SMART TRANSPLANTATION

Fever of Unknown Origin after Stem Cell Transplantation as a Model for a

Knowledge-Based Decision Support System in Medicine

Gerrit Meixner

, Nancy Thiels

University of Kaiserslautern, Gottlieb-Daimler-Strasse, Kaiserslautern, Germany

²German Research Center for Artificial Intelligence (DFKI), Trippstadter Strasse, Kaiserslautern, Germany

Ingo Haschler

, Andreas Wicht

and Ulrike Klein

University of Heidelberg, Im Neuenheimer Feld, Heidelberg, Germany

Keywords: Human-Computer Interaction in Health Care, Usability of Medical Information Systems, Human Aspects of

Future Technologies in Health Care, Cognitive Task Analysis, Usability Engineering, Stem cell

transplantation, Decision support system.

Abstract: Public health care has to make use of the potentials of IT to meet the enormous demands on patient

management in the future. Embedding artificial intelligence in medicine may lead to an increase in health

care quality and patient safety. One possibility in this respect is the use of knowledge-based decision

support systems which facilitate the practice of evidence-based medicine. Conditions for such a system are

structured data sources to extract relevant data for the proposed decision. Therefore, the demonstrator “allo-

tool” was designed. To develop the allo-tool a user-orientated process was applied and future users of the

later software were integrated in each step of the development process. The concept of introducing a

"Medical decision support system based on the model of Stem Cell Transplantation" was developed

afterwards. The global objectives of the planned system are (1) to improve patient safety (2) to support

patient autonomy and (3) to optimize the work flow of medical personnel.

1 INTRODUCTION

In many areas of human life, computer-based

Information Technology (IT) has prevailed and has

become essential for the coordinated and efficient

organization of work flow. This offers numerous

advantages for the future, but will also lead to

problems for the people confronted with it.

Especially in the field of health care, interaction

between human beings and information technology

is a sensitive subject. Physicians have immense

reservations and apprehensions of being made the

slaves of information scientists and of their

programmed computer system. Nevertheless,

medicine has to become more scientific in patient

management. The importance of interdisciplinary

advice and discussion which is a prerequisite for the

best possible decision on a treatment strategy has to

be reflected in the application of new information

technologies.

Embedding artificial intelligence in medicine

may lead to an increase in quality and safety and to a

decrease in costs significantly. In the future, IT-

systems are supposed to have the ability to extract

the relevant knowledge to filter irrelevant

information and focus on significant information and

to present it available to the user. Inevitably, people

working in health care will have to make use of the

potentials of IT in order to meet the enormous

demands on patient management in the future.

One possibility in this respect is a knowledge

based decision support system which facilitates the

practice of evidence-based medicine. The idea of

medical decision support systems is not new. In the

past, however, these systems have not become

popular and utilized. The reason was not that the

technology had failed, but that the implementation

was inadequate. A prerequisite for success is that we

understand "how medicine thinks" in order to be

298

Meixner G., Thiels N., Haschler I., Wicht A. and Klein U. (2008).

SMART TRANSPLANTATION - Fever of Unknown Origin after Stem Cell Transplantation as a Model for a Knowledge-Based Decision Support System

in Medicine.

In Proceedings of the First International Conference on Health Informatics, pages 298-304

 SciTePress

able to create a "decision-supporting", and not a

"decision-making" system.

Decision support systems need an evaluation of

their applicability and of their cost-benefit

relationship as well as the evaluation of precision,

quality, and effects of their recommendations.

Against the background of the challenges mentioned

above, the concept of introducing a "Medical

decision support system based on the model of Stem

Cell Transplantation", named allo-tool, was

developed.

Whenever bringing a decision support system

into clinical practice, one has to consider the

possible harms caused by the system. Concretely

this means to consider what happens if the system

recommends a wrong decision or presents

incomplete or inadequate information at the point of

care to the physician. We therefore plan to integrate

a feedback mechanism. In the case that a physician

has the opinion that the system could recommend

something wrong, he will submit an incident report

to the maintenance team. In that way, the users (i.e.

mainly the physicians) will retain control over the

system and possible harm to the patients caused by

the system will be minimized.

The global objectives of the planned system are

(1) to improve patient safety (2) to support patient

autonomy and (3) to optimize the work flow of

medical personnel. It might lead to more efficient

use of resources without detrimental effects on the

relationship between physician and patient or on the

physician's autonomy to decide. The allogeneic

haematopoietic stem cell transplantation is

extremely well suited as a model for this type of

system because of repeating standard procedures, the

well defined span of time required and the

predictable recovery period as well as recurrent side

effects after transplantation.

2 STEM CELL

TRANSPLANTATION

Hematopoietic stem cell transplantation (HSCT) or

bone marrow transplantation (BMT) is a medical

procedure in the field of hematology and oncology

(Korbling, 1986). It is most often performed for

people with diseases of blood or bone marrow

(Gratwohl, 2007; Goldschmidt, 2000). HSCT

remains a risky procedure and has always been

reserved for patients with life threatening diseases.

Since the availability of stem cell growth factors,

most hematopoietic stem cell transplantation

procedures have been performed with stem cells

collected from the peripheral blood (Montgomery,

2007). Most recipients of HSCTs are patients with

leukemia or aggressive hematological tumors who

would benefit from treatment with high doses of

chemotherapy or total body irradiation. Other

patients who receive bone marrow transplants

include pediatric cases where patients have an

inborn defect and were born with defective stem

cells. Other conditions that bone marrow transplants

are considered for include inherently diseases of the

bone marrow. More recently non-myeloablative, or

so-called "mini transplant," procedures have been

developed which do not require such large doses of

chemotherapy and radiation (Djulbegovic, 2003).

This has allowed HSCT to be conducted in older

patients and as a matter of principle without the need

for hospitalization. There are two major types of

stem cell transplantation maneuvers: Autologous

HSCT involves isolation of HSC from a patient,

storage of the stem cells in a freezer, high-dose

chemotherapy to eradicate the malignant cell

population at the cost of also eliminating the

patient's bone marrow stem cells, then return of the

patient's own stored stem cells to their body.

Autologous transplants have the advantage of a

lower risk of graft rejection, infection and graft-

versus-host disease. Allogeneic HSCT, as the second

type, involves two people, one is the healthy donor

and one is the recipient. Allogeneic HSC donors

must have a tissue type that matches the recipient

and, in addition, the recipient requires

immunosuppressive medications. Allogeneic

transplant donors may be related or unrelated

volunteers.

The number of performed HSCT, autologous and

allogeneic, is increasing. Due to better anti-infective

medication the life-threatening side effects of

infectious complications are decreasing but they still

remain as the main risk factor for life threatening

side effects (Walker, 2007; Afessa, 2006).

Therefore one intention of the allo-tool is the

improvement of patient safety through a decision

support system for choosing an anti-infective

therapy based on evidence based advices.

Bacterial, viral or fungal infections are severe

side effects of high dose chemotherapy and stem cell

transplantation and can result in life threatening

complications. Infections are the most important

causes of morbidity and mortality in patients

undergoing allogeneic stem cell transplantation.

That is why immediate anti-infective therapy in case

of fever is mandatory (Afessa, 2006). Because of the

resistance to widely used antibiotics and a shift of

SMART TRANSPLANTATION - Fever of Unknown Origin after Stem Cell Transplantation as a Model for a

Knowledge-Based Decision Support System in Medicine

299

Figure 1: Risk chart graphic.

causative pathogens towards multi-resistant bacteria,

anti-infective therapy remains to be a challenge.

Beside this drug-interaction, kidney function, known

allergies, the cause of fever and former anti-infective

therapies needs to be included into the decision of

the treatment of choice. Therefore the model of a

decision support system was developed. The idea of

programming and implementing such a system is not

new. Attempts of this type had already been initiated

in the Eighties (then called medical expert systems)

(Shortliffe, 1975). In the past, however, these

medical expert systems have not become popular

and utilized. The reason was not that the technology

had failed, but that the implementation was

inadequate.

Beside this the support of patient autonomy

through modern forms of communications like web-

based access to specific patient data and the

optimization of the work flow in the complex

process of an allogeneic transplantation is the goal

of the allogeneic tool “smartTransplantation”.

3 USEWARE ENGINEERING

PROCESS

The level of acceptance and efficiency of a modern

user interface are determined by the ease of use of

the interface. Primary considerations in this

evolutionary development process are always the

requirements and needs of the user, for whom the

user interface is being developed. The process

consists of analysis, structural design, rough and

detailed design, implementation, and evaluation

(Zühlke, 2004). As a continuation of the analysis,

evaluation occurs concomitantly to the development.

A risk chart (see Fig. 1) has been developed to

get an understanding of the transplantation

procedure. It explains the work flow of the stem cell

transplantation in detail. The preparation of the

analysis phase derived five different user groups.

From these user groups ten different persons were

chosen to be questioned to find out the tasks and

needs of users for the later user interface.

After the questioning and characterizing, the

collected data were compiled (Bödcher, 2007).

Additional results included an analysis of

weaknesses identified in the existing user interfaces

and documentation of the “wish list” expressed by

the users. Problems to be found within the analysis

of the clinical situation for the stem cell

transplantation were published.

Based on the task models of the different user

groups, the use structure for the future user interface

was developed. The modeling language useML

(Reuther, 2003) was used in structural design. The

result of the structural design phase is a platform-

independent model and it provides the foundation

for the later design phase. This model formed the

base for the development of the prototype.

4 PROTOTYPE: ALLO-TOOL

The intention of the allo-tool is not only the

optimization of the work flow in the complex

process of an allogeneic transplantation, but also the

provision of a structured platform for relevant data.

This aim is reached by data integration from

different existing information sources: clinical

information system, drug information system and

paper patient files as well as domain knowledge

formalized in knowledge bases (described in more

detail in chapter 5). The digitized aggregated data is

displayed in a clearly structured way.

The tool shall be able to extract medical

information from different sources, structure the

HEALTHINF 2008 - International Conference on Health Informatics

300

Figure 2: Patient history screen in the detailed view.

information and automatically generate discharge

letters and further documents, e.g. drug plans.

Each user group is able to work consistently with

the allo-tool. The tool consists of four main views:

calendar, memos, clinical trials and patients. The

memo view consists of a text field, where the user

has the possibility to take notes. In the clinical trial

view the user is able to administrate information

about current clinical trials.

For allogeneic transplantations information about

conditioning, transplantation, immunosuppression,

GVHD prevention (graft versus host disease) and

donor lymphocyte transfusions are stated within

these clinical trials.

According to the work flow of an allogeneic

transplantation, the detailed patients view consists of

five different chronological parts in the main work

area of the tool: patient history, donator-search,

preliminary examinations, inpatient stay and follow-

up treatment. These five parts are clearly separated

phases in the time line of an allogeneic

transplantation (see Fig. 1). The graphical

representation of a time line (see Fig. 2) enables

physicians to have a better understanding (Norman,

1990) of the current date according to the complex

process of an allogeneic transplantation. The tool

supports physicians in keeping deadlines according

to a physical examination time schedule.

The patient’s history phase (see Fig. 2) consists

of a structured overview of the patient’s master data.

In the course of diseases previous examinations and

medical results are displayed in a table view.

Secondary diagnoses are an important part of the

decision supporting system and help to minimize

adverse effects.

The preliminary examination phase contains

information about the accomplished examinations

according to the physical examination time schedule.

The results of external examinations are entered into

the clinical information system. The allo-tool shall

be able to extract the results of external

examinations.

In the phase of hospitalization, the patient

receives high dose chemotherapy and some patients

a whole body irradiation. After that the allogeneic

stem cell transplantation is performed. During the

whole phase patient data (e.g. vital signs, blood

counts or organ functions) are monitored very

closely. Beside others, one important task of the

attending physician is to administrate drug plans.

Via an implemented interface, physicians are

able to use an existing drug information system

(Kaltschmidt, 2004;

Pruszydlo, 2006).

SMART TRANSPLANTATION - Fever of Unknown Origin after Stem Cell Transplantation as a Model for a

Knowledge-Based Decision Support System in Medicine

301

In the follow-up phase the physician is reminded to

accomplish examinations according to the physical

examination time schedule. The time schedule

consists of a large number of physical examinations

and complex execution logic. The prototype

transfers the execution logic from the physicians to

the software.

Patients in the follow-up phase have the

possibility to access test results via internet and can

directly communicate with their allocated physician

via email. This will be reached by developing a web-

based access for patients. Patients will have the

possibility to view their own data (e.g. blood test

results, X-ray photographs). Test results, which can

be viewed by patients over internet are evaluated

and released by their allocated physician. Patients in

the follow-up phase have to control e.g. blood

pressure values. They can measure blood pressure

on their own and email the results over the web-

based access directly to their allocated physician. So

patients don’t have to call the hospital via telephone

or have to visit the hospital on their own.

5 KNOWLEDGE-BASED

SUPPORT

In the future, physicians will be supported by a

knowledge-based system with data mining

capabilities. The detailed functionality is explained

in the following.

Monitoring and interpretation of several

parameters such as vital signs or laboratory results

will be one main function of the knowledge-based

system component. The system thus detects specific

clinical situations and pushes unsolicited warnings

or reminder and starts the according work flow.

In everyday clinical practice physicians are often

faced with an information overflow rather than a

lack of information, they have to spend valuable

time in looking for relevant findings. To reduce this

time a “Semantic Information Extraction” is

provided. The system looks for findings in the

patient’s history which could fit into the context of

the current clinical picture. Furthermore, it

visualizes the classification of the findings, which

has previously to be provided by a physician:

“abnormal”, “no abnormality”, “unclear”. So the

physician gets a clearly arranged listing of relevant

findings matching the current issue.

For a defined amount of clinical pictures,

therapy and diagnostic recommendations are

provided. These recommendations are on the one

hand based on so-called domain knowledge which is

patient-independent knowledge on e.g. diseases or

processes. Domain knowledge will mainly derived

from clinical practice guidelines (CPG) or in-house

standard operating procedures (SOP). On the other

hand they are based on available patient data such as

current parameters as well as the patient’s general

data (if any parameters are missing or out of date,

the system will ask the user to enter these), the

system should give exact medication and dosing

recommendations. Once the system has generated a

recommendation, it will also provide the user with a

reasonable explanation. In this context it may also be

helpful that the user is guided to external knowledge

resources, which match the current clinical picture,

such as local SOP-documents, relevant study

protocols or other medical knowledge bases, if

further information is requested.

Recommending a (drug) therapy at first requires

a knowledge base which encapsulates the SOPs and

the guidelines’ knowledge. The process of

formalizing guidelines and SOPs is a challenging

and time-consuming task (Kaiser, 2007). We will try

to build a semi-automatic, peer-reviewed process,

i.e. the system will try to import the unstructured

material and recognize as much structure as

possible. In the next step a specially trained medical

professional will then review and complete

structural or semantic issues. Afterwards, another

medical professional will review the work of the

first one to ensure high quality of the formalized

knowledge.

Once the therapy decision was accepted and

started, the system supports the medical staff by

monitoring not only the over-all treatment work

flow, but also the flow of the specific therapy.

Concerning the over-all treatment work flow, the

system should give reminders or warnings if

scheduled diagnostic procedures (e.g. ECG) are not

done or assigned yet or scheduled medications (e.g.

antibiotic prophylaxis) are not ordered. Regarding

the therapy monitoring, the system should detect if a

patient does not respond to the treatment (e.g.

patient remains febrile) and recommend necessary

steps and/or alternative therapies as well as indicated

diagnostic procedures. On the other hand, the system

should detect if a therapy was apparently successful

and recommend further steps (e.g. to stop antibiotics

and to start a prophylaxis again). In this context the

system presents - as already mentioned before -

reasonable explanations and guides the user to

external knowledge resources if requested.

As medical knowledge is subject to ongoing

changes, the knowledge base has to be maintained

regularly. Changes can be triggered either by

external factors (e.g. update of a guideline) or by

internal factors. We therefore need a user feedback

mechanism which enables a physician to mark a

system’s decision as potentially wrong. The user

HEALTHINF 2008 - International Conference on Health Informatics

302

feedback has to be analyzed and the need to change

the knowledge base has to be assessed. All changes

to the knowledge base must be carried out within a

“bullet-proof” process like the one described above.

A critical issue when it comes to the

implementation of decision support systems is user

compliance. In our environment we face mainly

physicians as users, but also patients in an advanced

state of the software (e.g. test results via web-based

access). We have to ensure usability of the tool (as

described earlier) and ensure that physicians get a

benefit from it. The patients are most likely to do

everything to improve their therapy outcome

because of their severe illness.

One reason why decision support systems often

do not prevail in clinical practice is poor work flow

integration (Bates, 2003). Since our software will

cover the SCT-treatment process as a whole we can

map the practical work flow. Physicians will more

likely use a decision support system if they see a

clear benefit from it. This means mainly time-saving

as well as convenient access to all relevant

information.

Because the decision support system will be

seamlessly embedded as a component into the allo-

tool, we can reach an optimal solution to this

obstacle. All of the knowledge-based features we

presented above will appear within the current work

flow context. We provide the medical user with only

the information he or she needs at a given clinical

situation. Therefore we can expect a reasonable time

saving for the users. By ensuring that we do not miss

relevant information on the other side, we may raise

quality of treatment and patient safety.

6 CONCLUSIONS

It is incontestable that people working in health care

will have to make use of the potentials of IT in order

to meet the enormous demands on patient

management in the future. Beside this the quality of

work can be supported by intelligent software which

is able to extract, rate and provide the user with

relevant data. Not least patients require more

autonomy of their own health information data. To

meet this challenges the demonstrator of the allo-

tool was developed.

Time consuming data search, redundant

information and vast numbers of needed software

applications are reduced by displaying all data in

one tool. User interfaces, which are designed in

close relationship to known software products,

developed with the support of different users during

the whole process and consulting of usability experts

facilitate an easy-to-use application. Time schedules,

reminder of deadlines and coherent information

about study procedures enable medical staff to work

efficiently. Taking these analysis results as a basis,

conditions for a medical decision support system are

accomplished. In order to meet the exploding

number of scientific perception, decision support

systems are needed in the future to maintain the

quality of medical decisions.

Through web-based access to selected health

information, patients obtain more autonomy and

responsibility. Summarizing the potentialities of the

planned allo-tool, the goals mentioned at the

beginning, (1) improvement of patient safety (2)

support of patient autonomy and (3) optimizing the

work flow of medical personnel are illustrated.

Studies to evaluate these potentialities are needed to

prove these advantages of the allo-tool.

ACKNOWLEDGEMENTS

This work was supported by the Gottlieb Daimler-

and Karl Benz-Foundation.

REFERENCES

Afessa, B.; Peters, S.G., 2006. Major complications

following hematopoietic stem cell transplantation,

Semin Respir Crit Care Med, 2006, 27(3), 297-309.

Bates, D. W.; Kuperman, G. J.; Wang, S.; Gandhi, T.;

Kittler, A.; Volk, L.; Spurr, C.; Khorasani, R.;

Tanasijevic, M.; Middleton, B., 2003. Ten

commandments for effective clinical decision support:

making the practice of evidence-based medicine a

reality, J Am Med Inform Assoc, 2003, 10, 523-30.

Bödcher, A., 2007. Methodische Nutzungskontext-Analyse

als Grundlage eines strukturierten USEWARE-

Engineering-Prozesses, Fortschritt-Berichte pak, no.

14, Kaiserslautern: University of Kaiserslautern.

Djulbegovic, B.; Seidenfeld, J.; Bonnell, C.; Kumar, A.,

2003. Nonmyeloablative allogeneic stem-cell

transplantation for hematologic malignancies: a

systematic review, Cancer Control, 2003, 10(1), 17-41.

Goldschmidt, H.; Egerer, G.; Ho, A.D., 2000. Autologous

and allogeneic stem cell transplantation in multiple

myeloma, Bone Marrow Transplant, 2000; 25, Suppl.

2, 25-26.

Gratwohl, A.; Baldomero, H.; Frauendorfer, K.; Urbano-

Ispizua, A.; Niederwieser, D., 2007. Results of the

EBMT activity survey 2005 on haematopoietic stem

cell transplantation: focus on increasing use of

unrelated donors, Bone Marrow Transplant, 2007,

39(2), 71-87.

SMART TRANSPLANTATION - Fever of Unknown Origin after Stem Cell Transplantation as a Model for a

Knowledge-Based Decision Support System in Medicine

303

Kaiser, K.; Akkaya, C.; Miksch, S., 2007. How can

information extraction ease formalizing treatment

processes in clinical practice guidelines? A method

and its evaluation, Artif Intell Med, 2007, 39, 151-

163.

Kaltschmidt J.; Gallin S.; Haefeli W.E., 2004. Essential

functional requirements for an effective electronic

drug information system in a hospital, Int J Clin

Pharmacol Ther, 2004, 42:615.

Korbling, M.; Dorken, B.; Ho A.D.; Pezzutto, A.;

Hunstein, W.; Fliedner, T.M., 1986. Autologous

transplantation of blood-derived hemopoietic stem

cells after myeloablative therapy in a patient with

Burkitt's lymphoma blood, J AM Society of

Hematology, 1986, 67(2), 529-532.

Montgomery, M.; Cottler-Fox, M., 2007. Mobilization and

collection of autologous hematopoietic

progenitor/stem cells, Clin Adv Hematol Oncol, 2007,

5(2), 127-136.

Norman, D., 1990. The Design of Everyday Things,

Currency.

Pruszydlo, M.G.; Kaltschmidt, J.; Schmitt, S.P.W.;

Markmann, A.; Bertsche, T.; Haefeli W.E., 2004.

Successful implementation of computer-aided

prescribing in a hospital providing tertiary care, Int J

Clin Pharmacol Ther, 2006, 44:529.

Reuther, A., 2003. useML – Systematische Entwicklung

von Maschinenbediensystemen mit XML, Fortschritt-

Berichte pak, no. 8, Kaiserslautern: University of

Kaiserslautern.

Shortliffe, E. H.; Davis, R.; Axline, S. G.; Buchanan, B.

G.; Green, C. C.; Cohen, S. N., 1975. Computer based

consultations in clinical therapeutics: explanation and

rule acquisition capabilities of the MYCIN system,

Comput Biomed Res, 1975, Aug, 8(4), 303-320.

Walker, C.M.; van Burik, J.A.; De For, T.E.; Weisdorf,

D.J., 2007. Cytomegalovirus infection after allogeneic

transplantation: comparison of cord blood with

peripheral blood and marrow graft sources, Biol

Blood Marrow Transplant, 2007, 13(9), 1106-1115.

Zühlke, D., 2004. Useware-Engineering für technische

Systeme, Springer.

HEALTHINF 2008 - International Conference on Health Informatics

304