T,he collection of individual patient data requires

Transcription

1 The development of patient groupings for more effective management of health care HUGH SANDERSON, LEONIE MOUNTNEY The concepts of groupings of patients and treatment episodes are fundamental to better analysis of data for health services management. This article discusses the need to separate out condition groupings (of patients) from resource groupings (of treatment episodes) and demonstrates the way in which the use of these two types of groups could assist in the development of health care planning/purchasing and in the monitoring of the efficiency/effectiveness/appropriateness of services. The use of these groupings for data analysis is the logical development of previous work on programme planning and budgeting, but this has been handicapped in the past by a lack of suitable data. The current developments in clinical information systems and clinical coding mean that we can look forward to richer and more complete data and this will require new grouping systems. Existing grouping systems may not be appropriate for dealing with more detailed information and work to develop new grouping concepts and groupers is required. Key words: groupers, diagnosis related groups, severity, planning, purchasing T,he collection of individual patient data requires methods of classification and grouping to enable analysis. Ever since data on individuals and patients have been collected, they have been grouped in various ways, sometimes for epidemiological purposes, in tabulating mortality and morbidity rates [e.g. the International Classification of Diseases (ICD) 1 ] and sometimes for management purposes in identifying health care activity. In recent years, there has been an increased interest in the area of patient classification as emphasis is placed upon the efficiency and effectiveness of a complex, expensive and politically sensitive health care system. CONCEPTS OF PATIENT GROUPINGS The most widely known patient classification for management purposes is diagnosis related groups (DRGs) 2 which, on the basis of diagnosis and/or procedure, classifies acute hospital episodes into groups of similar expected resource use. DRGs have undergone considerable modification since their first introduction and inspired several imitators (Australian DRGs 4, Canadian CMGs 5, French GHMs 6 and English HRGs 7 ) but these classifications are all similar in that they seek to identify iso-resource groups in the acute care setting. Similarly, iso-resource groupings have been developed for other care settings, for example resource utilization groups (RUGs) for long-stay care, ambulatory visit groups (AVGs) 9 and ambulatory patient groups (APGs) 10 for out-patient/ambulatory settings and so on. Again these seek to distinguish similar resource cost groupings. _ * H.F. Sanderson', LM. Mountney 1 1 National Casemlx Office, IMG (G), NHS Executive, Winchester. UK Correspondence: H.F. Sanderson, National Casemlx Office, NHS Executive, Hlghcroft, Romsey Road, Winchester, Hants, England SO22 5OH, UK, 3 tel , fax *4711 The driving force for these groupings has been a managerial approach to health care which perceives the delivery of service as a production process in which products (health care episodes) are delivered to consumers (patients). The groupings of products into similar resource groups enables comparison of the efficiency with which the products are delivered and a way of identifying how much health care providers should be paid for delivering each unit of the produce This indeed has been the major way in which DRGs and their analogues have been used." However, these groupings of health care activities into similar resource groups do not indicate whether the product was delivered to the right consumer, nor do they indicate how well the product was delivered. Rather like die apocryphal Soviet boot factories which filled their boot production quotas with size 5 left boots, a hospital could (in theory) deliver very poor quality, unnecessary care at low cost and high efficiency. In order to consider these issues of appropriateness and quality, it is necessary to re-examine the concepts on which patient groups are based, in order to allow grouping designs which are adapted to focusing on the needs of the population, the services required to meet those needs and the outcome of those services. CONDITION AND TREATMENT GROUPS The simplest model of health care can be stated as (patient with condition) + (health care action) * outcome In this model there are 2 separate elements. Patients (or individuals) who have conditions (which may be symptoms, dependencies, impairments or fully defined as diagnoses), episodes of illness and a prognosis.

2 Development of patient groupings Condition groupings (states) Treatment groupings (actions) A B C D E Figure 1 The Condition/Treatment Grouping Matrix Health care actions which are packages of resources used in consultations, investigations, treatments, support or palliation. They may be episodes of care. Using 2 axes of classification, 'states' and 'actions', to develop groupings of conditions and groupings of care we can examine the needs for care, the efficiency of care provision and the outcome. In principle, we can draw up matrices which lay out 'state groups' and 'action groups' (figure 1). The cells within the matrix represent all die possible combinations of needs and care, although only a few of diese potential combinations are appropriate. It is of course possible to elaborate the model by distinguishing elements of die clinical process. Thus, problem presentation leads to examination/diagnostic activity, diagnosis leads to treatment planning and treatment/support and, finally, die condition after treatment leads to follow-up observations. In each of diese stages a patient state can be matched to a relevant action. This could lead to die production of 3 sets of matrices. Diagnostic matrix - widi condition/presenting problem mapped to the examination/diagnostic investigation actions. Treatment matrix - with condition/diagnosis mapped to die treatment/care/support actions. Follow-up matrix - widi die post-treatment state mapped to the observation/monitoring/further treatment actions. An important additional element could be to identify a pre-symptomatic/preventive phase, which includes die normal population, diose at risk and those widi early asymptomatic disease. In these, health promotion, prevention and screening may be appropriate actions. The level of elaboration however may confuse radier dian clarify die picture and for die purpose of diis discussion die focus will be on a single condition/treatment matrix. The concept of representing die combination of needs and care packages separately is not in itself new, 12 ' 1-5 but diis approach has not generally been used in die development of die most widely used iso-resource treatment groups. In diese (for instance, in DRGs and HRGs), die groups are constructed widi a mixture of bodi condition information (diagnoses) and treatment information (procedures). Not only are diagnoses used in die construction of die groups, but in many instances patients are divided into diose widi and those widiout complications and co-morbidities and dien assigned to different resource groups. The reason for this use of diagnostic information in die construction of treatment groups is two-fold. Firstly, it makes statistical sense, because normally diere is a relationship between die diagnosis and die treatment. Secondly, because diere is no alternative, using currently available routine information, for medical patients where diere is no procedure to identify die treatment given. Using die diagnosis in diis way makes die implicit assumption that the diagnosis serves as a proxy for die resource package of treatment. For instance, it assumes dial all patients widi an uncomplicated myocardial infarction receive a similar package of care. This assumption may be broadly true, but die unfortunate side effect is a blurring of die distinction between the condition and die treatment. If die data on die package of care was available dien it would be more correct (and probably more accurate) to use information about die package of care to form die groupings. At present however, diere is no option and, despite diis compromise widi the basic concepts, DRGs and HRGs work reasonably well as groupings of packages of care diat are expected to consume similar resources. THE PURPOSE OF DEVELOPING CONDITION AND TREATMENT GROUPS Given rliat diese in-patient treatment packages work reasonably well, why is it important to define condition groups and care groups separately? The answer is to help to obtain a view of die appropriate relationships between die patients' condition and die care provided. This will enable purchasers/payers and providers of care to distinguish between die patients and die type of care diat should be provided. The applications include die following. Turning epidemiological information into an assessment of die types of care and resources required to meet the needs of the community. Assessing die appropriateness of die current care provided and exploring die implications of different packages of care. Auditing die actual results of treatment against die expected outcomes. Assessing die actual costs of care against the expected costs of care. For purchasers/payers diis means diat purchasing and planning of services can be driven, not by die services currently provided, but by die epidemiology of conditions in die population. For example, this distinction would enable die assessment of die prevalence of disabling hip disease in die population so diat die appropriate numbers of hip replacements can be purchased, rather than using information about die existing numbers of hip replacements as die basis for purchasing and planning. This type of epidemiological approach has been advocated for several years, the healdi programme initiatives by the WHO'"* are one example and much of the development of epidemiology for healdi services manage-

3 EUROPEAN JOURNAL OF PUBLIC HEALTH VOL NO. 2 ment has been undertaken with this intention. These initiatives have had relatively little success in the past for a number of reasons. i) The exercises undertaken have been piecemeal and it has been difficult to obtain the resources to undertake the comprehensive work required across the whole spectrum of disease and health services to provide truly integrated service planning. However, incentives to handle these problems are developing in many European countries where recent changes in the organization of health care are leading to the creation of internal markets. At the same time the development of information systems is starting to provide answers to these problems and make it possible to plan work to establish a comprehensive grouping of conditions and care packages, ii) It has not been clear that there is a need to undertake the exercise in all disease areas. Conditions which present as emergencies will generally receive appropriate care and activity information is sufficient to plan services. Conditions treated on an elective basis have a greater need for this approach and, whilst very important in some areas, do not represent a major part of all clinical activity. On the odier hand the opportunities and demands of the introduction of internal markets, whether funded by central government, a sickness fund or private insurance, together with the separation of the purchasing from the providing function, means that purchasers need to be able to view the needs of their population separately from the activities that are provided. The emphasis is now on being able to establish what care should be provided and negotiating witli providers on the most effective and economical way of delivering that care. This kind of dialogue is difficult to pursue just using activity data. A view of all the needs of the community is necessary in order to purchase effectively and equitably and this requires a way of integrating all epidemiological information, iii) The epidemiological information widi which to support the development and use of the model is only sparsely available. The cost of collecting it has been perceived to be greater than its value in planning and monitoring services. In addition, for it to be useful over time, die information needs to be collected continuously so diat changes in the epidemiology can be monitored. Obtaining this epidemiological information is still difficult, because most health service information is captured as a function of the care of patients. It is also difficult to link die activities provided to a single individual during different episodes or by different providers. Some of diis will change widi the introduction of person-based information systems which use a single unique person identifier and the possibility of using primary care computer systems for capturing epidemiological information is being actively explored. If technically and organizationally feasible, this could revolutionize the capture of 'quasi-epidemiologicap information. iv) Prescription of die care required for identified conditions can be seen to intrude upon decision making by clinicians and die agreement of appropriate packages of care which are supported by professional associations can be very difficult to obtain. The degree to which specification of die appropriate patterns of care for particular conditions limits the ability of clinicians to exercise professional judgement is still uncertain. However, it needs to be emphasized diat some patterns of care for particular conditions are widely supported (odierwise there would be no medical textbooks) and diat definitions of appropriate care on a population basis are based on probabilities, radier dian on prescriptions about individuals, v) A terminology for describing bodi conditions and care packages has not been available, which has made it difficult to integrate the efforts of different initiatives, each of which may have covered complementary areas, but set out in ways which make it difficult to match die classification of conditions and care processes together. The last issue relates to die terminology for describing conditions and care packages. There are a number of approaches to tliis fundamental issue of capturing clinical language. The Systematised Nomenclature of Medicine (SNOMED) 15 has been expanded to cover most terms and concepts widiin clinical medicine and the Clinical Terms Projects in the.uk 1 " have similarly devised terms, which cover not only medical but nursing and paramedical conditions of activity. Similar activities within Europe include die GALEN (Generalised Architecture for Languages, Encyclopaedias and Nomenclature in Medicine) project, developing ways of processing natural language to capture die information content of medical records and The Good European Health Record (GEHR) 18 which seeks to identify appropriate structures for recording and understanding information. The pace of development is bodi rapid and frustrating. Criticisms of Version 2 of Read Coded Clinical Terms 19 have largely been met by the development of Version 3, but die implementation of die newer version will take considerable time to spread. Thus, die wide application of diese terms will take some time. However, their existence makes it possible to start work on defining groups which will be widely understandable and transferable. DEVELOPING CONDITION/TREATMENT GROUPINGS If this strategy for developing groupings to assist in health care planning/ purchasing is to succeed, dien a number of issues need to be considered. Numbers of groups In order to be useful, any kind of grouping system needs to strike a balance between minimizing die number of groups at die same time as maximizing die homogeneity of cases widiin die groups. These 2 goals are generally contradictory, die fewer die groups, die more likely it is diat different kinds of cases will be within any particular group. The greater die homogeneity wiriiin a group, the more groups that are required. It is difficult to predict what diis means in terms of condition and treatment groups for purchasing, but, a priori, discussions between purchaser/payer and provider of any contract widi more dian 50 condition treatment

4 Development ofpaaent groupings combinations would be unmanageable and it would, perhaps, be better to restrict the numbers to not more than 20. Assuming that the contract discussions will take place (at least in the future) on a specialty and/or service basis, this would imply something between 500 and 1,000 group combinations from the service/specialty/types which may be available. This estimate of the numbers of groups for contract negotiations does not preclude further subdivision of the groups where this is necessary for more detailed clinical discussion or in the examination of the outcomes of care for a particular condition/care combination. Variables involved in the definition of groups Since the purpose of the exercise is to enable purchasers/payers and providers to debate the numbers of patients and levels of care activity, the condition groups themselves need to reflect the appropriate type of care. They may therefore need to be based not only upon the presenting complaint/diagnosis, but also other relevant characteristics of the patient (for instance, their general fitness, important co-morbidities, social circumstances or attitudes to care). These variables are not likely to be collected as parts of a minimum data set for some time; however, the case for expansion of the minimum data set might be driven by the usefulness of this method of grouping. Validation of groupings The construction of condition groups which map to treatment groups will be based largely upon clinical judgement and consensus views, both from the literature and working groups. Validation of these group definitions can only come from peer review although some confirmation could be obtained by comparison with existing patterns of care. However, this will only provide confirmation that the groups of patients with similar needs are reasonable, rather than the best combinations of condition and treatment that can be defined. The groups will, however, also be used to predict the expected outcome of care and for this purpose can be validated against real data. The increasing interest in 'evidence-based practice' will provide further support for the development of these groups. Defining the condition axis In order to systematize the approach, condition categories need to be defined which can be broken down into a number of groups, based upon the appropriate package of care. These groups are likely to be related to the severity or stage of progression of the condition. For the purpose of defining these groups, both presenting complaints and diagnoses will need to be incorporated (unless separated into different matrices) but this does imply that when used with real data, care will have to be taken to prevent double counting of cases which can be considered as having both condition and diagnosis (for instance mobility problems and osteoarthritis of the hip). Potential groupings can be derived from a number of sources. There are already diagnosis-based groups in HRGs and DRGs which can be developed and mapped to the appropriate care package; equally, the procedurebased groups in HRGs and DRGs can be examined to determine the conditions for which these care processes are appropriate. There are also published clinical practice guidelines, 21 in a few instances recommended protocols of care 22 and a substantial amount of work done at a local level in the UK in developing clinical profiles as part of the resource management initiative. 2 -' Alternative patient groupings have sought to define the condition axis by concentrating on the patient [as in disease staging" and computerized severity index (CS1) ] and some, such as patient management categories (and patient management pathways) 2 ** have sought to explicitly define conditions and the treatment packages associated with them. In some of these, definitions are based upon the ICD-9-CM, but in others clinical findings and results are used to define the groupings. These approaches go a considerable way to constructing useful patient groupings, but are mainly based on diagnostic information and so are less useful for the broader spectrum of primary, community and continuing care in which symptoms, disabilities and impairments may be more important statements of the patient's condition and need for services. In addition, they are less able to deal with the subdivision into preventive, diagnostic, treatment and follow-up matrices which may be required. Defining the care axis As previously noted, DRGs (and their analogues) implicitly define items of treatment, but there is considerable bundling of the individual components of these packages. Whilst this is feasible for acute in-patient stays in which a reasonably standard set of care processes may be expected, extension of the concept to community and nonacute in-patient care is much less straightforward. Even in acute in-patient care, the prediction of resource use is much better where the grouping is based on a procedure, because this drives the major resource component of theatre time and in-patient days. For medical admissions, the diagnosis per se is less likely to have a consistent relationship with the packages of care consumed, because of the individual variations of patients within a single diagnosis. In treatments spanning more than one admission or where the admission is only part of a multistage process of care, the complexity of defining a single typical resource package is much greater. They may be based on multiple admissions, out-patient attendances, community nursing visits, etc. and all be a coherent part of a single treatment plan. To address this problem a better definition of episode of care types and better ways of capturing information are required so that discrete episodes can be linked together to form useful care packages. The other key issue is the definition of packages of care for non-surgical patients. There is no adequate classification for this purpose at present. If we are to have treatment groupings, we will need to define and classify the components of care with which to build them.

5 EUROPEAN JOURNAL OF PUBLIC HEALTH VOL NO. 2 CONCLUSIONS Up to now our ability to explore different ways of grouping patients and treatments has been limited, because the only data readily available are the diagnosis on discharge and the surgical procedures carried out. This continues to be the case at present, but the logic of developing information strategies (as identified in the Information Management and Technology Strategy of the NHS in England 27 ) points to the gradual adoption of electronic medical records. Such systems should permit the capture of information for statistics as a by-product of the operational systems supporting the process of medical care although there may need to be greater attention to the quality of clinical records before they can be used reliably for statistics. 28 If that can be overcome it will then be possible to create precise groups of patients whose detailed resource requirements may be reasonably accurately predicted. We are therefore on the threshold of a new generation of ways of grouping patient records, made possible by changes in the availability of data and made necessary by the changes in the delivery of care. The development is certain, the time scale is not, but we can be sure that the task of defining condition and treatment groups will be an important way to use clinically rich data to manage health services more sensibly and appropriately. Our thanks are due to all the staff of the NCMO and other colleagues who have helped to develop these concepts through discussion over many months. Thanks are also due to Lesley Morris for patient editing of the various drafts. 1 World Health Organization international classification of diseases. 9th revision. Geneva: WHO, Fetter RB, Shin Y, Freeman JL, et al. Case mix definition by diagnosis related groups. Med Care 1980;18(SuppQ: McGuire TE. DRG evolution. In: Casas M, Wiley MM, editors. Diagnosis related groups in Europe, uses and perspectives. Berlin: Springer-Verlag, 1993: McGuire TE. DRGs: labels, information and uncertainty. Austr Hlth Bull 1993;16(2): Pi I la J, Hindle D. Adapting DRGs, the British, Canadian and Australian experiences. Austr J Hlth Informat Manage 1994,24: Frutiger P, Fessler JM. Le programme de medicalisation des systemes d'information a I'Assistance Publique Hopitaux de Paris (PMSI-AP) (The medicalization programme for information systems in the Public Assistance Hospitals of Paris). Soz Praevenfrvmed 1989,34: Sanderson HF, Anthony P, Mountney LM. Healthcare resource groups. Version 2. J Public Hlth Med 1995;17(3): Fries BE, Schneider DP, Foley WJ, et al. Refining a casemix measure for nursing homes: resource utilization groups (RUG-lll). Med Care 1994,32: Schneider KC, Lichtenstein JL, Fetter RB, et al. The new ICD-9-CM ambulatory visit groups classification scheme: definitions manual. New Haven, CT: Yale University, Averill RF, Goldfield Nl, McGuire TE, et al. Ambulatory patient groups: definitions manual. Wallingford, CT: 3M Health Information Systems, Bardsley M, Coles J, Jenkins L DRGs and health care: the management of case mix London: Kings Fund, Hornbrook MC. Hospital case mix: its definition, measurement and use. Med Care Rev 1983;39: Eskin F, Bull A. Squaring a difficult circle. Hlth Serv J 1991;101(5233): WHO Regional Office for Europe. Guidelines for health care practice in relation to cost-effectiveness: report on a WHO Workshop. EURO Reports and Studies 53. Copenhagen: WHO Regional Office for Europe, Cote RA, editor. Systematized nomenclature of medicine SNOMED International. Northfield, IL College of American Pathologists, Chisholm J. The Read Clinical Classification. BMJ 1990;(300): Rector AL, Nowlan WA, Glowinski A. Goals for concept representation in the GALEN project In: Safran C, editor. Proceedings of SCAMC '93. New York: McGraw-Hill, 1993: Griffith SM, Kalra D, Lloyd DSL, et al. A portable communicative architecture for electronic healthcare records: the Good European Healthcare Record project In: Gre«ves RA, etal., editors. Medinfo 95 Proceedings of the 8th World Congress on Medical Informatics; 1995 July 23-27; Vancouver, Canada. Alberta, Canada: Healthcare Computing and Communications, Inc. 1995: Smith N, Wilson A, WeekesT. Use of codes in development of a standard data set. BMJ 1995;(311): Payne C. Version 3 of Read Codes addresses many difficulties [letter]. BMJ 1995:311: Department of Health. Implementing clinical practice guidelines: can guidelines be used to improve clinical practice. Effect Hlth Care 1994;8: Department of Health Standing Medical Advisory Committee. Management of lung cancer: current clinical practices. Report of a Working Group. London: Department of Hearth, NHS Management Executive. The clinical profiles of care-workshop proceedings. Cheadle, UK: British Association of Medical Managers, Gonella JS, Hornbrook MC, Louis DZ. Staging of disease: a case mix measurement. JAMA 1984,241: Horn S. Measuring severity of illness: comparisons across institutions. Am J Public Hlth 1983;73: Young WW, Swinkola RB, Zorn DM. The measurement of hospital case mix. Med Care 1982;20: Information Management Group, National Health Service Management Executive, Department of Health. Getting better with information: IMST strategy overview London: NHSME, Boydell L, Grandidier H, Rafferty C, McAteer C. Reilry P. General practice data retrieval: the Northern Ireland project J Epidemiol Commun Hlth 1995;49:(Suppl 1):22-5. Received 2 June 1995, accepted 20 February / 996