detailed Description: The system provides an integrated patient record that includes essential data elements, plus re- sults of psychological test instruments, nurse's progress notes, general physical, medications, and recordings of diagnoses.21> f2^,^1 Special Features: Experimental programs for prediction of patient behavior, optimum classification for therapeutic purposes, and various customized searches and reports in support of research and management objectives. System costs are entirely recovered from patient fees at $2.60 per day. Clinical Computer Applications: University of Utah, Salt Lake City. Overall Description: Many subsystems have been created within the Department of Biophysics and Bioengineering of the
207 University of Utah, under the direction of Dr. Homer Warner and his associates, with implementation within Latter Day Saints Hospital. The systems operate (generally) on CDC computers un- der the Med Lab operating system. Detailed Description: The medical application areas and functions include: a computerized integrated computer patient record with input from physician's diagnosis; automated multi- phasic screening; computerized electrocardiogram interpretation; automated compilation of various laboratory tests, including blood gases; pulmonary functions; clinical chemistry; analysis and interpretation of cardiac catheterization; and physiological monitoring of intensive care units. Output of the subsystems is used primarily by physicians and nurses for direct patient care. Output of the integrated record system is also used for quality of care monitoring by hospital committees. Functions include many physician assistance systems related to diagnosis and interpretation of measurements.2 ,Â°Â° Special Features: The HELP system is an advanced attempt to formalize the medical logic that interprets and integrates the numerous data elements incorporated with the automated pa- tient record system. Massachusetts General Hospital Computer System, Boston Overall Description: There are a number of operational sys- tems and subsystems that were created by the Laboratory for Com- puter Science under the direction of Dr. G. Octo Barnett and associates. They all operate from terminals to DEC computers and were written in the MUMPS language. Detailed Description: COSTAR is a computer-based system for ambulatory patients served by the prepaid Harvard Community Health Plan. It provides the only record for about 60,000 pa- tients. 6,44 In addition, other subsystems in the hospital it- self handle the data acquisition and recording for clinical chemistry and bacteriology. Another subsystem provides physician assistance functions in an anticoagulation clinic. Others as- sist undergraduate medical education through automated testing and patient simulation. Special Features: The roost outstanding special feature of the programs themselves is their ability to share and modify files using the MUMPS language. Commercially Offered Systems Systemedics prepared a survey of commercially offered automated hospital information systems for the Health Resources Administra- tion in October l976.62 Battelle Laboratories conducted an ex- tensive evaluation study of one particular commercially offered
208 automated hospital information system: that of Technicon Cor- poration, installed in El Camino Hospital, Mountain View, Cali- fornia.8 The present report draws from these previous studies and from vendor literature. It should be noted that all the systems referred to in this section operate at a level of medical sophistication that is far below that seen in the noncommercial systems designed for particular institutions. The latter are generally university-based, and often serve a research as well as a service function. The merit of the commercial systems is that they are designed, potentially at least, to be replicatable in other institutions. An additional strong point of the com- mercial systems is that they have succeeded in implementation across a substantial number of hospital departments. Thus they are made up of a relatively large number of the subsystem com- ponents that are the building blocks of the ultimate mature medi- cal information systems of the future. Along with the relative breadth of coverage across the hospital, which is offered by the commercial systems, there is a tendency for a somewhat shallow approach to any individual medical area. Medical information systems are offered commercially by Bur- roughs Corporation, Data Care, IBM, McDonnell-Douglas, National Data Communications, Inc./Honeywell Corporation, and Technicon Corporation. Historically it should be noted that the McDonnell- Douglas McAuto System was developed jointly with the Sister of the Third Order of Saint Francis. The NDC system was originally developed by National Data Communciations, Inc., later marketed jointly with Honeywell (since it utilizes Honeywell computers), and later marketed once again solely by NDC. The Technicon MIS system was purchased from Lockheed Corporation and subsequently enhanced. There are currently six installations of the Technicon system. There are four installations of the medical version of the McAuto system and four installations of the NDC/Honeywell system. Bur- roughs, IBM, and Data Care each have one or two hospital instal- lations of medical information systems. The actual count on any given day of just how many MIS commercial systems are operational is always subject to dispute. This is because installations will come and go, and there is no central registry for recording new sales and/or failures. Companies competing in this field also come and go. Since the Systemedics survey was completed, two new companies have announced medical information systems. These are Shared Medical Systems of King of Prussia, Pennsylvania, and Medicus of Chicago, Illinois. The last reason for ambiguity con- cerning the actual number of such installations is that it is the marketing policy of some companies, for instance Shared Medical Systems, to encourage the installation of incremental functional modules over and above the basic accounting packages, so as gradu- ally to convert towards a medical information system.57 This is the apparent policy of Medicus Systems as well.47
209 All six vendors offer application options that include the following hospital areas: admissions, medical records, pharmacy, laboratory, radiology, nursing stations, dietary, administration, business office, emergency room, and outpatient departments. Three of the vendors also offer packages for heart station, util- ization review, and surgery. The extent to which these syterns can support extensive and sophisticated functions at all of the hospital areas named is said by the vendors to be essentially unlimited. They do caution, however, that the systems must be tailored to local hospital standards and procedures. This is a reasonable limitation in cases such as the standard battery of orders and treatments. Such commercial companies, as, for ex- ample, Technicon, take the position that they do not propose to offer standardized medical treatments as part of their systems. They insist that these items be undertaken by the hospital staffs. In some respects this is not only wise but advantageous. The Technicon system in El Camino offers the very desirable op- tion that medication orders may be completely tailored to each individual physician. That is, after identifying himself through a keyboard entry, the doctor can call up on the cathode-ray tube his own particular set of commonly used orders and/or drug speci- fications. The system provides him with the capability of enter- ing these to begin with, and of altering them at will. Each of the commercial systems varies, but all have roughly the same operational characteristics. One of the consequences of the necessarily disciplined approach to data entry is to enhance completeness. The systems have the ability to prompt. The six systems differ in the data entry modalities employed. The Technicon system utilizes CRT's and a light pen. The NDC/ Honeywell VITAL system also utilizes a CRT terminal, but this includes special function buttons and a badge reader. McAuto and IBM systems utilize a variety of standard terminals. The first system, Technicon, is designed to encourage direct data entry by the physician; the last three systems present this pos- sibility, but do not normally operate with physician entering either data or orders. Some of the commercial systems are quite sizable. The number of computer terminals employed is: Technicon Corporation, Maine Medical Center, Portland, l07 terminals; Technicon Corporation, El Camino Hospital, Mountain View, California, 56 terminals; NDC/ Honeywell, Deaconess Hospital, Evansville, Indiana, 83 terminals; McAuto, Missouri Baptist Hospital, St. Louis, Missouri, 40 ter- minals. The cost of these systems is said to vary between $4 and $6 per day per patient. There are limitations upon all the commercial systems de- scribed. In no case do the systems accept physicians' progress notes, the patient history, nor the results of the general
2l0 examination by the physician. Since these three types of infor- mation are central to the traditional medical record, it appears that the present-day commercial systems fall quite short of being truly medical information systems. Systems of Historical Interest The G.E. Medinet Division was established in June l966, and its completed computer center and permanent offices dedicated on May ll, l967. It proposed to offer computer services of a wide variety to health care institutions anywhere in the country via time-shared computers and connected by dedicated telephone lines and transmission networks. The parent company, General Electric, already had the advantage of an extensive telecommunications net- work on a nationwide scale. Medinet)s initial plans called for services to be provided in the following hospital and clinic application areas: admission and discharge, laboratory reporting, doctors' orders, pharmacy and medications, medical record statistics, inventory, patient billing, and hospital payroll.48 At its peak it employed l06 individuals in the categories of programmer and hospital or systems analysts. Hardware develop- ment included production of 400 custom computer terminals based on the KSR33 teletype. A central computing complex was built up based on the GE-485 with smaller Honeywell process control com- puters in the proposed message network. Software developments included creation of a new string-processing interpretative lan- guage and special file procedures. The system was demonstrated at the American Hospital Associa- tion meeting on August l2, l967. At this time, a press release announced that "nationwide availability of the system to the medi- cal community would not come until late l968 or early l969." A major administrative reorganization was announced publicly in December l967. Early in l968 it ceased to make offerings of medi- cal computer services and elected to make offerings of business office and administrative services, aiming at a potential of 600 hospitals. At this time, Medinet had spent about $l6 million on systems development, mostly aimed at the medical information sys- tems market. The Medinet Division was eliminated as an administrative entity in April l975. Kaiser-Permanente MIS: By far the most advanced of all American general MIS's was that at Kaiser-Permanente. The destruction of what had been built can most simply be attributed to unique and unfortunate circumstances.^
2ll The original multiphasic screening system had been built as a response to internal company needs and client demands. Twelve years of development were financed internally and successfully before any government research funding was accepted. At its furthest development, this system included extensive general outpatient screening and follow-up data; data for a spe- cial, prospective study on benefits of periodic health examina- tion on a population of l0,000; an emergency room patient record and physician's assistance function; prospective records on pa- tient pharmacy records; and a developing system of hospital ter- minals for orders, record-keeping, and reporting. At a critical juncture, when there were very serious technical problems during expansion, the two major sources of federal fund- ing were precipitously withdrawn. These were: designation as one of a dozen Health Service Research Centers, followed by the elimination of the federal program for all such centers, and can- cellation of a major contract with FDA for important prospective drug reaction and toxicity studies. The Medical Methods Research Group was in essence pulled off- balance by the elimination of research support of developments that had been premised upon a balance of service and research funding. Missouri laboratory systems: The laboratory system based upon on-line computer terminals at the University of Missouri was an early one, developed in l963. This system succeeded an even earlier off-line automated reporting system. The off-line sys- tem was developed with institutional funds. The on-line system was developed with NIH research grant support. Taking the on-line system as an example, one may note that the technology was successfully implemented within budget and on schedule in 3 years. Follow-up and internal evaluation studies were completed.37 The host institution assumed the entire cost of the operation of the system. In many ways the system achieved all of its objectives, and succeeded in satisfying the then cur- rent NIH standards for success. Namely, it "worked," and it was adopted by the hospital. The system continues to this day (with evolutionary improve- ment) . It has had no federal research support for ll years. On the negative side, however, is the fact that the absence of research support for this application has meant that no re- search has been done on the system per se for ll years. Con- sequently, it is now patched, inefficient, out of date, and incompatible with the many advances and changes in laboratory technology which have arisen in the past ll years. The programs are still doing their job. There is no parallel manual arrange- ment. Yet this system is no longer innovative. Consequently, it is no longer able to exert a beneficial influence upon the
2l2 development of laboratory systems elsewhere, nor to enhance the diffusion of MIS technology more generally. BARRIERS TO THE DEVELOPMENT OF MEDICAL INFORMATION SYSTEMS Operational Difficulties Throughout many system developments, implementations, and evalu- ations, certain difficulties have been reported by the authors involved with surprising consistency. While it is difficult to classify each and every problem, it is clear that these bar- riers are technical, social, and managerial. For this reason at least some of the barriers to MIS development could be con- sidered to be similar to those associated with any large and com- plex system effort. Nonetheless, they will be recapitulated briefly, since these are the raw data of our review. After this we will examine the possibility that two categories of obstacles are more or less inherent in the particular medical application area. A Variety of Problems Described by MIS Designers Many investigators identify their problems in terms that echo Collen's formulation.13 After an analysis of successful and un- successful attempts to produce operational medical information systems, Collen concluded that the reasons for failure were: "(l) a suboptimal mix of medical and computer specialists . . . (2) inadequate commitment of capital for long term investment ... (3) a suboptimal system approach" (that is, either in- ability to fuse incompatible subsystems or a too grand initial total design) "(4) . . . unacceptable keyboard terminals . . . (5) inadequate local management." In his view, factors that would tend toward success would be correction of all the above deficiencies. He feels that for success "a hospital (or group of hospitals) of sufficient size is required, with effective organization and management by technically sophisticated men who can make reliable decisions after considering technological alternatives." In a recent survey of automated ambulatory record systems, Henley and Wiederhold summarized the "problems encountered" at each of the system sites visited. Their list is remarkably sim- ilar to Collen's, and like his is a combination of technologic difficulties and social and management problems. Examples of the former are: "man-machine interaction limited" and "diver- sity of ... record formats" and "voluminous text." Examples
2l3 of social barriers include "lack of interest from . . . manage- ment," "poor management interface between [university and health care facility]" and "city funding intermediary has other pri- orities."26 Another viewpoint on barriers and limitations is presented by Friedman and Gustafson.22 Among other things they conclude that a critical fault has been in "not succeeding in producing appli- cations which exceed the capabilities of the physician without the computer." In a review of 28 computer projects in health care, Giebink and Hurst in l975 report in each case on the developmental and operational problems as perceived by directors of the projects.23 This valuable list contains the same mixture of technical and social problems. Examples of the former are: "moving head disk hardware problems," "no back-up computer system," "slow response time of terminals and unreliable terminals," "major technical problems in computer representation of extensive medical logic," and "overly optimistic expectations for speedy implementation." Explicit examples of social barriers are: "clinics' resistance to change, and social problems resulting from change," "poor acceptance of the system because of 'top down' decision making in its creation," and "insecurity of future funding." Difficulty with the management of complex systems was also a theme repeatedly reported. One project manager said that the systems finally became so complex that he could not implement pro- gramming changes and additions without intolerable delays.13 Re- grettably, that particular project manager is himself experienced and extraordinarily competent. Much simpler difficulties in the management of technical work forces baffle other projects. There are excellent managers for highly technical large tasks (witness the NASA successes), but they are rare and apparently not often associated with MIS developments. The Particular Problems of Computer Hardware and Software While there have been vast improvements in computer hardware during the past 20 years, there are still obvious major defects and obstacles. Awkward, slow, expensive computer terminals have been an impediment to all computer system builders. Likewise, all users suffer from the malfunctioning of moving head disk in- formation storage devices. In spite of encouraging increases in disk storage capacity, large medical record files still frequently exceed storage capacity of many systems. In a sense this issue is something of a trade-off against costs. That is, at a greater cost one can often obtain an increase in storage capacity. Yet the combination of costs and direct access memory capacity re- mains a general problem.
2l4 Computer reliability has likewise improved enormously. Yet people expect more of computers now. Clinical systems must be reliable, just as many other critical control systems must be. In the case of missile launches, the costs are acceptable. In the hospital, they are not currently acceptable. One example of this problem, and a workable solution, is the Technicon MIS at El Camino Hospital. A backup computer is avail- able at a service center for reliability. The records given Wiederhold26 showed that the backup machine was used an average of 33 minutes per day. When billed by the minute, the redundancy is quite inexpensive. In this case, Technicon is able to use the backup machine profitably in other tasks. Had the hospital been obliged to keep a second machine, as Kaiser had to for their pharmacy and ward terminal system, the costs of the system would have been doubled. Reliability is still a major impediment to medical information system implementation in most settings. Software development costs remain high, and progress remains slower than in hardware systems. The software interface is still with the computer professional, rather than with the physician user, or with the other subject matter experts. This tends to increase personnel costs and to make management of system rede- sign difficult. Worse yet, it tends to separate the health care professionals from direct participation in the creative aspects of the application development. Medical Barriers to Medical Information Systems Development There is nothing about the computer techniques used in medical information systems that makes them in any way fundamentally dif- ferent from such systems in nonmedical fields. There are, how- ever, two special nontechnical barriers that have to some extent been inherent to the medical application. These are: limitations on the state of medical knowledge about illness and health, and limitations on the state of medical systems management. Limitations of Medical Knowledge There simply is still much art in medicine. Often this is inherent to our ignorance of basic bodily mechanisms and mental processes. Friedman and Gustafson warn that we must try to make computer systems that do things physicians cannot do.22 Yet at least one barrier to this happening certainly is to identify the person with the idea. There are ways to encourage more people, and even more imaginative people, to join the field. Still, the fundamental barrier is the idea for the significant new medical function that can be accomplished with the help of the computer.
2l5 It is the occasional insights into useful new information systems tasks that create major benefits from computer-based systems. The alternative is merely automation of current practices. In the use of data base systems for patient and medical rec- ords, this difficulty is most apparent. For example, after one knows or guesses that the variables "first trimester pregnancy" and soporific "Thalidomide" are relevant to the diagnosis "phocomelia," it is technically easy to construct the appropri- ate data base system for patient records. The same is true of preparturient estrogen therapy and endometrial adenocarcinoma, or, of leukocyte antigen typing and spondylolisthesis, and also of a variety of industrial carcinogen exposures and drug-drug interactions. In the absence of such fundamental health knowl- edge, enormous complexity is required of the data base systems in order to "shoot in the dark" searching for relevancy among the patient record variables. Undesirable medical practices stem from this incomplete knowl- edge. These include irregular terminology and ad hoc identifica- tion systems. Both present serious barriers to automation. The language of medicine is simply unstructured. This lack of stan- dard vocabulary is a considerable obstacle to the creation of medical information systems, as well as to the transplantation of a given successful system to other locations. The second difficulty is comparable but not so easily solved: the problem of identification of individuals, their medical samples, and observations about both in a computer-based infor- mation system. Corn flake boxes and railroad cars are now made with "zebra stripes"; people are much more difficult to identify. Limitations of Management of Medical Systems The second barrier that is peculiar to medical information sys- tems is the medical environment, or what is now more properly called the health care systems environment. The U.S. health care system is made up of thousands of relatively autonomous units, centering on large hospitals, which are themselves made up of relatively autonomous divisions and departments. There is no common ownership nor meaningfully common accounting system. In addition there is an apparent shortage" here too of individuals capable of managingâor even rearrangingâcomplexity. To the extent that health care institutions do not work smoothly and sensibly with one another, the medical information system cannot be shared or transplanted. To the extent that health care insti- tutions are balkanized into small administrative parcels, the information systems must of necessity be small as well. It is quite clear why minicomputers are so popular in medicine, and why
2l6 large data base systems are so rare. The mini system matches the miniadministrative fiefdom. The large data base systems rep- resent one of the many institutional goals to which the institu- tions often cannot manage their way. Socioeconomic Barriers to Medical Information Systems Development Health Professions' Response to Information Systems Technology Clearly technical changes can and generally do occur more rap- idly than society's adjustment to them.49,63 This has certainly been true of computers in general, medicine in general, and the combination of the two fields in particular. Evidence for this can be seen in the shortages even today of skilled and experienced computer technologists and the even smaller number of medically trained individuals who have experience or cross-training in com- puter or information science. Medical school curricula have been notoriously slow to change, so that none currently provide formal courses in computing to students of medicine as a part of their regular curricula. Likewise no testing for information-processing skills is included on the examinations of the National Board of Medical Examiners. This is in spite of the fact that the M.D. f\ Q exam itself is totally processed and scored by computer. Postdoctoral training in medical computer work is provided at eight institutions.50 The oldest program is 5 years old. Problems with operational medical information systems also attest to the claim that social engineering proceeds less rapidly than hardware engineering. There are repeated mentions of dif- ficulty in getting communication between medical and computing personnel on the same research team, and in establishing communi- cation and cooperation between health care institutions in the same city. What is demanded of the development of medical information sys- tems is creativity and technological innovation. No one knows very surely how to manage the creative process, in science or else- where. The building of medical information systems is known to require teamwork by a multidisciplinary group, which complicates matters by adding a substantial management problem. Furthermore, the activity is expected to proceed in the face of unstable fi- nancing, intermittent encouragement from government, and disincen- tives from medical specialty societies. None of the latter have requirements for nor "give credit" for computer and information systems experience as part of their postgraduate educational cur- riculum. Hence, they strongly penalize those who invest post- graduate training years in obtaining computer skills.
2l7 All in all, there is a long way to go before social adaptation to the computer has caught up with the technical state of the art. The Computer Industry's Response to Medical Needs There has at times been moderate enthusiasm for developing and marketing medical information systems. In l973 Ball reviewed l5 commercial systems.4 By l977, six of the companies were not in the MIS business. Major companies leaving the computer field altogether have included RCA and GE. IBM, an early enthusiast for medical inform- ation systems, has virtually abandoned medicine as a high-priority marketplace. There are of course still large numbers of computers going into hospital business offices to do accounting jobs. One can only con- clude that industry takes its mandate from stockholders to maximize profits over the long run, and that medical applications, espe- cially medical information systems, are not judged to be the most profitable investment. Yet there are profits to be made elsewhere in the medical com- puting field. Witness the installation in the United State alone of 320 computer tomography units (i.e., CAT scanners) at about $500,000 each with another 224 units approved and the orders on backlog as of July l976. Witness ll U.S. companies in the market- place, and 8 foreign competitors. Note that all this has trans- pired since Hounsfield's first demonstration in England in l971 and in the United States in l972. CT and MIS developments have much in common. The computer tomography problems are both chal- lenging and scientific. They involve multidisciplinary teams for advances in the state of the art. The state of the art has ad- vanced, indeed quite rapidly. Third-generation equipment has been announced while there is still a backlog of orders for first- generation EMI scanner. Contrasting the problem of slow development of medical informa- tion systems with all too rapid diffusion of the technology of computerized tomography presents the question: Why the differ- ence? One can only conclude that industry does not "respond to economic and scientific challenge" at all. Industry responds to opportunities to sell hardware at a profit. Customers likewise seem to buy more of those hardware items with which they in turn can increase earnings. Computer tomography units increase hos- pital and professional earnings. Automated laboratory instruments increase hospital and professional earnings. Medical information systems cost everyone a great deal of money and anguish. If the desired outcome of the development of medical informa- tion systems is really recognition of new diseases, or creation
2l8 of community medical data bases, or more complete reporting to PSRO or Medicare intermediaries, or shorter hospitalizations for illness, or recognition of occult disease in ambulatory population, then someone is needed to pay for these outcomes. All the outcomes are desirable (and even cost justifiable) on the basis of society as a whole and within a time frame of dec- ades. Yet each of them either decreases the revenues of an in- dividual hospital, or increases health care overhead in the short term, or both. In no case do any of the most desirable outcomesâ so long pursuedâpresent the opportunity to offer the manufac- turers the kind of hardware sales "challenge" that compares with either simple automation or new and costly measurement technol- ogy, nor even just selling a few more accounting systems. IMPACT OF PUBLIC POLICIES ON THE DEVELOPMENT, ADOPTION, AND DIFFUSION OF MEDICAL INFORMATION SYSTEMS Federal Research Policies Research Support for Computers in Medicine The creation of hospital accounting and business office computer-based information systems has proceeded on the basis of local funding and commercial development and sales. This appli- cation area for computers has been recognized for many years as more or less analogous to business office and accounting func- tions in other institutions. Hence it has always been an oppor- tunity for entrepreneurial development, the success of which could be measured by cost reductions, labor savings, or at least some kind of suitable cost displacement. In contrast, attempts to develop research applications of computer-based information systems in medical areas has neces- sarily had to be contingent upon funding cum research. Since the end of World War II, the federal government has become far and away the largest funder of medical research. Consequently, it was of great importance to the development of medical information systems that such efforts were formally recognized as legitimate research. This recognition came in l960 with the establishment at the National Institutes of Health of the Advisory Committee on Computers in Research, which was charged to define general areas of biomedical computing and to stimulate interest in them.43 This group was established as a regular Study Section on Computer Research in l964. It became the Computer and Biomathematical Sciences Study Section in l970. The Study Section was abolished on June 30, l977. With the creation of the Health Services and Mental Health Ad- ministration and its National Center for Health Services Research
2l9 and Development in l969, additional study sections were created that had the ability to support some aspects of the development, diffusion, and evaluation of the technology of medical informa- tion systems. Certain special aspects of such systems have also been supported by the National Library of Medicine through the Biomedical Library Review Committee. The NLM has also been the main support for training programs to provide the special education and experience, both pre- and postdoctoral, to individ- uals entering this field from medicine, as well as from the computer-related disciplines. It should be emphasized that, regardless of whether one consid- ers the priorities and funding policies of these institutions to have been wise and/or consistent, they did nonetheless give le- gitamacy to attempts to explore and define research users of medical information systems. Government-Sponsored Computer Centers The initial research grants from NIH in this area took the form of facility support awards. These were made to encourage and subsidize the creation and operation of computing facilities in selected major medical centers. The purpose of the facilities was by no means specific to medical information system develop- ment. Rather, they were to provide appropriate and convenient computational support to biomedical investigators in the local or regional environment. Efforts to develop patient data base sys- tems came relatively early. The most general rationale was rea- sonable enough: namely, to render the records of patient care suitable to be the subject of research. The NIH supported computer centers were roughly comparable with the general university computer centers that were obtaining financial support at the same time from the National Science Foundation. Both agencies soon found that the country's appetite for such funds was large, indeed beyond the agencies' means. Both began encouraging the computer centers to shift to "fee for service" mode of operation, so as to be self-sustaining after the initial federal subsidies were withdrawn. University and medi- cal facilities managers were encouraged to take a strong adminis- trative hand, so as to shift the operational costs to the individual institutions. Similarly, individual investigators were obliged to budget for, request, and defend computer use charges as an integral part of their individual grant applica- tions. The various scientific study sections were instructed to honor these requests (when the research proposals themselves were meritorious), because computing was no longer free. The net effect of these moves was to encourageâalmost to com- pelâthe development of large central computing facilities at
220 major medical centers. This has turned out to be a mixed bless- ing. At least, however, it was accompanied by a legitimization of attempts to utilize information systems in support of medical research, this is to say, to support the asking of new questions about human health and disease, and to support attempts to do new things in health care by using the emerging computer-based information processing technology. Effects on Medical Information System Development Biomedical research support has been provided through rela- tively short-term funding of generally modest size. Regrettably, this pattern has never been suitable for computer projects. Lusted said that even in l960 it was apparent that computer grant applications differed from all other kinds of grand re- quests by being for larger sums of money and by including the puchase of computers (certainly a long-term commitment) within the request. Development of information systems has never been compatible with year-to-year funding mechanisms. The fact that the professional literature is not full of such pleadings is merely an artifact of the policies of refereed journals and the chameleon nature of information system developers. After conducting a survey of health computing projects, Giebink and Hurst2^ conclude that funding policies placed a "pre- mium on quick transitions from applied research to operational demonstration." They identified two serious undesirable con- sequences of this policy. First, systems were declared opera- tional before they actually had been fully developed. Second, "much research essential to subsequent development" was never performed. Their implication is that there was not time within the original grant award period to complete all essential re- search, and that it could never subsequently be justified for funding because the investigator was obliged to "go operational." Encouragement of Medical Information Systems Development In the light of ever clear hindsight, one must conclude there has not been a systematic federal government plan for deployment of the technology of medical information systems. Several is- sues require special notice. These are: (l) mention of formal views concerning the sequence by which technologic innovations can progress, (2) the time frame for such accomplishments, and (3) the magnitude of developmental c6sts.
22l Sequence of Technologic Innovation Students of history, sociology, and engineering note a number of stages that technical innovations go through before becoming accepted as traditional marketplace items or services. Fre- quently the sequence is represented to be: research, develop- ment, demonstration, commercial prototyping, production. Some writers show "invention" or "discovery" preceeding this list. Some show "marketing" as following the list. Whatever terms best describe this process, it is clear that transitions between the stages occur at irregular intervals and that total costs for each phase typically increase as one moves toward the market. Certainly it is clear that there is a major hidtus between a scientifically interesting and valid research system and any- thing that could be considered commercially practical. For medical information systems, the hiatus between scien- tific success and successful demonstration in a somewhat more practical environment is mirrored geographically by the distance from the NIH campus at Bethesda and the HSMHA (later Health Re- sources Administration) in Rockville, Maryland (later moved to Hyattsville). Projects have existed in which demonstrations of information systems (and other health care innovations) required support. The Health Resources Administration has been authorizedâal- though not always sufficiently well fundedâto support such dem- onstration and evaluation projects. The transition from research support by NIH to HRA support for hospital demonstration can be rough, but some projects have survived it. The transition from HRA support to commercial viability is unheard of for anything as large as a medical information system. Officials of funding agencies and peer review groups who undertake review of research grant proposals have been unaware of the problems of transition. It has always been considered a wise plan to propose withdrawal of federal funding and sub- stitution of hospital or institutional support for systems once they have achieved their research objectives. On the other hand, one must acknowledge that very frequently the hospital support has not been forthcoming. The results of scientific research, that is, new knowledge do not alwaysâeven oftenâsave money for the institution in which they were developed. In a noncomputer field, the discovery and creation of the polio vaccines did not save any money and surely did not make money for hospitals. Indeed they eliminated the patronage of polio victims. Likewise, a computer system that has.'solved some aspect of medical records processing, for instance how to record automatically a patient interrogation history, cannot save
222 money for the teaching hospital in which the system will typi- cally have been developed. Consequently, the funds for operat- ing, transplanting, exporting, and/or expanding such a system cannot be demanded of the host hospital. In brief, there is no clear path for a research system to follow that can provide transition to a practice setting, unless that system makes money for its original host institution. In the absence of a strong national policy to manage technical in- novation, the development must add to health costs or it cannot â¢ succeed. Time Frame for Accomplishments In the simplest possible terms, it has been the mission of NIH to support research, and it chose to include medical infor- mation system research in this mission. The support has always terminated once the scientific success of the project has been declared (or earlier in the case of a failure). Research grants have been typically l to 3 years, never more than 5 years. This time frame may be consistent with the conduct of the research phase of a project. It clearly is not consistent with the time frame needed to bring a medical information system to the stage of even a prototype commercial system. Such systems have been shown to require up to l0 years for prototype development.^,26 Completing a study "on time" in the terminal year of a grant has presented a serious problem: How to provide for transition of the system to a self-sustaining basis. Often this translates to: How to get the hospital administration to pick up the costs. The Magnitude of Developmental Costs Since no real example exists of a full Medical Information Sys- tem in a practical environment, it may be illusory to speak de- finitively of developmental costs. Nonetheless one can reason from the substantial developmental costs that have been reported for existing partial systems. Henley and Wiederhold report on development costs for nine operational systems that had been de- signed to be full MIS's for ambulatory patients.9'26 Costs for the nine ranged from $230,000 to $l0,000,000. Five of the nine had development costs greater than $l,000,000. The five had an- nual costs for continuing development that ranged from $l54,000 to $539,000. This study did not include hospital MIS's. In this category, the National Data Communications/Honeywell system cost $l2,000,000 to develop.62 Development costs for the Technicon MIS were reported to be $25,000,000. It is a serious problem that no operational unit of the De- partment of Health, Education, and Welfare actually has grant budgets of the magnitude required to support the big systems.
223 Research grants typically run $30-$50 thousand per year. Re- search grants for computer work typically are somewhat larger: perhaps $25-$l50 thousand per year, with really large ones reach- ing $400 thousand per year. In the aggregate, of course, DHEW spends substantially for support of biomedical research. The amounts available to single functional offices, however, are typically small compared with the magnitude of commitments that appear to be required to "see through" the development of a medical information system. Health Care Reimbursement Policies The most profound effect upon medical information systems cer- tainly was the initiation of the Medicare system under PL 89-97 (42 USCA l395 et seq.). This program, with its series of en- titlements of an ever larger number of individuals, provides for reimbursement to hospitals (and under Medicaid to physicians and other providers) of actual costs of care for citizens 65 and older (plus other entitled groups). The enumeration of charges, the justification of costs, the certification of entitlement, and the huge cash flow problems associated with delayed and partial reimbursements have forced all hospitals to devote greatly increased resources to these business office matters. The expenses are "reimbursable" and as such are folded into the rising per diem bed charges. With space at a real premium in most hospitals and trained clerical personnel never plentiful, most hospitals have been quite willing to shift these costs to support of computer installations or services, and to support of the administrative portions of medical information systems. In- deed these federal programs have created whole industries within the computer field that prosper mightily in computing, printing, and even reading the documentation of the health care services required for "third party" reimbursement. An example of the' new computer services (outside the field of medical information systems per se) is Electronic Data Systems Corporation in Dallas. Computer processing of Medicare claims by this company constituted a $l33 million business in l976. 6lf With respect to the effect of this legislation upon hospitals, figures are a bit hard to come by concerning the increases in business office personnel associated with Medicare per se. Everyone agrees they are substantial. One isolated example may be offered. The University of Missouri Medical Center in Colum- bia found that, as a direct effect of Medicare, it had to in- crease its business office claims processing staff from 25 to 32 and increase the record room staff from 22 to 26. In l976, 82 percent of the 2,800 hospitals with l00 beds or more had com- puters or used computer service bureaus.62
224 This strong increase in demand for accounting-oriented in- formation systems for hospitals and clinics has had absolutely no effect upon the development of the medical components of medical information systems. How can this be? It's really quite reasonable. A system does not need to con- tain any medical information in order to print a valid and cur- rent bill, any more than the Texaco Company needs to understand one's vacation plans in order to forward gasoline charge tickets. The hospital accounting systems are not really simple, but they are well within the state of the art. No research is needed. Their purpose is to collect bills, either from the patient (who used to pay far smaller and simpler statements than he now re- ceives) or from the federal government's Medicare intermediaries. Adding medical information to the accounting systems would be costly. It is not required and it will not be done under the present system. Indeed the cost of adding medical information would probably not be considered "reimbursable" or "allowable" by the same intermediaries who currently wonder at increasing hospital costs. Regulatory Legislation Monopoly and Restraint of Trade The only obvious effect upon medical information systems from our antimonopoly policies has been to preclude the development of systems jointly between equipment manufacturers. This is prob- ably not a major disadvantage to the field except in the medical application areas using remote terminals and telecommunication systems. Here it is a definite disadvantage. Our strongly com- petitive system has resulted in much deliberate built-in incom- patibility. Manufacturer A's terminals simply are personna non grata on Manufacturer B's teleprocessing unit. In spite of standards for the hardware interface (e.g., EAI pin specifica- tions) and standards for the data format (e.g., ASCII character definitions), there are many other means to assure incompatabil- ity (transmission protocol being the neatest) . Mixed systems do not happen frequently. If it is true that the developer of a medical information system should not be attempting fundamen- tal computer research, then his only reasonable recourse is to develop systems based upon the equipment of a single hardware manufacturer. This is almost universally the pattern. This is a marriage of great inconvenience. It has the pre- dictable consequence that transportability of any system to another location is dependent upon compatible (i.e., identical) computers being available at the two or more locations involved.
225 In the case of good systems or subsystems, it puts the developer in the role of being an unpaid hardware salesman. Patents This issue is discussed in a separate report. PROBLEMS IN EVALUATING MEDICAL INFORMATION SYSTEMS Some students of technology have distinguished between three stages of work: invention, innovation, and diffusion. In this scheme, invention is more or less equivalent to the early or fundamental phase of basic research. Innovation is taken to be the first successful application of the basic research. This is equivalent to the stages in other schemata that are variously called technological development and demonstration. The dif- fusion stage is taken to be the widespread repetition of the innovations. This corresponds with stages that are called by others clinical trials and implementation, or prototyping and production. One writer notes that "only invention defies cost- ing or economic fitting."18 She feels that it is proper to attempt measurement of economic costs during the stages of in- novation and diffusion. Cost-Effectiveness and Benefit Costs The cost-effectiveness of a given methodology is traditionally measured by comparing it with an alternative methodology. Given that the tasks accomplished are roughly the same, then it is not conceptually difficult to put a relative cost on the two opera- tions. A familiar and happy example is a manual procedure versus an automated procedure. It is frequently the case that the auto- mated procedure can be shown more cost-effective. Many of the components or subsystems of the medical informa- tion system can be compared in this way with some predecessor technology or a manual method. Often this is an entirely fair comparison and an entirely sufficient basis for preferring the cost-effective method. It should be noted of course that human decisions are based on many factors, and that cost-effectiveness analysis is only intended to assist in formalizing the relative economic merits of the question to be decided. An example of a choice that could reasonably be based upon cost-effectiveness analysis is that between automated clinical chemistry analyzers as compared with manual methods. Briefly,
226 the automated system does more tests per hour at a small fraction of the unit cost for manual procedures. Assuming that observational data are available for both modal- ities, then a number of comfortable choices exist for ways to conclude the analysis. A reasonable decision criterion might be "net present benefit," although "return on investment" might be chosen if this seemed more appropriate because of limitation of investment funds. For the case of any particular institutionâas opposed to the general case based on average dataâthere are complications that ' need to be taken into account in the performance specifications. Many subsystems of what we now think of as a medical informa- tion system can be evaluated by the cost-effectiveness method, whether in the general case based on average data, or with re- spect to a specific institution. This method for evaluation is suitable for all those subsystems of the medical information sys- tem that have a manual counterpart. Some of the comparisons must be contingent upon some condition (such as patient acceptance or equal safety, etc.), but under these conditions the evaluations and choices will still be reasonable from an economic point of view. The method of cost-effectiveness is unsuitable for evaluation of the entire medical information system because there is no al- ternative methodology. The raison d'etre for MIS's is to bring together all available parts of the patient record, i.e., to be more than the sum of the parts. Hence, there is not a valid basis for accepting a cost-effectiveness comparison. There will also very likely be difficulty in getting solid observational data or estimations. Cost-benefit analysis is suitable, and cost-effectiveness anal- ysis is unsuitable, if one is attempting to answer the question, should a new technology (such as MIS) be employed? It is reason- able to question whether to expend resources on MIS's versus some other purpose. This is a valid application of cost-benefit anal- ysis. The analysis will, however, require measurement of the costs and benefits of all the schemes being considered, including the nonhealth related ones. One medical information can, of course, be evaluated against another MIS, if they are doing the same job. The evaluation mea- sure, i.e., the basis for the comparison, could be cost, or thor- oughness, reliability, precision, training requirements, space utilizationâone of these or all, weighted or unweighted. Such scales are generally measures of the process. In the case of methodologies that have no counterpart or predecessor method- ologyâand even in many other circumstancesâone is urged lately to prefer measures that compare the outcome of employing each method, or of employing the method in question (e.g., an MIS)
227 versus not employing the method. This strategem is useful partly because it tends to ask two questions at once: namely, which method tends to reach the goal best, and what is the va- lidity of the goal in any case. It is especially desirable to estimate health outcomes for such a comparison. One writer warns, however, that: "The growing literature on evaluating medical care and health status clearly indicates that studies rigorously documenting relationships between changes in health delivery process and changes in health status are diffi- cult and costly to undertake."62 In the case of medical information systems the outcomes or benefits to be measured may be specified in terms that are cen- tered upon the patient, the institution, society, or all of these and others. The evaluator can be left to weigh the relative im- portance of each outcome measure either before or after the anal- ysis has been made. One obvious difficulty is that outcome measures that have been expressed in these various terms will often not have the same unit of measure (and hence are difficult to weigh or combine). An example of such benefits might be: im- proved access to care, improved productivity, and patient satis- faction. Second, some outcomes that are thought to be important may not have any readily apparent unit of measure at all. The most classical example of such a benefit is improved quality of care. The last difficulty is that some outcomes are not measur- able immediately after the methods have been tested, and often not within the same system under consideration. An example of the third difficulty, externalities, arises in consideration of a universally desired benefit: cost containment. If the cost saving of a given procedure is accrued to some soci- etal unit above that of the institution under study (e.g., to a community or a state), then the benefit may never be measured within the institution. The obverse, that is, cost escalation, presents a like problem. The patient who is denied a computer tomography scan at Institution A will not have saved society's health care dollars by traveling across town or across the state to get one at Institution B. Specifically, with respect to the type of MIS's providing for ambulatory patients, Henley and Wiederhold9'26 looked for out- come benefits. In spite of the obvious difficulties in quanti- tation, they concluded in over half of the l7 systems visited for the purpose of evaluation that "quality of care had been improved." They were not convinced that MIS's affected the problem of initial access to care, although "improved secondary access was a major benefit at many sites." They found consistent evidence of medical information systems contributing to improved institutional manage- ment. Patient satisfaction was found to be difficult to measure.
228 Few of the particular systems they chose to visit were imple- mented with research as the primary goal. None did patient ed- ucation as an MIS activity. After this exercise in informal and nonquantitative evaluation, it was clear that there were identifiable benefits, no negative benefits, and readily calculable dollar costs. To make a formal cost-benefit analysis would require the following: that one ex- tend the time frame for studying benefits; extend the domain in which benefits and costs are measured so as to include the ex- ternalities (e.g., other institutions in the communities studied); express benefits and costs in commensurate terms (i.e., dollars); and select a decision criterion. It would then only be left to decide if the costs involved were of greater or lesser value than the benefits in the cases studied, and over the range of costs and benefits that had been identified. At the end of this process, one would have identified the rela- tive economic merits of the question. If there were other rele- vant aspects of the question of evaluation of MIS systems (e.g., social, political, or moral), then these could be considered against the background of the formalization of the economic anal- ysis. Potential Benefits From Medical Information Systems Utilization There is some reason to suspect that the benefit from medical in- formation systems is a summation of a number of small gains, plus the very large potential gain in the additional desirable activ- ities that an MIS will permit. This means that we cannot look for all the benefit in one place. More importantly, however, we should not look for major bene- fits of MIS support of clinical decision making necessarily to be reflected in dollars and cents. The MIS approach at Duke, for instance, collects data from many sources to build an integrated record for a special class of cardiac patient. The desired bene- fit is to base the decision for surgical bypass procedures on the computer matching of the outcomes of like patients treated by medical and by surgical regimens. The system does appear to re- sult in measurable improvement in clinical decision making and in improved patient outcomes. It may actually be possible to asso- ciate a savings in dollars at least with the cases who avoid sur- gery. It may. But the real benefit of such an MIS is the new knowledge relevant to prognosis and therapy that it produces for use by the entire profession. The medical information system in radiology at the University of Missouri has been thoroughly cost-justified merely on fiscal
229 operational grounds, but this misses the point of the real bene- fits.19,33 These exist in the research data base, which has permitted formalization of the logic for medical diagnoses, physician assistance functions, and the substantial but untouted education benefits for residents and students that the rigor of the system imposes. The same kinds of benefits are associated with the HELP component of the MIS at the University of Utah.23 Operation of the medical information system in the multiphasic screening clinic at Kaiser-Permanente has benefited medicine in many places besides those patients seen at the Kaiser Clinic.14 The benefits include evaluation of screening tests such as pain tolerance, anthropometries, and breast thermography; refinement of normal ranges; evaluation of statistical measures; validation of screening concepts; and creation of a valuable health care management model. Even when one agrees that such outcomes or goals are benefi- cial, however, it is not easy to fit the data to a cost-benefit analysis. As a practical matter, it is too easy to fix upon financial measures. Traditional accounting methods and economic variables often come to dominate system evaluations. A potential way around these difficulties may be to pay more attention to the evolutionary stage in which each system is eval- uated, and to insist upon using only a set of measures that is relevant to the stage of development of the system in question. Figure l presents an example of such a scheme for evaluations of health care technology at various stages in an evolutionary pro- cess. The scheme was proposed by Wallace and Fairman.67 The description of diffusion of technology often provides for five stages. Since past analyses have often been more focused on hardware developments than health care methods, the last two stages have often been referred to as "industrial development and marketing." In this scheme, the last two stages have been called "clinical trials" and "implementation." The important point is that the general classes of evaluation criteria that are appropriate will depend upon the stage of the system being examined. The individual measurement to be made (within the class of evaluation measures) would not usually be determined by the individual or local peculiarities of each study site. Effects of Major Technological and Economic Developments Medicine will continue to be the beneficiary or the victim of the large forces of technology and our economy that medicine itself does not control or even influence. If the computer hardware were suddenly to become absolutely free, the costs of developing medical information systems would
.i! I Â£i3 Sy nd ra .2 â¢ Â¥' p â tf> C fill Â§ 1 s = I 5 Â§, r s! 2 Â« â¢5 6 S-E) 385 It 3 t oj O S. â¢, sllii^ Â° = s E - 2 O O w iS 1 uj o Â£ < c illll! silllili II III w ra o o -2 4J iH I B1 s o 5 H id i -* II c J ll.fr I0 c i| Is ill ii2 s 230
23l fall more than 25 percent and less than 50 percent. That is, one would be left with the personnel costs (and relatively mini- mal supply costs) and the communication costs. This ignores facility costs entirely. Computer project budgets made up of slightly less than 50 percent hardware and communication costs combined and about 50 percent personnel costs have ben observed for many years in such projects during the developmental phases. , Once operational, personnel costs are expected to fall to l0-25 percent of the total budget. If hardware performance were to become extremely reliable be- cause of new technological advances, the hardware cost portion of medical information systems could be cut roughly in half be- cause of the possibility of eliminating the redundancy that is now demanded in order to achieve acceptable reliability. If a new technology were to bring the much prophesied tril- lion bit direct-access memories at roughly the cost of present- day memories, then implementation of medical information systems in small or moderate sized clinics might become possible without remote time-sharing to a large central computer. The consequence again would probably be a reduction in cost, by eliminating tele- phone line charges and the associated communication devices. If the computer costs did not change, and based on current cost ex- periences , the reduction in cost could be as large as 50 percent of the computer hardware budget. A special problem of time-shared computer facilities is commu- nications costs. These frequently equal the size of the computer rental bill. Communication costs will be less if time-sharing is done within a city. They approach or exceed equality with the computer costs when time-sharing occurs within a state. Intra- state line charges are controlled by state utility commissions. They are very high as compared with interstate long-line charges, which are controlled by the Federal Communications Commission. A change in this policy would create quite a stir in the computer field in general. It would definitely represent a major influ- ence upon the costs of shared computer facilities in support of medical information systems. The future prospects for computer communications companies is consequently also much entwined with the future relative cost, and hence the benefit-cost, of medical information systems. Under the marketing policy adopted by com- munication companies such as Tymnet, the effect of the pricing policy is essentially to eliminate the importance of the relative distance of various users from the computer service provider.65 The fate of such ventures is really determined by public policy (largely federal)ânot by technology, and certainly not by medi- cal computational considerations. Another major factor in the future value of all forms of auto- mation in medical services, including MIS's, is the cost of labor.
232 One need not be a seer to recognize that wages in the computer and health fields, relative to anything, have been climbing steadily. Perhaps the greatest single factor increasing hospi- tal costs in the last two decades was the Fair Labor Practices Act of l967. This bill did indeed put an end to some unfair practices, including the extreme underpaying of nonprofessional hospital personnel. It also essentially imposed a 40-hour work week, which was a substantial reduction from the minimum 48-hour and above that formerly had been demanded. Whether or not there will be general hospital labor unions in the future, the actual cost of labor in health care facilities will be determined by general market conditions outside of our control. By and large, more expensive labor has made automation in other fields increas- ingly attractive. If this trend continues, it will encourage computer systems, even medical information systems, simply as labor savers and even at the level of mere automation of manual procedures. Effects of Regulation One last large factor confounds projections of benefit-costs of medical information systems. This is federal government policy specifically directed to MIS's. Certainly this is not a simply economic factor, but, in such a field as health, it may well be the only factor that really matters. We can encourage or dis- courage deployment of this technology (and quite effectively) simply by regulation. The requirements for certification for Medicare would be a possible means, although there are many oth- ers. Implementation of such systems could be encouraged and made attractive simply by making specified services reimbursible costs. It would be especially encouraging if these services (e.g., phy- sician assistance functions, quality assurance analyses, screen- ing studies, risk estimates, prognoses, treatment plans, or patient educational services) were made billable as services outside the negotiated per diem. Hospitals and practitioners have traditionally implemented those services for which there was a true marketplace demand. No one can doubt that such a policy would create this demand and strongly influence the dif- fusion of the technology. The medical computer market is not sufficiently large as to be able to influence any of the large technologic determinants. Consequently, to evaluate how economical or how practical are certain applications is to guese contingent upon these major ex- ternal factors.
233 FUTURE MANAGEMENT OF MEDICAL INFORMATION SYSTEMS TECHNOLOGY Medical information systems technology can be managed under a variety of governmental attitudes. The consequences are esti- mated for three separate paradigmatic views of the field. These are essentially the judgmental, the observational, and the man- agerial. The Consequences of Assuming That We Must Examine, Evaluate, Appraise, and Permit The present de facto federal policy for managing the development of medical information systems is to shut off research support for the further development of MIS's or the creation of new ones, and to emphasize evaluation of existing systems. Examination and appraisal is proceeding, often with barely favorable appraisals. Some secondary gains have appeared, mostly in the form of well- written documentations and analyses that are a great benefit to others in this and related fields. The anticipated decision to permit or not to permit will unfortunately be based on considera- tions that may be quite relevant to some aspects of the health planning process but that may be quite unrelated to the priorities and potentials of the medical information systems themselves. Of- ten the analyses must lean heavily upon financial measures. Often there are substantial unmeasured benefits in other areas. If survival of the various forms of MIS's must be dependent upon the best we can do now by way of formal cost-benefit anal- yses, then there is a clear danger. Systems of borderline merit, which concentrate primarily upon business office and institutional management functions, will be "permitted." Some systems that are of great merit because of their clinical features (e.g., physician assistance, education, prospective com- munity data base building, etc.) will be immune to evaluation based on dollar accounting,' and will be harmed or destroyed. If one believes that the correct paradigm is to examine, evalu- ate, and permit, then one must surely also include the need for further research in the methodology of evaluation. Cost-benefit analysis, in health fields especially, merits further research. The Consequences of Assuming That We Should Observe and Predict A laissez-faire strategy of management of these developments (i.e., not much management) will mean that ideas for innovative research based potentially on medical information systems will either hibernate or will seek support as traditional, small, discipline-specific research projects funded over short time
234 periods. As such they will (even if successful) not be capable of supporting the development of true full medical information systems. A research program cannot be funded by a collection of research projects. An alternative for some research ideas is to emerge repackaged as unnecessarily large clinical trials. Medical information systems that are primarily oriented to hospital administration will pay their own ways by producing small savings for individual institutions. There is no known case in which a business office system has even evolved into a medical information system. The concept of the medical informa- tion system under a laissez-faire paradigm will simply not be developed. The Consequences of Viewing Such a System As a National Intent This paradigm implies managing the development and maturation of the concept of the medical information system. It is clear that the MIS concept, like many other innovations, has to find its proper place in its problem space. That is, one must determine by experimental exploration of the problem domain just which areas are feasible and fruitful. Next, at least some sample of the various kinds of medical information systems must be selected to pass as far as possible through the known sequence of the phases of maturation. That is, we must be prepared to see systems that are moved from re- search phase to development to demonstration to clinical trial use to full implementation in the market. Each of the stages will need evaluation, but each will require quite a different set of measures. Each phase should have distinct limits but not a predetermined time for each phase. One must plan for costs to increase as a project moves through such a sequence. Different phases could properly be supported by separate branches of gov- ernment, but it should be government's responsbility to provide for a smooth transition between phases. There must, of course, be strict criteria for evaluation, and unsuccessful projects must be selected out. Transition or selecting out should be accompa- nied by formal reporting from the project so as to document the experiment and its results. The process of managing the development and documenting the growth and changes in the field, and relevant other fields, should rest with a permanent government office or agency. The consequences of this paradigm would be to complete matura- tion of the concept, exploration of appropriate problem areas, and transition to commercial availability of those systems for which such a metamorphosis is appropriate.
235 In short, this paradigm sees medical information systems as a concept of potentially great national value. It sees govern- ment as having the opportunity and responsibility to manage the development of this technology and to assure that society gains the benefits. REFERENCES l. Abrahamsson, Sixten, and Larsson, Rare. "Danderyd Hospital Computer System." Computers and Biomedical Research 3(l970): 30-46; 4(l97l):l26-40. 2. Aikwa, Jerry K. "The Cost-Effectiveness of the C.U. Com- puterized Clinical Laboratory System." Biomedical Science Instrumentation l0 (l974):89-92. 3. American Hospital Association. CT Scanners: A Technical Report. Chicago: American Hospital Association, l977. 4. Ball, Marion J. "Fifteen Hospital Information Systems Available." Pages l0-27 in How to Select a Computerized Hospital Information System. Marion J. Ball, ed. Basel: S. Karger, l973. 5. Barnett, G. Octo. "Massachusetts General Hospital Computer System (Boston)." Pages 5l7-45 in Hospital Computer Systems. Morris F. Collen, ed. New York: John Wiley & Sons, l974. 6. Barnett, G. Octo. Personal communication, March l, l977. 7. Barnett, G. Octo. "The Modular Hospital Information System." Pages 243-66 in Computers in Biomedical Research, Vol. IV. Ralph W. Stacy and Bruce D. Waxman, eds. New York: Academic Press, l974. 8. Barrett, James P., Barnum, R. A., Gordon, B. B., and Pesut, R. N. Final Report on Evaluation of the Implementation of a Medical Information System in a General Community Hospital. Columbus, Ohio: Battelle Memorial Institute, l975. NTIS Document PB-248 340. 9. Bender, Allen E. "Draft Summary Report of an Analysis of Automated Ambulatory Medical Record Systems, University of California, June, l975." Rockville, Md. June, l976. l0. Caceres, C. A. "Electrocariographic Analysis by a Computer System." Archives of Internal Medicine 3(l963):l96-202. ll. Caceres, C. A.. Steinerg, C. A., Abraham, S., Carberry, W. J., McBride, J. M., Tolles, W. E., and Rikli, A. E. "Com- puter Extraction of Electrocardiographic Parameters." Cir- culation 25(l962):l96-202. l2. Collen, Morris F. "General Requirements for a Medical In- formation System (MIS)." Pages l-l6 in Proceedings of a Conference on Medical Information Systems, January 28-30, l970. U.S. Department of Health, Education, and Welfare.
236 l3. Collen, Morris F. "Reasons for Failures and Factors Making for Success in Public Health in Europe." In Health Plan- ning and Organization of Medical Care. Copenhagen: World Health Organization, l972. l4. Collen, M. F., Davis, L. S., and Van Brunt, E. E. "The Computer Medical Record in Health Screening." Methods of Information in Medicine l0 (3) (l97l):l38-42. l5. Craig, L., Golenzer, F., and Laska, E. "Computer Con- structed Narratives." Pages 59-80 in Computers and Elec- tronic Devices in Psychiatry. N. S. Kline and E. Laska, eds. New York: Grune and Stratton, l968. l6. Cronkhite, Leonard W., Jr. "Computer Brings Order to Clinic Scheduling System." Hospitals 43(l969):55-57. l7. Davis, Lou S. "A System Approach to Medical Information." Methods of Information in Medicine l2 (l973):l-6. l8. Davis, Ruth M. "Meeting the Real Costs of the Computer as a Research Tool in the Life Sciences." Pages 235-38 in Computers in Life Science Research. William Silver and Donald A. B. Lindberg, eds. FASEB. Bethesda and New York: Plenum Press, l974. l9. Fairman, William, and Dickhaus, Elizabeth A. "Technology EvaluationâA Case Study of MARS." Medical Care (in press). 20. Fowler, R. D., Jr. "Computer Interpretation of Personality Tests: The Automated Psychologist." Comprehensive Psychi- atry 8(l967):455. 2l. Friedman, Gary D., Collen, M. F., Harris, L. E., Van Brunt, E. E., and Davis, L. S. "Experience in Monitoring Drug Re- actions in Outpatients." Journal of the American Medical Association 2l7(l97l):567-72. 22. Friedman, Richard B., and Gustafson, David H. "Computers in Clinical Medicine: A Critical Review." Computers and Biomedical Research l0(l977):l-6. 23. Giebink, Gerald A., and Hurst, Leonard L. Computer Projects in Health Care. Ann Arbor: Health Administration Press, l975. 24. Glueck, Bernard C., Ericson, R. Peter, and Stroebel, Charles F. "The Use of a Psychiatric Patient Record System." Pre- sented at FASEB Conference on the Computer as a Research Tool in the Life Sciences, Aspen, Colorado, June 27, l974. 25. Glueck, Bernard C. In Progress in Mental Health Information Systems: Computer Applications. Jeffrey L. Crawford, ed. Cambridge, Mass.: Ballinger Publishing Company, l974. 26. Henley, Ronald R., and Wiederhold, Gio. An Analysis of Auto- mated Ambulatory Medical Record Systems, Volumes l and 2. Technical Report No. l3. , San Francisco: University of Cali- fornia, San Francisco Medical Center, l975.
237 27. Hodge, M. H. "Large Scale Medical Data SystemsâThe In- tegrated Approach." Pages ll-lâll-2l in Journees D'In- formatique Medicale (Symposium on Medical Data Processing). Toulouse, March 4-7, l975. 28. Hubbard, John P. Measuring Medical Education. Philadelphia: Lea and Febiger, l97l. 29. Hulse, Russell K., Clark, S. J-, Jackson, J. C., Warner, H. R., and Gardner, R. M. "Computerized Medication Monitor- ing System." American Journal of Hospital Pharmacy 33(1976): l06l-64. 30. Jensen, R. E., Shubin, H., Meagher, P. F., and Weil, M. H. "On-line Computer Monitoring of the Seriously Ill Patient." Medical Biological Engineering 4(l966):265-72. 3l. Jessiman, A. G., and Erat, K. "Automated Appointment Sys- tem to Facilitate Medical Care Management." Medical Care 8(l970):234-46. 32. Laventurier, Marc F., Talley, R. B., Hefner, D. L. , and Kennard, L. H. "Drug Utilization and Potential Drug-Drug Interactions." Journal of the American Pharmaceutical As- sociation (New Series) l6 (l976):77-8l. 33. Lehr, J. L., Lodwick, G. S., Nicholson, B. F., and Birzneiks, F. B. "Experience with MARS (Missouri Automated Radiology System)." Radiology l06 (l973):289-94. 34. Lindberg, Donald A. B. The Computer and Medical Care. Springfield, Ill.: Charles C Thomas, l968. 35. Lindberg, Donald A. B. "Electronic Processing and Trans- mission of Clinical Laboratory Data." Missouri Medicine 62(l965):296-302. 36. Lindberg, Donald A. B. "Collection, Evaluation and Trans- mission of Hospital Laboratory Data." Pages 375-97 in Proceedings of 7th IBM Medical Symposium. White Plains, N.Y.: IBM, l965. 37. Lindberg, Donald A. B., Schroeder, J. J., Jr., Rowland, L. R., and Saathoff, J. "Experience with a Computer Lab- oratory Data System." In Multiple Laboratory Screening. New York: Academic Press, l969. 38. Lindberg, Donald A. B., Van Peenen, Hubert J., and Couch, Rex D. "Patterns in Clinical Chemistry: Low Serum Sodium and Chloride in Hospitalized Patients." American Journal of Clinical Pathology 44 (l965):3l5-2l. 39. Arthur D. Little, Inc. Introduction to Use of Cost-Benefit Analysis in Planning for Emerging Health Care Technologies. Cambridge, Mass.: Arthur D. Little, Inc. February 1977. 40. Lodwick, Gwilym S. "Information Management in Radiology." Pages 206-40 in Hospital Computer Systems. Morris F. Collen, ed. New York: John Wiley & Sons, l974.
238 4l. Lodwick, Gwilym S. "Radiographic Diagnosis and Grading of Bone Tumors, with Comments on Computer Grading." Pages 369-80 in Proceedings of Fifth National Cancer Conference. September l964. Philadelphia: J. B. Lippincott Co. 42. Lodwick, G. S., Keats, T. E., and Dorst, J. P. "The Coding of Roentgen Images for Computer Analysis as Applied to Lung Cancer." Radiology 8l(l963):l85-200. 43. Lusted, Lee B. "Computers in MedicineâA Personal Per- spective." Journal of Chronic Diseases l9 (l966):365-72. 44. Massachusetts General Hospital. "COSTAR, Computer Stored Ambulatory Record, A Progress Report." Boston, Mass.: Laboratory of Computer Science, Massachusetts General Hos- pital. 45. McDonald, Clement J. "Protocol-Based Computer Reminders, The Quality of Care and the Non-Perfectability of Man." New England Journal of Medicine 295(l976):l35l-55. 46. McKeown, Michael J., and Domizi, Dario. "Computer Enhance- ment of Obstetrical Intensive Care." In Proceedings of Biomedical Symposia. San Diego, California, April l970. University of Chicago. 47. Medicus Systems Corporation. Spectra 2000 Medical Infor- mation System. Chicago: Medicus Systems Corporation, l977. 48. Medinet. System Description Materials by Medinet Company, a Department of General Electric Company. Watertown, Mass.: Medinet, l968. 49. Mowshowitz, Abbe. The Conquest of Will: Information Pro- cessing in Human Affairs. Reading, Mass.: Addison-Wesley Publishing Co., l976. 50. National Library of Medicine. "Health Sciences and Com- puter Technology. Report of Training Directors." May ll- l2, l976. 5l. New Haven Health Care, Inc., and Yale University. Proceed- ings of the First National Symposium on Emergency Medical Technician Evaluation and Emergency Para-Professional Util- ization. New Haven, Conn., l975. 52. Osborn, J. J., et al. "Measurement and Monitoring of Acutely 1ll Patients by Digital Computer." Surgery 64 (l968):l057-70. 53. Pipberger, H. V. "Computer Analysis of the Electrocardio- gram." Pages 377-404 in Computers in Biomedical Research. Ralph W. Stacy and Bruce D. Waxman, eds. New York: Aca- demic Press, l965. 54. Pipberger, H. V., Stallmann, F. W., Yano, K., and Draper, H. W. "Digital Computer Analysis of the Normal and Abnormal Electrocardiogram." Progress in Cardiovascular Diseases 5(l963):378-92.
239 55. Rikli, Arthur E., Allen, Scott I., and Alexander, Samuel N. "Study Suggests Value of Shared Computers." The Modern Hospital (May l966):l00-l08. 56. Rosati, Robert A., McNeer, J. F., Starmer, C. F., Mittler, B. S., Morris, J. J., and Wallace, A. G. "A New Informa- tion System for Medical Practice." Archives of Internal Medicine l35 (l975):l0l7-24. 57. Shared Medical Systems. A Data Processing Service for Hospitals. King of Prussia, Pa.: Shared Medical Systems. 58. Sheppard, L. C., Kouchoukos, N. T., Kurtis, M. A., and Kirklin, J. W. "Automated Treatment of Critically 1ll Pa- tients Following Operation." Annals of Surgery l68(l968): 596. 59. Sletten, I. W., Altman, H., and Ulett, G. A. "Routine Diag- nosis by Computer." American Journal of Psychiatry l27 (l97l):ll47. 60. Smith, R. E., and Hyde, C. M. "Computer Analysis of the Electrocardiogram in Clinical Practice." In Electrical Activity of the Heart. Edited by G. W. Manning and S. P. Ahuja, eds. Springfield, Ill.: Charles C Thomas, l969. 6l. Stroebel, Charles F., Bennett, W., Ericson, P., and Glueck, B. C., Jr. "Designing Psychiatric Computer Information Systems: Problems and Strategy." Comprehensive Psychi- atry 8(l967):49l-508. 62. Systemedics, Inc. "Automated Hospital Information Systems, Case Study Report." Princeton, New Jersey, April 29, l976. 63. Toffler, Alvin. Future Shock. New York: Random House, l970. 64. "Texas Firm Moving into Position to Control U.S. Health Care." St. Louis Post Dispatch, January 30, l977, p. l. 65. Tymnet, Inc. "Network Services." Cupertino, Calif., l977. 66. Van Brunt, Edmund E., David, Lou S., and Collen, Morris F. "Kaiser-Permanente Hospital Computer System (San Francisco)." Pages 70l-53 in Hospital Computer Systems. Morris F. Collen, ed. New York: John Wiley & Sons, l974. 67. Wallace, Richard L., and Fairman, William. Personal com- munication, August, l976. 68. Warner, H. R., Gardner, R. M., and Toronto, A. F. "Computer- Based Monitoring of Cardiovascular Functions in Postoperative Patients." Circulation 37 (l968):68-74.