Status of the Guardian Angel Project

April 28, 1995

Peter Szolovits

After many months of delay, ARPA has finally funded (in part) our proposal for Guardian Angel. Our project is one of about a dozen that are being funded by the HIIP (Health Information Infrastructure Program), and ARPA is very interested in having us work together as a single team rather than as a group of independent projects.[1] HIIP is supporting three (or maybe four) kinds of projects:

  1. site projects at Bethesda Naval Hospital and at Walter Reed Medical Center,
  2. component projects (applications and tools) such as Guardian Angel (GA) that must, by the groundrules, be demonstrated at and integrated into the test sites, and
  3. generalization (meta-level) projects that focus on the process of knowledge acquisition, construction of common domain models, and project management itself.

A fourth project category, support projects, tries to apply the technology being developed for HIIP to the problem of project management.[2] Based on previous discussions, I have decided that our initial "military" application of GA technology is to be for women with high-risk pregnancies, at the OB clinic at Bethesda. My overall impression of the HIIP project is that it will be "research by demonstration": everyone will have to show interesting stuff that works in situ or face the prospect of gutted funding beyond the first year.

The focus of HIIP in the short run will be to bring the OB/GYN clinic at Bethesda and the Urology clinic at Walter Reed from the stone age into the modern era of computer technology. In addition, Bethesda is opening a new Breast Disease clinic in July, and that will be supported from the start by new technologies. In subsequent years, the implementation is supposed to spread to other clinics and then other institutions. John Silva, the program manager, wants all this to happen very rapidly. The DEC team, led by John McDermott, is responsible for site management and integration at Bethesda, and SAIC for Walter Reed. The component projects include two that are application-oriented: GA and Barbara Hayes-Roth's Patient Advocate (PA).[3] A couple of projects focus on what I would call "knowledge engineering"--called SEP (scenario-based engineering protocol) in this jargon. Many of the other pieces are to build technical components that should make it easier to build the systems at the test sites: tools to build "smart forms", database interfaces for legacy and heterogeneous systems, text processing systems, and protocol managers. I am attaching at the end of this memo brief descriptions of the various projects, for your information.

DEC's short-term plans are to install some modern hardware and software into the OB clinic: mostly Pentium-based PC's and basic productivity tools like Microsoft Office and network terminal programs to allow people access on the same machines to CHCS (the military's hospital computer system). CHCS currently provides access (thought to be clunky by many) to labs, radiology reports, ancillary services and schedules. Ultrasound, fetal heart monitoring, and MSAFP (alphafetoprotein) screens are not currently available in CHCS. In addition, most within-clinic information is missing from the computer. Basic clinical records are kept on a so-called "Form 533", which is a pen and ink chart. There are also flowsheets, progress notes, problem lists, and management plans that are not computerized. DEC's ambition is to do at least a first cut at all this "by summer" (1995!), and there appears to be great enthusiasm in the clinic to buy in, even at the cost of having to key in much more information than at present. Much of these data will not be coded, though they have worked out coding schemes for OB history and would be interested, in the longer term, in using UMLS or any ideas that develop from the EMRS project.

The DEC team has created a generic model of a clinic, including workflows, and has some tools that generate database schema from these. The SmartForms project is meant to run off these schema and to automate the generation of interfaces. Current technology uses Visual Basic and an Access database, though they are moving to SQL-server to allow broader database access.

In the long term, DEC would like to support a much more proactive approach to the clinic. They see this as centered around identifying or creating and customizing protocols for various common conditions: e.g., how the clinic treats gestational diabetes or ecclampsia. They want to focus on how the clinic will track problems, and to take advantage of a movement already under way to create a formal structure of "care managers" in the clinics, who are to be responsible for the ongoing longitudinal integration of care for individuals. The new breast clinic will open with two such people (trained as nurses) on staff, and they expect similar positions to develop in other clinics as well.[4] They see the principal new role of computer systems as supporting such case managers.

Obstetrics Clinic

In mid-April I was able to meet with Drs. Gallagher and Miller (incoming, I think, and outgoing chiefs of OB at Bethesda), and with Diane Bloom, a wonderfully dedicated nurse (with various graduate degrees) who deals with all the high-risk patients. They are very much interested in applying our GA ideas to high-risk pregnancy, and Dr. Gallagher especially was taken with the "sci fi" scenario in our proposal. I began by trying to understand what Dr. Gallagher considered to be the most important problem that makes a pregnancy high-risk. Based perhaps partly on the influence of our scenario, he identified gestational diabetes as the right place to start. They get 2-3 such new patients per week, so it should be easy to recruit experimental subjects. At the Bethesda clinic, a routine health screen is conducted at 28 weeks of pregnancy[5] looking for possible diabetic problems in patients not earlier noted to be at high risk. People thought to be at higher risk are screened earlier during pregnancy. Questions to the expectant mother cover family history, whether she has had previous episodes of gestational diabetes, whether she has gained more than 20 lbs during the pregnancy, and whether she has had unexplained fetal losses in the past. She is then administered a 1 hr glucose tolerance screening test, on which a result <140 is considered OK. If the result is over 140, then a 3 hr test is ordered. This involves taking a fasting glucose, then administering 100g of glucose, and taking measurements after 1, 2 and 3 hours. Exceeding different thresholds at these different times indicates possible problems, and over-threshold measurements on two of the four samples is considered an abnormal outcome.

Patients with abnormal outcomes on the 3 hr test are contacted by Nurse Bloom and referred to a dietitian (Mrs. Stouffer?) who explains the basic dietary approach to controlling gestational diabetes. After the class and five days following dietary advice, a further glucose test is administered,[6] and the nurse gives patients a lecture on the basics of gestational diabetes, including the risks to mother and child caused by this condition and the approach to control that they follow. In addition, she gives the patient a selection of educational literature, whose sophistication depends on the interests and educational background of the patient.[7]

Two of the complications of GD are:

  1. polyhydramnios, an abnormal increase in the quantity of amniotic fluid. As in adult diabetics, the fetus is stimulated to pee a lot. This increases the amniotic fluid, and can lead to premature labor.
  2. insulin dependency. Because glucose crosses the placenta but insulin does not, the fetus is stimulated to produce an abnormally large amount of insulin. When the child is born, its own system cannot sustain the high glucose levels that it is accustomed to, but its insulin production is in high gear. Thus, at least until its system comes into a new equilibrium, it is overproducing insulin.

I brought back three publications (which Heather is in the process of turning into computer text) that explain various aspects of gestational diabetes and discuss dietary strategies.

Nurse Bloom also, at this time, starts a (paper) registration form that includes mother's age, due date, previous pregnancies, whether there have been previous large babies, personal and family histories, etc. She also orders four weeks of labs for the patient, explains the expected results from those labs, and asks the patients to call her at 3pm on each day that they have a set of lab tests. Each time, she looks at those results and does further dietary counseling and, depending on the test results, recommends further interventions. She accumulates paper reports from these tests, stapled to the paper registration form. She also gives copies to the patient.

If diet does not adequately control the gestational diabetes (i.e., if glucose readings are consistently high: 130, rather than the normal 120), then the patient is given a glucometer to take home and asked to track her serum glucose four times per day.[8] During two weeks with the glucometer, the nurse calls twice a week, transcribes the glucometer data to her flowsheet, discusses any dietary lapses (ice cream, cokes, etc.) that may have caused high readings, and helps to motivate the patient to follow her diet. If adequate control still cannot be achieved, then the patient is told to take insulin injections one or two times per day. She is also encouraged to "slack off" if she gets four normal glucose levels in a row, because dietary control is considered better than outright insulin administration.

Most women with gestational diabetes return to normal after pregnancy, but some significant fraction (?) become insulin-dependent diabetics and must continue on insulin indefinitely.

According to Dr. Gallagher, we could also choose to focus on a more complex set of cases: women who begin pregnancy as diabetics.

Nurse Bloom thinks that almost all of her patients are and would be highly motivated to participate more in their own care. She thinks that a GA agent in the patient's hands could allow the patient to check on lab results, report in glocometer readings, etc., and would thus save her a lot of time telephoning people simply to tell them that a reading was normal and that nothing special needs to be done. She also believes that fewer cases would "fall through the cracks" if the patient herself had access to the relevant data: they do now occasionally fail to notify a patient of abnormal test results, or fail to get a patient to see the dietitian, etc.

Dr. Miller agreed with a lot of these impressions, emphasizing that diabetic patients tend to be very compliant, especially when motivated by their pregnancy. He also thought that tracking one's health "gives them something to do" during pregnancy, and he felt that they would respond very positively to something like GA that would give them needed information and rationale for their therapy.

Another routine intervention this OB clinic follows is to ask expectant mothers to conduct fetal activity monitoring every evening after 28 weeks of gestational age. This monitoring consists of timing how long it takes to count ten noticeable movements of the fetus in the evening. If the time is longer than an hour, the patient is told to call her doctor or nurse immediately. Then follows a sequence of successively more intensive tests: (a) non-stress test, (b) biophysical profile, (c) challenge stress test, and (d) direct fetal blood sampling. There is evidence that this monitoring reduces the stillbirth rate. This is based, if I understand correctly, on prospective cohort studies, but not on randomized clinical trials. I wondered if there is any reason to think that trend analysis of the activity monitoring data before it crosses the one hour threshold would lead to early recognition of possible problems and thus be useful. There are no studies that would shed light on this.

Other studies suggest that daily contact from nurses helps reduce the incidence of prematurity, especially for pregnancies with multiple births. Unexpectedly, fetal monitoring (I presume with full-time external detectors?) turned out not to have any appreciable effect.

Measuring Outcomes

We then turned to a discussion of how one would go about measuring the outcome of an intervention such as providing a gestational diabetes GA to a selected group of patients. Seriously addressing this issue is clearly going to be critical, because our funding agency is interested almost exclusively in demonstrable results, and the ethos of decision-making in the military is now very much outcomes oriented.

It appears that showing dramatic improvements in clinical outcome are unlikely to be possible. The doctors' impression is that for most cases of gestational diabetes, the ill effects of minimal treatment are likely to be relatively minor: characterized by Dr. Miller as "a short waterfall." They expect that a treatment regimen that simply said "Here's a diet you should follow; see you!" would result in a few additional C-sections, bigger babies, and perhaps a day in the ICU for the child. The long-term outcome is unlikely to differ much, however.

Another way to quantify possible measurable outcome results is to look at the possible improvement in morbidity or mortality. Currently, perinatal mortality for pregnancies that run past twenty weeks[9] is about 12/1000. This number varies, and is as high as 18-20/1000 in Washington D.C. area hospitals. The doctors' impression is that about half of these losses are impossible to prevent even with perfect and unlimited care. These are fetuses with severe developmental abnormalities, such as anencephaly. Of the remaining half, they believe that perhaps half of those (3/1000) could be prevented by perfect care, and the other 3 are "potentially preventable" (which means they don't know). If it is true that at best 3-6 perinatal fatalities of 1000 births can be prevented, it would be difficult to demonstrate significant improvements based on any reasonable sized trial.

Most morbidity in newborns is apparently associated with prematurity. The five major causes of morbidity are:

Bethesda Naval Hospital, along with all the Naval Hospitals, has been engaged, with some consultants, in trying to generate an outcomes model for the hospitals' conduct of obstetrics.[10] The OB model defines an aggregate outcome for mother and child, which aggregates and weights different aspects of maternal and pediatric outcomes from a pregnancy. The dominant term for the mother is the presence and severity of post-partum infections, though other factors also enter. The dominant term for the child is status, as measured by things like APGAR scores, need for ICU treatment, etc. The overall outcomes model considers three types of variables:

I should be able to obtain a report containing the analysis of this study, which is just being completed.[11] The study has developed a logistic regression model for how risk factors affect outcomes. There are some oddball results,[12] though most make sense. This model may be able to provide us the background data needed to demonstrate some health effectiveness of Guardian Angel-based interventions.

Finally, there is a much better chance that we may be able to demonstrate effects on the process of care, the overloading of medical and/or nursing staff, and the cost of care. Nurse Bloom and Dr. Gallagher emphasized the tremendous pressure on them under the current system. The nurse's commitment to caring for difficult pregnancies is amazing, and both worried about her ability to sustain this level of energy indefinitely. Any way to offload even routine efforts (e.g., telling patients that test outcomes were nominal) would be much appreciated. Dr. Gallagher says that an OB's average time for personal contact with a patient during a visit is seven (!!!) minutes. The patient is scheduled for fifteen, but about eight of those minutes involve the doctor's reviewing the patient's chart and writing up orders and visit results. To complicate matters, there has been a recent decision to fold the Walter Reed OB service into the Bethesda Naval Hospital's OB clinic, and everyone expects that this will involve further crowding and a worsening of the doctor/patient ratio. Under these circumstances, measurable improvements in time efficiency can be valuable outcomes.

In addition, the Navy has a "work unit" model of how much effort is involved in medical care. For estimation purposes, a work unit translates into the expenditure of $31, though of course Navy medical costs tend to be much more granular and difficult to change in small increments. The baseline total cost of a pregnancy, including delivery, comes in this model to about $3,000. (C.f., comparable cost at a university teaching hospital is about $10,000.) Gestational diabetes adds a little over $2,000 (66 work units ) of additional expense. This is a reasonably well-quantified number, and we may be able to demonstrate eventual cost savings with GA.

Plans

My plans, so far, are very tentative, and I hope that this memo and my seminar presentation will generate useful discussions to help clarify, redirect and refine them. I plan initially to focus on the short-term issues in the project because I think HIIP requires some short-term results and we need to get our "hands dirty" in order to figure out our longer term objectives in any case.

I would like to develop, by this Fall, a prototype Guardian Angel for gestational diabetes that will assist the patient in following the path described above. It must also assist the obstetricians and nurses in delivering at least as good care as currently, and perhaps better. It must also interface with the moving target of the information system(s) that are being implemented at Bethesda by DEC, so that much of the flow of data that is now transmitted by telephone and stored redundantly in computer and on paper can be automated. The GA must also contain, perhaps in innovatively linked forms, the patient educational data that are currently being used in the clinic. I would like to view this first prototype as just that, but to have it be complete enough and robust enough that we can plan to put one or two units into the hands of actual patients by around the end of this calendar year.

Some of the critical tasks to achieve these ambitious goals are:

  1. Determine what data will be available, and in what form, from computers at Bethesda.
  2. Decide on an implementation platform for us to use. Alternatives include a Newton or similar PDA, a laptop computer, or a desktop machine.
  3. Codify details of what our GA will provide, in collaboration with the clinicians at Bethesda.
  4. Then, the "simple matter of programming."
  5. Plan some evaluation measures, even if based on very small amounts of actual use.

Breast Disease Clinic

Quite independently of the Obstetrics clinic, we have a "target of opportunity" (can't resist military talk in this context) in the fact that Bethesda is planning to open a breast disease clinic, much of whose business is expected to be care for breast cancer. As the genetics of breast cancer is coming to be elucidated, many issues concerning the heritability of breast cancer, individual and family counseling, etc., touch on capabilities we are building anyway in the Geninfer project. It would be foolish not to take advantage of this synergy.

I plan to get involved in some of the planning for the breast clinic at Bethesda also, and to discover how we can best offer our Geninfer program or some derivative for use in their setting.

Administrative

Because we were funded at a lower budget level than we had requested and because we were funded to do work that is substantially different from what we had proposed, we need to figure out how best to use the budget that we have gotten and the abilities and interests of the entire group of collaborators. High on my agenda is to hold discussions with our collaborating doctors and with Gensym about exactly how to define their roles, contributions, and funding from this project.

We should probably seek out the participation in this project of some Boston-area specialists in high-risk obstetrics. Dr. Gallagher recommended John Repke, M.D., at Brigham and Women's, and Sara Garmel, M.D., at the Division of Maternal-Fetal Medicine, Dept. Ob/Gyn at Tufts. Both may be good choices for other reasons as well. Garmel's institutional connection with Dr. Pauker's group at Tufts, and Dr. Repke's connection with Brigham & Women's, where it might make sense to connect up with Dr. Greenes' project in formulation and implementation of protocol-based care, especially in high-risk obstetrics.

Appendix:

Materials for the HIIP PI Meeting, San Diego CA, March 23-24

(psz note: This material was prepared for the above meeting, at which all HIIP participants gathered for two relentless days to discuss overall organization of the project and to figure out strong and weak relationships among components.)

These 22 project descriptions characterize the work that is currently being done or about to be started within HIIP. The focus of the PI meeting in San Diego March 23-24 will be to figure out how to enable ourselves to work effectively together. Please review the descriptions before the meeting so we all start from a common base.

Note that the 22 projects fall into 4 project types; let's try to improve the following definitions as we discuss the 22 projects:

Site Projects

The purpose of the site projects is to

Component Projects

The purpose of the component projects is to

Generalization Projects

The purpose of the generalization projects is to

Support Projects

The purpose of the support projects is to


Project Descriptions as of 3/20/95

Bethesda Ob/Gyn (site project)

The goal of the Bethesda OB/GYN Site project is to incrementally introduce information technology into the clinic to solve their clinical information management problems. Our approach involves

We will engage the clinic in several vision-in-action cycles, each culminating in a useful, cooperatively-designed piece of functionality in the following areas

The result will be a Clinical Associate, generating potentially widespread value to the ambulatory care domain.

NNMC Breast Disease Center (site project)

The goal of the Bethesda Breast Disease Center Site project is to incrementally introduce information technology into the center to solve their clinical information management problems. Our approach involves

We will engage the clinic in several vision-in-action cycles, each culminating in a useful, cooperatively-designed piece of functionality in the following areas

The result will be a Clinical Associate, generating potentially widespread value to the ambulatory care domain.

WRAMC Urology Clinic (site project)

The goal of the Walter Reed Urology Clinic Site project is to incrementally introduce information technology into the clinic to solve their clinical information management problems. Our approach involves

We will engage the clinic in several vision-in-action cycles, each culminating in a useful, cooperatively-designed piece of functionality in the following areas

The result will be a Clinical Associate, generating potentially widespread value to the ambulatory care domain.

HIIP Consortia (site project)

This project will use the SEP methodology to understand and help improve and support the work practices of parts of HIIP. Our focus will be on how to support activities such as program management, the generation of work practice descriptions, the identification of change-relevant areas, compelling-case making, and vision-in-action design. Components that can likely make it easier to do HIIP right are currently being developed in other projects (eg, a VNS-based PME that can be used to support HIIP program management, a Process Handbook that can be used by HIIP KA people to generate useful work practice descriptions).

Our initial focus will be on SEPing HIIP from a program management perspective. After that, we will continue to SEP HIIP, but from the perspective of understanding better how to support KA. Our focus after that will be decided on after we have had some experience with the program management and KA perspectives.

SmartForms (component project)

The SmartForms project aims to develop an advanced architecture for applications whose user interfaces are based on computerized forms. SmartForms are not simply passive screens where users enter information. SmartForms are aware, to some degree, of the meaning of the information they collect and how that information is used in the workplace. This awareness may be manifested in various ways. For example, a SmartForm in a medical application might highlight lab test results that are outside of the normal range. Or, if the variance is particularly extreme, the form might try more actively to contact a doctor to bring the results to their attention. Other SmartForms might know where they should be sent once they are filled out, and transmit themselves. Or, they might automatically fill parts of themselves out based on values on other forms. For example, a SmartForm might help a doctor select ICD-9 codes based on clinical data entered on standard medical chart forms.

SmartForms systems will be prototyped and deployed in selected military hospital clinics, such as the obstetrics and breast clinics at the National Naval Medical Center in Bethesda, MD. Our experience with the prototypes will be used to develop guidelines for creating effective SmartForms systems. Research institutions associated with the SmartForms project will create and demonstrate software that facilitates the creation of SmartForms-based applications.

Mecano (component project)

Development of human-computer interfaces by traditional methods involves piecing together low-level components (widgets) that have no explicit relationship to the task of the interface user or to the application domain for which the software is being constructed. The Mecano project seeks to develop new methodologies for construction of user interfaces that are informed explicitly by declarative models of the user's task and of the application domain. We are exploring methods for allowing developers to specify task models and domain models, and for programmatically generating user interfaces based on those models. The approach is consistent with the philosophy of domain-specific software architectures (DSSA). Our work is being tested by development of a new version of the PROTEGE-II system that runs under the Windows NT operating system.

Clinical Spreadsheet (component project)

The Clinical Spreadsheet (CSS) will be a clinical management information tool that allows health care personnel to quickly view a large number of data elements with a minimum number of keystrokes. CSS is intended to improve the management, retrieval, and presentation formats of data and information essential for the care of patients by nurses, HCPs and other members of the health care delivery team. The CSS user interface will allow the user to specify what data elements are to be retrieved from CHCS and how these data elements are to be displayed into a spreadsheet style format. By choosing a window or group of windows, the user is enabled to define the level of specificity desired. The intent is that, as a tool, CSS will have the flexibility to be a user-defined interface as well as an interface which utilizes pre-defined templates to assist the user in obtaining desired information.

Example: A user may wish, for example, to retrieve general information about a particular patient. The user might therefore define a spreadsheet-like matrix in which patients are listed as row names and order types as column headers. By having the capability to move in any direction within this user-defined spreadsheet, the user selects one or more column headings (order types) and one or more rows (patients). A pop-up menu might then show a list of the activities available within the selected order types. A selection is made to review or result orders, for example. Then that request would be processed for the patients that had been selected and the data displayed for the user.

Arachne (component project)

Components for specific functions of need to physicians and other health care workers can be built in reusable form. These can include such functions as clinical data access and display, image access and manipulation, form-based data entry, and information resource access and navigation. The latter include information resources such as WWW, electronic books, bibliographic data bases (e.g., MEDLINE), expert systems, and clinical guideline libraries. This project will develope prototype tools for image access and manipulation, guideline navigation, DXplain access, electronic book browsing, WWW access, and form-based data entry, initially in a single-platform mode and not truly component-based. We will work with the HIIP collaborators and the SEP process to identify high priority tools for the testbed, and to focus further development on implementation of components to deliver that functionality.

SIMS (component project)

The goal of this project is to support the knowledge retrieval needs associated with the work of a genetics counselor at the NNMC OB/GYN Clinic. We plan to bring up a SIMS server that will support transparent access to the following information sources, some of which are (and will continue to be) accessed by phone:

Update capabilities will be added to the system to support entry of test results and other data. If additional database software is installed at NNMC (eg, the Benetech package), we will work to integrate access to it into the knowledge server as well. If electronic networking is set up between NNMC and Malcolm Grow Medical Center, we will add to the system the ability to access the CHCS system there as well.

MEDTRANS (component project)

The MEDTRANS project aims to facilitate communication in the Health Care process by "translating" among the sublanguages used by different providers (such as physician, nurse, hospital administrator, insurance clerk, etc.). The system converts incoming data/records into a neutral internal form, selects (for each appropriate recipient) all relevant information, and translates that information into the sublanguage dialect of the recipient. Two complementary technologies are used: domain-specific software architecture (DSSA) modules convert standardized information while machine translation (MT) modules convert freer-form natural language expressions. Terms of the internal form are defined in several large terminology banks, both medical and lay, taxonomized to support translation processing.

The intial phase of MEDTRANS involves recognition of certain classes of inputs and conversion of these classes into appropriate internal form. In the first test application, appropriate portions of patient records will be identified and their ICD-9 or CPT codes recognized, with associated degrees of confidence.

Patient Advocate (component project)

The Patient's Associate (PA) project will design, build, demonstrate, and evaluate a reusable technology for a class of associate systems to support patients' management of their own health-related behavior on a day-to-day basis. In general, PAs will help patients to comply with physicians' instructions and advice, exercise health-related common sense, and monitor themselves for condition-specific danger signs. PAs will assist in clinical practice by helping patients decide when they need unscheduled consultation with health-care professionals, helping them to identify and report relevant symptoms, and consulting directly with health care professional as needed.

Our reusable technology will incorporate and extend the domain-specific software architecture previously developed in our ARPA-sponsored Guardian Project for ICU patient monitoring. In particular, we will reuse Guardian's underlying architecture and 4-5 of its most reliable reasoning components (e.g.,monitoring, fault detection, diagnosis, prediction, and planning). We will extend it with appropriate medical domain knowledge and new reasoning components as needed (within available resources). We will build, demonstrate, and evaluate a series of increasingly sophisticated prototype PA systems for a well-defined class of obstetrics patients requiring both well-mother care and monitoring for specific complications.

Guardian Angels (component project)

The use of integrated patient-centered medical information systems can improve the collection and maintenance of health-care records, improve communication between the patient and the health-care system, enhance patient education, involve the patient in cooperative decision-making, improve compliance, and ultimately lead to better preventive care, earlier intervention, and improved clinical outcomes. We believe that we will be able to construct an early prototype of this kind of system, restricted to a single category of patients -- pregnant women. Both clinical staff and patients will find the system usable and will perceive its benefits. A system will be constructed that outpatient pregnant women can take home with them (either in the form of a PDA or software to run on a home computer). This interactive system will solicit information from the patient on the progress of her pregnancy, alert her to any conditions that appear to require immediate medical attention, and present material to her describing the progress of her pregnancy, risk factors she may face, planned diagnostic tests and options, and any interactions between her pregnancy and other medical conditions. We plan to conduct surveys addressing the usability and utility of the system, both to patients and health-care personnel. Initial surveys will be informal, formative evaluations to help refine system design.

Document Management (component project)

In ambulatory clinics there are a broad range of situations in which information comes into the clinic on paper; in many of these situations, dealing with the paper is problematic. The problem is sometimes that the paper canÕt be in two places at once; it is sometimes that the paper documents are frequently misplaced, it is sometimes that it is hard to find the information that is needed, and it is sometimes that the paper documents become worn and thus partly illegible. This project will focus on developing reuseable document management technology that

Program Management Environment (component project)

This project will produce PME -- an environment for program management in large, geographically distributed research programs. We will extend our Virtual Notebook Systesm (VNS) in the following ways:

In developing the PME, we will work with DEC as they SEP the HIIP project.

Text Analysis and Access Techniques (component project)

The goal of this project is to develop techniques that will make the information in health-related text databases accessible to a broad range of users and for a variety of purposes. Specifically, we will assess the value of human language technologies (HLT), i.e. advanced text retrieval, routing, categorization, extraction, and browsing techniques, for medical information including health policy documents, medical literature, and, particularly, clinical text such as patient records. The project will carry out a range of research projects to evaluate these techniques and will deliver software that can be used for the Health Affairs Analyst Associate (HAAA) and other applications such as retrieving clinical information, coding documents, and supporting outcome analysis. As part of carrying out this program, we plan to generate test databases that can serve as benchmarks for future research and development efforts. This work is an extension of the research and development carried out under the ARPA TIPSTER program.

SEP Development (component project)

The projectÕs goal is the development of methods, tools, and expertise to enable productive design engagements in complex, changing workplaces. We will:

DomainModel/ReferenceArchitecture Creation (generalization project)

This project will create an aggregated model for ambulatory healthcare containing at least all the information in the HIIP models for the OB/GYN Clinic at NNMC, the Urology Clinic at WRAMC, and the Breast Disease Center at NNMC; it will then report on the attempts to use this aggregated model:

Process Handbook (generalization project)

This project will demonstrate the utility of the Process Handbook. The Process Handbook consists of a process model Repository and a set of Viewers, each of which highlights a particular facet of a model. It also provides a method to import and export portions of a model via the Process Interchange Format (PIF). Potential demonstrations include integrating the Handbook with a set of Commerical off the Shelf (COTS) applications such as Workflow Managers and Process Simulators. Other uses of the model include providing the information required for application forms generation and application database generation.

This project will also enhance the existing Process Handbook software as required. Expected activities include extending the Repository schema and constructing additional Viewers.

The project is currently making use of two types of models as examples:

The ambulatory healthcare domain model, as an example of a model which highlights the important characteristics of a set of site models, will be used to demonstrate the process design capabilities of the Handbook.

SmartForms Generation (generalization project)

Creation of user interfaces for data presentation and data acquisition for clinical applications can take advantage of the existing PROTEGE-II system, which constructs static form-based user interfaces directly from declarative domain models. In this project, we will experiment with the representation of both site models and domain models that entail the data requirements of clinical tasks. We will also construct new problem-solving methods that can analyze process models to determine the attendant data requirements for the represented processes. The goal is to represent the processes that take place in clinical environments, and to generate programmatically from those process representations specifications for user interfaces that can display and acquire the corresponding patient data.

Geode (generalization project)

Care guidelines for ambulatory patients with a particular health problem can be defined in sufficient detail that care providers can be reminded of what actions are appropriate at any time given specific details of the patientÕs situation. We will create state diagrams describing what care is appropriately provided to complicated OB patients depending on the current state of their pregnancy, past treatments, and physiological factors. This library of state diagrams will be used in workflow management to recommend to care providers specific actions to perform during each patient visit.

Program Integration and Infrastructure (support project)

This project supports the coordination of HIIP. The work involves helping to plan how the products of the various HIIP primes and subcontractors will be integrated as part of useful systems at NNMC and WRAMC. It also involves reporting both internally and externally on the progress of this integrated work. It is also responsible for seeking opportunities to demonstrate and transition the integrated products of this R&D effort into mainline military applications.

The coordination activities occur during site visits to component providers, to demonstration sites and during a weekly conference phone call of the Core Team.

Technical Infrastructure (support project)

This project involves creating and maintaining appropriate computing environments at all of the HIIP sites (eg, NNMC, WRAMC, and HIIP) and creating and maintaining at UTA, Digital, and SAIC computing environments that mirror the environments at the sites.


  1. Note: we are funded by contract, not grant, to help assure this!
  2. This attempt seems problematic to me, because it is unclear how managing an R&D and deployment project like HIIP is enough like managing health care. But this idea seems close to the hearts of some of the program leaders, so on we go with it.
  3. Incidentally, as far as I can tell, PA is has about exactly the same vision as GA. Barbara's strategy, however, is to begin by showing the applicability of her process monitoring software (blackboard architecture for managing ventillation) to this domain: so the actual work is likely to be quite different.
  4. Arguably, the nurse in the OB clinic already fills this role, though without official recognition. Vide infra.
  5. The fetus develops glucose receptors and begins to have a functioning system for producing its own insulin at about this time. It is thought that elevated levels of serum glucose in the fetus before this gestational age do not matter much.
  6. I am unsure of the details of this, but it appears to involve a fasting glucose and then a measurement two hours after breakfast; I need to determine the exact regimen and what are considered the threshold values.
  7. Some women apparently want only a rudimentary explanation of issues, whereas others have advanced degrees in biochemistry and want to know the latest research articles on the topic.
  8. They use a LifeScan OneTouch II glucometer, which can apparently be fitted with a serial line interface. There is a PC program called Merlin which is able to upload data from the glucometer and do some trend analyses on it. Lifescan's tech support is 1-800-227-8862, and we should at some appropriate time learn more about this instrument.
  9. Most spontaneous abortions occur early in pregnancy, and the 20-week start time eliminates these.
  10. I suspect this is part of a larger outcomes effort, but I heard only about the OB part.
  11. Dr. Miller's recommendation is to ask Dr. Raymond S. Crawford, III, of FMAS Corp's Medical Information Services division, in Rockville, MD. He is, I think, conducting the study.
  12. For example, contrary to expectations, being a single mother has a strongly positive predictive contribution to good outcomes. Some thought about the population can make sense of this apparently odd result. Most women being cared for by the Navy's OB clinics are wives of servicemen. A few are servicewomen themselves. Dependent wives cannot, almost by definition, be single, although dependent daughters might be. Servicewomen, however, don't depend on a spouse for membership in the healthcare system, and can therefore be single. These latter women, being in the active military, are likely to be in considerably better physical condition than the average woman, thus contributing to the somewhat unexpected regression result.