Health Policy 115 (2014) 172–179
Contents lists available at ScienceDirect
Health Policy journal homepage: www.elsevier.com/locate/healthpol
Cross-country comparisons of health-care costs: The case of cancer treatment in the Nordic countries Jorid Kalseth a,b,∗ , Thomas Halvorsen a , Birgitte Kalseth a , Kjartan Sarheim Anthun a , Mikko Peltola d , Kirsi Kautiainen d , Unto Häkkinen d , Emma Medin e , Jonatan Lundgren e , Clas Rehnberg e , Birna Björg Másdóttir f , Maria Heimisdottir f , Helga Hrefna Bjarnadóttir f , Jóanis Erik Køtlum g , Janni Kilsmark h , Vidar Halsteinli b,c a
Department of Health Research, SINTEF Technology and Society, Trondheim, Norway Department of Public Health and General Practice, Norwegian University of Science and Technology, Trondheim, Norway c Central Norway Regional Health Authority, Stjørdal, Norway d National Institute for Health and Welfare (THL), Helsinki, Finland e Karolinska Institutet (KI), Solna, Sweden f Landspítali—National University Hospital of Iceland, Reykjavik, Iceland g Ministry of Health Affairs, Tórshavn, Faroe Islands h Danish Institute for Health Services Research (DSI), Copenhagen, Denmark b
a r t i c l e
i n f o
Article history: Received 22 March 2013 Received in revised form 18 December 2013 Accepted 2 January 2014
Keywords: Health-care costs National comparisons Cancer treatment
a b s t r a c t The objective of this study is to perform a cross-country comparison of cancer treatment costs in the Nordic countries, and to demonstrate the added value of decomposing documented costs in interpreting national differences. The study is based on individual-level data from national patient and prescription drug registers, and data on cancer prevalence from the NORDCAN database. Hospital costs were estimated on the basis of information on diagnosis-related groups (DRG) cost weights and national unit costs. Differences in per capita costs were decomposed into two stages: stage one separated the price and volume components, and stage two decomposed the volume component, relating the level of activity to service needs and availability. Differences in the per capita costs of cancer treatment between the Nordic countries may be as much as 30 per cent. National differences in the costs of treatment mirror observed differences in total health care costs. Differences in health care costs between countries may relate to different sources of variation with different policy implications. Comparisons of per capita spending alone can be misleading if the purpose is to evaluate, for example, differences in service provision and utilisation. The decomposition analysis helps to identify the relative influence of differences in the prevalence of cancer, service utilisation and productivity. © 2014 Elsevier Ireland Ltd. All rights reserved.
1. Introduction
∗ Corresponding author at: SINTEF Technology and Society, Health Research, Klæbuveien 153, Trondheim, Norway. Tel.: +47 73 59 03 00/+47 92 88 50 80. E-mail address: [email protected] (J. Kalseth). 0168-8510/$ – see front matter © 2014 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.healthpol.2014.01.003
The rising cost of health care is a central challenge facing most developed countries, with growth in health care spending exceeding the rate of economic growth [1]. In this situation, it is interesting to observe that per capita spending on health care can vary substantially, even between
J. Kalseth et al. / Health Policy 115 (2014) 172–179
neighbouring countries [1]. Monitoring national differences might provide important bench-mark information for the development of policies aimed at controlling future costs. Furthermore, concentrating on health care spending at a disaggregated level could reinforce the policy relevance of cross-country analyses because comparisons of total spending may be hampered by differences in the service delivery system and may also conceal differences between patient groups. This study is a cross-country comparison of the costs of cancer treatment in the Nordic countries. Cancer is of particular interest because it is one of the world’s major diseases, placing heavy burdens on patients and their families, as well as on the health system and economy in general [2–4]. Cancer rates are on the rise and are expected to rise further in the future as populations age [3,5–7]. The Nordic countries are well suited for comparison, not only because of their geographical closeness but also because they are similar demographically and in the organisation of their political systems and welfare and health care institutions [8]. We also aim to demonstrate the value of decomposing documented costs in interpreting national differences. Differences in health care costs between countries may relate to different sources of variation with different policy implications. Use of decomposition analysis allows for a simple and coherent comparison of contributing factors. Three sources of variation in deflated per capita cancer costs are identified in our analysis: • productivity, • service utilisation, and • morbidity.
2. Materials and methods Per capita costs can be decomposed in several stages depending on the purpose of the study and data availability. Our study uses a two-stage approach. Price and service volume are two basic components of health care costs: ci =
Ci C Q = i i = pi qi Ni Qi Ni
where ci is the per capita cost in country i, pi is the unit cost of health care provision, and qi equals the activity rate, that is, the level of activity per capita (Ci is the costs, Ni is the population and Qi is the level of activity). Differences in unit costs (pi ) may be related to differences in input prices and productivity: pi =
(1)
wi Xi = wi i Qi
(2)
where wi is input price and i is the quantity of inputs per unit of activity and hence (the inverse of) productivity (Xi is the units of inputs). Identification of productivity differences will, of course, have significant policy relevance. The starting point in our decomposition analysis is deflated national cost estimates. Using an input-based deflator, the resulting cost measure corresponds to resource use (level of inputs). A two-stage decomposition analysis can then be performed on the deflated costs ci . In stage one, the deflated per capita costs are decomposed to identify productivity and activity rates: ci =
ci = i qi wi
(3)
In stage two, the activity rate is decomposed into two separate components capturing service needs and service use, respectively, namely disease prevalence rates and service utilisation rates: qi =
Our approach is based on the same methodology as that used in cost driver analysis, which decomposes spending growth into different factors using mathematical identities as the starting point [9–15], and also has much in common with the analysis of geographical variation in medical care expenditures carried out by Dunn et al. [16]. The novelty of our study is the application of a stringent decomposition analysis in a cross-country comparison of per capita health care costs taking into account differences in morbidity. The inclusion of morbidity as a separate component is important because it is an expression of health care needs that is a major influencing factor on service volumes [12,13,15,17], and because differences in health care needs should be taken into account when service levels are compared. Restricting the analysis to one major patient group in five relatively homogeneous countries and applying a common methodology in the data collection should contribute to both the robustness and applicability of the results [18].
173
Mi Qi = mi q˜ i Ni Mi
(4)
where mi is the prevalence rate and q˜ i is the utilisation rate (Mi is the number of prevalent cases). When the analysis covers patient groups for which treatment patterns and costs vary, as is the case for cancer, then the composition (case-mix) of patient groups will affect the total treatment costs at the country level. Hence, both prevalence rates and prevalence composition need to be taken into account in comparisons of cost levels. So, instead of using the raw prevalence rate to express the expected burden of morbidity on treatment costs at the national level – that is, giving each case the same weight – information on estimated costs (deflated) per prevalent case at the diagnostic group level can be used to generate weights that are subsequently used to adjust the raw rates. To do this, the first step is to construct a case-mix index (CMI) for each country based on the average cost for all countries: CMIi =
˙j cj × Mji /Mi ˙j cj × Mj /M
(5)
where cj is the average cost per prevalent case of diagnostic group j for all countries, Mji is the number of prevalent cases of diagnostic group j in country i, Mi is the total number of prevalent cases in country i, Mj is the sum of prevalent cases of diagnostic group j for all countries, M is the sum of prevalent cases for all countries. A CMI above (below) 1 indicates a more (less) costdemanding patient group composition than the average for
174
J. Kalseth et al. / Health Policy 115 (2014) 172–179 adj
all countries. A case-mix-adjusted prevalence rate (mi ) and service utilisation rate (q˜˜ i ) can then be calculated as: adj
mi
q˜˜ i =
= CMIi × mi
(6)
qi
(7)
adj
mi
This enables us to calculate indices measuring the relative performance of a country by dividing each factor by the (weighted) average for all countries (indicated with overbar) and to decompose the relative per capita real costs: ci c¯
adj
(A) =
m i q q˜˜ (B) × i (C) = B × i (D) × i (E) ¯ q¯ ˜q˜¯ madj
=B×D×
mi (E1) × CMIi (E2) ¯ m
(8)
Note that we distinguish between activity rates defined as activity (service volume) per capita and service utilisation rates defined as service volumes per case-mix-adjusted prevalence. The data used in our analysis were obtained through a collaboration between national research groups from Denmark, Finland, Iceland, Norway, Sweden and the Faroe Islands. The study was initiated and financed by the Nordic Cancer Union (NCU) [19]. The data from the Faroe Islands were excluded because of a lack of cancer prevalence information. The estimated costs include costs of hospital services and costs of medication taken outside hospital (prescription drugs). We estimate annual cancer-related treatment costs at the country level, that is, including for each cost type, costs accrued in the chosen year irrespective of the patient’s point in the course of their illness. In cost-of-illness terminology, this is equivalent to applying a prevalence approach to identify cancer costs, which should be appropriate given the purpose of this paper [7,20]. Costs in 2007 were estimated from patient-level data sourced from national patient registers and national prescription drug registers. The exception is the Icelandic data, which were collected directly from hospital registers (see Table 1). Cancer-related cases in the patient registers were identified by primary diagnosis according to the ICD-10 classifications C00-97 (Malignant neoplasms). In addition, we included main diagnoses with the z-procedure codes Z510 (Radiotherapy session), Z511 (Chemotherapy session for neoplasm), and Z515 (Palliative care) when a cancer diagnosis was the secondary diagnosis. The identification of cancer costs in the data for prescription drugs was based on different approaches according to the type of information available in the national registers. DRGs formed the basis of the hospital treatment cost calculations. For the sample of cancer-related cases in the patient registers, information on country-specific DRG codes for each case was retrieved and the cases were aggregated according to the DRG. The numbers of in-patient and day patient discharges and outpatient visits in each DRG were combined with the respective national set of relative DRG cost weights and national unit costs to calculate cost estimates. Cost estimates for prescription drugs
were based on pharmacy retail prices for medications used by cancer patients in each country. The details on the data and estimation procedures used, including countryspecific information on data collection, DRGs, cost weights, unit price calculations, and comments on data limitations and comparability are described in the original report [19]. Costs were measured in 2007 EUR (D ). For hospital costs, an input-based deflator developed in a previous Nordic study [21] was used to convert each country’s cost estimates into comparable values. The hospital deflator is a combination of cost indices for wage costs (60 per cent) and other costs (40 per cent), where the wage-cost index is based on wage statistics (including employer taxes and pensions) for personnel working in the health sector, while the non-wage hospital cost is based on the GDP-PPP (purchasing power parity) deflator from the Organisation for Economic Co-Operation and Development (OECD). Only exchange rate corrections were applied to medicine costs, as specific deflators for prescription drugs could not be found (details on the above corrections can be found in [19]). National estimates of the per capita cost of cancer treatment (ci ) were then calculated by summarising the deflated hospital and prescription drug costs and dividing by the population number. We henceforth use the term costs as synonymous with deflated costs. Hospital treatment includes several types of activities such as in-patient, day patient and outpatient treatment. For each country, we constructed a single activity measure (Qi ) as the weighted sum of bed-days (for in-patients and day patients) and outpatient visits. In this aggregation, the estimated cost per bed-day for in-patients and day patients and the estimated cost per outpatient visit for cancer patients at the Nordic level were used as weights. Cost per bed-day was calculated by summarising hospital costs for in-patient and day patient cancer cases for the five countries and dividing by the sum of in-patient and day patient bed-days for cancer cases. We used beddays instead of discharges because the latter is much more influenced by differences in medical practices and hospital organisation and structure. A similar procedure was used for calculating cost per outpatient visit for cancer cases at the Nordic level. The cost of prescription drugs was then added to the cost of outpatient visits. The inclusion of prescription drugs is important because the systems for financing drug treatment differ between the countries in terms of the weight put on hospital-administered drugs and prescription drugs. The weighted activity measure was used to calculate activity per capita (qi ). National productivity estimates (i ) can now be calculated from the estimated (deflated) costs for cancer treatment and the related estimated activity. The hospital discharge data did not allow identification of individual cancer patients for all countries. Five-year prevalence was therefore used as a proxy for the number of cancer patients, assuming that this represents the average number of patients receiving cancer-related treatment corresponding to the costs identified at a macro level within a given year. This is in line with Pisani et al. [22], who argue that “For most cancer sites, cases surviving 5 years from diagnosis experience thereafter the same survival as the general population, so most of the workload is therefore
J. Kalseth et al. / Health Policy 115 (2014) 172–179
175
Table 1 Population, five-year cancer prevalence, hospital activity and cost estimates (EUR) for cancer treatment. Sum Nordic countries 2007 (numbers are in 1000s). Share for each country (%). Sum (in 1000s)
Populationb Five-year cancer prevalencec In-patient dischargesd In-patient daysd Day patients daysd Outpatient visitsd Weighted activity Hospital cost estimate Prescription drug cost estimatee Sum treatment costs
25 011 388 314 2994 132 2292 3036 112 2642 226 393 886 3036 112
Share (%) a
Denmark
Finland
Iceland
Norway
Sweden
21.9 21.9 19.1 18.0 14.4 29.8 22.6 25.5 0.8 22.3
21.2 19.9 24.2 (4.9) 27.1 (11.9) 14.4 (0.8) 24.9 (0.3) 25.9 (7.0) 19.0 (2.3) 27.7 20.1 (2.0)
1.3 1.0 0.9 1.0 7.8 1.2 1.2 1.2 0.6 1.1
18.9 18.9 21.4 19.3 57.8 16.5 19.2 20.2 24.5 20.8
36.7 38.3 34.4 34.6 5.5 27.7 31.1 34.1 46.3 35.7
a
The share of sum activity/costs at the Nordic level attributed to the Finnish health centres in parentheses. Data sources: National bureau of statistics in each country. Data sources: NORDCAN project. d Data sources: Finnish Hospital Discharge Register and Finnish Hospital Bench-marking Project (Finland), Norwegian Patient Register (Norway), National Patient Register (Sweden), National Patient Register (Denmark), and Patient Registers from Landspitali University Hospital and Region Hospital of Akureyri (Iceland). e Data sources: Finnish National Insurance Institution (Finland), Norwegian Prescription Database (Norway), Swedish Prescription Database (Sweden), Register of Medicinal Product Statistics (Denmark), and Landspitali University Hospital pharmacy database (Iceland). b c
due to medical acts within these first 5 years”. Data on cancer prevalence were collected from the database of the NORDCAN project [23,24]. Cancer comprises a range of different diagnostic groups (sites). In calculating CMIs for cancer prevalence, 11 specific cancer sites (colorectal, lung, skin, breast, cervix uteri, corpus uteri, prostate, testis, kidney-bladder, non-Hodgkin lymphoma and acute leukaemia) were identified in the patient registers and in the NORDCAN database based on the ICD-10 classification. Cost estimates for each site were calculated using the procedure described above. Costs and prevalence for the other cancer diagnoses were calculated residually, as the difference between the costs/prevalence of all cancer diagnoses and the sum of the costs/prevalence of the 11 cancer sites specified. Site-specific costs and prevalence were then summarised for the five countries and used to calculate the cost per prevalent case for each site at the Nordic level. Finally, the CMI and service levels per case-mix adjusted prevalence for each country were calculated according to Eqs. (5) and (7), respectively. National differences in the organisation of health care services are a challenge in cross-country comparisons, even within the Nordic context. The DRG-sorted patient data from Finland, for example, also include activities performed in health centres—activities that in many cases are not likely to be performed in hospitals in the other countries. Nevertheless, some health centre activity is treatment that would be carried out within hospitals in the other Nordic systems. The special case of the health centres is reflected in the long average stay for cancer-related activity in Finnish health centres and the relatively short average stay in Finnish hospitals. We therefore chose to present estimates for Finland with health centre activity both included and excluded. The collected data cover a total of 314 000 discharges and nearly three million days of in-patient cancer treatment, 132 000 days for day patients, and almost 2.3 million outpatient visits (Table 1). Prescription drugs constitute 13 per cent of the total estimated costs. The data indicate
national differences in cancer treatment patterns, both with regard to the reliance on day patient and outpatient treatment (as opposed to in-patient treatment) and in the use of prescription drugs versus providing medication within hospitals. Denmark has, for instance, a high share of outpatient activity and most of the cancer drugs are provided by the hospitals. This is in contrast to Finland, Norway and Sweden where the cost share of prescription drugs is about 20 per cent. 3. Results 3.1. Country differences in estimated per capita treatment costs The estimated costs of cancer treatment in the Nordic countries in 2007 were three billion euros. The total population of the five countries is approximately 25 million. The estimated average per capita cost in the Nordic region was therefore about D 120. Sweden represents almost 40 per cent of the total population in the Nordic region. Iceland, on the other hand, represents only about 1 per cent. It follows that the results for Sweden heavily influence the estimate for the region as a whole. The country-specific per capita estimates range from D 133 in Norway to D 110 in Iceland. The estimated cost for Finland is D 104 without health centre activity and D 115 when health centre activity is included. Sweden and Denmark lie in the middle with estimated per capita costs of D 118 and D 124, respectively. Differences in per capita cancer treatment costs between the Nordic countries are therefore as high as 30 per cent. 3.2. Decomposition of per capita costs The results of the decomposition of per capita costs in accordance with Eq. (8) are shown in Table 2. We comment on the results for each country in turn, starting with Finland.
176
J. Kalseth et al. / Health Policy 115 (2014) 172–179
Table 2 Per capita treatment cost of cancer (EUR). Decomposition of treatment cost per capita. Indicators measured relative to Nordic weighted average (=1) 2007.
Cost per capita (EUR) Decomposition [Eq. (8)] Cost per capita A=B×C Productivity (cost per activity) B Activity rate (activity per capita) C=D×E D Service utilisation rate (activity per adj. preval.) E = E1 × E2 Adjusted prevalence rate E1 Raw five-year prevalence rate E2 Cancer prevalence case-mix index (CMI)
The size of the Finnish estimates is dependent on whether health centres are included or not. Health centre activity amounted to 27 per cent of total estimated activity for Finland, and the activity of the health centres was characterised by long stays at a low cost per day. The per capita costs were 5 per cent below the Nordic average when health centres were included and 15 per cent below the average when they were not. In both cases, costs were on the low side in a Nordic context. The decomposition analysis indicated that high productivity (low costs per activity) is an important explanatory factor. High productivity is present in Finland regardless of the inclusion or exclusion of health centre activity, even though estimated productivity was almost 20 per cent higher when the health centres were included. The estimated per capita activity was high when health centre activity was included, more than 20 per cent above the Nordic average. Excluding health centre activity, the estimated activity rate was low (about 10 per cent below the average), and this contributed to low per capita costs in this case. The case-mix-adjusted prevalence rate was close to the Nordic average even though the raw prevalence rate was low. Hence, the relative service utilisation rate in Finland was on a par with the relative position of its per capita activity. The per capita cost for Iceland was 9 per cent below the Nordic average. This follows from the high productivity and low case-mix-adjusted prevalence rate. Nevertheless, the service utilisation rate was high. Sweden was in a median position with regard to per capita costs, lying 3 per cent below the Nordic average. Two counteracting factors appear. First, productivity was low (cost per activity was high), but this was counteracted by the relatively low case-mix-adjusted prevalence rate and estimated service utilisation rate. Denmark had the second highest estimated per capita costs at 2 per cent above the Nordic average. Productivity was also close to average (1 per cent above), but the casemix-adjusted prevalence rate was high. The Danish service utilisation rate was relatively low. The Norwegian cost per capita was as much as 10 per cent above the Nordic average, which mainly relates to relatively low productivity. The case-mix-adjusted prevalence rate and service utilisation rate were close to the Nordic average. 4. Discussion By employing a common methodology in the data collection and analysis, we identified differences in the per
Denmark
Finland
Iceland
Norway
Sweden
124
115
110
133
118
1.02 0.99 1.03 0.94 1.10 1.00 1.10
0.95 0.78 1.22 1.21 1.01 0.94 1.07
0.91 0.94 0.97 1.08 0.90 0.83 1.09
1.10 1.08 1.01 1.03 0.99 1.00 0.99
0.97 1.15 0.85 0.89 0.95 1.05 0.91
capita treatment cost of cancer of up to 30 per cent between the Nordic countries. These estimates show relatively modest differences compared with previous estimates for Sweden and Finland based on non-comparable data and methodologies [25]. The ranking of the countries is largely the same as the OECD statistics for total health care expenditure per capita (Table 3, columns i and iii). Although the differences in per capita health care costs are narrower for cancer than costs in general, our results still suggest that Norway has the highest per capita health care costs of all the Nordic countries. Nevertheless, health care spending as a share of GDP is lower for Norway than for the other Nordic countries, except for Finland [1]. This suggests that the high spending level in Norway is related to the high per capita GDP. This is also a more general observation: national income level is found to be the strongest determinant of per capita health care spending [26]. The ranking of total per capita health expenditure among the Nordic countries is the same as for per capita GDP, taking expenditure on long-term care into account. More modest differences in cancer costs compared with total spending on health care may reflect the fact that cancer patients are highly prioritised in all countries, but it may also reflect the challenges of comparing total health care costs as such. The OECD data on per capita health care spending often serve as a reference in health policy analysis and comparative research. We have learned, however, that the organisation and delivery of health care services may affect the reported health care expenditures. This limits the comparability of international statistics on health care Table 3 Per capita cancer treatment costs and per capita total health care costs (OECD). Nordic countries measured relative to Norway 2007.
Denmark Finland Iceland Sweden
Cancer treatment Ia (i) (%)
Cancer treatment IIb (ii) (%)
Health care OECDb,c (iii) (%)
Health care OECD excluding LTCb,c,d (iv) (%)
93 78e 83 89
85 71f 75 84
77 60 69 70
78 70 74 86
a Deflator: Previously developed deflator for hospital costs [21] and only exchange rate adjustment for prescription drugs. b Deflator: GDP-PPP (Eurostat-OECD methodological manual on Purchasing Power Parities. 2012 edition). c OECD.StatExtracts. d LTC-services of long-term nursing care as defined in A System of Health Accounts. 2011 edition (OECD). e 87% if health centres are included. f 79% if health centres are included.
J. Kalseth et al. / Health Policy 115 (2014) 172–179
spending levels. Take, for instance, the different practices involved in estimating expenditures of long-term nursing care. The reported expenditures in the OECD statistics on health care and social services, respectively, appear to differ quite substantially. The share of total health care spending on long-term care, on the one hand, was between 19 and 25 per cent in Iceland, Denmark and Norway, while it was only 12 per cent in Finland and was as low as 8 per cent in Sweden in 2007 (OECD.StatExtracts). On the other hand, Sweden in particular, but also Finland, report high levels of long-term care as social expenditure, as opposed to Norway, Denmark and Iceland [1]. Excluding long-term care from health care expenditure reduces the differences in total spending levels between Norway and the other Nordic countries (Table 3, column iv). The choice of deflator also affects estimated cost differences. In the OECD statistics, health care expenditures measured in national currencies are converted into comparable health expenditures, using the PPP index for the entire economy (GDP). This measure is sensitive to any deviation in price levels in health care from the rest of the economy. The estimated cancer costs were adjusted for differences in input prices according to medical personnel wage differences in this study. The estimated cost difference of cancer treatment between Norway and the other countries is about 5–8 per cent higher if we use the GDPPPP deflator for all cost components (Table 3, column ii). When the application of sector-specific deflators (rather than GDP-PPPs) narrows the differences between Norway and the other Nordic countries, it implies that high wage levels in the health sector contribute substantially to the high spending on health care in Norway. All in all, when we take into account the differences in the deflators and exclude long-term care expenditure from the OECD health care spending data, the estimated national differences in cancer treatment costs are quite similar to the differences in reported expenditures on health care in the OECD statistics. The decomposition of per capita costs illustrates a trivial but still important point: looking only at per capita spending may be very misleading for policy purposes. Differences in per capita spending for each country may reflect differences in underlying factors that have very different policy implications. High spending levels do not necessarily reflect “superior” health care in the form of high service utilisation rates, and vice versa. It may, for example, reflect differences in productivity rather than differences in accessibility or standards of care. Furthermore, there may be differences for each country in morbidity or particular national health care needs that might legitimise the observed differences. High costs because of high morbidity may represent a fiscal challenge and a public health problem, but are typically weakly related to the performance of the health system [27]. Using per capita measures alone may therefore also be problematic for comparisons of service levels. Cancer prevalence rates as well as the composition of cancer prevalence differ between the Nordic countries. Some of the differences in five-year prevalence rates are quite striking [19]. The raw five-year prevalence rate was 26 per cent higher in Sweden than in Iceland. On the other
177
hand, Sweden had low shares of total prevalence for cancer sites with high estimated cost per prevalence (acute leukaemia, non-Hodgkin lymphoma, lung cancer, and the residual category) and high shares of prostate and skin cancer, both of which have large patient groups in terms of prevalence but a low estimated cost per prevalence. For Iceland, and partly for Denmark and Finland, the opposite is the case. Prostate cancer was the largest group in terms of five-year prevalence at the Nordic level, and the estimated cost per prevalent case was half of the average cost for all sites. The share of total prevalence for prostate cancer varied from 14 per cent in Denmark to 27 per cent in Sweden. When resources are scarce and becoming scarcer, exploiting available resources optimally is crucial. Productivity differences are therefore of particular interest. The results for cancer treatment seem to reflect more general features of hospital productivity in the Nordic countries. High productivity in Finland compared with Norway, Sweden and Denmark has already been found for general hospitals [21,28]. The ranking of the countries in terms of productivity in cancer treatment mirrors that of hospitals in general. Differences in productivity between Nordic hospitals seem to be due to country-specific factors. The best practice hospitals in Finland perform better than best practice hospitals in Denmark, Norway and Sweden, while the variations within each country are not significantly different [29]. This suggests that the hospitals are operating under different external conditions or different country-specific institutional frameworks. The study has several limitations. First, the number of five-year prevalent cases was used as a proxy for the number of patients receiving treatment because no patient identification was available in the dataset. Next, no distinctions were made concerning phase of disease. It is well known that cancer costs are phase sensitive [7] and, given a U-shape of phase-specific costs [30], it is possible that some of the people included who are living with a cancer diagnosis received limited or no hospital services at all. Biased results may arise if there are differences between countries in phase distributions within the cancer sites or in the share of five-year prevalent cases actually receiving treatment. It is also worth noting that the inferences drawn regarding service utilisation rates are sensitive to the choice of five-year prevalence as an indicator of morbidity and the method of case-mix adjustment. This is particularly true for the relative positions of Denmark and Iceland (compared with activity per capita). A third limitation is that the case-mix adjustment may also be affected by the relatively large residual group of cancer diagnoses for which identical costs per prevalence are assumed. The analysis should in this respect be regarded as an attempt to illustrate the importance of taking differences in morbidity into account in the cross-country comparisons of health care costs. Even though the case-mix-adjusted five-year prevalence rate may not be a perfect proxy, it does seem to capture differences in activity and costs resulting from differences in health care needs. We expect the differences in health care needs to be at least partly reflected in the level of activity per capita, and we find a positive correlation between activity per capita and the case-mix-adjusted
178
J. Kalseth et al. / Health Policy 115 (2014) 172–179
five-year prevalence rate. The results suggest that the countries can be divided into three groups with Norway in a middle position with regard to both adjusted prevalence and activity, Denmark and Finland having higher rates than Norway, and the adjusted prevalence rates and activity rates for Sweden and Iceland being lower. We do not find a similar relationship between the raw prevalence rate and activity. Hence, case-mix adjustment of prevalence rates seems to be clearly warranted. Although we have put considerable effort into ensuring that our estimates of cancer treatment activity and costs are comparable, differences in data accessibility for each country because of different national data registration and patient classification systems, as well as differences in health care organisation, may still bias the results. Limitations on data availability prohibited the inclusion of direct medical costs incurred by primary health care and longterm care to treat and care for cancer patients in our cost measure. Although costs related to outpatient primary care are relevant, estimates for Norway, for which we were able to collect data for cancer patients, indicate that outpatient primary care is likely to play only a limited role in cancer treatment in the Nordic countries because it amounted to only about 2 per cent of estimated cancer treatment costs. The lack of data on cancer-related costs of long-term institutional or home care is more likely to be influential, as illustrated both by the different shares of reported longterm care in the OECD statistics and by the sensitivity of the results for Finland to the inclusion or exclusion of health centre activity. 5. Conclusions The present study demonstrates that differences in health care costs may be due to several factors that have very different policy implications. It is important to be mindful of this when health care is compared across countries because the conclusions may depend heavily on the indicators used. This study shows, for instance, that per capita spending levels can be a poor proxy for service utilisation rates. To gain insight into the underlying factors, cost comparisons should involve a decomposition of costs into individual influencing factors: morbidity, service levels, productivity and, preferably, quality and outcomes. Such a decomposition analysis may be more feasible if the study is limited to a particular service or patient group. Close attention should also be given to the choice of conversion factors for health care costs because this may also have a significant bearing on the results in cross-country comparisons. Acknowledgements The authors would like to thank two anonymous referees and the journal editors for their contributions to this paper. Their perceptive comments and suggestions have been tremendously important in the revisions of the manuscript. The usual disclaimer applies. References [1] OECD. Health at a glance 2011: OECD indicators. OECD Publishing; 2011.
[2] World Health Organization. Global status report on noncommunicable diseases 2010. Geneva: World Health Organization; 2011. [3] World Health Organization. The global burden of disease: 2004 update. Geneva: World Health Organization; 2008. [4] American Cancer Society and Livestrong. The global economic cost of cancer 2010. [5] Møller B, Fekjær H, Hakulinen T, Tryggvadóttir L, Storm HH, Talbäck M, Haldorsen T. Prediction of cancer incidence in the Nordic countries up to the year 2020. European Journal of Cancer Prevention 2002;11(Suppl 1):1–96. [6] Rahimi A, Møller B, Bray F. Predicting cancer prevalence in 2025 for the Nordic countries. Norwegian Cancer Registry 2010. [7] Yabroff KR, Lund J, Kepka D, Mariotto A. Economic burden of cancer in the United States: estimates, projections, and future research. Cancer Epidemiology, Biomarkers, & Prevention 2011;20(10):2006–14, http://dx.doi.org/10.1158/1055-9965.EPI-11-0650. [8] Magnussen J, Vrangbæk K, Saltman R, editors. Nordic health care systems: recent reforms and current policy challenges. London: Open University Press; 2009. [9] Schieber GJ, Poullier JP. International health spending and utilization trends. Health Affairs 1988;7(4):105–12, http://dx.doi.org/10.1377/hlthaff.7.4.105. UG, Lundin D. Why did drug spending [10] Gerdtham increase during the 1990s?: a decomposition based on data. Pharmacoeconomics 2004;22(1):29–42, Swedish http://dx.doi.org/10.2165/00019053-200422010-00003. [11] Bongaarts J. Population aging and the rising cost of public pensions. Population and Development Review 2004;30(1):1–23, http://dx.doi.org/10.1111/j.1728-4457.2004.00001.x. [12] Thorpe KE, Howard DH. The rise in spending among Medicare beneficiaries: the role of chronic disease prevalence and changes in treatment intensity. Health Affairs 2006;25(5):378–88, http://dx.doi.org/10.1377/hlthaff.25.w378. [13] Dunn A, Liebman E, Shapiro AH. Decomposing medical-care expenditure growth. Working paper 2012-26. Federal Reserve Bank of San Francisco; November 2012. [14] Panopoulou E, Pantelidis T. Convergence in per capita health expenditures and health outcomes in the OECD Applied Economics 2012;44(30):3909–20, countries. http://dx.doi.org/10.1080/00036846.2011.583222. [15] Thorpe KE. Treated disease prevalence and spending per treated case drove most of the growth in health care spending in 1987–2009. Health Affairs 2013;32(5):851–8, http://dx.doi.org/10.1377/hlthaff.2012.0391. [16] Dunn A, Shapiro AH, Liebman E. Geographic variation in commercial medical-care expenditures: a framework for decomposing price and utilization. US Bureau of Economic Analysis; 2012 January. [17] Thorpe KE, Howard DH, Galactionova K. Differences in disease prevalence as a source of the U.S.—European health care spending gap. Health Affairs 2007;26(6):678–86, http://dx.doi.org/10.1377/hlthaff.26.6.w678. [18] Ó Céilleachair AJ, Hanly P, Skally M, O’Neill C, Fitzpatrick P, Kapur K, Staines A, Sharp L. Cost comparisons and methodological heterogeneity in cost-of-illness studies: the example colorectal cancer. Medical Care 2013;51(4):339–50, of http://dx.doi.org/10.1097/MLR.0b013e3182726c13. [19] Kalseth J, Halsteinli V, Halvorsen T, Kalseth B, Anthun K, Peltola M, Kautiainen K, Häkkinen U, Medin E, Lundgren J, Rehnberg C, Másdóttir BB, Heimisdottir M, Bjarnadóttir HH, Køtlum JE, Kilsmark J. Costs of cancer in the Nordic countries. A comparative study of health care costs and public income loss compensation payments related to cancer in the Nordic countries in 2007. Rapport SINTEF A19395. 2011. [20] Tarricone R. Cost-of-illness analysis. What room in economics? Health Policy 2006;77:51–63, health http://dx.doi.org/10.1016/j.healthpol.2005.07.016. [21] Kittelsen SAC, Anthun KS, Kalseth B, Kalseth J, Halsteinli V, Magnussen J. En komperativ analyse av spesialisthelsetjenesten i Finland, Sverige, Danmark og Norge: Aktivitet, ressursbruk og produktivitet 2005-2007. Rapport SINTEF A12200. 2009. [22] Pisani P, Bray F, Parkin DM. Estimates of world-wide prevalence of cancer for 25 sites in the adult population. International Journal of Cancer 2002;97(1):72–81, http://dx.doi.org/10.1002/ijc.1571. [23] Engholm G, Ferlay J, Christensen N, Bray F, Gjerstorff ML, Klint Å, Køtlum JE, Ólafsdóttir E, Pukkala E, Storm HH. NORDCAN: cancer incidence, mortality, prevalence and prediction in the Nordic countries, Version 3.7. Association of the Nordic Cancer Registries. Danish Cancer Society; 2010. http://www.ancr.nu.
J. Kalseth et al. / Health Policy 115 (2014) 172–179 [24] Engholm G, Ferlay J, Christensen N, Bray F, Gjerstorff ML, Klint Å, Køtlum JE, Ólafsdóttir E, Pukkala E, Storm HH. NORDCAN-a Nordic tool for cancer information, planning, quality control and research. Acta Oncologica 2010;49(5):725–36, http://dx.doi.org/10.3109/02841861003782017. [25] Wilking N, Jönsson B, Högberg D. Comparator Report on Patient Access to Cancer Drugs in Europe. 15 February 2009. [26] Hartwig J. What drives health care expenditure?—Baumol’s model of ‘unbalanced growth’ revisited. Journal of Health Economics 2008;27:603–23, http://dx.doi.org/10.1016/j.jhealeco.2007.05.006. [27] Smith PC, Mossialos E, Papanicolas I. Performance measurement for health system improvement: experiences, challenges and prospects. World Health Organization 2008 and World Health Organization, on behalf of the European Observatory on Health Systems and Policies; 2008.
179
[28] Medin E, Häkkinen U, Linna M, Anthun KS, Kittelsen SAC, Rehnberg C. International hospital productivity comparison: experiences from the Nordic countries. Healt Policy 2013;112(1-2):80–7, http://dx.doi.org/10.1016/j.healthpol.2013.02.004. [29] Kalseth B, Anthun KS, Hope Ø, Kittelsen SAC, Persson BA. Spesialisthelsetjenesten i Norden. Sykehusstruktur, styringsstruktur og lokal arbeidsorganisering som mulig forklaring på kostnadsforskjeller mellom landene. Rapport SINTEF A19615. 2011. [30] Brown ML, Riley GF, Schusseler N, Etzioni R. Estimating health care costs related to cancer treatment from SEER-Medicare data. Medical Care 2002;40(8 (Supplement:IV-104-IV-117)), http://dx.doi.org/10.1097/00005650-200208001-00014.
Contents lists available at ScienceDirect
Health Policy journal homepage: www.elsevier.com/locate/healthpol
Cross-country comparisons of health-care costs: The case of cancer treatment in the Nordic countries Jorid Kalseth a,b,∗ , Thomas Halvorsen a , Birgitte Kalseth a , Kjartan Sarheim Anthun a , Mikko Peltola d , Kirsi Kautiainen d , Unto Häkkinen d , Emma Medin e , Jonatan Lundgren e , Clas Rehnberg e , Birna Björg Másdóttir f , Maria Heimisdottir f , Helga Hrefna Bjarnadóttir f , Jóanis Erik Køtlum g , Janni Kilsmark h , Vidar Halsteinli b,c a
Department of Health Research, SINTEF Technology and Society, Trondheim, Norway Department of Public Health and General Practice, Norwegian University of Science and Technology, Trondheim, Norway c Central Norway Regional Health Authority, Stjørdal, Norway d National Institute for Health and Welfare (THL), Helsinki, Finland e Karolinska Institutet (KI), Solna, Sweden f Landspítali—National University Hospital of Iceland, Reykjavik, Iceland g Ministry of Health Affairs, Tórshavn, Faroe Islands h Danish Institute for Health Services Research (DSI), Copenhagen, Denmark b
a r t i c l e
i n f o
Article history: Received 22 March 2013 Received in revised form 18 December 2013 Accepted 2 January 2014
Keywords: Health-care costs National comparisons Cancer treatment
a b s t r a c t The objective of this study is to perform a cross-country comparison of cancer treatment costs in the Nordic countries, and to demonstrate the added value of decomposing documented costs in interpreting national differences. The study is based on individual-level data from national patient and prescription drug registers, and data on cancer prevalence from the NORDCAN database. Hospital costs were estimated on the basis of information on diagnosis-related groups (DRG) cost weights and national unit costs. Differences in per capita costs were decomposed into two stages: stage one separated the price and volume components, and stage two decomposed the volume component, relating the level of activity to service needs and availability. Differences in the per capita costs of cancer treatment between the Nordic countries may be as much as 30 per cent. National differences in the costs of treatment mirror observed differences in total health care costs. Differences in health care costs between countries may relate to different sources of variation with different policy implications. Comparisons of per capita spending alone can be misleading if the purpose is to evaluate, for example, differences in service provision and utilisation. The decomposition analysis helps to identify the relative influence of differences in the prevalence of cancer, service utilisation and productivity. © 2014 Elsevier Ireland Ltd. All rights reserved.
1. Introduction
∗ Corresponding author at: SINTEF Technology and Society, Health Research, Klæbuveien 153, Trondheim, Norway. Tel.: +47 73 59 03 00/+47 92 88 50 80. E-mail address: [email protected] (J. Kalseth). 0168-8510/$ – see front matter © 2014 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.healthpol.2014.01.003
The rising cost of health care is a central challenge facing most developed countries, with growth in health care spending exceeding the rate of economic growth [1]. In this situation, it is interesting to observe that per capita spending on health care can vary substantially, even between
J. Kalseth et al. / Health Policy 115 (2014) 172–179
neighbouring countries [1]. Monitoring national differences might provide important bench-mark information for the development of policies aimed at controlling future costs. Furthermore, concentrating on health care spending at a disaggregated level could reinforce the policy relevance of cross-country analyses because comparisons of total spending may be hampered by differences in the service delivery system and may also conceal differences between patient groups. This study is a cross-country comparison of the costs of cancer treatment in the Nordic countries. Cancer is of particular interest because it is one of the world’s major diseases, placing heavy burdens on patients and their families, as well as on the health system and economy in general [2–4]. Cancer rates are on the rise and are expected to rise further in the future as populations age [3,5–7]. The Nordic countries are well suited for comparison, not only because of their geographical closeness but also because they are similar demographically and in the organisation of their political systems and welfare and health care institutions [8]. We also aim to demonstrate the value of decomposing documented costs in interpreting national differences. Differences in health care costs between countries may relate to different sources of variation with different policy implications. Use of decomposition analysis allows for a simple and coherent comparison of contributing factors. Three sources of variation in deflated per capita cancer costs are identified in our analysis: • productivity, • service utilisation, and • morbidity.
2. Materials and methods Per capita costs can be decomposed in several stages depending on the purpose of the study and data availability. Our study uses a two-stage approach. Price and service volume are two basic components of health care costs: ci =
Ci C Q = i i = pi qi Ni Qi Ni
where ci is the per capita cost in country i, pi is the unit cost of health care provision, and qi equals the activity rate, that is, the level of activity per capita (Ci is the costs, Ni is the population and Qi is the level of activity). Differences in unit costs (pi ) may be related to differences in input prices and productivity: pi =
(1)
wi Xi = wi i Qi
(2)
where wi is input price and i is the quantity of inputs per unit of activity and hence (the inverse of) productivity (Xi is the units of inputs). Identification of productivity differences will, of course, have significant policy relevance. The starting point in our decomposition analysis is deflated national cost estimates. Using an input-based deflator, the resulting cost measure corresponds to resource use (level of inputs). A two-stage decomposition analysis can then be performed on the deflated costs ci . In stage one, the deflated per capita costs are decomposed to identify productivity and activity rates: ci =
ci = i qi wi
(3)
In stage two, the activity rate is decomposed into two separate components capturing service needs and service use, respectively, namely disease prevalence rates and service utilisation rates: qi =
Our approach is based on the same methodology as that used in cost driver analysis, which decomposes spending growth into different factors using mathematical identities as the starting point [9–15], and also has much in common with the analysis of geographical variation in medical care expenditures carried out by Dunn et al. [16]. The novelty of our study is the application of a stringent decomposition analysis in a cross-country comparison of per capita health care costs taking into account differences in morbidity. The inclusion of morbidity as a separate component is important because it is an expression of health care needs that is a major influencing factor on service volumes [12,13,15,17], and because differences in health care needs should be taken into account when service levels are compared. Restricting the analysis to one major patient group in five relatively homogeneous countries and applying a common methodology in the data collection should contribute to both the robustness and applicability of the results [18].
173
Mi Qi = mi q˜ i Ni Mi
(4)
where mi is the prevalence rate and q˜ i is the utilisation rate (Mi is the number of prevalent cases). When the analysis covers patient groups for which treatment patterns and costs vary, as is the case for cancer, then the composition (case-mix) of patient groups will affect the total treatment costs at the country level. Hence, both prevalence rates and prevalence composition need to be taken into account in comparisons of cost levels. So, instead of using the raw prevalence rate to express the expected burden of morbidity on treatment costs at the national level – that is, giving each case the same weight – information on estimated costs (deflated) per prevalent case at the diagnostic group level can be used to generate weights that are subsequently used to adjust the raw rates. To do this, the first step is to construct a case-mix index (CMI) for each country based on the average cost for all countries: CMIi =
˙j cj × Mji /Mi ˙j cj × Mj /M
(5)
where cj is the average cost per prevalent case of diagnostic group j for all countries, Mji is the number of prevalent cases of diagnostic group j in country i, Mi is the total number of prevalent cases in country i, Mj is the sum of prevalent cases of diagnostic group j for all countries, M is the sum of prevalent cases for all countries. A CMI above (below) 1 indicates a more (less) costdemanding patient group composition than the average for
174
J. Kalseth et al. / Health Policy 115 (2014) 172–179 adj
all countries. A case-mix-adjusted prevalence rate (mi ) and service utilisation rate (q˜˜ i ) can then be calculated as: adj
mi
q˜˜ i =
= CMIi × mi
(6)
qi
(7)
adj
mi
This enables us to calculate indices measuring the relative performance of a country by dividing each factor by the (weighted) average for all countries (indicated with overbar) and to decompose the relative per capita real costs: ci c¯
adj
(A) =
m i q q˜˜ (B) × i (C) = B × i (D) × i (E) ¯ q¯ ˜q˜¯ madj
=B×D×
mi (E1) × CMIi (E2) ¯ m
(8)
Note that we distinguish between activity rates defined as activity (service volume) per capita and service utilisation rates defined as service volumes per case-mix-adjusted prevalence. The data used in our analysis were obtained through a collaboration between national research groups from Denmark, Finland, Iceland, Norway, Sweden and the Faroe Islands. The study was initiated and financed by the Nordic Cancer Union (NCU) [19]. The data from the Faroe Islands were excluded because of a lack of cancer prevalence information. The estimated costs include costs of hospital services and costs of medication taken outside hospital (prescription drugs). We estimate annual cancer-related treatment costs at the country level, that is, including for each cost type, costs accrued in the chosen year irrespective of the patient’s point in the course of their illness. In cost-of-illness terminology, this is equivalent to applying a prevalence approach to identify cancer costs, which should be appropriate given the purpose of this paper [7,20]. Costs in 2007 were estimated from patient-level data sourced from national patient registers and national prescription drug registers. The exception is the Icelandic data, which were collected directly from hospital registers (see Table 1). Cancer-related cases in the patient registers were identified by primary diagnosis according to the ICD-10 classifications C00-97 (Malignant neoplasms). In addition, we included main diagnoses with the z-procedure codes Z510 (Radiotherapy session), Z511 (Chemotherapy session for neoplasm), and Z515 (Palliative care) when a cancer diagnosis was the secondary diagnosis. The identification of cancer costs in the data for prescription drugs was based on different approaches according to the type of information available in the national registers. DRGs formed the basis of the hospital treatment cost calculations. For the sample of cancer-related cases in the patient registers, information on country-specific DRG codes for each case was retrieved and the cases were aggregated according to the DRG. The numbers of in-patient and day patient discharges and outpatient visits in each DRG were combined with the respective national set of relative DRG cost weights and national unit costs to calculate cost estimates. Cost estimates for prescription drugs
were based on pharmacy retail prices for medications used by cancer patients in each country. The details on the data and estimation procedures used, including countryspecific information on data collection, DRGs, cost weights, unit price calculations, and comments on data limitations and comparability are described in the original report [19]. Costs were measured in 2007 EUR (D ). For hospital costs, an input-based deflator developed in a previous Nordic study [21] was used to convert each country’s cost estimates into comparable values. The hospital deflator is a combination of cost indices for wage costs (60 per cent) and other costs (40 per cent), where the wage-cost index is based on wage statistics (including employer taxes and pensions) for personnel working in the health sector, while the non-wage hospital cost is based on the GDP-PPP (purchasing power parity) deflator from the Organisation for Economic Co-Operation and Development (OECD). Only exchange rate corrections were applied to medicine costs, as specific deflators for prescription drugs could not be found (details on the above corrections can be found in [19]). National estimates of the per capita cost of cancer treatment (ci ) were then calculated by summarising the deflated hospital and prescription drug costs and dividing by the population number. We henceforth use the term costs as synonymous with deflated costs. Hospital treatment includes several types of activities such as in-patient, day patient and outpatient treatment. For each country, we constructed a single activity measure (Qi ) as the weighted sum of bed-days (for in-patients and day patients) and outpatient visits. In this aggregation, the estimated cost per bed-day for in-patients and day patients and the estimated cost per outpatient visit for cancer patients at the Nordic level were used as weights. Cost per bed-day was calculated by summarising hospital costs for in-patient and day patient cancer cases for the five countries and dividing by the sum of in-patient and day patient bed-days for cancer cases. We used beddays instead of discharges because the latter is much more influenced by differences in medical practices and hospital organisation and structure. A similar procedure was used for calculating cost per outpatient visit for cancer cases at the Nordic level. The cost of prescription drugs was then added to the cost of outpatient visits. The inclusion of prescription drugs is important because the systems for financing drug treatment differ between the countries in terms of the weight put on hospital-administered drugs and prescription drugs. The weighted activity measure was used to calculate activity per capita (qi ). National productivity estimates (i ) can now be calculated from the estimated (deflated) costs for cancer treatment and the related estimated activity. The hospital discharge data did not allow identification of individual cancer patients for all countries. Five-year prevalence was therefore used as a proxy for the number of cancer patients, assuming that this represents the average number of patients receiving cancer-related treatment corresponding to the costs identified at a macro level within a given year. This is in line with Pisani et al. [22], who argue that “For most cancer sites, cases surviving 5 years from diagnosis experience thereafter the same survival as the general population, so most of the workload is therefore
J. Kalseth et al. / Health Policy 115 (2014) 172–179
175
Table 1 Population, five-year cancer prevalence, hospital activity and cost estimates (EUR) for cancer treatment. Sum Nordic countries 2007 (numbers are in 1000s). Share for each country (%). Sum (in 1000s)
Populationb Five-year cancer prevalencec In-patient dischargesd In-patient daysd Day patients daysd Outpatient visitsd Weighted activity Hospital cost estimate Prescription drug cost estimatee Sum treatment costs
25 011 388 314 2994 132 2292 3036 112 2642 226 393 886 3036 112
Share (%) a
Denmark
Finland
Iceland
Norway
Sweden
21.9 21.9 19.1 18.0 14.4 29.8 22.6 25.5 0.8 22.3
21.2 19.9 24.2 (4.9) 27.1 (11.9) 14.4 (0.8) 24.9 (0.3) 25.9 (7.0) 19.0 (2.3) 27.7 20.1 (2.0)
1.3 1.0 0.9 1.0 7.8 1.2 1.2 1.2 0.6 1.1
18.9 18.9 21.4 19.3 57.8 16.5 19.2 20.2 24.5 20.8
36.7 38.3 34.4 34.6 5.5 27.7 31.1 34.1 46.3 35.7
a
The share of sum activity/costs at the Nordic level attributed to the Finnish health centres in parentheses. Data sources: National bureau of statistics in each country. Data sources: NORDCAN project. d Data sources: Finnish Hospital Discharge Register and Finnish Hospital Bench-marking Project (Finland), Norwegian Patient Register (Norway), National Patient Register (Sweden), National Patient Register (Denmark), and Patient Registers from Landspitali University Hospital and Region Hospital of Akureyri (Iceland). e Data sources: Finnish National Insurance Institution (Finland), Norwegian Prescription Database (Norway), Swedish Prescription Database (Sweden), Register of Medicinal Product Statistics (Denmark), and Landspitali University Hospital pharmacy database (Iceland). b c
due to medical acts within these first 5 years”. Data on cancer prevalence were collected from the database of the NORDCAN project [23,24]. Cancer comprises a range of different diagnostic groups (sites). In calculating CMIs for cancer prevalence, 11 specific cancer sites (colorectal, lung, skin, breast, cervix uteri, corpus uteri, prostate, testis, kidney-bladder, non-Hodgkin lymphoma and acute leukaemia) were identified in the patient registers and in the NORDCAN database based on the ICD-10 classification. Cost estimates for each site were calculated using the procedure described above. Costs and prevalence for the other cancer diagnoses were calculated residually, as the difference between the costs/prevalence of all cancer diagnoses and the sum of the costs/prevalence of the 11 cancer sites specified. Site-specific costs and prevalence were then summarised for the five countries and used to calculate the cost per prevalent case for each site at the Nordic level. Finally, the CMI and service levels per case-mix adjusted prevalence for each country were calculated according to Eqs. (5) and (7), respectively. National differences in the organisation of health care services are a challenge in cross-country comparisons, even within the Nordic context. The DRG-sorted patient data from Finland, for example, also include activities performed in health centres—activities that in many cases are not likely to be performed in hospitals in the other countries. Nevertheless, some health centre activity is treatment that would be carried out within hospitals in the other Nordic systems. The special case of the health centres is reflected in the long average stay for cancer-related activity in Finnish health centres and the relatively short average stay in Finnish hospitals. We therefore chose to present estimates for Finland with health centre activity both included and excluded. The collected data cover a total of 314 000 discharges and nearly three million days of in-patient cancer treatment, 132 000 days for day patients, and almost 2.3 million outpatient visits (Table 1). Prescription drugs constitute 13 per cent of the total estimated costs. The data indicate
national differences in cancer treatment patterns, both with regard to the reliance on day patient and outpatient treatment (as opposed to in-patient treatment) and in the use of prescription drugs versus providing medication within hospitals. Denmark has, for instance, a high share of outpatient activity and most of the cancer drugs are provided by the hospitals. This is in contrast to Finland, Norway and Sweden where the cost share of prescription drugs is about 20 per cent. 3. Results 3.1. Country differences in estimated per capita treatment costs The estimated costs of cancer treatment in the Nordic countries in 2007 were three billion euros. The total population of the five countries is approximately 25 million. The estimated average per capita cost in the Nordic region was therefore about D 120. Sweden represents almost 40 per cent of the total population in the Nordic region. Iceland, on the other hand, represents only about 1 per cent. It follows that the results for Sweden heavily influence the estimate for the region as a whole. The country-specific per capita estimates range from D 133 in Norway to D 110 in Iceland. The estimated cost for Finland is D 104 without health centre activity and D 115 when health centre activity is included. Sweden and Denmark lie in the middle with estimated per capita costs of D 118 and D 124, respectively. Differences in per capita cancer treatment costs between the Nordic countries are therefore as high as 30 per cent. 3.2. Decomposition of per capita costs The results of the decomposition of per capita costs in accordance with Eq. (8) are shown in Table 2. We comment on the results for each country in turn, starting with Finland.
176
J. Kalseth et al. / Health Policy 115 (2014) 172–179
Table 2 Per capita treatment cost of cancer (EUR). Decomposition of treatment cost per capita. Indicators measured relative to Nordic weighted average (=1) 2007.
Cost per capita (EUR) Decomposition [Eq. (8)] Cost per capita A=B×C Productivity (cost per activity) B Activity rate (activity per capita) C=D×E D Service utilisation rate (activity per adj. preval.) E = E1 × E2 Adjusted prevalence rate E1 Raw five-year prevalence rate E2 Cancer prevalence case-mix index (CMI)
The size of the Finnish estimates is dependent on whether health centres are included or not. Health centre activity amounted to 27 per cent of total estimated activity for Finland, and the activity of the health centres was characterised by long stays at a low cost per day. The per capita costs were 5 per cent below the Nordic average when health centres were included and 15 per cent below the average when they were not. In both cases, costs were on the low side in a Nordic context. The decomposition analysis indicated that high productivity (low costs per activity) is an important explanatory factor. High productivity is present in Finland regardless of the inclusion or exclusion of health centre activity, even though estimated productivity was almost 20 per cent higher when the health centres were included. The estimated per capita activity was high when health centre activity was included, more than 20 per cent above the Nordic average. Excluding health centre activity, the estimated activity rate was low (about 10 per cent below the average), and this contributed to low per capita costs in this case. The case-mix-adjusted prevalence rate was close to the Nordic average even though the raw prevalence rate was low. Hence, the relative service utilisation rate in Finland was on a par with the relative position of its per capita activity. The per capita cost for Iceland was 9 per cent below the Nordic average. This follows from the high productivity and low case-mix-adjusted prevalence rate. Nevertheless, the service utilisation rate was high. Sweden was in a median position with regard to per capita costs, lying 3 per cent below the Nordic average. Two counteracting factors appear. First, productivity was low (cost per activity was high), but this was counteracted by the relatively low case-mix-adjusted prevalence rate and estimated service utilisation rate. Denmark had the second highest estimated per capita costs at 2 per cent above the Nordic average. Productivity was also close to average (1 per cent above), but the casemix-adjusted prevalence rate was high. The Danish service utilisation rate was relatively low. The Norwegian cost per capita was as much as 10 per cent above the Nordic average, which mainly relates to relatively low productivity. The case-mix-adjusted prevalence rate and service utilisation rate were close to the Nordic average. 4. Discussion By employing a common methodology in the data collection and analysis, we identified differences in the per
Denmark
Finland
Iceland
Norway
Sweden
124
115
110
133
118
1.02 0.99 1.03 0.94 1.10 1.00 1.10
0.95 0.78 1.22 1.21 1.01 0.94 1.07
0.91 0.94 0.97 1.08 0.90 0.83 1.09
1.10 1.08 1.01 1.03 0.99 1.00 0.99
0.97 1.15 0.85 0.89 0.95 1.05 0.91
capita treatment cost of cancer of up to 30 per cent between the Nordic countries. These estimates show relatively modest differences compared with previous estimates for Sweden and Finland based on non-comparable data and methodologies [25]. The ranking of the countries is largely the same as the OECD statistics for total health care expenditure per capita (Table 3, columns i and iii). Although the differences in per capita health care costs are narrower for cancer than costs in general, our results still suggest that Norway has the highest per capita health care costs of all the Nordic countries. Nevertheless, health care spending as a share of GDP is lower for Norway than for the other Nordic countries, except for Finland [1]. This suggests that the high spending level in Norway is related to the high per capita GDP. This is also a more general observation: national income level is found to be the strongest determinant of per capita health care spending [26]. The ranking of total per capita health expenditure among the Nordic countries is the same as for per capita GDP, taking expenditure on long-term care into account. More modest differences in cancer costs compared with total spending on health care may reflect the fact that cancer patients are highly prioritised in all countries, but it may also reflect the challenges of comparing total health care costs as such. The OECD data on per capita health care spending often serve as a reference in health policy analysis and comparative research. We have learned, however, that the organisation and delivery of health care services may affect the reported health care expenditures. This limits the comparability of international statistics on health care Table 3 Per capita cancer treatment costs and per capita total health care costs (OECD). Nordic countries measured relative to Norway 2007.
Denmark Finland Iceland Sweden
Cancer treatment Ia (i) (%)
Cancer treatment IIb (ii) (%)
Health care OECDb,c (iii) (%)
Health care OECD excluding LTCb,c,d (iv) (%)
93 78e 83 89
85 71f 75 84
77 60 69 70
78 70 74 86
a Deflator: Previously developed deflator for hospital costs [21] and only exchange rate adjustment for prescription drugs. b Deflator: GDP-PPP (Eurostat-OECD methodological manual on Purchasing Power Parities. 2012 edition). c OECD.StatExtracts. d LTC-services of long-term nursing care as defined in A System of Health Accounts. 2011 edition (OECD). e 87% if health centres are included. f 79% if health centres are included.
J. Kalseth et al. / Health Policy 115 (2014) 172–179
spending levels. Take, for instance, the different practices involved in estimating expenditures of long-term nursing care. The reported expenditures in the OECD statistics on health care and social services, respectively, appear to differ quite substantially. The share of total health care spending on long-term care, on the one hand, was between 19 and 25 per cent in Iceland, Denmark and Norway, while it was only 12 per cent in Finland and was as low as 8 per cent in Sweden in 2007 (OECD.StatExtracts). On the other hand, Sweden in particular, but also Finland, report high levels of long-term care as social expenditure, as opposed to Norway, Denmark and Iceland [1]. Excluding long-term care from health care expenditure reduces the differences in total spending levels between Norway and the other Nordic countries (Table 3, column iv). The choice of deflator also affects estimated cost differences. In the OECD statistics, health care expenditures measured in national currencies are converted into comparable health expenditures, using the PPP index for the entire economy (GDP). This measure is sensitive to any deviation in price levels in health care from the rest of the economy. The estimated cancer costs were adjusted for differences in input prices according to medical personnel wage differences in this study. The estimated cost difference of cancer treatment between Norway and the other countries is about 5–8 per cent higher if we use the GDPPPP deflator for all cost components (Table 3, column ii). When the application of sector-specific deflators (rather than GDP-PPPs) narrows the differences between Norway and the other Nordic countries, it implies that high wage levels in the health sector contribute substantially to the high spending on health care in Norway. All in all, when we take into account the differences in the deflators and exclude long-term care expenditure from the OECD health care spending data, the estimated national differences in cancer treatment costs are quite similar to the differences in reported expenditures on health care in the OECD statistics. The decomposition of per capita costs illustrates a trivial but still important point: looking only at per capita spending may be very misleading for policy purposes. Differences in per capita spending for each country may reflect differences in underlying factors that have very different policy implications. High spending levels do not necessarily reflect “superior” health care in the form of high service utilisation rates, and vice versa. It may, for example, reflect differences in productivity rather than differences in accessibility or standards of care. Furthermore, there may be differences for each country in morbidity or particular national health care needs that might legitimise the observed differences. High costs because of high morbidity may represent a fiscal challenge and a public health problem, but are typically weakly related to the performance of the health system [27]. Using per capita measures alone may therefore also be problematic for comparisons of service levels. Cancer prevalence rates as well as the composition of cancer prevalence differ between the Nordic countries. Some of the differences in five-year prevalence rates are quite striking [19]. The raw five-year prevalence rate was 26 per cent higher in Sweden than in Iceland. On the other
177
hand, Sweden had low shares of total prevalence for cancer sites with high estimated cost per prevalence (acute leukaemia, non-Hodgkin lymphoma, lung cancer, and the residual category) and high shares of prostate and skin cancer, both of which have large patient groups in terms of prevalence but a low estimated cost per prevalence. For Iceland, and partly for Denmark and Finland, the opposite is the case. Prostate cancer was the largest group in terms of five-year prevalence at the Nordic level, and the estimated cost per prevalent case was half of the average cost for all sites. The share of total prevalence for prostate cancer varied from 14 per cent in Denmark to 27 per cent in Sweden. When resources are scarce and becoming scarcer, exploiting available resources optimally is crucial. Productivity differences are therefore of particular interest. The results for cancer treatment seem to reflect more general features of hospital productivity in the Nordic countries. High productivity in Finland compared with Norway, Sweden and Denmark has already been found for general hospitals [21,28]. The ranking of the countries in terms of productivity in cancer treatment mirrors that of hospitals in general. Differences in productivity between Nordic hospitals seem to be due to country-specific factors. The best practice hospitals in Finland perform better than best practice hospitals in Denmark, Norway and Sweden, while the variations within each country are not significantly different [29]. This suggests that the hospitals are operating under different external conditions or different country-specific institutional frameworks. The study has several limitations. First, the number of five-year prevalent cases was used as a proxy for the number of patients receiving treatment because no patient identification was available in the dataset. Next, no distinctions were made concerning phase of disease. It is well known that cancer costs are phase sensitive [7] and, given a U-shape of phase-specific costs [30], it is possible that some of the people included who are living with a cancer diagnosis received limited or no hospital services at all. Biased results may arise if there are differences between countries in phase distributions within the cancer sites or in the share of five-year prevalent cases actually receiving treatment. It is also worth noting that the inferences drawn regarding service utilisation rates are sensitive to the choice of five-year prevalence as an indicator of morbidity and the method of case-mix adjustment. This is particularly true for the relative positions of Denmark and Iceland (compared with activity per capita). A third limitation is that the case-mix adjustment may also be affected by the relatively large residual group of cancer diagnoses for which identical costs per prevalence are assumed. The analysis should in this respect be regarded as an attempt to illustrate the importance of taking differences in morbidity into account in the cross-country comparisons of health care costs. Even though the case-mix-adjusted five-year prevalence rate may not be a perfect proxy, it does seem to capture differences in activity and costs resulting from differences in health care needs. We expect the differences in health care needs to be at least partly reflected in the level of activity per capita, and we find a positive correlation between activity per capita and the case-mix-adjusted
178
J. Kalseth et al. / Health Policy 115 (2014) 172–179
five-year prevalence rate. The results suggest that the countries can be divided into three groups with Norway in a middle position with regard to both adjusted prevalence and activity, Denmark and Finland having higher rates than Norway, and the adjusted prevalence rates and activity rates for Sweden and Iceland being lower. We do not find a similar relationship between the raw prevalence rate and activity. Hence, case-mix adjustment of prevalence rates seems to be clearly warranted. Although we have put considerable effort into ensuring that our estimates of cancer treatment activity and costs are comparable, differences in data accessibility for each country because of different national data registration and patient classification systems, as well as differences in health care organisation, may still bias the results. Limitations on data availability prohibited the inclusion of direct medical costs incurred by primary health care and longterm care to treat and care for cancer patients in our cost measure. Although costs related to outpatient primary care are relevant, estimates for Norway, for which we were able to collect data for cancer patients, indicate that outpatient primary care is likely to play only a limited role in cancer treatment in the Nordic countries because it amounted to only about 2 per cent of estimated cancer treatment costs. The lack of data on cancer-related costs of long-term institutional or home care is more likely to be influential, as illustrated both by the different shares of reported longterm care in the OECD statistics and by the sensitivity of the results for Finland to the inclusion or exclusion of health centre activity. 5. Conclusions The present study demonstrates that differences in health care costs may be due to several factors that have very different policy implications. It is important to be mindful of this when health care is compared across countries because the conclusions may depend heavily on the indicators used. This study shows, for instance, that per capita spending levels can be a poor proxy for service utilisation rates. To gain insight into the underlying factors, cost comparisons should involve a decomposition of costs into individual influencing factors: morbidity, service levels, productivity and, preferably, quality and outcomes. Such a decomposition analysis may be more feasible if the study is limited to a particular service or patient group. Close attention should also be given to the choice of conversion factors for health care costs because this may also have a significant bearing on the results in cross-country comparisons. Acknowledgements The authors would like to thank two anonymous referees and the journal editors for their contributions to this paper. Their perceptive comments and suggestions have been tremendously important in the revisions of the manuscript. The usual disclaimer applies. References [1] OECD. Health at a glance 2011: OECD indicators. OECD Publishing; 2011.
[2] World Health Organization. Global status report on noncommunicable diseases 2010. Geneva: World Health Organization; 2011. [3] World Health Organization. The global burden of disease: 2004 update. Geneva: World Health Organization; 2008. [4] American Cancer Society and Livestrong. The global economic cost of cancer 2010. [5] Møller B, Fekjær H, Hakulinen T, Tryggvadóttir L, Storm HH, Talbäck M, Haldorsen T. Prediction of cancer incidence in the Nordic countries up to the year 2020. European Journal of Cancer Prevention 2002;11(Suppl 1):1–96. [6] Rahimi A, Møller B, Bray F. Predicting cancer prevalence in 2025 for the Nordic countries. Norwegian Cancer Registry 2010. [7] Yabroff KR, Lund J, Kepka D, Mariotto A. Economic burden of cancer in the United States: estimates, projections, and future research. Cancer Epidemiology, Biomarkers, & Prevention 2011;20(10):2006–14, http://dx.doi.org/10.1158/1055-9965.EPI-11-0650. [8] Magnussen J, Vrangbæk K, Saltman R, editors. Nordic health care systems: recent reforms and current policy challenges. London: Open University Press; 2009. [9] Schieber GJ, Poullier JP. International health spending and utilization trends. Health Affairs 1988;7(4):105–12, http://dx.doi.org/10.1377/hlthaff.7.4.105. UG, Lundin D. Why did drug spending [10] Gerdtham increase during the 1990s?: a decomposition based on data. Pharmacoeconomics 2004;22(1):29–42, Swedish http://dx.doi.org/10.2165/00019053-200422010-00003. [11] Bongaarts J. Population aging and the rising cost of public pensions. Population and Development Review 2004;30(1):1–23, http://dx.doi.org/10.1111/j.1728-4457.2004.00001.x. [12] Thorpe KE, Howard DH. The rise in spending among Medicare beneficiaries: the role of chronic disease prevalence and changes in treatment intensity. Health Affairs 2006;25(5):378–88, http://dx.doi.org/10.1377/hlthaff.25.w378. [13] Dunn A, Liebman E, Shapiro AH. Decomposing medical-care expenditure growth. Working paper 2012-26. Federal Reserve Bank of San Francisco; November 2012. [14] Panopoulou E, Pantelidis T. Convergence in per capita health expenditures and health outcomes in the OECD Applied Economics 2012;44(30):3909–20, countries. http://dx.doi.org/10.1080/00036846.2011.583222. [15] Thorpe KE. Treated disease prevalence and spending per treated case drove most of the growth in health care spending in 1987–2009. Health Affairs 2013;32(5):851–8, http://dx.doi.org/10.1377/hlthaff.2012.0391. [16] Dunn A, Shapiro AH, Liebman E. Geographic variation in commercial medical-care expenditures: a framework for decomposing price and utilization. US Bureau of Economic Analysis; 2012 January. [17] Thorpe KE, Howard DH, Galactionova K. Differences in disease prevalence as a source of the U.S.—European health care spending gap. Health Affairs 2007;26(6):678–86, http://dx.doi.org/10.1377/hlthaff.26.6.w678. [18] Ó Céilleachair AJ, Hanly P, Skally M, O’Neill C, Fitzpatrick P, Kapur K, Staines A, Sharp L. Cost comparisons and methodological heterogeneity in cost-of-illness studies: the example colorectal cancer. Medical Care 2013;51(4):339–50, of http://dx.doi.org/10.1097/MLR.0b013e3182726c13. [19] Kalseth J, Halsteinli V, Halvorsen T, Kalseth B, Anthun K, Peltola M, Kautiainen K, Häkkinen U, Medin E, Lundgren J, Rehnberg C, Másdóttir BB, Heimisdottir M, Bjarnadóttir HH, Køtlum JE, Kilsmark J. Costs of cancer in the Nordic countries. A comparative study of health care costs and public income loss compensation payments related to cancer in the Nordic countries in 2007. Rapport SINTEF A19395. 2011. [20] Tarricone R. Cost-of-illness analysis. What room in economics? Health Policy 2006;77:51–63, health http://dx.doi.org/10.1016/j.healthpol.2005.07.016. [21] Kittelsen SAC, Anthun KS, Kalseth B, Kalseth J, Halsteinli V, Magnussen J. En komperativ analyse av spesialisthelsetjenesten i Finland, Sverige, Danmark og Norge: Aktivitet, ressursbruk og produktivitet 2005-2007. Rapport SINTEF A12200. 2009. [22] Pisani P, Bray F, Parkin DM. Estimates of world-wide prevalence of cancer for 25 sites in the adult population. International Journal of Cancer 2002;97(1):72–81, http://dx.doi.org/10.1002/ijc.1571. [23] Engholm G, Ferlay J, Christensen N, Bray F, Gjerstorff ML, Klint Å, Køtlum JE, Ólafsdóttir E, Pukkala E, Storm HH. NORDCAN: cancer incidence, mortality, prevalence and prediction in the Nordic countries, Version 3.7. Association of the Nordic Cancer Registries. Danish Cancer Society; 2010. http://www.ancr.nu.
J. Kalseth et al. / Health Policy 115 (2014) 172–179 [24] Engholm G, Ferlay J, Christensen N, Bray F, Gjerstorff ML, Klint Å, Køtlum JE, Ólafsdóttir E, Pukkala E, Storm HH. NORDCAN-a Nordic tool for cancer information, planning, quality control and research. Acta Oncologica 2010;49(5):725–36, http://dx.doi.org/10.3109/02841861003782017. [25] Wilking N, Jönsson B, Högberg D. Comparator Report on Patient Access to Cancer Drugs in Europe. 15 February 2009. [26] Hartwig J. What drives health care expenditure?—Baumol’s model of ‘unbalanced growth’ revisited. Journal of Health Economics 2008;27:603–23, http://dx.doi.org/10.1016/j.jhealeco.2007.05.006. [27] Smith PC, Mossialos E, Papanicolas I. Performance measurement for health system improvement: experiences, challenges and prospects. World Health Organization 2008 and World Health Organization, on behalf of the European Observatory on Health Systems and Policies; 2008.
179
[28] Medin E, Häkkinen U, Linna M, Anthun KS, Kittelsen SAC, Rehnberg C. International hospital productivity comparison: experiences from the Nordic countries. Healt Policy 2013;112(1-2):80–7, http://dx.doi.org/10.1016/j.healthpol.2013.02.004. [29] Kalseth B, Anthun KS, Hope Ø, Kittelsen SAC, Persson BA. Spesialisthelsetjenesten i Norden. Sykehusstruktur, styringsstruktur og lokal arbeidsorganisering som mulig forklaring på kostnadsforskjeller mellom landene. Rapport SINTEF A19615. 2011. [30] Brown ML, Riley GF, Schusseler N, Etzioni R. Estimating health care costs related to cancer treatment from SEER-Medicare data. Medical Care 2002;40(8 (Supplement:IV-104-IV-117)), http://dx.doi.org/10.1097/00005650-200208001-00014.
