Much has been written of the effect of the Baby Boomer generation on our societal and economic systems. However, their effect on the American healthcare system has yet to be felt. This historically large population is now 50, and will soon be 65 years old. If there is concern in contemporary medicine about appropriate treatments and outcomes today, just wait until tomorrow. How can we cope in a constrained financial market, when we face the largest number of elderly ever? The impact on the provision of care, in a contracting medical marketplace and a constrained economic environment, will be profound.

The infrastructure to provide care at this scale does not exist [Ginzberg,1998; Smith,1998]. In 2011, the first of 77 million baby boomers reach retirement age; the age when health care costs start to escalate [Schneider,1999]. In 2000, Medicare will have an estimated 40 million enrollees; in 2040, Medicare will have an estimated 81 million. It is clear that the economy will not be able to support all treatments for all the people all the time. We will have to squarely face the issue of appropriate outcomes for chronic diseases with chronic treatments.

The common thread connecting the problems of healthcare in the 21st century is a model of outcomes: health status functionality, quality of life, and, ultimately, of health outcome measures. Measuring outcomes gives the medical definition of "true health" for an individual or a population [Donabedian,1980]. Yet there is little agreement as to how health and disease outcomes are defined [Eddy,1996]. It was easier in medicine when treatment plans were focused on the surgical management of acute illness. Then, measurements of survival from the surgery and the incidence of post-operative complications were sufficient to determine quality of care and treatment delivered, and therefore the health outcome, for the patient. Mortality and morbidity are good and useful measures when an operation can cure the disease, but leave room for improvement when chronic conditions and illness are discussed.

But what to do as the population is older and larger, with an increasing incidence of chronic disease in aging Baby Boomers? The clear determinants for acute illness are less applicable for chronic illness characterized by a plethora of physical, psychological, social, economic, and personal factors [Wagner,1998; Applegate,1990]. How does one determine the effectiveness of treatment for minor depression in a nursing home patient who is visited by a physician, at best, once a month? How does one measure the outcome of arthritis medication if the patient has personal and economic constraints that prohibit visiting a health care facility more frequently than once a year? And how does one measure the incidence of chronic disease in the urban areas of poverty or dispersed pockets of the 40 million Americans without health insurance?

A new model of health status, functionality, quality of life and health outcomes, particularly for chronic conditions, is needed. The model proposed here provides a feasible method whereby patients as individuals can be continuously tracked and their health measured over time. Subtle changes in health status will be detected as they cannot be today. Treatments can then be continuously course-corrected, through interactions with physicians and other healthcare providers, to provide individualized healthcare. In addition, the collection of health tracking records will enable the detailed tracking of entire populations. Finally, this methodology will approach measurements of population health on a national scale for the first time, and enable treatment guidelines on a customized basis for each individual, similar cohorts, and populations based upon similar cases within that population.

Outcomes management would draw on four already rapidly maturing techniques. First, it would place greater reliance on standards and guidelines that physicians can use in selecting appropriate interventions. Second, it would routinely and systematically measure the functioning and well-being of patients, along with disease-specific clinical outcomes, at appropriate time intervals. Third, it would pool clinical and outcome data on a massive scale. Fourth, it would analyze and disseminate results from the segment of the database most appropriate to the concerns of each decision maker. This should also allow the entire outcomes management system to be modified continuously and improved with advances in medical science, changes in people’s expectations, and alteration in the availability of resources." -- Paul Ellwood, Shattuck Lecture, 1988 [Ellwood,1989,p1551]

Outcomes have traditionally been provided by physicians or other medical professionals to describe a patient’s, or population’s, course of disease or response to therapy [Donabedian,1980; Coker,1998]. This situation relies on physicians who perform a brief examination to determine the patient status. Because most patients visit a healthcare provider rarely or randomly, a medical outcome description, if encompassing health status functionality and quality of life, relies on an interpretation by the provider, based on limited evidence.

Attempts to ameliorate this situation often take the form of questionnaires that the patient completes during diagnosis or treatment. General-health questionnaires, such as SF-36 [Ware,1992; Hays,1993], provide accurate answers to the questions posed. These surveys attempt to measure outcomes in an organized and standard fashion. Disease-specific questionnaires, such as AIMS2 [Mennan,1982,1992] for arthritis, provide similar concrete assessment at a more detailed level. However, problems with any of these questionnaire instruments remain. Are the questions appropriate for the patient in their current situation? Are they asked often enough? Are they sampling at an atypical period in the condition? [Reuben,1995]

Health outcome data is currently not used in routine management of the patient [Meadows,1998]. Structural constraints make the underlying reasons clear. The number of questions in a standardized questionnaire is geared towards the attention span of a patient and the practical time constraints of the patient-healthcare facility interaction. This often limits the length of the questionnaire to 3 pages, yielding 30-40 questions. There is simply not enough information for an accurate diagnosis. The status of elderly patients with chronic illness are particularly poorly captured, since they commonly have multiple conditions whose interaction is missed by a general-health questionnaire or a single disease-specific one. In addition, questionnaires have become so numerous and complex that they might best be administered by experts who know their range of effectiveness [Applegate,1990].

Most patient perceptions of their personal health can be described by 10 broad categories [Evans,1994; Kindig,1998]. These categories encompass the major physical health factors, including: disease, health care, health function, genetic endowment, physical environment, social environment, individual response, behavior, biology, well-being, prosperity. Similarly, there are only a few important broad categories that encompass the major mental health factors. A standard list of 10 factors [Hersch,1996] includes: level of consciousness, emotional state, orientation, attention, memory, language, calculations, praxis, visual-spatial function, reasoning/abstractions. Individuals relate to the health of their bodies through the 10 physical factors and the 10 mental factors.

The standard questionnaires today are largely concentrated on disease diagnosis and physical manifestations. For example, arthritis questionnaires ask about ability to move joints and limitations placed on daily activity. But they omit genetic background of the patient, which may be more important to assessing health status. As an example, if both of the patient’s parents were dehabilitated by arthritis at age 65, then the patient likely will be, no matter what the patient’s joint status at age 50.

How many questions would be necessary to completely and accurately capture the health status of a patient? There are roughly 1000 common diseases described in a diagnosis manual for general practitioners [Woolliscroft,1998]. Each of the 10 physical and 10 mental factors may require 5 to 10 subcategories to elicit adequate information about the status. For example, the visual-spatial mental health factor has subcategories [Hersch,1996,Table 21-16-1] of: "orientation impaired, can’t drive, can’t use tools, route-finding problems, dressing difficulty, can’t copy figures, agnosia, apraxia".

Coverage for the health factors would thus require 50-100 physical and 50-100 mental subcategories. Accounting for the health factors for every disease in a large dataset would require 1000 data items (one for each disease) times each of the 100 physical and mental subcategories. A patient’s dataset would therefore include 100,000 data points at any one time. Collecting data daily on a patient’s health status would yield some 36,500,000 data points in a year. This dataset would be approximate answers to a SF-36M (where M is a Million).

This large calculation does not include data on the 1000 genetic illnesses, future individual data input from the human genome project, or medical input from physicians concerning evaluations, diagnoses, treatments, medications and the like. Additionally, data would be collected over time yielding a cumulative dataset of like quantity per year. Although some subcategories might require fewer datapoints and questions, additional questions would be needed for interactions across categories. Thus, a full health status dataset would lead to an interaction health status/quality of life questionnaire covering all category factors administered on a regular, daily basis. This dataset would include millions of health status datapoints, and millions of physician and healthcare data inputs.

The need to ask a million questions will cause a radical paradigm shift in how health assessment is done. To eliminate the variation of the current patient-physician interaction in obtaining health status/quality of life data, all data must be standardized and accurately entered by patients and physicians alike. To enable a patient at home to provide correct answers about health status within their attention span, questions must be asked on a situational basis, personalized to the present condition and the past answers of the particular patient.

The logistics of economically providing situational questions and data input require networked solutions via home computers. Since the patients are describing their health on a daily basis at their convenience, the descriptions will be more detailed and comprehensive than the current infrequent and rushed interactions with healthcare providers.

The proposed health status and quality of life outcomes model is based upon the continual monitoring of the physical and mental health factors of the individual patient. These health monitors are similar in function to a heart monitor, which enables an individual patient to track their own cardiological functions and seek treatment when the functions fall outside safe parameters [Carey,1995]. In the more general health case, the patient must interact with a tracking system to record their various health parameters on a continual basis. The dynamic descriptions of health parameters can then be used to provide individualized treatments. During the course of a chronic illness, for example, communications with health care providers could result in treatments that could be rapidly varied to match the episodic nature of chronic disease.

The Internet provides the technology, for the first time, to record continuously an individual’s health, via direct interaction with each patient. The elderly in general and the baby boomers in particular are a population that is comfortable with computers and with accessing information over the Internet [Jeffrey,1998]. An interactive program would be used by each individual to generate a daily record of their health parameters. This would make possible the charting through the good days and the bad days of life, through the days when medicines are taken or not, through the days when a physician is visited or not.

Rand's SF-36 is a tool for measurement of general health status that is applicable across a wide variety of health care conditions and types of encounter. Meanwhile, other researchers have developed more disease-specific assessment instruments targeted at symptoms, outcomes, and experiences associated with particular diagnoses. Important advances in disease-specific measurement have occurred, with applicability to such [chronic] conditions as arthritis, coronary heart disease, depression, and neurological impairment. Altogether, today's health services research has transformed the simple and elegant measurements of survival proposed by Codman in the early twentieth century into a robust and elegant ‘tool kit’ for the measurement of quality in its many and subtle dimensions." -- Donald Berwick in New Rules [Brennan and Berwick, 1996, pp115-116].

Chronic illness is now the dominant feature of health care [Kane,1998]. Some 22% of primary care patients report a major complaint of chronic pain, and this impact will grow with the aging of the population [Gureje,1998]. Arthritis is a chronic illness typical of the aging baby boomer population. Some 12% of the US population, some 30 million people, have arthritis [Harvey,1988]. Over the age of 65, essentially everyone has arthritis.

Like other chronic illness, arthritis has acerbations and remissions – the intensity and location of the pain changes over time [Schnitzer,1993]. It is known that the disease and symptoms wax and wane within the day [Bellamy,1991], that disabilities one year may disappear the next, that one joint involvement may become several, or alter in significance over time [Bellamy,1990].

There are excellent arthritis questionnaires. However, their static administration lacks the regular interaction needed to provide a true outcome of treatment response and measure of ability to function with the current state of the disease [Bellamy,1995]. A continual tracking of a patient's functional status would monitor and measure ability to use joints and remain active. Continual tracking would significantly enhance the utility of existing arthritis questionnaires [Bellamy,1997].

An information infrastructure is needed to focus clinicians’ attention on changes in patients’ status [Kane,1998]. Rather than returning to a physician's office only when a medical regime fails, a patient could instead interact electronically with a personal database that would track progress, including response to therapy and problems with disease. Frequent and reassuring interaction would achieve a truer picture of the episodic nature of chronic disease and lead to a more sophisticated treatment pattern personalized to the particular individual.

By daily interaction with variants of traditional questionnaires, monitoring can be done of an individual patient’s varying ability over time to cope with their disease. For example, an adaptation for daily interaction of [Clark,1998] and [Brandao,1998] might include: "Do you have stiffness in your knees? Is it greater or lesser than yesterday?" "Could you do your errands in the neighborhood today?" "Do you have stiffness in your hands? Is it greater or lesser than yesterday?" "Could you do your housework without help today?"

From the point of view of the healthcare delivery system, the interactive data model will produce an evolving picture of a patient at any given time, taking into account diagnoses and treatments, successes and setbacks. The system would assess the current level of functionality and interactively coach the patient to higher levels of functionality. The consistency of administration via computer would eliminate much of the inaccuracy from the current random interactions between patients and physicians.

Periodically, this data would be reviewed (determined by medical parameters and health plan factors) by trained professionals adept at such data evaluation. Such a record of a patient's and disease's condition would identify changes that might be of importance to a patient and early intervention by a health care professional could be recommended.

Collecting continual interaction with detailed patient status over the Internet will eventually develop a comprehensive healthcare database. Such a database would allow a similarity match between a patient’s clinical situation and a cohort of similarly described patients. This cohort would conceivably be quite large for some diagnoses and permit the formation of unique subgroups for analysis. A patient’s situation would be viewed in light of the cohort of similar cases, rather than the current environment where clinical trials are performed on a small number of patients and the results of such trials extrapolated to individuals [Liang,1997].

The national database would locate the appropriate cohort from which to draw conclusions for an individual patient. This cohort is the patient sub-population relevant to the particular patient’s demographics and conditions. This identification would give a realistic expectation of disease progression and treatment outcome, as measured over the population at large.

Virtual Town Doctors

Potential advantages for health communication efforts include: Improved opportunity to find information ‘tailored’ to the specific needs or characteristics of individuals or groups of users, …, Increased access to information and support on demand, because these resources can be used at any time and from numerous locations. … Increased opportunity for users to interact with health professionals or to find support form others similarly situated through the use of networking technologies, which enable direct communication between individuals despite distance or structural barriers." -- IHC report summarized in [Robinson,1998,pp1264-65]

The new technology for the Internet that has arisen in the past ten years has standardized many necessary components for constructing a national patient database from individual patient interactions [Weinberger,1997]. Existing patient record systems are largely physical rather than digital. Even for the few hospital and clinics with electronic patient records [McDonald, 1992], the political and sociological difficulties of federating across existing information systems has proven an impossible obstacle to widespread patient databases.

The document protocols in the World-Wide Web, however, have standardized the data formats traditionally handled by diverse management information systems [Schatz,1994]. PC-based Web-forms to database servers are rapidly substituting for traditional MIS database entry to mainframes in many applications, including patient record systems [McDonald, 1998].

The rise of the Internet has made it economically feasible to support national-scale health monitors from home personal computers. Personal computers are now widely enough deployed throughout the general public that millions of Americans routinely browse information on the Web. According to national surveys, healthcare is the single most referenced topic, with some two-thirds of users having accessed such information [Hafner,1998]. A Harris Poll in February 1999 estimated that 60 million Americans accessed health information on the Web in 1998, with 10% of the respondents searching for arthritis.

A virtual doctor in the convenience of your home would thus be widely popular. This "doctor" would have built-in medical knowledge of common diseases and electronic access to national databases of similar patients so that patients could access information if they desired. Patients are eager to learn about potential treatments and share experiences with similar patients. For example, experience with CHESS [Gustafson,1999] showed that patients with severe diseases, such as AIDS, are quite willing to spend an hour a day on the computer, even if only to read targeted brochures and chat with similar patients. Elderly women with breast cancer are willing to carry out daily interactions over the Net [Gustafson,1998].

Research technology is available now that can support a functional Virtual Town Doctor (VTD). Such a "doctor" will have both local knowledge of your particular situation and global knowledge of similar situations nationally. Every person will interact daily, or periodically, to a virtual doctor to input information to their personal health status dataset from the convenience of their own home. This doctor would be like a personal trainer who would gather personal information, assess your current level of performance, and, as coordinated by a physician, coach you to higher levels of performance. Specialized questionnaires would drive the coaching for each health factor, with the doctor inferring the status levels from the patient’s answers.

A Town Doctor is an effective metaphor, indicating that the Net is an enabling technology for the average person in the modern world, with its global scale and fast pace. The longing for the "good old days" of small towns and friendly neighbors can be addressed by an always-available always-knowledgeable virtual doctor. The town doctor was more effective than an HMO physician because he knew what was happening with you and with your world.

A VTD will record the interactions, to create an accurate and current database, and coach the patient to better health, based on general medical knowledge and specific database records. The home setting will be particularly conducive to eliciting daily details of condition progress and the life-style environment around the patient. The local advisor on the computer screen will be backed up by expert analysts of the national patient database.

Initial users for VTD will likely be elderly patients, who already monitor their health continually. Having a doctor always available in the natural setting of their homes will help reassure them, which may in itself aid in their feeling better. The most similar current situation is a triage nurse, who consults with patients over the telephone, using a heuristic computer program covering the most common complaints [McKeon,1998].

A Virtual Town Doctor is better, in that the source has a much wider set of medical knowledge available and is much more readily available (plenty of time to listen and plenty of information to dispense). The information is far more detailed than a CD-ROM medical encyclopedia and the information is far more personalized, due to the continual record. In time, VTD will create a comprehensive national database as more patients come online and record their complaints and their feelings, covering situational effectiveness of particular treatments.

Initial technologies for VTD will likely be interactive patient questionnaires, backed by searchable medical manuals. The user interface will eventually develop into computer-generated virtual doctors listening and talking in natural speech, using built-in references for standard medical situations and recorded patient databases for custom medical situations. The session transcripts will implicitly generate the individual values for the patient questionnaires. A national grid of health monitors will create a patient database similar in statistical detail and coverage to the house database created by electricity meters for the national power grid.

It will be technologically feasible in less than five years to generate realistic-looking people on computer screens, whose talking heads can interact with patients in a natural fashion (in this limited environment and situation). Patients can choose whoever is most reassuring to them, be it their favorite grandfather or their favorite movie star, be it Andrew Weil or Marcus Welby. The virtual doctor will react sympathetically to their facial emotions and provide personalized interactions, based on the connotation as well as the denotation.

Living in the Interspace

-- Bruce Schatz in [Schatz,1997,p333]

Virtual Town Doctors will shortly become standard information infrastructure. Feasibility of sophisticated user interfaces for VTDs is only part of the reason. More significantly, information infrastructure is evolving to support concept search across heterogeneous sources. The lack of concept search is the primary technological obstacle that has prevented federation of electronic patient record systems in the past [Lincoln,1999]. Concept search can retrieve different records discussing the same topic (concepts) but using different terminology (words).

In less than ten years, the Internet will have transformed into the Interspace, where users navigate connections across concept spaces of information rather than transmit data across packet networks of computers [Schatz,1997]. This will enable the federation of all the different information sources generated by different healthcare stakeholders into a single uniform source for search and navigation.

For example, a single search for "limited knee mobility due to osteoarthritis in elderly women" should retrieve relevant items from VTD interactions across the country entered by different patients with different terminologies and analyzed by different doctors for different treatments. Vocabulary navigation across the Interspace will eliminate patient variability for answers to questions, much as standardized questionnaires across the Internet will eliminate physician variability in asking the questions initially.

The CANIS (Community Architectures for Network Information Systems) Laboratory at the University of Illinois at Urbana-Champaign is developing large-scale models of the Interspace and deploying prototypes with medical information to practicing clinicians. Last year, we generated concept spaces for all of MEDLINE, in the largest computation ever in information science [Chung,1999; Alper,1998b]. The utility of these concept spaces is now being evaluated by physicians in the HealthAlliance regional HMO, using a web interface from a home computer to the CANIS research prototype of the Interspace.

An Interspace session with concept spaces on MEDLINE illustrates the power of concept spaces in interactively translating vocabulary terminology across different stakeholders. For example, an arthritis patient would interact with her VTD, using coaching expertise from arthritis questionnaires and contextual expertise from previous interactions. The patient would state that "my knee mobility is fine but I have stabbing pains in my knee" and "my medicine causes stomach irritation". The VTD would recommend switching to Tylenol instead of the current prescription drug and notify the healthcare provider. The explanation is the situation has changed from reducing the inflammation to reducing the pain.

Such an interactive session is feasible due to vocabulary navigation in the Interspace. A physician had previously analyzed concept spaces for MEDLINE to develop treatment guidelines for particular situations. GI (gastrointestinal) bleeding with NSAIDs (non-steroidal anti-inflammatory drugs) was a common complaint. An article discussing "maintenance therapy for rheumatoid arthritis" made clear that simple analgesics such as acetaminophen were the preferred treatment when pain management was the primary goal. This article did not explicitly mention "GI Bleeding" or "NSAIDs". It was located using concept spaces to navigate within the arthritis subspace from "GI Bleeding" to the related phrase "maintenance therapy".

Similar vocabulary navigation produces the inferences in the interactive session for the patient. The concept spaces for previous VTD interactions on arthritis interactively transform "stomach irritation" into "gastrointestinal bleeding". The previous patient records indicate "my medicine" is an NSAID. So the established treatment guideline is used to suggest "acetaminophen", which is transformed by the Interspace into "Tylenol", as a commonly occurring related phrase.

Cross-correlation across information sources is the next wave of infrastructure in the Net. This information infrastructure will be supported by navigating across concept spaces in the Interspace. The indexing techniques to support text mining of patterns for large document collections are already operational in research laboratories [Alper,1998a]. They are computationally feasible, since they only record phrases that occur together frequently, which can be utilized to navigate between related phrases. These transformations are simple terminology variations, using the document context, rather than deep semantic inferences.

The next generation of personal computers will be powerful enough to compute concept spaces for all the knowledge of a community-scale collection. Each clinic in each community can collect their local patient interactions and perform their own analysis to identify local patterns. These local patterns can be assembled into global patterns, using concept switching technologies across community collections, which will generate a national patient database.

CANIS is developing models of virtual town doctors. Deployment of a medical Interspace (MEDSPACE) is in stages: from 10 users in the laboratory to 1000 in the region. A thousand users is the scale achieved in the Illinois Digital Library project for experimental systems with new technology on engineering collections [Schatz,1999]. The MEDSPACE project [www.canis.uiuc.edu] is building an experimental testbed for clinical medicine, by indexing and integrating concept spaces on medical literature and patient records.

During the twenty-first century, the packet switching gateways of the Internet will transform into the concept switching gateways of the Interspace. Patients can act locally by continually monitoring their own health and reporting their conditions in their homes via Virtual Town Doctors. Physicians can then think globally by analyzing the resultant national patient database, containing detailed local information, and perform diagnosis by locating similar patients. The enabling technology to Act Locally Think Globally will support the friendliness of small towns with the reach of national healthcare plans. A revolution in the quality and economy of healthcare infrastructure will arise from this revolution in information infrastructure.

References

Bellamy, N., et. al. (1995) Health status instruments/utilities, J. Rheumatology 22(6):1203-7 (Jun).