Research in medical informatics

8 downloads 0 Views 69KB Size Report
their own medical informatics research disci- plines. ... Medical informatics is concerned with the applica- ..... mately 27 per 100,000 compared to China at.
Health Informatics Journal http://jhi.sagepub.com

Research in medical informatics S. Abdul-Kareem, A. Baba and M.I.A. Wahid HEALTH INFORMATICS J 2000; 6; 110 The online version of this article can be found at: http://jhi.sagepub.com/cgi/content/abstract/6/2/110

Published by: http://www.sagepublications.com

Additional services and information for Health Informatics Journal can be found at: Email Alerts: http://jhi.sagepub.com/cgi/alerts Subscriptions: http://jhi.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav

Downloaded from http://jhi.sagepub.com at PENNSYLVANIA STATE UNIV on April 17, 2008 © 2000 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.

Research in medical informatics S. Abdul-Kareem, S. Baba and M.I.A. Wahid

Medical informatics is concerned with the application of computers in the medical and biological sciences and has been considered a field of research in its own right for more than twenty years. In Malaysia, however, there are very few published efforts in this area. To keep up with research activities carried out worldwide and to create expertise that will be in great demand once the Malaysian Multimedia Super Corridor (MSC) Telemedicine Flagship project is implemented, it is time that Malaysians involved themselves in medical informatics research activities. For this reason, we are proposing a project that will involve the application of an artificial neural network in the domain of cancer. As a prelude to our own research, we review current research in medical informatics. This paper subsequently proposes the use of an artificial neural networks as an alternative tool for investigating cancer survival.

Keywords: Medical Informatics, artificial neural networks, cancer, survival

INTRODUCTION S. Abdul-Kareem Lecturer Faculty of Computer Science and Information Technology University of Malaya S. Baba Director, Multimedia Development Centre University of Malaya M.I.A. Wahid Head, Clinical Oncology Unit University Hospital, University of Malaya Address for correspondence: Sameem Abdul-Kareem Lecturer Faculty of Computer Science and Information Technology University of Malaya Pantai Valley Kuala Lumpur 50603 Malaysia Tel: 603-7696371 Fax: 603-7579249 Email: [email protected]

Medical informatics, as defined by Shortliffe, encompasses a wide range of issues including the use of computer and telecommunication technology in the management and use of biomedical information, which includes medical computing and medical information [1]. The simplistic definition of medical informatics is the application of computers in medical and healthcare. Medical informatics has been considered a field of study in its own right for the past two decades [2]. Research in medical computing in the United States can be traced back to the 1960s, when the field was known as medical information sciences. Subsequently, this field of research has been known by a myriad of other names such as healthcare informatics, healthcare information systems and medical informatics. The first Symposium on Computer Applications in Medical Care was held

Health Informatics Journal (2000) 6, 110-15

in 1970, and its success attracted a large group of followers and led to the launch of several medical informatics journals. In the late 1980s a comprehensive textbook was written as an introductory informatics course [1]. Medical informatics research activities have spanned the entire globe from Europe to Australia. Closer to home, both Singapore and Thailand are actively involved in establishing their own medical informatics research disciplines. In Malaysia, however, activities in medical informatics are still lagging behind, despite the fact that the government is intensifying information technology usage through its Multimedia Super Corridor (MSC) projects. The MSC’s Telemedicine Flagship project is divided into four main applications; namely, lifetime health plan (LHP), teleconsultation, mass customized/personalized health information and education (MCPHIE), and continuing medical education (CME). The implementation of these projects will result in the massive use of computing technology in the field of medicine and healthcare covering research areas of medical computer applications, hospital information systems, electronic medical records, remote consultation, remote medical eduation, and so on [3]. Millions of ringgits have been allocated to the project and more will be spent when the pilot project goes nationwide. Vendors, whose priority would invariably be to see that it becomes a huge commercial success, will carry out the bulk of the project. Although there are a few isolated individuals who are keen on the subject, there is no medical informatics group in Malaysia; medical informatics is not currently offered in any university faculty in Malaysia. There are very few Malaysians (if any) who can be considered experts in the field. It is therefore important that Malaysians in both the medical and information technology community understand the importance of medical informatics as a field in its own right. The establishment of a research group in this field is imperative because most projects in medical informatics can only take place through a collaborative effort amongst the various specialties, involving hospitals and universities as well as government institutions. Universities should play a more prominent role in stepping up efforts to promote this discipline amongst academicians and medical and healthcare professionals as well as students. Thus, to keep up with research efforts carried out worldwide and to create expertise that would be in great need once the MSC’s Telemedicine Flagship project is implemented, it is time that Malaysians involved themselves in medical informatics research activities. After reviewing some current research activities carried out worldwide in medical informatics, and

Downloaded from http://jhi.sagepub.com at PENNSYLVANIA STATE UNIV on April 17, 2008 © 2000 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.

Research in medical informatics

discussions with medical and technical experts on the resources currently available, the present authors are proposing a research project in the area of artificial neural networks in medicine. We are specifically looking into the possibility of using neural networks in the field of cancer research.

BACKGROUND Current research in medical informatics The scope of informatics in medicine is huge. Research efforts range from applications of artificial intelligence to problems in biomedicine [4] [5]. Various issues with regard to medical informatics have been raised and are being addressed by different healthcare groups. Although some of the issues may be connected, researchers may sometimes not be aware of this; usually the problems are concerned with the general acceptance of informatics. Current medical informatics research projects around the globe include studies of [1] [2] [6]: ● ● ● ● ● ● ● ● ●

Medical computer applications Hospital information systems Nursing informatics Electronic medical records The impact of IT on the patient–provider relationship The impact of IT on patient care The impact of IT on continuing medical education The impact of IT on learning in healthcare organizations Telemedicine implementation issues

Medical computer applications These are classified as communication systems and advice systems. Communication systems store, retrieve and disseminate medical information while advice systems participate in the diagnosis and management of patient care. Examples of communication systems are databases and archives while advice systems include consultation and critiquing systems. A more detailed discussion of medical computer applications involving artificial neural networks will be found below. Hospital information systems These may be defined as total and integrated systems encompassing admission/discharge, laboratory information, a pharmacy system, and reporting and billing. The use of computer technology in medical and healthcare

111

applications is said to improve healthcare delivery while reducing healthcare costs. The Malaysian LHP project includes the complete computerization of government hospitals involved in the pilot project. Although it will be interesting to follow the development of this project, at present much of it is still at the confidential paperwork stage. Nursing informatics This is an area of research dealing with the science of nursing, computer science and information technology, and is considered a branch of medical informatics. Nursing informatics aids in the identification, collection, processing and management of data and information to support nursing practice, administration, education, research and the expansion of nursing knowledge [7]; it is one of the areas actively being researched in the medical world. This area does not, however, seem to be receiving any interest in Malaysia. Electronic medical records Medical records are considered an important aspect of medical care. Many organizations including the American Institute of Medicine, Medical Records Institute (MRI), American Medical Informatics Institute (AMIA) and the Canadian Institute of Health Informatics (CIHI) have shown keen interest and have invested considerably in research in electronic medical record systems [8]. The two terms ‘computer-based patient record’ (CPR) and ‘electronic medical record’ (EMR) are used interchangeably, although there are subtle differences between the terms. CPR refers to medical records that usually reside in a central computer system, such as a desktop or intranet server. EMR, on the other hand, are intrinsically distributed over the network and can be accessed by a variety of databases [9]. In this paper, as in most of the literature, we shall not distinguish between these two terms. The Medical Records Institute (MRI) of America identifies five phases of the computerization of medical records [10]. It is envisaged that the implementation of these five phases will result either in the complete dissipation or entire clarification of the distinction between CPR and EMR. The Institute of Medicine (IOM) of America defines the computer-based patient record (CPR) as an electronic record that resides in a system specifically designed to support users through [11]: ● ● ●

Availability of complete and accurate data Practitioner reminders and alerts Clinical decision support systems

Downloaded from http://jhi.sagepub.com at PENNSYLVANIA STATE UNIV on April 17, 2008 © 2000 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.

112

Health Informatics Journal

● ●

Links to medical knowledge Other aids

The CPR also enhances patient care through a longitudinal patient record. This may be defined as consisting of records from different times, providers and sites of care that are linked to form a view over time of a patient’s healthcare encounters. Research in computer-based medical records is a long-term process. In certain places, such as Duke University in the US, research has been ongoing since 1968. This initiative has resulted in what Duke claims is the premier computer-based medical record system in the United States [12]. Although much work has been done in this area many research challenges remain, before complete, practical and flexible systems can be produced. The development of automated records is riddled with issues concerning standardization arising from the complex nature of medical data, uncertainty in the observation of data, precision of time and differences in terminology [13] [14] [15]. As a solution to these problems various suggestions have been proposed, such as the development of models of medical records, models of medical information, protocols for data interchange and user interfaces for presenting and acquiring information. Many commercial and localized CPR systems exist at present because of the efforts of various organizations; it is thus unlikely that these organizations, having invested heavily in proprietary systems, would abandon these in favour of the adoption of a standard but different system. This may therefore prove to be the main obstacle in the challenging task ahead for a standard CPR system. It will be interesting to see how the vendors try to resolve these issues for the Malaysian EMR, which forms the core of the LHP project. The impact of IT The impact of information technology on the patient–provider relationship, on patient care, continuing medical education and on learning in healthcare organizations are research areas that are currently being actively pursued by groups and individuals worldwide. Research is based on social–technical issues, implementation and evaluation issues of the use of IT – how IT is perceived by healthcare workers, how it affects the work ethics of care providers and how it might be subjected to abuse are some topics which are currently the focus of attention. There is a lot of potential in this area of research as more and more organizations are looking at the computerization of

their clinics and hospitals and this could be one of the areas future local researchers in Malaysia can look into after the implementation of the MSC telemedicine project. Telemedicine implementation issues Telemedicine involves the provision of medical diagnosis and other services to patients over a distance, through the use of telecommunications technology. The concept encompasses everything from the use of a standard telephone service through to the high speed, wide-bandwidth transmission of digitized signals, fibre optics, satellites, video conferencing, electronic mail and other sophisticated peripheral equipment and software. Studies of the success and failure of telemedicine projects may be found in the literature over the past twenty years. The advent of telemedicine brings a number of benefits to the medical community, but before these benefits can be realized to their full potential, problems associated with telemedicine will have to be ironed out. Some of the issues raised through the use of telemedicine are confidentiality and secrecy, fraud, legal issues and the dispersion of liability [16] [17]. These areas of research would be of paramount interest to the Malaysian community as the Ministry of Health is at present developing its own telemedicine project. The research areas described above are by no means discrete; one may see an overlap or some degree of association between one area of research and another.

Artificial intelligence in medicine Artificial intelligence has been playing a prominent role in medical research activities, and the importance granted to its applications in medicine may be observed through the formation of the Artificial Intelligence in Medicine (AIM) research group as well as through conferences and the publication of journals and books dedicated to this subject. Some of the projects that have been carried out in the past include medical expert systems, clinical knowledge bases, the use of natural-language processing in capturing free text in medical records research, data mining of medical knowledge and the use of artificial neural networks and case-based reasoning in medical diagnosis [18] [19] [20].

Artificial neural networks in medicine Artificial neural networks (ANN) are sets of interconnected processing units capable of

Downloaded from http://jhi.sagepub.com at PENNSYLVANIA STATE UNIV on April 17, 2008 © 2000 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.

Research in medical informatics

self-organizing in response to the training data that are given. They are good classifiers and pattern recognition engines, offering ideal solutions to problems such as speech, character and signal recognition, as well as functional prediction and system modelling. The main advantage of a neural networks are their ability to generalize to new situations: after being trained on a number of examples, ANN can interpolate and extrapolate from the examples to induce a certain pattern or relationship. Another advantage to using ANN is that they are fault tolerant; they can continue to function although in a reduced fashion, even if some of their components are faulty [21] [22]. Neural networks are easy to apply and are domain independent in the sense that they can be used in a variety of applications, from banking and finance to handwritten character recognition and the detection of explosives in airline luggages. Neural networks would probably be the best approach to a problem that is complex and has plenty of data [23] [24] [25] [26] [27]. Neural networks have been used in medicine since the 1980s as an aid to diagnosis and treatment. Since then, both clinicians and computer scientists have been researching the use of ANN in providing clinical diagnosis. Their application has been producing good results in the identification of thyroid diseases, the prediction of the survival analysis of AIDS patients and the prediction of the outcomes of coronary heart disease [28] [29] [30] [31] [32]. Cancer is another disease where ANNs have been used as an aid to identification, prediction and diagnosis, and there are many successful applications of ANNs in cancer: Renal cell carcinoma. The problem of deciding whether the mass from ultrasound data is renal cancer, a renal cyst or some other condition is a common dilemma for radiologists. Backpropagation ANN have been successfully trained to correctly diagnose renal cell carcinoma [33]. Classification of bone tumours. In this investigation ANN were trained to classify bone tumours; the performance of ANN was compared to that of experienced clinicians (three or more years of radiology training) and inexperienced clinicians (less than one year of radiology training). Although the experiment showed that ANN were able to classify bone tumours it was undecided whether their performance was accurate enough to assist radiologists in clinical practice [33].

113

Spread (metastasis) and prognosis in breast cancer. ANN successfully predicted lymph node involvement in breast cancer thus preventing unnecessary dissection of the breast. The ANN also used to predict the long-term prognosis of patients with breast cancer thus facilitating the management and planning strategies of treatments of breast cancer patients [34]. Comparing the prediction accuracy of statistical models and artificial neural networks in breast cancer. The ability to predict the survival of cancer patients is important as it determines the patient’s therapy. Five types of ANN were used to predict five-year survival of patients with breast cancer [36]. Prediction of the planning target volume in radiotherapy for cancer treatment. Target volume in conformal radiotherapy is usually defined by physicians of radiotherapy. It has been observed that the volume varies from physician to physician and, even when the same physician administers the therapy, the volume varies significantly at different points of time. These inaccuracies make conformal radiotherapy less appealing. A feed-forward neural network was trained to predict the planning target volumes with results that show promise as a first step research work [37].

THE PROPOSED PROJECT Medical informatics has been considered a field of research in its own right for more than twenty years. In Malaysia, researchers have yet to formally recognize it as such. There is a need for Malaysians, especially those concerned with medical and healthcare research, as well as those in the world of information technology, to be more actively involved in medical informatics research; for this reason, after reviewing some of the current projects carried out worldwide, we would like to propose a research project in the area of artificial neural networks in medicine with particular focus on cancer research. We hope that this would be a first step towards establishing a research group actively involved in carrying out local research in the field of medical informatics.

Objectives and scope We would like to analyse cancer survival data and hence develop a computer-based cancer prognosis predictor that can be used at an

Downloaded from http://jhi.sagepub.com at PENNSYLVANIA STATE UNIV on April 17, 2008 © 2000 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.

114

Health Informatics Journal

individual level, based on the parameters of individual risk factors, tumour stage and treatment by using the technology of artificial neural networks (ANN). We believe that this predictor would be more accurate than current statistical methods. The tumour model we have chosen for our study is nasopharyngeal carcinoma (NPC) since its occurrence is high amongst the Chinese and South-East Asian ethnic communities. The incidence of NPC amongst the Chinese community of Malaysia is approximately 27 per 100,000 compared to China at 40 per 100,000 and Hong Kong at 35 per 100,000 population. In the United States and in Europe the incidence is only about 1 per 100,000. Secondly, we have a comprehensive set of data available at the University Hospital in Kuala Lumpur. The present authors hope to extend the results obtained from our system and use them to predict the outcome of new nasopharyngeal carcinoma cases. With the help of computerized systems such as the ANN we believe that a fairly accurate evaluation of a patient’s prognosis and survival can be made.

METHODOLOGY

relationships, which are not otherwise apparent. Statistics have proved to be a useful tool in analysing large amounts of data, especially in the field of cancer survival. However, statistics apply to large groups of people and may be meaningless for an individual; they are based on estimates taken from a sample. These estimates will differ from the figures obtained if a complete census had been taken using the same survey procedures. Thus, statistics do not and cannot predict events in the future with any certainty [38] [39]. Neural network technology is relatively new in the field of cancer survival; however, based on its proven success in other fields it has the potential to be explored as an alternative technique to currently used statistical methods. In the University Hospital we have a comprehensive set of data on nasopharyngeal carcinoma (NPC) that has been collected over more than ten years. Bearing in mind that ANN are considered one of the best models for complex and data-rich problems, we would like to show that artificial neural networks can be used as an effective modelling tool in cancer survival analysis. We would also like to show that ANN are able to predict survival at an individual level.

Hypothesis We propose the following hypotheses and intend to develop a system that will prove them: 1. ANN are an effective and efficient tool for the analysis of cancer survival data. 2. ANN can predict survival time at an individual level. 3. ANN produce a better estimate of survival time than statistical methods. We believe that the implementation of our system would help in answering the following questions: 1. What is the survival rate of a particular patient? 2. Does the type of treatment influence survival and outcome? 3. When does metastasis develop? 4. Given a new NPC patient, could the neural network be used to predict how long the patient will survive?

Reasons for our methodology Neural networks are considered one of the most powerful techniques in the area of data mining. The learning algorithms of neural networks can probe through data and learn

SUMMARY Research activities in medical informatics are practiced worldwide from Europe to Australia and includes neighbouring countries such as Singapore and Thailand. Activities range from the implementation of medical computer applications and systems, to the impact of the use of information technology on healthcare. In Malaysia, the Ministry of Health Telemedicine initiative, which has been allocated millions of ringgits has been secured predominantly by vendors. Published research work in medical informatics amongst the health professionals and academicians in Malaysia is minimal. We are thus proposing a research project that would involve professionals from across the discipline. We hope that this project would act as a stepping stone towards establishing a local research group in medical informatics. The project we are proposing involves the use of artificial neural networks in analysing survival data and hence in predicting survival and outcome. The tumour model we have chosen is nasopharyngeal carcinoma, which is commonly seen here in Malaysia, and for which we have a comprehensive set of data

Downloaded from http://jhi.sagepub.com at PENNSYLVANIA STATE UNIV on April 17, 2008 © 2000 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.

Research in medical informatics

collected at the University Hospital, University of Malaya, Kuala Lumpur.

References [1] Shortliffe E H, Perreault L E, Wiederhold G, Fagan L M (eds.) Medical Informatics: Computer Applications in Health Care, Reading, MA: Addison-Wesley, 1990. See also http://www.imbi. uni.freiburg.de/medinf/mi_list.htm. [2] Nerur, S, Raghupathi W. Healthcare information systems: a review of issues toward research themes and agendas into the 21st century. http://hsb.baylor.edu/ramsower/ais.ac.96/papers/ HEALTH1.htm [3] The Government of Malaysia: Ministry of Health. Telemedicine Standards, July 1997. [4] Ohno-Machado L. ‘Medical applications of neural networks: connectionist models of survival’. PhD thesis, Section in Medical Informatics, Stanford University, 1996. [4] Coiera E. The guide to medical informatics. In: The Guide to Medical Informatics, the Internet and Telemedicine. Bath, Chapman and Hall, 1998. http://www.coiera.com. [5] Shortliffe E H. Medical informatics training at Stanford University School of Medicine. In: The IMIA Yearbook of Medical Informatics, Stuttgart: Schattauer, 1995. [6] Green R, Shortliffe E H. Medical Informatics – an emerging discipline and institutional priority, JAMA 1990; 263 (8): 11141120. [7] American Nurses Association Task Force. The Scope of Practice for Nursing Informatics. Washington: American Nurses Publishing, 1994. [8] Nierbergal D. Benefit realisation of computer-based patient records in medical group practices. Proceedings of the COACH Conference 22, Vancouver 1997. [9] http://www.medicalcomputingtoday.com/0nvemrproj. html [10] http://www.aafp.org/fpr/july96/computer/records.html [11] Schreiner R C. ‘The computer-based patient record: the struggle for specifications and standards’, 1996. http://www.sph. unc.edu/courses/hpaa155/cpr.html [12] http://dmi-www.mc.duke.edu/ [13] http://www.acl.lanl.gov/sunrise/Medical/cpr/IOM.html [14] http://www.ha.osd.mil/cbpr/cbpr.html [15] Norman J. ‘Building the computer-based patient record’. Technical report of the Section on Medical Informatics, Stanford University School of Medicine, 1995. [16] Scannell K M, Perednia D A, Kissman H M. Telemedicine, Past, Present, Future. US Department of Health and Human Services, 1995. [17] http://www.duke.edu/~sjd1/pageone.html

115

[18] http://www.ai.univie.ac.at/research.html [19] http://www.ai.mit.edu/projects/cbcl/projects.html [20] http://camis.stanford.edu/projects/uncertain.html [21] Fausett L. Fundamentals of Neural Networks. New Jersey: Prentice Hall, 1994. [22] http://www.nd.com/welcome/whatisnn.htm. [23] Smith L. An Introduction to Neural Network. Centre for Cognitive and Computation Neuroscience, Department of Computing and Mathematics, University of Stirling, 1996. http:// www.cs.stir.ac.uk/~lss/NNIntro/InvSlides.html [24] Tsoi A C. An introduction to neural network, Lecture notes, University of Malaya, 1994. [25] Gurney K. ‘Neural Nets’,1996. http://www.shef.ac.uk/ psychology/gurney/notes/l1/l1.html [26] ftp://ftp.sas.com/pub/neural/FAQ.html#A_intro [27] http://www.zsolutions.com/soyou.htm [28] Ohno-Machado L. ‘Identification of low frequency patterns in backpropagation neural network’, Technical Report, Section in Medical Informatics, Stanford University, 1994. [29] Ohno-Machado L, Walker M G, Musen M A. ‘Hierarchical neural networks for survival analysis’. Technical Report, Section in Medical Informatics, Stanford University, 1994. [30] Ohno-Machado L, Musen M A. ‘Modular neural network for medical prognosis: quantifying the benefits of combining neural networks for survival prediction’. Section in Medical Informatics, Stanford University, 1996. [31] Ohno-Machado L, Musen M A. ‘Sequential versus standard neural networks for temporal pattern recognition: an example using the domain of coronary heart disease’. Section in Medical Informatics, Stanford University, 1996. [32] Ohno-Machado L. ‘Medical applications of neural networks: connectionist models of survival’. PhD thesis, Section in Medical Informatics, Stanford University, 1996. [33] Maclin P S, Dempsey J, Brooks J, Rand J. Using neural networks to diagnose cancer. Journal of Medical Systems 1991; 15 (1): 11-19. [34] Piraino D W, Amartur S C, Richmond B J, Schills J P, Thome J M, Belhobek G H, Schlucter M D. Application of neural network in radiographic diagnosis. J Digit Imaging 1991; 4 (4): 226-32. [35] http://smi-web.stanford.edu/people/pratt/test/bc-prognosis/497396516 [36] Burke H B, Rosen D B, Goodman P H. ‘Comparing the prediction accuracy of statistical models and artificial neural network in breast cancer’. NIPS94 Post-Conference Workshop, 1994. [37] Kaspari N, Michaelis B, Gademann G. Using artificial neural network to define the planning target volume in radiotherapy. Journal of Medical Systems 1997; 21 (6). [38] http://nces.ed.gov/pubsold/D95/dhilite.html [39] http://www.vanhedge.com/stats.htm

Downloaded from http://jhi.sagepub.com at PENNSYLVANIA STATE UNIV on April 17, 2008 © 2000 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.