Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Cost estimates for hospital inpatient care in Australia: evaluation of alternative sources

Cost estimates for hospital inpatient care in Australia: evaluation of alternative sources Hospital Services Research Group, Monash University Health Economics Unit introduction he growing use of economic evaluation of health care interventions has led to increased attention to the measurement of outcomes of care, but relatively less attention to the measurement of cost. Standard works on costing of health care emphasise the economic concepts underlying cost r n e a ~ u r e m e n t ' ~ ~ ~ ~ are but pragmatic in their acceptance of the poor quality of available cost data, and offer little study design advice on the selection of costing methods to optimise data quality. This paper proposes five criteria for evaluating the quality of cost data, and reviews three broadly-defined approaches to cost estimation against these criteria. The relative importance of the different criteria obviously varies with the decision context and with the magnitude and dispersion in the costs of components of care. Evaluation against these criteria should inform research projects using cost estimates at two stages: in the design of cost measurement efforts, and in the performance of sensitivity analyses. A review of Australian secondary sources of cost data is presented, including a description of how cost estimates are derived in two cornputerised approaches which rely on accounting data. Documentation of the validity and reliability of cost estimates is a minimal expectation of research reports which use cost data, and this paper seeks to improve the quality of such reports in Australia. THE DECISION CONTEXT Cost information is used for a variety of purposes: economic evaluation of health care interventions, documenting the costs of a particular condition or illness, comparing the efficiency of one hospital with another, and in some states, as the basis of relative resource weights for payment. Almost inevitably, the need to understand the costs of hospital care arises in the context of a practical decision at the clinical or policy level, and this decision context shapes the kinds of costs and the approaches to cost estimation which are most relevant. Designing a study with a costing component entails an optimisation exercise. Relevant questions are: How 'big' are the hospital costs as a proportion of the total costs under consideration? How sensitive are findings likely to be to imprecision in the cost estimates? How fine-grained a picture of patient groups and component costs is required to answer the research question? How important is it that particular cost types (capital equipment costs, for example, or start-up/once only costs) be included or excluded from the analysis? How much is it worth investing in better quality cost data? On the latter point, Young4 observes that decision makers "must balance the benefit of improved information with the additional cost and complexity of obtaining it." [p. 711. Shepard' suggests that researchers evaluate these questions at the research design phase in the form of a 'back of the envelope' marginal analysis of the costs of additional data acquisition. Sensitivity analysis to test whether Submitted: December 1999 Revision requested: February 2000 Accepted: April 2000 Correspondence to: Dr Terri Jackson, Hospital Services Research Group, Monash University Health, Economics Unit, P 0 Box 477, W. Heidelberg VIC 3081 Fax: (03) 9496 4424. Email: Terri.Jackson@buseco.monash.edu.au AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2000 VOL. 24 NO. 3 Cost estimates for hospital inpatient care in Australia and how much study conclusions would change with variations in the estimates of costs should also be undertaken in order to demonstrate that decisions in the design phase resulted in optimisation of data acquisition effort.6 Approaches to estimation of hospital costs There are two aspects to cost estimation: counting the number of services each patient receives and placing a dollar value per unit of service. Although 'cost' is often treated as an observable fact, it is by nature a construct. The problem of 'joint costs' (resources which are used to support a number of simultaneous patient care activities) requires the use of simplifying assumptions.' This is generally acknowledged for the overhead costs of a hospital (administration, quality assurance, etc), but also characterises the costing of direct patient care. Decision rules are needed to allocate, for example, shares of a nurse's salary amongst the patients in the recovery suite of an operating theatre. There are three basic ways of estimating the costs of hospital services and, when combined, many variants. MICROCOSTING and, while the classification aims to create resourcehomogeneous categories, the Australian DRG classes have been defined solely on the basis of length of stay data." Various researchers have costed the treatments of patients who fall into these LOS-defined groups and this gives rise to different sets of DRG costs or, when standardised,cost weights. The protocol approach may be combined with information from specific patient groups, with time and motion recording to value specific services, or with recorded utilisation of specific services" to yield closer approximations of per-patient cost. HOSPITAL ACCOUNTING DATA The first approach is primary patient-specific data collection or 'microcosting'. Because labour costs form such a large proportion of total costs of care, this generally entails recording of staff time used in performing various health care tasks (thus termed 'time and motion'), and enumeration of the materials needed to perform those tasks. This is the most precise and detailed approach to costing' and also the most costly. Data may be collected by dedicated research staff or, more commonly, by care-providers recording their activities on survey forms. As with any survey data, estimates may be flawed by low response rate or incomplete recording of information, and the collection effort may entail delays affecting the timeliness of estimates. COSTING TO A CLINICAL PROTOCOL These published sources rely on the third basic method of cost estimation: secondary analysis of hospital cost accounting system data. Historically, hospital costs were estimated from financial accounting data whose primary purpose was to ensure accountability for funds to charitable donors and later government. In the last three decades, two developments in hospital accounting have radically changed the nature and amount of data available: the growth of cost accounting (as distinguished from financial accounting)I3 and the development of better product descriptors (such as DRGs).14 These have improved the quality of accounting data and led to the development of 'top down' product costing and 'bottom up' patient costing. Problems remain in the costing of products which are not directly related to patient care (such as teaching and research) or where products are ill-defined (hospital health promotion, for example). When published sources of cost estimates rely on hospital accounting data, their quality will, in part, reflect the strengths and weaknesses of the estimation method. Derivation of accounting system estimates Estimates using financial accounting data have been derided by many economists because accounting conventions frequently distort which costs are included in financial account^,'^ and such costs often bear little relationship to the use of resources in patient care.'. I('. PER DIEM COSTS A second approach is to use a clinical protocol to define the set of services used, with standard costs (the cost of a lab test or a day of hospitalisation, for example) to value each ~ervice.~.'" differs from primary data collection in that it It is not specific to a particular patient, and measures of cost will not normally reflect variability between patients for any given protocol. When protocols require only estimates of the total hospital episode cost, they may rely on published sources of cost estimates. The longest-standing of these is the use of published hospital expenditure data calculated as per diem estimates. In the absence of better cost data, total hospital expenditures are divided by the number of occupied bed days to yield a per diem estimate of hospital costs. These costs are then attributed to a patient or group of patients on the basis of recorded length of stay (LOS). Published sources have also been developed in support of casemix funding or other comparative efficiency policies, and thus they are often referred to as 'DRG costs'. This is problematic because it confuses the DRG classification system with the costing system. DRGs have no inherent cost Both economic theory and empirical studies support the observation that length of stay (LOS) is a poor proxy for hospital costs, particularly in the context of changing length of stay and intensity of treatment per day of stay. Many of the resources formerly provided over a 6-7 day stay are now concentrated in shortened 3-4 day stays. With the increased availability of high cost technologies in operating suites and intensive care units, and expensive investigations such as CT and MRI, LOS becomes a relatively poor predictor of cost differences between patients.l 2 Patient diagnosis has an important effect on the use of resources which per diem estimates generally do not reflect." Donaldson' suggests refining the per diem method by attributing hospital costs down to the specialty or ward level and then allocating these total costs per patient day to partially take account of casemix. Published DRG cost estimates can also be used to calculate more casemix- 2000 VOL. 24 NO. 3 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Jackson sensitive per diem costs. Jacobs & Bachynsky'' have identified a third refinement, the 'Hotel and Ancillaries' (H & A) approach, where costs incurred on the ward are allocated per diem, but other specific services (termed 'ancillaries') are counted at the patient level. A unit cost per ancillary is assigned, and the product of a patient's ancillary costs is added to theirper diem hotel costs to yield a total cost at the patient level. This improves the precision of per diem costing, but requires considerable primary data collection. Crude forms of clinical or patient costing (considered below) combine Donaldson's specialty refinement for medical costs with the H & A method for ward nursing and other cost types. COST MODELLING across hospitals rather than across patients. Although standard error figures are published, it is not possible to estimate how widely variable the patient sample might be. The Commonwealth Casemix Development Program made large investments to disseminate the cost-modelling approach, supporting the development of national DRG cost weights and service weights, and facilitating adoption by individual hospitals. For researchers seeking cost data, this means that large and nationally-representative hospital samples are relatively easy to define. CLINICAL OR PATIENT COSTING Cost modelling begins with a hospital's total expenditures and allocates them to various 'products' of the hospital (treatments or DRG classes) through a cascading series of allocations. This is referred to as a 'top down' allocation of costs, and has been described in detail by a number of authors.20~2'~22~23~24,25crudest forms, In its modelling divides a hospital's total expenditures by the number of weighted patients (where DRG weights derived from an external source are multiplied by the number of the hospital's patients in each class). More commonly, allocations are first made to inpatient and outpatient activities (the so-called IFRAC or inpatient fraction estimation), then to 'intermediate' departments (radiology, theatre, etc.) and then to the DRG using 'service weights,' externally-derived sets of relative values which estimate the DRG 'shares' of radiology expenditures to be attributed to each DRG. The precision of the estimates from a cost model relies on the extent to which the model is related to actual resource utilisation in the institution being costed, both the IFRAC estimation and the DRG service relativities. In the early days of cost modelling in Australia, the service weights used to distribute intermediate product costs (eg, radiology) were based on relativities in charges for hospital care from the US State of Maryland. Although this data source was one of the most accurate in terms of measured utilisation for its time:' cost relativities inevitably reflected practice patterns from the US. Subsequent work has been sponsored to define specific Australian service weights based on studies of DRG relativities in the use of intermediate products in samples of Australian hospitals.26 Differences between patients are averaged-out at the DRG level, and differences in use of intermediate products of care (x-ray, ICU days, etc.) reflect the service weight used rather than actual patterns of care. Thus differences between hospitals show up only in the total dollars (expenditures) allocated by these statistics rather than by differences in patterns of care. When data from a number of hospitals are combined, cost modelling estimates can be reported with a standard error of the mean cost per DRG, but this reports the dispersion of hospital-level rather than patient-level cost, because the DRG mean is calculated Clinical costing uses what is known as a 'bottom up' approach to costing, whereby the cost per patient is built up from recorded utilisation of each intermediate prodUCt.27,28,29,30 Relative values or weights may be used in the estimation of the unit cost of these intermediate products (as in cost modelling) but these are modelled not on relativities amongst DRGs but on the products themselves. For example, a pathology department relative value scale reflects the relative resource intensity of the individual pathologytests rather than DRG-sharesof the pathology budget. Overheads are assigned to intermediate departments usually after first having been allocated to each other (using simultaneous equations to record the proportion of payroll office costs to be allocated to the housekeeping department and the proportion of housekeeping department costs to be allocated to the payroll office). Records of the use of tests for a specific period are weighted by the relative value scale being used, and a standard cost per test is estimated by dividing total departmental expenditure by the number of weighted tests. In order to do patient-level costing, most intermediate departments must be equipped with a means of recording individual patient utilisation, usually in computerised form (so-called 'feeder' systems). Examples are automated testordering systems, workflow tracking systems, nursing patient dependency systems and theatre recording systems. To estimate the costs of individual patient episodes, utilisation data on each patient is tracked to the admission through the use of data matching on medical record or episode number, and costed using the standard costs. This 'bottom up' approach preserves information about the variability inherent in individualised medical treatment, while making possible estimates of average resource use at clinician, service and hospital levels. Clinical costing requires the investment of additional effort in precise measurement, and results in ''cost data . .. as accurate as the current state of technology allows'' 31 [p. 4771. Both patientlevel and DRG-level costs may be estimated, but in clinical costing, the DRG-level cost always derives from the average of individual patients, whereas in cost modelling, the reverse is true, with the DRG average cost extrapolated to individual patients. A clinical costing system can report the range and standard deviation of the mean cost in a DRG; a cost modelling system cannot. Clinical activities are not all costed with the same level AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2000 VOL. 24 NO. 3 Cost estimates for hospital inpatient care in Australia of precision even within a single hospital's clinical costing system, and systems vary markedly between hospitals. Medical costs are typically poorly measured and often assigned at the specialty level on a length-of-stay basis because there is no easy way to routinely record differences between patients in the amount of sessional or salaried doctor time devoted to their care. The IFRAC problem is obviated by the use of payroll data which distinguishes inpatient and outpatient sessions for registrars and sessional consultants. Pathology and other requests are identified to individual admitted or non-admitted patients, providing an empirical basis for the allocation of costs for these departments. When the research question is sensitive to differences in the amount of medical time required, few clinical costing systems will pick up such differences. Nursing costs, however, may be measured using either simple LOS allocation, or more sophisticated nursing dependency and skill-mix allocation^.^^ Detailed knowledge about the basis for allocation of important or sensitive intermediate services is thus required in order to judge the degree of precision available. The sophistication of the clinical costing approach, however, comes at a higher cost. Where they have not done so, hospitals must make the initial investment in information technology to track individual utilisation of services. In addition to hardware and software investments, there are significant 'set up' costs in implementing and documenting such systems, and ongoing costs for the salaries of suitably qualified information technology staff to maintain them. 'resolution,' used here as in microscopy, that is, how finely detailed a picture can you get from the data? In terms of patients, are they aggregated and costed at the DRG-level, the ICD-9/10 diagnosis code level, or at the level of identified patients? In terms of dollars assigned to patients, is this done on the basis of broad averages or more finelydistinguished services? The fewer the categories in a classification, the more likely it is that dissimilar patients with dissimilar costs will be grouped together. There are some 10,000 ICD-9 codes, but only 667 DRGs. Some DRGs are more a 'hotch potch' of clinical codes than others: the rarer a condition, and the more typical its average LOS, the more likely it will be included with other conditions in the same body system, rather than forming its own DRG. High volume treatments, and very high cost treatments (even when rare) are more likely to have their own DRG, eg, multiple organ transplants. Clinical costing has shown that the dispersion around the mean cost varies greatly by DRG,35with highly variable DRGs most likely combining different clinical subgroups. The way DRG categories have been defined may mean that they are at an inappropriate resolution for some research questions, even when the 'swings and roundabouts' justification for DRG variations may apply in payment policy. Many DRGs separate similar patient groups on the basis of whether or not patients undergo an operating room procedure; others result in classification and costing categories for dissimilar patients, for example, those patients who require tracheostomy, regardless of original diagnosis. Cost identification may also be at an inappropriate resolution. Whynnes & Walker l 2 used patient-level costs for treatment of colorectal cancer to compare the simulated results of 'crude' costing approaches with results from their original, more detailed data. They found that the crude approach understated costs by more than 10% for some patient subgroups, and overstated costs by more than 13% for others, while the mean for all patients differed by only 1.2%. Obviously patient costs will vary around the mean, but without the capacity to identify patient characteristics associated with those variations, findings may be extrapolated to inappropriate patients or settings. TIMELINESS Criteria for evaluation of cost estimates All approaches to hospital cost estimation entail trading off desirable characteristics to optimise the estimate for the decision being made. I am proposing five criteria for evaluating cost data: precision of the estimate, 'resolution' of the estimate, timeliness, generalisability, and affordability of data acquisition. PRECISION OF THE ESTIMATE In some decision contexts, an estimate of hospital costs can be very imprecise without seriously affecting the decision. But the more precisely costs are allocated, the smaller the difference in efficiency which can be detected between two institutions, for example, or between the cost effectiveness of two therapies. Methods which measure utilisation of services at the patient level, provide estimates of cost which vary both with total hospital expenditure for each service department and with different patterns of patient care. This enables clinical and other decision-makers to understand whether differences in treatment costs arise from variations in unit cost or from variation in service intensity,33 and to understand the distributional form of the cost data on which estimates are based.34 RESOLUTION Closely related to precision of the cost estimate is the 2000 VOL. 24 NO. 3 Patterns of hospital and medical care are constantly changing: new tests and treatments are introduced, substitute sites of care (such as outpatient surgery) replace inpatient care and thus change the severity mix of the patients treated in hospital, average lengths of stay have been declining over the past decade or more. Estimates of cost can be adjusted readily for differences in the inflation rate from one year to the next, but it is not feasible to make such adjustments for complex changes in patterns of care. Thus, the timeliness of costing source data is important in many decision contexts. The paradox in applying this criterion, however, is that the more detailed and precise the data, the longer it takes to assemble, and thus, the greater the risk that the estimate will not reflect current practice. AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Jackson ~~ ~ ~ TABLE 1:OVERVIEW OF STRENGTHS AND LlMlTATiONS O f APPROACHES TO THE ESTIMATION OF AUSTRALIAN HOSPITAL COSTS Precision of estimate' Primary data collection (microcosting) Protocol costing using published sources Bedlday cost estimates National DRG Cost Weights Victorian DRG Cost Weights Hospital accounting data Bedlday cost estimates Cost-modelling estimates Clinical costing estimates Affordability of data collection" Resolutionb Timeliness" Generalisabilityd _-++ --- - 1 +' ++ ++ - 1 +' ++ ++ ++ ++ - 1 +' -++ __++ - 1 +' ++ - / +' - 1 +' - 1 +' Notes: a How close to the hospital's true resource use does the cost estimate come? b At what level of detail are patient groups and cost components identified? c How old or outdated are the estimates? How likely is it that the estimate reflects contemporary clinical practice? d How representative are the costdpractice pafferndpatients in the hospitals providing cost estimates? e 'Affordability' refers to low cost of data acquisition f Depends on decision context; see text for fuller description GENERALISABILITY A fourth criterion for evaluating sources of cost data is how readily estimates can be extrapolated to the relevant decision context. Costs estimated in a single hospital may not be representative of the costs of the same service in the wider universe of hospitals." Patients in particular kinds of hospitals may be different from those in others, even after casemix adjustment. A particular problem of external validity arises from the 'trimming' of data sets, particularly those used for cost weight estimation. Conventionally, 'outlier' cases are defined to distinguish cases with abnormally long lengths of stay or abnormally high cost. Cost weights are then estimated using the remaining 'inlier' cases, which represent a more homogeneous group for payment policy but exclude extreme cases which may be relevant to the decision at hand. Trimming may be done on statistical criteria such as multiples of the standard deviation from the mean or on length of stay measures such as the 'L3H3' approach used in Victoria and elsewhere.36 Applying this criterion entails another trade-off of desirable characteristics. The larger the research effort in terms of the sample of hospitals, the more costly and less timely estimates are likely to be. Focussed data collections (limited to the sample of hospitals or patients of interest) may require special surveys rather than use of secondary or published data. DATA AFFORDABILITY motion studies are the most costly to undertake and would be wasteful to use in answering many economic questions in health care. But use of low cost 'off the shelf estimates may also be ultimately costly, if the validity or reliability of the estimate are such that they distort research findings. Australian sources of hospital cost estimates Table 1 provides an overview of approaches to the estimation of hospital costs in Australia with rankings of these against the proposed criteria for evaluation. For some cells of the table, the positive or negative rankings will strongly depend on the decision context. PRIMARY DATA COLLECTION (MICROCOSTING) When the importance or sensitivity of the research questions warrants it, primary time and motion data collection provides the most precise and fine-grained estimates of the costs of care for an identified group of patients. Because of the cost of this form of data collection (either for paid recorders, or the opportunity cost of professional health care staff recording utilisation data, such efforts may be limited to a smaller than desirable sample of institutional settings. They are also vulnerable to non-response and delays which may affect data timeliness. PROTOCOL COSTING USING PUBLISHED ESTIMATES Researchers must continually balance the costs of answering research questions against the importance or magnitude of the question itself.4'5 * 37 Epidemiologists have placed increasing emphasis in recent years on the use of power calculations in determining the appropriate sample size to answer research questions: too large a sample wastes research effort and resources, too small a sample is wasteful because it may well yield an equivocal result. Time and Protocol costing entails varying levels of direct data collection on the utilisation of health care services with published cost data. These have the general advantage that they are a very low-cost method of deriving cost estimates, but may be available only at high levels of aggregation (the DRG, for example) and thus offer relatively poor resolution. Drummond 38 has observed that while the use of such 'standard costs' may be useful in some circumstances, they are only ' I . . .satisfactory so long as there are not systematic differences between the various standard cost figures and true costs. For example, if the standard costs were 2000 VOL. 24 NO.3 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Cost estimates for hospital inpatient care in Australia overestimates for certain hospital items, but were underestimates for certain physician services, then biases in comparison between technologiescould be introduced" [p. 13I]. Per diem costs can be estimated from a range of readily accessible financial accounting sources including statutory hospital annual reports, various State health authority reports, and reports from the Australian Institute of Health and Welfare.39They provide the roughest estimate of the costs of inpatient care (lacking both precision and resolution), because patients are assigned the same daily cost, regardless of diagnosis, severity of illness or patterns of care, unless DRG-specific costs and length of stay data are used. Financial data are generally published each year, which ensures that estimates reflect any changes in input prices, but the larger the collection, the more likely it is that there will be publication lags. Estimates published as the National Cost Weights derive from varying-sized national samples of hospitals using cost modelling estimation techniques. The original study was undertaken in 1993 in a large national sample of public and private hospitals.'6 The Commonwealth Department of Health and Family Services subsequently sponsored service weight studies to refine ('Australianise') the modelling of the use of five intermediate services (nursing, intensive care, theatre, imaging and pathology). The 'national weights' are the ones mandated by the Pharmaceutical Benefits Advisory Committee for use in protocol-based economic evaluations of new pharmace~ticals.~~ Mechanisms have been set in place to establish an ongoing national cost data collection4'with plans to include up to 200 sites across the country for publishing annual cost estimates. In the interim, the National Hospital Morbidity (Casemix) Database is used to derive annual national cost estimates using hospital expenditures allocated to DRGs on the basis of the revised 1995 National Cost Weights.42 These estimates thus reflect current expenditure and DRG volume patterns (1997/8 data available at June 2000), but because estimates rely on the mid-1990s service weights, the costs associated with intensity of care (ALOS and relative use of inputs to care) reflect earlier practice patterns. Evaluating the timeliness of the data for a particular decision context thus depends on the relative importance of current dollar costs (+) versus underlying practice patterns (-); estimates for DRGs subject to changing clinical practices will be poorer than those where practice is relatively stable. The strengths of this data source are its affordability and its generalisability,with a large national sample of hospitals of various sizes and roles. Tables of component costs by DRG are available, with the caveat that these estimates reflect cost relativities locked in by the once-measured service weights. State-specific estimates are also available, as are estimates for public versus private hospitals. Use of the data should take particular note of inclusions and exclusions in these two sectors, with no medical costs reported for private hospitals and under-reporting of capital costs for public hospitals. Attention has been paid to standardising the overhead allocation statistics across the various samples, enhancing precision of the estimates, although the extent to which this compensates for the imprecision attributable to the IFRAC estimation and cost modelling in general is unclear. The data reflect the weaknesses of top-down allocation methods, with estimates available only at the DRG level and offering no information about inter-patient variability. A second published source of DRG-level cost data is the series of annual Victorian Cost Weight Studie~~~~'~,''.~'.'~~'~ which commenced in 1994 using 1992/93 cost data. Victoria was the first of the State health authorities to invest in patient-level costing systems, and these studies use data from opportunistic samples of public hospitals with fullyimplemented systems. The estimates have the inherent advantages of the precision of patient-level or clinical costing, and annual updating of weights reflects changes both in input prices and in clinical practice. The sample for the collection has increased from 5 hospitals to 18 over the period, with some discontinuity in hospital participation from year to year. Early samples comprised only 6 months of data, aggregated at the DRG level, but from the 1995 study onwards, full year samples have been used, and validation is undertaken using patientlevel data. Generalisability to Victorian inpatient care is very good, with recent studies representing over half of total Victorian separations. The limitation of the sample to larger metropolitan and regional hospitals (which are able to maintain such systems), and any biases attributable to distinctive Victorian practice patterns or cost structures, pose obvious problems in generalising the DRG costs reported. Techniques for data validation at both the hospital level and through preliminary analysis of cases have evolved over the with documentation of hospital level and case level findings published for each study. Beginning with the 1997 study, conformance of hospital costing systems with the Clinical Costing Standards Association of Australia4' standards on allocation of overheads has been assessed and reported. The payment formula under Victoria's casemix funding system generates 'costs' which are expressed as WIES (for Weighted Inlier Equivalent Separation) payment amounts. These may be useful for hospital profitability analyses in Victoria, but should not be used for other economic applications. While the 'weights' used in the calculations are those described above, the formula reflects other factors such as the considerable discounting implicit in recent Victorian budget cuts for public hospitals, and WIES calculations may not incorporate the DRG-based fixed cost component which is also funded under this policy. HOSPITAL ACCOUNTING DATA For many research questions, 'off the shelf cost information will be inadequate. The same three approaches to cost estimation which result in published figures (as discussed above) can also be used from unpublished sources. Bedday estimates again perform poorly against the proposed criteria, having only convenience/affordabilityto argue for their use. 2000 VOL. 24 NO. 3 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Jackson ~~ ~ commonwealth investment in widespread use of cost modelling means that many hospitals have some capacity to generate these estimates. Strengths and limitations of costmodelled estimates from samples of hospitals are similar to those for the National Cost Weight estimates, but with two further caveats: smaller samples of hospitals will limit generalisability, and if estimates from a single State are used, there may be some state-specific factors which impair wider generalisability. When this level of specificity is required for a decision context, small samples (or a single hospital) can provide estimates which accurately reflect their particular expenditure patterns and casemix, though not local practice pattern differences. Data acquisition costs are generally low, although negotiation of permission to use these data may have a time cost for the researcher. With the wider use of casemix funding by States, increasing numbers of hospitals have installed clinical (patient) costing systems. These provide all the benefits of the increased precision of published clinical costing results with potential for better resolution. Data can be interrogated at the 'product' level to document different patterns of patient care (number of nursing shifts with higher nursing dependency, for example, or patterns of use of high-cost diagnostic tests). Patients can be grouped below the DRG level of classification to identify variations in resource use attributable to specific diagnoses. When used in the context of economic evaluations alongside clinical trials, patient cost outcomes can be reported with the same resolution and for the same groups of patients as clinical outcomes. These data must be carefully validated, however, as subsystems vary in the precision and resolution of both costs and patient identifi~ation.~~ decision context If the requires measuring differences in the costs of pharmaceuticals used for two groups of patients, data from a particular hospital's system may not be at sufficient resolution to find such a difference, even if it really exists, Tests of data quality should be undertaken prior to aggregation so that costing errors can be identified. Although improving, documentation of these systems has been poor, and it may be difficult to establish the ways in which overheads (particularly capital equipment) have been allocated. Generalisability of estimates from small samples have both the positive and negative aspects identified for small samples in cost modelling. They are also comparable in terms of data acquisition costs. For studies requiring larger or broader samples, the Victorian Department of Human Services considers applications for use of the Victorian Cost Weights Study databases for specific projects, allowing better resolution (outlier cases, particular diagnoses, particular hospitals) and specificity than is possible using the published DRG-level reports. This data file reports component costs only in 12 'service cost groups' representing intermediate products of care such as imaging, theatre, ICU, and ward nursing, and hospital identity may be encrypted. Conclusion The choice of estimation method entails an optimisation analysis for each decision context. 'Time and motion' survey techniques remains the most valid approach to cost estimation, but are prohibitively costly and this may, in turn, limit the sample of institutions or patients costed. When precision and resolution are important objectives, clinical costing approaches provide the most valid inpatient cost estimates at a reasonable data cost. When external validity is important, use of National Cost Weights-derived estimates may be preferred, and where standardisation of hospital costs is desired (as in the context of economic evaluations of outpatient pharmaceuticals), published cost estimates may be useful. There is little justification for continued use of crude per diem cost estimates. Both primary and secondary sources of cost data must withstand challenges to their internal and external validity. The 'resolution' (or precision) of cost estimates and the relative costs of collection must also be considered, with investigators explicitlyjustifying the use of particular data sources. Acknowledgements An earlier version of this paper was presented in a seminar at the Centre for Health Economics Research and Evaluation, University of Sydney. The author is indebted to the Schneider Institute for Health Policy (The Heller School, Brandeis University) for research support as a Visiting Scholar during the writing of the paper, and to three anonymous reviewers for their helpful suggestions for restructuring the paper. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Australian and New Zealand Journal of Public Health Wiley

Cost estimates for hospital inpatient care in Australia: evaluation of alternative sources

Download PDF
8 pages

Loading next page...
 
/lp/wiley/cost-estimates-for-hospital-inpatient-care-in-australia-evaluation-of-MCCXiTBp4m

References (53)

Publisher
Wiley
Copyright
Copyright © 2000 Wiley Subscription Services, Inc., A Wiley Company
ISSN
1326-0200
eISSN
1753-6405
DOI
10.1111/j.1467-842X.2000.tb01562.x
Publisher site
See Article on Publisher Site

Abstract

Hospital Services Research Group, Monash University Health Economics Unit introduction he growing use of economic evaluation of health care interventions has led to increased attention to the measurement of outcomes of care, but relatively less attention to the measurement of cost. Standard works on costing of health care emphasise the economic concepts underlying cost r n e a ~ u r e m e n t ' ~ ~ ~ ~ are but pragmatic in their acceptance of the poor quality of available cost data, and offer little study design advice on the selection of costing methods to optimise data quality. This paper proposes five criteria for evaluating the quality of cost data, and reviews three broadly-defined approaches to cost estimation against these criteria. The relative importance of the different criteria obviously varies with the decision context and with the magnitude and dispersion in the costs of components of care. Evaluation against these criteria should inform research projects using cost estimates at two stages: in the design of cost measurement efforts, and in the performance of sensitivity analyses. A review of Australian secondary sources of cost data is presented, including a description of how cost estimates are derived in two cornputerised approaches which rely on accounting data. Documentation of the validity and reliability of cost estimates is a minimal expectation of research reports which use cost data, and this paper seeks to improve the quality of such reports in Australia. THE DECISION CONTEXT Cost information is used for a variety of purposes: economic evaluation of health care interventions, documenting the costs of a particular condition or illness, comparing the efficiency of one hospital with another, and in some states, as the basis of relative resource weights for payment. Almost inevitably, the need to understand the costs of hospital care arises in the context of a practical decision at the clinical or policy level, and this decision context shapes the kinds of costs and the approaches to cost estimation which are most relevant. Designing a study with a costing component entails an optimisation exercise. Relevant questions are: How 'big' are the hospital costs as a proportion of the total costs under consideration? How sensitive are findings likely to be to imprecision in the cost estimates? How fine-grained a picture of patient groups and component costs is required to answer the research question? How important is it that particular cost types (capital equipment costs, for example, or start-up/once only costs) be included or excluded from the analysis? How much is it worth investing in better quality cost data? On the latter point, Young4 observes that decision makers "must balance the benefit of improved information with the additional cost and complexity of obtaining it." [p. 711. Shepard' suggests that researchers evaluate these questions at the research design phase in the form of a 'back of the envelope' marginal analysis of the costs of additional data acquisition. Sensitivity analysis to test whether Submitted: December 1999 Revision requested: February 2000 Accepted: April 2000 Correspondence to: Dr Terri Jackson, Hospital Services Research Group, Monash University Health, Economics Unit, P 0 Box 477, W. Heidelberg VIC 3081 Fax: (03) 9496 4424. Email: Terri.Jackson@buseco.monash.edu.au AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2000 VOL. 24 NO. 3 Cost estimates for hospital inpatient care in Australia and how much study conclusions would change with variations in the estimates of costs should also be undertaken in order to demonstrate that decisions in the design phase resulted in optimisation of data acquisition effort.6 Approaches to estimation of hospital costs There are two aspects to cost estimation: counting the number of services each patient receives and placing a dollar value per unit of service. Although 'cost' is often treated as an observable fact, it is by nature a construct. The problem of 'joint costs' (resources which are used to support a number of simultaneous patient care activities) requires the use of simplifying assumptions.' This is generally acknowledged for the overhead costs of a hospital (administration, quality assurance, etc), but also characterises the costing of direct patient care. Decision rules are needed to allocate, for example, shares of a nurse's salary amongst the patients in the recovery suite of an operating theatre. There are three basic ways of estimating the costs of hospital services and, when combined, many variants. MICROCOSTING and, while the classification aims to create resourcehomogeneous categories, the Australian DRG classes have been defined solely on the basis of length of stay data." Various researchers have costed the treatments of patients who fall into these LOS-defined groups and this gives rise to different sets of DRG costs or, when standardised,cost weights. The protocol approach may be combined with information from specific patient groups, with time and motion recording to value specific services, or with recorded utilisation of specific services" to yield closer approximations of per-patient cost. HOSPITAL ACCOUNTING DATA The first approach is primary patient-specific data collection or 'microcosting'. Because labour costs form such a large proportion of total costs of care, this generally entails recording of staff time used in performing various health care tasks (thus termed 'time and motion'), and enumeration of the materials needed to perform those tasks. This is the most precise and detailed approach to costing' and also the most costly. Data may be collected by dedicated research staff or, more commonly, by care-providers recording their activities on survey forms. As with any survey data, estimates may be flawed by low response rate or incomplete recording of information, and the collection effort may entail delays affecting the timeliness of estimates. COSTING TO A CLINICAL PROTOCOL These published sources rely on the third basic method of cost estimation: secondary analysis of hospital cost accounting system data. Historically, hospital costs were estimated from financial accounting data whose primary purpose was to ensure accountability for funds to charitable donors and later government. In the last three decades, two developments in hospital accounting have radically changed the nature and amount of data available: the growth of cost accounting (as distinguished from financial accounting)I3 and the development of better product descriptors (such as DRGs).14 These have improved the quality of accounting data and led to the development of 'top down' product costing and 'bottom up' patient costing. Problems remain in the costing of products which are not directly related to patient care (such as teaching and research) or where products are ill-defined (hospital health promotion, for example). When published sources of cost estimates rely on hospital accounting data, their quality will, in part, reflect the strengths and weaknesses of the estimation method. Derivation of accounting system estimates Estimates using financial accounting data have been derided by many economists because accounting conventions frequently distort which costs are included in financial account^,'^ and such costs often bear little relationship to the use of resources in patient care.'. I('. PER DIEM COSTS A second approach is to use a clinical protocol to define the set of services used, with standard costs (the cost of a lab test or a day of hospitalisation, for example) to value each ~ervice.~.'" differs from primary data collection in that it It is not specific to a particular patient, and measures of cost will not normally reflect variability between patients for any given protocol. When protocols require only estimates of the total hospital episode cost, they may rely on published sources of cost estimates. The longest-standing of these is the use of published hospital expenditure data calculated as per diem estimates. In the absence of better cost data, total hospital expenditures are divided by the number of occupied bed days to yield a per diem estimate of hospital costs. These costs are then attributed to a patient or group of patients on the basis of recorded length of stay (LOS). Published sources have also been developed in support of casemix funding or other comparative efficiency policies, and thus they are often referred to as 'DRG costs'. This is problematic because it confuses the DRG classification system with the costing system. DRGs have no inherent cost Both economic theory and empirical studies support the observation that length of stay (LOS) is a poor proxy for hospital costs, particularly in the context of changing length of stay and intensity of treatment per day of stay. Many of the resources formerly provided over a 6-7 day stay are now concentrated in shortened 3-4 day stays. With the increased availability of high cost technologies in operating suites and intensive care units, and expensive investigations such as CT and MRI, LOS becomes a relatively poor predictor of cost differences between patients.l 2 Patient diagnosis has an important effect on the use of resources which per diem estimates generally do not reflect." Donaldson' suggests refining the per diem method by attributing hospital costs down to the specialty or ward level and then allocating these total costs per patient day to partially take account of casemix. Published DRG cost estimates can also be used to calculate more casemix- 2000 VOL. 24 NO. 3 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Jackson sensitive per diem costs. Jacobs & Bachynsky'' have identified a third refinement, the 'Hotel and Ancillaries' (H & A) approach, where costs incurred on the ward are allocated per diem, but other specific services (termed 'ancillaries') are counted at the patient level. A unit cost per ancillary is assigned, and the product of a patient's ancillary costs is added to theirper diem hotel costs to yield a total cost at the patient level. This improves the precision of per diem costing, but requires considerable primary data collection. Crude forms of clinical or patient costing (considered below) combine Donaldson's specialty refinement for medical costs with the H & A method for ward nursing and other cost types. COST MODELLING across hospitals rather than across patients. Although standard error figures are published, it is not possible to estimate how widely variable the patient sample might be. The Commonwealth Casemix Development Program made large investments to disseminate the cost-modelling approach, supporting the development of national DRG cost weights and service weights, and facilitating adoption by individual hospitals. For researchers seeking cost data, this means that large and nationally-representative hospital samples are relatively easy to define. CLINICAL OR PATIENT COSTING Cost modelling begins with a hospital's total expenditures and allocates them to various 'products' of the hospital (treatments or DRG classes) through a cascading series of allocations. This is referred to as a 'top down' allocation of costs, and has been described in detail by a number of authors.20~2'~22~23~24,25crudest forms, In its modelling divides a hospital's total expenditures by the number of weighted patients (where DRG weights derived from an external source are multiplied by the number of the hospital's patients in each class). More commonly, allocations are first made to inpatient and outpatient activities (the so-called IFRAC or inpatient fraction estimation), then to 'intermediate' departments (radiology, theatre, etc.) and then to the DRG using 'service weights,' externally-derived sets of relative values which estimate the DRG 'shares' of radiology expenditures to be attributed to each DRG. The precision of the estimates from a cost model relies on the extent to which the model is related to actual resource utilisation in the institution being costed, both the IFRAC estimation and the DRG service relativities. In the early days of cost modelling in Australia, the service weights used to distribute intermediate product costs (eg, radiology) were based on relativities in charges for hospital care from the US State of Maryland. Although this data source was one of the most accurate in terms of measured utilisation for its time:' cost relativities inevitably reflected practice patterns from the US. Subsequent work has been sponsored to define specific Australian service weights based on studies of DRG relativities in the use of intermediate products in samples of Australian hospitals.26 Differences between patients are averaged-out at the DRG level, and differences in use of intermediate products of care (x-ray, ICU days, etc.) reflect the service weight used rather than actual patterns of care. Thus differences between hospitals show up only in the total dollars (expenditures) allocated by these statistics rather than by differences in patterns of care. When data from a number of hospitals are combined, cost modelling estimates can be reported with a standard error of the mean cost per DRG, but this reports the dispersion of hospital-level rather than patient-level cost, because the DRG mean is calculated Clinical costing uses what is known as a 'bottom up' approach to costing, whereby the cost per patient is built up from recorded utilisation of each intermediate prodUCt.27,28,29,30 Relative values or weights may be used in the estimation of the unit cost of these intermediate products (as in cost modelling) but these are modelled not on relativities amongst DRGs but on the products themselves. For example, a pathology department relative value scale reflects the relative resource intensity of the individual pathologytests rather than DRG-sharesof the pathology budget. Overheads are assigned to intermediate departments usually after first having been allocated to each other (using simultaneous equations to record the proportion of payroll office costs to be allocated to the housekeeping department and the proportion of housekeeping department costs to be allocated to the payroll office). Records of the use of tests for a specific period are weighted by the relative value scale being used, and a standard cost per test is estimated by dividing total departmental expenditure by the number of weighted tests. In order to do patient-level costing, most intermediate departments must be equipped with a means of recording individual patient utilisation, usually in computerised form (so-called 'feeder' systems). Examples are automated testordering systems, workflow tracking systems, nursing patient dependency systems and theatre recording systems. To estimate the costs of individual patient episodes, utilisation data on each patient is tracked to the admission through the use of data matching on medical record or episode number, and costed using the standard costs. This 'bottom up' approach preserves information about the variability inherent in individualised medical treatment, while making possible estimates of average resource use at clinician, service and hospital levels. Clinical costing requires the investment of additional effort in precise measurement, and results in ''cost data . .. as accurate as the current state of technology allows'' 31 [p. 4771. Both patientlevel and DRG-level costs may be estimated, but in clinical costing, the DRG-level cost always derives from the average of individual patients, whereas in cost modelling, the reverse is true, with the DRG average cost extrapolated to individual patients. A clinical costing system can report the range and standard deviation of the mean cost in a DRG; a cost modelling system cannot. Clinical activities are not all costed with the same level AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH 2000 VOL. 24 NO. 3 Cost estimates for hospital inpatient care in Australia of precision even within a single hospital's clinical costing system, and systems vary markedly between hospitals. Medical costs are typically poorly measured and often assigned at the specialty level on a length-of-stay basis because there is no easy way to routinely record differences between patients in the amount of sessional or salaried doctor time devoted to their care. The IFRAC problem is obviated by the use of payroll data which distinguishes inpatient and outpatient sessions for registrars and sessional consultants. Pathology and other requests are identified to individual admitted or non-admitted patients, providing an empirical basis for the allocation of costs for these departments. When the research question is sensitive to differences in the amount of medical time required, few clinical costing systems will pick up such differences. Nursing costs, however, may be measured using either simple LOS allocation, or more sophisticated nursing dependency and skill-mix allocation^.^^ Detailed knowledge about the basis for allocation of important or sensitive intermediate services is thus required in order to judge the degree of precision available. The sophistication of the clinical costing approach, however, comes at a higher cost. Where they have not done so, hospitals must make the initial investment in information technology to track individual utilisation of services. In addition to hardware and software investments, there are significant 'set up' costs in implementing and documenting such systems, and ongoing costs for the salaries of suitably qualified information technology staff to maintain them. 'resolution,' used here as in microscopy, that is, how finely detailed a picture can you get from the data? In terms of patients, are they aggregated and costed at the DRG-level, the ICD-9/10 diagnosis code level, or at the level of identified patients? In terms of dollars assigned to patients, is this done on the basis of broad averages or more finelydistinguished services? The fewer the categories in a classification, the more likely it is that dissimilar patients with dissimilar costs will be grouped together. There are some 10,000 ICD-9 codes, but only 667 DRGs. Some DRGs are more a 'hotch potch' of clinical codes than others: the rarer a condition, and the more typical its average LOS, the more likely it will be included with other conditions in the same body system, rather than forming its own DRG. High volume treatments, and very high cost treatments (even when rare) are more likely to have their own DRG, eg, multiple organ transplants. Clinical costing has shown that the dispersion around the mean cost varies greatly by DRG,35with highly variable DRGs most likely combining different clinical subgroups. The way DRG categories have been defined may mean that they are at an inappropriate resolution for some research questions, even when the 'swings and roundabouts' justification for DRG variations may apply in payment policy. Many DRGs separate similar patient groups on the basis of whether or not patients undergo an operating room procedure; others result in classification and costing categories for dissimilar patients, for example, those patients who require tracheostomy, regardless of original diagnosis. Cost identification may also be at an inappropriate resolution. Whynnes & Walker l 2 used patient-level costs for treatment of colorectal cancer to compare the simulated results of 'crude' costing approaches with results from their original, more detailed data. They found that the crude approach understated costs by more than 10% for some patient subgroups, and overstated costs by more than 13% for others, while the mean for all patients differed by only 1.2%. Obviously patient costs will vary around the mean, but without the capacity to identify patient characteristics associated with those variations, findings may be extrapolated to inappropriate patients or settings. TIMELINESS Criteria for evaluation of cost estimates All approaches to hospital cost estimation entail trading off desirable characteristics to optimise the estimate for the decision being made. I am proposing five criteria for evaluating cost data: precision of the estimate, 'resolution' of the estimate, timeliness, generalisability, and affordability of data acquisition. PRECISION OF THE ESTIMATE In some decision contexts, an estimate of hospital costs can be very imprecise without seriously affecting the decision. But the more precisely costs are allocated, the smaller the difference in efficiency which can be detected between two institutions, for example, or between the cost effectiveness of two therapies. Methods which measure utilisation of services at the patient level, provide estimates of cost which vary both with total hospital expenditure for each service department and with different patterns of patient care. This enables clinical and other decision-makers to understand whether differences in treatment costs arise from variations in unit cost or from variation in service intensity,33 and to understand the distributional form of the cost data on which estimates are based.34 RESOLUTION Closely related to precision of the cost estimate is the 2000 VOL. 24 NO. 3 Patterns of hospital and medical care are constantly changing: new tests and treatments are introduced, substitute sites of care (such as outpatient surgery) replace inpatient care and thus change the severity mix of the patients treated in hospital, average lengths of stay have been declining over the past decade or more. Estimates of cost can be adjusted readily for differences in the inflation rate from one year to the next, but it is not feasible to make such adjustments for complex changes in patterns of care. Thus, the timeliness of costing source data is important in many decision contexts. The paradox in applying this criterion, however, is that the more detailed and precise the data, the longer it takes to assemble, and thus, the greater the risk that the estimate will not reflect current practice. AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Jackson ~~ ~ ~ TABLE 1:OVERVIEW OF STRENGTHS AND LlMlTATiONS O f APPROACHES TO THE ESTIMATION OF AUSTRALIAN HOSPITAL COSTS Precision of estimate' Primary data collection (microcosting) Protocol costing using published sources Bedlday cost estimates National DRG Cost Weights Victorian DRG Cost Weights Hospital accounting data Bedlday cost estimates Cost-modelling estimates Clinical costing estimates Affordability of data collection" Resolutionb Timeliness" Generalisabilityd _-++ --- - 1 +' ++ ++ - 1 +' ++ ++ ++ ++ - 1 +' -++ __++ - 1 +' ++ - / +' - 1 +' - 1 +' Notes: a How close to the hospital's true resource use does the cost estimate come? b At what level of detail are patient groups and cost components identified? c How old or outdated are the estimates? How likely is it that the estimate reflects contemporary clinical practice? d How representative are the costdpractice pafferndpatients in the hospitals providing cost estimates? e 'Affordability' refers to low cost of data acquisition f Depends on decision context; see text for fuller description GENERALISABILITY A fourth criterion for evaluating sources of cost data is how readily estimates can be extrapolated to the relevant decision context. Costs estimated in a single hospital may not be representative of the costs of the same service in the wider universe of hospitals." Patients in particular kinds of hospitals may be different from those in others, even after casemix adjustment. A particular problem of external validity arises from the 'trimming' of data sets, particularly those used for cost weight estimation. Conventionally, 'outlier' cases are defined to distinguish cases with abnormally long lengths of stay or abnormally high cost. Cost weights are then estimated using the remaining 'inlier' cases, which represent a more homogeneous group for payment policy but exclude extreme cases which may be relevant to the decision at hand. Trimming may be done on statistical criteria such as multiples of the standard deviation from the mean or on length of stay measures such as the 'L3H3' approach used in Victoria and elsewhere.36 Applying this criterion entails another trade-off of desirable characteristics. The larger the research effort in terms of the sample of hospitals, the more costly and less timely estimates are likely to be. Focussed data collections (limited to the sample of hospitals or patients of interest) may require special surveys rather than use of secondary or published data. DATA AFFORDABILITY motion studies are the most costly to undertake and would be wasteful to use in answering many economic questions in health care. But use of low cost 'off the shelf estimates may also be ultimately costly, if the validity or reliability of the estimate are such that they distort research findings. Australian sources of hospital cost estimates Table 1 provides an overview of approaches to the estimation of hospital costs in Australia with rankings of these against the proposed criteria for evaluation. For some cells of the table, the positive or negative rankings will strongly depend on the decision context. PRIMARY DATA COLLECTION (MICROCOSTING) When the importance or sensitivity of the research questions warrants it, primary time and motion data collection provides the most precise and fine-grained estimates of the costs of care for an identified group of patients. Because of the cost of this form of data collection (either for paid recorders, or the opportunity cost of professional health care staff recording utilisation data, such efforts may be limited to a smaller than desirable sample of institutional settings. They are also vulnerable to non-response and delays which may affect data timeliness. PROTOCOL COSTING USING PUBLISHED ESTIMATES Researchers must continually balance the costs of answering research questions against the importance or magnitude of the question itself.4'5 * 37 Epidemiologists have placed increasing emphasis in recent years on the use of power calculations in determining the appropriate sample size to answer research questions: too large a sample wastes research effort and resources, too small a sample is wasteful because it may well yield an equivocal result. Time and Protocol costing entails varying levels of direct data collection on the utilisation of health care services with published cost data. These have the general advantage that they are a very low-cost method of deriving cost estimates, but may be available only at high levels of aggregation (the DRG, for example) and thus offer relatively poor resolution. Drummond 38 has observed that while the use of such 'standard costs' may be useful in some circumstances, they are only ' I . . .satisfactory so long as there are not systematic differences between the various standard cost figures and true costs. For example, if the standard costs were 2000 VOL. 24 NO.3 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Cost estimates for hospital inpatient care in Australia overestimates for certain hospital items, but were underestimates for certain physician services, then biases in comparison between technologiescould be introduced" [p. 13I]. Per diem costs can be estimated from a range of readily accessible financial accounting sources including statutory hospital annual reports, various State health authority reports, and reports from the Australian Institute of Health and Welfare.39They provide the roughest estimate of the costs of inpatient care (lacking both precision and resolution), because patients are assigned the same daily cost, regardless of diagnosis, severity of illness or patterns of care, unless DRG-specific costs and length of stay data are used. Financial data are generally published each year, which ensures that estimates reflect any changes in input prices, but the larger the collection, the more likely it is that there will be publication lags. Estimates published as the National Cost Weights derive from varying-sized national samples of hospitals using cost modelling estimation techniques. The original study was undertaken in 1993 in a large national sample of public and private hospitals.'6 The Commonwealth Department of Health and Family Services subsequently sponsored service weight studies to refine ('Australianise') the modelling of the use of five intermediate services (nursing, intensive care, theatre, imaging and pathology). The 'national weights' are the ones mandated by the Pharmaceutical Benefits Advisory Committee for use in protocol-based economic evaluations of new pharmace~ticals.~~ Mechanisms have been set in place to establish an ongoing national cost data collection4'with plans to include up to 200 sites across the country for publishing annual cost estimates. In the interim, the National Hospital Morbidity (Casemix) Database is used to derive annual national cost estimates using hospital expenditures allocated to DRGs on the basis of the revised 1995 National Cost Weights.42 These estimates thus reflect current expenditure and DRG volume patterns (1997/8 data available at June 2000), but because estimates rely on the mid-1990s service weights, the costs associated with intensity of care (ALOS and relative use of inputs to care) reflect earlier practice patterns. Evaluating the timeliness of the data for a particular decision context thus depends on the relative importance of current dollar costs (+) versus underlying practice patterns (-); estimates for DRGs subject to changing clinical practices will be poorer than those where practice is relatively stable. The strengths of this data source are its affordability and its generalisability,with a large national sample of hospitals of various sizes and roles. Tables of component costs by DRG are available, with the caveat that these estimates reflect cost relativities locked in by the once-measured service weights. State-specific estimates are also available, as are estimates for public versus private hospitals. Use of the data should take particular note of inclusions and exclusions in these two sectors, with no medical costs reported for private hospitals and under-reporting of capital costs for public hospitals. Attention has been paid to standardising the overhead allocation statistics across the various samples, enhancing precision of the estimates, although the extent to which this compensates for the imprecision attributable to the IFRAC estimation and cost modelling in general is unclear. The data reflect the weaknesses of top-down allocation methods, with estimates available only at the DRG level and offering no information about inter-patient variability. A second published source of DRG-level cost data is the series of annual Victorian Cost Weight Studie~~~~'~,''.~'.'~~'~ which commenced in 1994 using 1992/93 cost data. Victoria was the first of the State health authorities to invest in patient-level costing systems, and these studies use data from opportunistic samples of public hospitals with fullyimplemented systems. The estimates have the inherent advantages of the precision of patient-level or clinical costing, and annual updating of weights reflects changes both in input prices and in clinical practice. The sample for the collection has increased from 5 hospitals to 18 over the period, with some discontinuity in hospital participation from year to year. Early samples comprised only 6 months of data, aggregated at the DRG level, but from the 1995 study onwards, full year samples have been used, and validation is undertaken using patientlevel data. Generalisability to Victorian inpatient care is very good, with recent studies representing over half of total Victorian separations. The limitation of the sample to larger metropolitan and regional hospitals (which are able to maintain such systems), and any biases attributable to distinctive Victorian practice patterns or cost structures, pose obvious problems in generalising the DRG costs reported. Techniques for data validation at both the hospital level and through preliminary analysis of cases have evolved over the with documentation of hospital level and case level findings published for each study. Beginning with the 1997 study, conformance of hospital costing systems with the Clinical Costing Standards Association of Australia4' standards on allocation of overheads has been assessed and reported. The payment formula under Victoria's casemix funding system generates 'costs' which are expressed as WIES (for Weighted Inlier Equivalent Separation) payment amounts. These may be useful for hospital profitability analyses in Victoria, but should not be used for other economic applications. While the 'weights' used in the calculations are those described above, the formula reflects other factors such as the considerable discounting implicit in recent Victorian budget cuts for public hospitals, and WIES calculations may not incorporate the DRG-based fixed cost component which is also funded under this policy. HOSPITAL ACCOUNTING DATA For many research questions, 'off the shelf cost information will be inadequate. The same three approaches to cost estimation which result in published figures (as discussed above) can also be used from unpublished sources. Bedday estimates again perform poorly against the proposed criteria, having only convenience/affordabilityto argue for their use. 2000 VOL. 24 NO. 3 AUSTRALIAN AND NEW ZEALAND JOURNAL OF PUBLIC HEALTH Jackson ~~ ~ commonwealth investment in widespread use of cost modelling means that many hospitals have some capacity to generate these estimates. Strengths and limitations of costmodelled estimates from samples of hospitals are similar to those for the National Cost Weight estimates, but with two further caveats: smaller samples of hospitals will limit generalisability, and if estimates from a single State are used, there may be some state-specific factors which impair wider generalisability. When this level of specificity is required for a decision context, small samples (or a single hospital) can provide estimates which accurately reflect their particular expenditure patterns and casemix, though not local practice pattern differences. Data acquisition costs are generally low, although negotiation of permission to use these data may have a time cost for the researcher. With the wider use of casemix funding by States, increasing numbers of hospitals have installed clinical (patient) costing systems. These provide all the benefits of the increased precision of published clinical costing results with potential for better resolution. Data can be interrogated at the 'product' level to document different patterns of patient care (number of nursing shifts with higher nursing dependency, for example, or patterns of use of high-cost diagnostic tests). Patients can be grouped below the DRG level of classification to identify variations in resource use attributable to specific diagnoses. When used in the context of economic evaluations alongside clinical trials, patient cost outcomes can be reported with the same resolution and for the same groups of patients as clinical outcomes. These data must be carefully validated, however, as subsystems vary in the precision and resolution of both costs and patient identifi~ation.~~ decision context If the requires measuring differences in the costs of pharmaceuticals used for two groups of patients, data from a particular hospital's system may not be at sufficient resolution to find such a difference, even if it really exists, Tests of data quality should be undertaken prior to aggregation so that costing errors can be identified. Although improving, documentation of these systems has been poor, and it may be difficult to establish the ways in which overheads (particularly capital equipment) have been allocated. Generalisability of estimates from small samples have both the positive and negative aspects identified for small samples in cost modelling. They are also comparable in terms of data acquisition costs. For studies requiring larger or broader samples, the Victorian Department of Human Services considers applications for use of the Victorian Cost Weights Study databases for specific projects, allowing better resolution (outlier cases, particular diagnoses, particular hospitals) and specificity than is possible using the published DRG-level reports. This data file reports component costs only in 12 'service cost groups' representing intermediate products of care such as imaging, theatre, ICU, and ward nursing, and hospital identity may be encrypted. Conclusion The choice of estimation method entails an optimisation analysis for each decision context. 'Time and motion' survey techniques remains the most valid approach to cost estimation, but are prohibitively costly and this may, in turn, limit the sample of institutions or patients costed. When precision and resolution are important objectives, clinical costing approaches provide the most valid inpatient cost estimates at a reasonable data cost. When external validity is important, use of National Cost Weights-derived estimates may be preferred, and where standardisation of hospital costs is desired (as in the context of economic evaluations of outpatient pharmaceuticals), published cost estimates may be useful. There is little justification for continued use of crude per diem cost estimates. Both primary and secondary sources of cost data must withstand challenges to their internal and external validity. The 'resolution' (or precision) of cost estimates and the relative costs of collection must also be considered, with investigators explicitlyjustifying the use of particular data sources. Acknowledgements An earlier version of this paper was presented in a seminar at the Centre for Health Economics Research and Evaluation, University of Sydney. The author is indebted to the Schneider Institute for Health Policy (The Heller School, Brandeis University) for research support as a Visiting Scholar during the writing of the paper, and to three anonymous reviewers for their helpful suggestions for restructuring the paper.

Journal

Australian and New Zealand Journal of Public Health – Wiley

Published: Jun 1, 2000

There are no references for this article.