Information and communication technology in health care: do we need feedback?

Transcription

1 International Journal of Medical Informatics 66 (2002) 75/83 Information and communication technology in health care: do we need feedback? Johan van der Lei * Department of Medical Informatics, Erasmus MC-University Medical Centre Rotterdam, PO Box 1738, 3000 DR Rotterdam, The Netherlands Abstract Applying information and communication technology (ICT) to a given medical domain is not merely adding a new technique. When applied to a medical domain, ICT has the potential to radically change processes in that domain. In their paper Health care in the information society: a prognosis for the year 2013 Haux and co-workers provide us with a set of predictions based on progress in three main areas: (1) patient-centred recording and use of medical data for collaborative care; (2) process-integrated decision support; and (3) comprehensive use of patient data for research and health care reporting. We complement their theses by predicting some of the feedback mechanisms that will develop as ICT is shaping health care. Feedback is return to the input of a part of the output of a mechanism; this part of the input constituting information that reports discrepancies between intended and actual operation and leads to a self-correcting action that can be utilised. We discuss feedback in the areas Haux identifies: the patient record, decision support, and the use of data for research and health care reporting. For each of these domains, we will discuss the output that serves as input, the discrepancies discovered, and the self-correction that will or should follow. # 2002 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Information and communication technology; Health care; Feedback 1. Introduction In their paper Health care in the information society: a prognosis for the year 2013 Haux and co-workers provide us with a set of predictions regarding the year Their predictions are based on progress in three main areas: (1) patient-centred recording and * Tel.: / ; fax: / address: vanderlei@mi.fgg.eur.nl (J. van der Lei). use of medical data for collaborative care; (2) process-integrated decision support through current medical knowledge; and (3) comprehensive use of patient data for research and health care reporting. Haux tries to be specific in their predictions; in a total of 30 theses they convey to the reader the scope and speed of the changes. Although one could argue some of the theses or the exact percentages Haux gives in these theses, I find myself in overall agreement with their predictions /02/$ - see front matter # 2002 Elsevier Science Ireland Ltd. All rights reserved. PII: S ( 0 2 )

2 76 J. van der Lei / International Journal of Medical Informatics 66 (2002) 75/83 In this paper, we will try to complement their theses by predicting some of the feedback mechanisms that will develop (or should be developed) as ICT is shaping health care. Biology teaches us the importance of feedback loops*/feedback loops are essential for any living organism. System theory argues that, also for systems, feedback is an essential ingredient. Webster [1] defines feedback as: The return to the input of a part of the output of a mechanism, this part of the input constituting information that reports discrepancies between intended and actual operation and leads to a self-correcting action that can be utilised. This definition identifies three components of feedback. First, a part of the output is fed back as input. Second, this input in turn reveals shortcomings in operation. Finally, the input results in self-correcting action. In the remaining of this paper I will discuss feedback in the areas Haux identifies: the patient record, decision support, and the use of data for research and health care reporting. For each of these domains, we will discuss the output that serves as input, the discrepancies discovered, and the self-correction that will or should follow. We do not intend to be complete in our description of feedback loops that will emerge, we will only describe a few mechanisms to illustrate the dynamic interaction between health care and ICT. But before we address these feedback loops, we will take a closer look at the interaction between ICT and health care. 2. Interaction The term Medical Informatics dates from the second half of the 1970s and is based on the French term informatique medicale. Research in informatics ranges from fundamental computer science to applied informatics. Definitions of medical informatics take both the scientific, fundamental aspect and the applied, pragmatic aspect into account. According to Shortliffe, medical informatics is... the science of using system-analytic tools... to develop procedures (algorithms) for management, process control, decisionmaking and scientific analysis of medical knowledge [2]. According to van Bemmel medical informatics comprises the theoretical and practical aspects of information processing and communication, based on knowledge and experience derived from processes in medicine and health care [2]. Medical informatics is determined by the intersection of the terms medicine and informatics. Medicine identifies the area of research. Informatics identifies the methodology used. In medical informatics, we develop and assess methods and systems for the acquisition, processing, and interpretation of patient data. Computers are the vehicles to realise these goals. The role of computers in medical informatics, however, varies. If the medical informatics research is applied, the objectiveis to develop a computer system that will be used by health-care professionals; for example, research aimed at the development of electronic medical records. If the medical informatics research is more fundamental, the computer plays a role as experimental environment for models that are developed; the objective is not to build a system, but to verify a hypothesis or to investigate the limitations of models. Applying information and communication technology (ICT) to a given medical domain is not merely adding a new technique. When applied to a medical domain, ICT has the potential to radically change processes in that domain. That change, however, may not be apparent at the beginning. When introduced

3 J. van der Lei / International Journal of Medical Informatics 66 (2002) 75/83 77 into an environment, ICT will initially often emulate or resemble the already existing processes. Typically, this is only a temporary stage. When workers and researchers in that domain begin to appreciate the potential of ICT, this initial stage is followed by more fundamental change? In that domain that take advantage of the potential of ICT. Electronic communication, for example, is a relative simple technology. The contents of a message may be structured (i.e. contain a predefined set of data) or free text. When introduced in the health-care process, electronic communication is used to replace existing paper documents. The names of the first electronic messages often even carry the names of their paper counterpart; electronic discharge letter, electronic prescription, etc. At first glance, little has changed when compared with the previous paper-based communication except the speed of delivery, In this stage, the infrastructure required for ICT has been installed, but the impact of ICT on the processes is still very limited. Subsequently, however, the ability to send data using electronic communication is used to support new forms of collaboration between health-care professionals. The emphasis shifts from replacing paper documents to sharing data between healthcare professionals. As physicians increasingly share data, issues such as the standardisation of the content of medical records are becoming important areas of research. In addition, the fact that data can be transferred easily over distances enables physicians to interpret data while the patient is located miles away (resulting in, for example, the so-called tele-diagnosis ) or to communicate with patients over longer distances using, for example, the internet. For our discussion of feedback loops, we will distinguish three stages in the use of ICT applications in health care. The first stage, the research stage, is dominated by researchers and characterised by little actual use of the systems in daily routine. The second stage, the dissemination stage, is characterised by an (often unpredictable) explosion of clinical use but focussed on replacing or augmenting existing paper-based procedures. The third stage, the interaction stage, is characterised by the interaction between ICT and its environment resulting in fundamental changes in the health-care delivery system. Predicting is difficult, especially when it concerns the future. To further complicate the issue, we need to distinguish between the dissemination stage of ICT use in which ICT resembles the existing, paper-based procedures, and the interaction stage in which the potential of ICT to fundamentally change the delivery of care is exploited. In the dissemination stage, when ICT resembles the paperbased procedures, the dominant feedback loops tend to centre on the improved efficiency of the existing procedures. In the interaction stage, more of the potential benefits of ICT become apparent, but also the limitations or unexpected side effects begin to show. As a result, more complicated feedback mechanisms begin to emerge. 3. Patient-centred recording and use of medical data for collaborative care In its early stages, the written medical record had the purpose of documenting the care given to a patient, and thus to facilitate continuity of that care. The entries in the medical record enabled the physician to recall previous episodes of illness and treatment. In recent years, however, medical records have been used increasingly for other purposes; the records are used as a data source for purposes ranging from billing the patient to performing epidemiological studies, and from performing quality control to defending oneself against

4 78 J. van der Lei / International Journal of Medical Informatics 66 (2002) 75/83 legal claims. One of the major barriers for using the data in paper medical records for purposes other than providing medical care to an individual patient is the inaccessible and of ten unstructured nature of the paper medical record. The introduction of computer-based medical records removes, to a certain degree, that barrier. The explicit purpose of automating medical records is to use the data in those records to support not only care of individual patients, but also applications such as decision support, quality control, cost control, or epidemiology [3]. The quality of medical-record data, however, has often been lamented. The reliability of clinical data, for example, has long been questioned, and tensions between reimbursement schemes and coding schemes have been discussed. Some researchers argue that the process of automation may further reduce the reliability of data. Bumum, for example, states: with the advent of the information era in medicine, we are pouring out a torrent of medical record misinformation [4]. Although we disagree with this pessimistic viewpoint, we acknowledge that medical data are recorded for a specific purpose and that this purpose has an influence on what data are recorded and how they are recorded. In developing systems that record medical data, designers of systems make decisions about how to model those data in order to perform a given task. For example, in designing the computer-based medical-record system Elias [5], the designers focused on issues such as ease of data entry and emulating existing paper records. The same designers subsequently discovered significant limitations in the Elias records when they developed a decision-support system that uses these records as a source of data [6]. Despite the limitations of the current computer-based medical records and the data contained in these records, many researchers believe electronic medical records will significantly change medical practice [3,7]. In my opinion, the use of electronic records is, broadly speaking, moving from the research stage into the dissemination stage. As experience through actual use accumulates, feedback loops are emerging. Feedback involves the return of some of the output as input so that the system can correct itself. The sheer volume of data drives an obvious feedback loop. When data are available in an electronic format, shipping large quantities of data becomes easy. This data, however, is input to the physicians who increasingly are confronted with information overload. A lifelong medical record becomes a nightmare when a vast sea of data hides relevant data. The feedback shows that the intended operation, adequate data, is not achieved. For the system to correct itself, methods for discarding or deleting data need to be developed. At present, however, such methods are hardly available. Haux and co-workers add an explicit statement: patient-centred recording and use of medical data for collaborative care. This prompts another feedback loop. The loop is again drive by the sheer volume of data. But the discrepancy shown by the feedback is that the physician is not able to identify the expectations or requests of other health-care workers. The intended operation, collaborative care, is not achieved. For the system to correct itself, methods that convey expectations and request need to be developed. At present, however, one of the major shortcomings in collaborative care is the absence of clear divisions of labour. As a result, sharing (large volumes of) medical (e.g. by means of electronic communication) does not automatically result in collaborative care. Electronic medical records will stimulate and enable other developments. Haux focuses on two of these developments: the develop-

5 J. van der Lei / International Journal of Medical Informatics 66 (2002) 75/83 79 ment and use of integrated decision support systems, and the creation of observational databases. 4. Integrated decision support In a clinical decision-support system, medical knowledge is modelled. That is, the designers of the system encode in formalisms the medical knowledge that is required to make decisions. Such formalisms or models have traditionally been divided into two main groups: the quantitative and the qualitative models. The quantitative models are based on well-defined statistical methods and may rely on training sets of patient data to train the model. Examples of such models are neural networks, fuzzy sets, Bayesian models or belief nets. Qualitative models are typically less formal, and often based on perceptions of human reasoning. Examples of such models are truth tables, decision trees or clinical algorithms. Increasingly, however, builders of decision-support systems will combine different models in a given system: a Bayesian network may be used to model the diagnostic knowledge and a decision tree may be used to model treatment decisions, and a pharmacokinetic model may be used to calculate dosage regimens. Clinical decision-support systems require patient data. Without patient data, patient specific advice cannot be generated. Some systems may require interaction between the system and the clinician. The clinician initiates a dialogue with the system and provides data to the system by entering symptoms or answering questions. Experience with this type of system has shown that the acceptance of this type of system by clinicians is relatively low. Other systems are integrated with electronic medical records, and use the data in these records as input. In such sellings, receiving decision support requires little or no additional data input on the part of the clinician. Finally, some systems are directly connected to the devices that generate the data, for example, systems that interpret electrocardiograms (ECOs) or laboratory data. By applying the medical knowledge to the patient data, the system generates patient specific advice. Some systems, especially systems integrated with electronic medical records, provide advice independent of a physician s request for advice-unsolicited advice. Examples are reminding systems that continuously screen patient data for conditions that should be brought to the clinician s attention (e.g. the patient s kidney function is decreasing, or the patient is eligible for preventive screening). Other systems, such as critiquing systems, may monitor the decisions of the clinician and report deviations from guidelines. Although hundreds of clinical decisionsupport systems have been reported in literature, only a few have been the subject of a rigorous clinical evaluation. Of the systems that have been evaluated in trials, however, the majority of the studies showed an impact on clinician performance [8]. Especially systems integrated with electronic medical records have been demonstrated to improve the quality of care [8]. In the light of the currently available evidence, clinical decision-support systems constitute a possible method to support the implementation of clinical guidelines into practice. Integrated decision support is still predominantly in the research stage. However, studies that document changes in physician behaviour in response integrated decision support will have a profound impact. For many years, researchers have argued the difficulties encountered in changing physician behaviour. The fact that integrated decision

6 80 J. van der Lei / International Journal of Medical Informatics 66 (2002) 75/83 support, albeit often in a research selling, has an impact on clinician performance will initiate a positive feedback resulting in an increased emphasis on decision-support systems as a method to implemented evidencebased medicine. As evidence accumulates regarding the impact of integrated decision support, panics interested in changing physician behaviour (e.g. government, health-care insurers or professional organisations) will increasingly use ICT as a mechanism to achieve their objectives. Physicians, when treating an individual patient, will be confronted with a reduction in their freedom of choice and will search for mechanisms to divert from standardised care. These feedback mechanisms are in conflict. The positive feedback is based recognition that ICT is a powerful mechanism to influence decisionmaking. The negative feedback originates from physicians who have to treat individual patients. In order to resolve the conflicting feedback, a mechanism that balances the individual against the collective has to be established. Developing such a mechanism, however, is beyond the scope of ICT itself and poses questions that society has to answer. 5. Comprehensive use of data Haux and co-workers argue that one of the goals is the comprehensive use of patient data for research and health care reporting. As a result of the increased use of electronic medical records, large observational databases containing data on millions of patients have become available for researchers. The data contained in these databases is not of the same quality as the data collected in, for example, a clinical trial. The data are collected in routine practice, and, although most observational databases attempt to standardise the recording of data and to monitor the quality of the data, only few observational databases require the physician to record additional information [9]. In the absence of a clear study design (e.g. a randomised control led trial to compare the effectiveness of possible treatment regimens) and specification of data required for that study, the data in observational databases are difficult to interpret due to possible confounding. The advantage of observational databases, however, is that they reflect current clinical practice. Moreover, the data are readily available, and the costs are not prohibitive. In settings where all the medical data are recorded in an electronic format, the opportunities for research are similar to studies carried out using paper charts. Compared with paper charts, these observational data bases provide an environment where the analysis can be performed quicker, the data are legible, normal practice can be studied, rare events can be studied, longitudinal follow-up of patients is possible, and subgroups for further study (e.g. additional data collection or patients eligible for a prospective trails) can be identified. In cases where a rapid analysis is required (e.g. a suspected sideeffect of a drug), observational databases provide a setting for a quick assessment of the question. Observational databases that rely on electronic records have limitations. Analysis of the contents of the records shows that information important for a researcher is often not recorded. Medical records typically contain data describing the patient s state (e.g. the results of laboratory tests), and the actions of the physician (e.g. prescribing medication). Relationships between data are often not recorded. The medical record mainly reflects what is done rather than why. A further complicating factor is that when data in the medical record describes the relationship between observations (or findings) and actions

7 J. van der Lei / International Journal of Medical Informatics 66 (2002) 75/83 81 (e.g. treatment), the information is often recorded in the form of free text. The physician s first and most important objective of keeping automated medical records is to document with the purpose of ensuring the quality and continuity of medical care. From the physician s perspective, free text is often an ideal method for expressing the patient s condition. Researchers, on the other hand, prefer coded data to facilitate the automatic processing of large numbers of patients. It is unrealistic, however, to expect physicians to code all relevant data; the time required to code medical data renders it impractical. In addition, coding is in essence a process of reducing the rich presentation of the patientphysician encounter to a limited set of predefined terms. The available data in an observational database may therefore not be sufficient to validly answer a specific research question. The completeness of data can only be discussed in the context of a specific study. It is not possible to predict all possible data that would be required for all possible studies. As a result, data in observational databases will be incomplete. Depending on the study question and the impact of incomplete information, additional data may need to be collected [9]. The analysis of large observational databases will create a feedback loop in which the results of those analyses will demonstrate the limitations of the data. Based on the same data, publications will increasingly report conflicting results. Although efforts are made to ensure complete and reliable data collection, researchers will discover that the requirements posed by different use of data are insufficiently addressed. Comprehensive use of data for research and healthcare reporting will only partly achieved. Methods will have to be developed that bridge the gap between patient care and research. 6. Closing the loop One of the fundamental changes when conventional paper-based medical records are replaced with computer-based records involve the ability to process the data in that record for different purposes. As electronic medical records arc becoming available, researchers use these records to change medical practice by providing decision support, and analyse observational databases to study the delivery of care. New usage of data, however, generates additional requirements. Thus the experience in developing decision support systems and analysing observational databases feeds back into the requirements for electronic medical records. And as new requirements for the electronic record are formulated, the electronic record begins to change. In the area of decision-support systems, researchers are combining reminder systems that rely solely on already recorded data in the electronic records with systems that request additional information from clinicians. The resulting systems rely on one hand on data already available in the electronic record to determine eligible patients, and subsequently interact with the clinician to assess, for example, whether the patient should be treated according to a certain protocol. The results of that interaction are recorded in the medical record. Researchers working on the development of observational databases are beginning to combine retrospective research with prospective research. Trials are translated into software, distributed electronically, and added to an electronic medical record. Based on data in an electronic medical record, the system automatically detects patients eligible for a trial. The electronic medical record informs the clinician that the patient is eligible, and request permission to include

8 82 J. van der Lei / International Journal of Medical Informatics 66 (2002) 75/83 the patient in the trail. The system subsequently performs the randomisation between treatment arms during patient consultation, and the electronic record supports subsequent data collection. As a result the boundaries between an electronic record, a decision-support system, and systems for clinical trials are beginning to fade. In the year 2013, feedback loops will have shaped the use of ICT in health care. By the year 2013, a large majority of the workers in health care have first hand experience of the consequences of data overload. Researchers have developed some methods to reduce large amounts of data. These methods, however, still require human supervision and have proven to be time-consuming. Health-care professionals have rediscovered the importance (and costs) of archiving data and have started to develop new formats for medical records. These new formats are still in a research stage-central theme is the notion that an individual care provider him/herself has the obligation to summarise his/her data for other health-care professionals. Parallel to efforts to develop new formats for medical record, a debate among the different medical disciplines focuses on what information has to be included in that summary. The underlying issue in this unresolved debate is how to determine the relevance of data when future use of that data is not known. By 2013, health-care professionals have recognised that sharing data does not necessarily mean better communication. Healthcare professionals have recognised the need to collaborate, and have been redesigning the delivery of care. They have augmented the medical record with data specifically aimed at communicating expectations and dividing workload. Many of the existing treatment protocols, for example, are multidisciplinary and involve a clear division of labour and specify when communication should occur. In 2013, integrated decision support is an accepted and widely used mechanism to distribute and execute guidelines and protocols. The delivery of care is tuned to these protocols to ensure an optimal use of time and money. Whereas 10 years before, in 2003, the emphasis of the decision support systems was to ensure adherence to protocols, an important function of the systems in 2013 is to identify situations in which the protocol should not be followed. Researchers in 2013 have access to very large observational databases. These obsensational databases have their prime source of data. In many cases, these observational databases do not contain all the data required for a specific study. Researchers, however, are able to contact both care provider and patient to obtain additional information. Moreover, naturalistic trials conducted in a routine setting are the most common method for clinical trials. In 2013, we have made significant progress in closing the ultimate feedback loop. Each patient/physician encounter, each investigation, each laboratory test, and each treatment in medical practice constitutes, in principle, an experiment. Ideally, we learn from each experiment. Paper as a medium to record data limits our ability to exploit that potential. Electronic medical records will facilitate research that relies on data recorded in routine medical practice. The potential, however, of ICT lays in its ability to close the loop between clinical practice, research, and education. References [1] Webster s Third New International Dictionary, Encyclopedia Britannica Inc., London, 1971.

9 J. van der Lei / International Journal of Medical Informatics 66 (2002) 75/83 83 [2] J.H. Van Bemmel, M.A. Musen (Eds.), Handbook of Medical Informatics, Springer, Heidelberg, Germany, [3] Institute of Medicine, Committee on Improving the Patient Record, The Computer-Based Patient Record: an Essential Technology for Health Care (Revised Edition), National Academy Press, Washington, DC, [4] J.F. Burnum, The misinformation era: the fall of the medical record, Ann. Int. Med. 110 (1989) 482/484. [5] J. van der Lei, J.S. Duisterhout, H.P. Westerhof, E. van der Does, P.V.M. Cromme, W.M. Boon, et al., The introduction of computer-based patient records in the Netherlands, Ann. Int. Med. 119 (1993) 1036/1041. [6] J. van der Lei, M.A. Musen, E. Van der Does, A.J. Man in t Veld, J.H. Van Bemmel, Comparison of computer-aided and human review of general practitioners management of hypertension, Lancet 338 (1991) 1505/1508. [7] J.A. Knottnerus, The role of electronic patient records in the development of general practice in The Netherlands, Methods Inform. Med. 38 (1999) 350/355. [8] D.L. Hunt, R.B. Haynes, S.E. Hanna, K. Smith, Effects of computer-based decision support systems on physician performance and patient outcomes: a systematic review, J. Am. Med. Assoc. 280 (15) (1998) 1339/1346. [9] A.E. Vlug, J. van der Lei, B.M.T. Mosseveld, M.A.M. Van Wijk, P.D. Van der Linden, M.C.J.M. Sturkenboom, et al., Postmarketing surveillance based on electronic patient records: the IPCI project, Methods Inform. Med. 38 (1999) 339/344.