How do Hospitals Respond to Price Changes? Evidence from Norway.

HEALTH ECONOMICS Health Econ. 25: 620–636 (2016) Published online 30 April 2015 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/hec.3179

HOW DO HOSPITALS RESPOND TO PRICE CHANGES? EVIDENCE FROM NORWAY JURGITA JANULEVICIUTEa,b, JAN ERIK ASKILDSENa,b,c, ODDVAR KAARBOEa,b,*, LUIGI SICILIANIa,d and MATT SUTTONa,e a

Health Economics Bergen, Bergen, Norway Department of Economics, University of Bergen, Bergen, Norway c Uni Research Rokkan Centre, Bergen, Norway d Department of Economics and Related Studies, University of York, York, UK e Manchester Centre for Health Economics, Institute of Population Health, University of Manchester, Manchester, UK b

ABSTRACT Many publicly funded health systems use activity-based financing to increase hospital production and efficiency. The aim of this study is to investigate whether price changes for different treatments affect the number of patients treated and the mix of activity provided by hospitals. We exploit the variations in prices created by the changes in the national average treatment cost per diagnosis-related group (DRG) offered to Norwegian hospitals over a period of 5 years (2003–2007). We use the data from Norwegian Patient Register, containing individual-level information on age, gender, type of treatment, diagnosis, number of co-morbidities and the national average treatment costs per DRG. We employ fixed-effect models to examine the changes in the number of patients treated within the DRGs over time. The results suggest that a 10% increase in price leads to about 0.8–1.3% increase in the number of patients treated for DRGs, which are medical (for both emergency and elective patients). In contrast, we find no price effect for DRGs that are surgical (for both emergency and elective patients). Moreover, we find evidence of upcoding. A 10% increase in the ratio of prices between patients with and without complications increases the proportion of patients coded with complications by 0.3–0.4 percentage points. Copyright © 2015 John Wiley & Sons, Ltd. Received 1 October 2013; Revised 19 December 2014; Accepted 24 February 2015 KEY WORDS:

hospitals; DRGs; prices; activity

1. INTRODUCTION In countries with publicly funded health systems, concerns about cost efficiency in the delivery of health services are frequently raised. Hospital payment has been changed or revised in several countries to improve efficiency. A policy instrument that is frequently introduced is activity-based financing (ABF) where hospitals with higher volumes of patients are rewarded with higher revenues. Often, such ABF mechanisms are linear with a fixed price for each additional patient. The ideas behind such a financing system are twofold: (i) to provide incentives for efficient hospital production, because hospitals are allowed to retain any surplus if they provide treatments with price above cost and (ii) to provide hospitals with incentives to treat more patients, because payments are proportional to the number of patients discharged. An ABF system is typically based on diagnosis-related groups (DRGs). A system based on DRGs classifies patients according to homogenous groups in terms of hospital stay and resources to treat patients. The DRG classification system can be used as the basis for ABF. For example, in 1983, Medicare adopted this system to serve as a basis of payment for US hospitals. Subsequently, it was introduced and is currently being used *Correspondence to: Department of Economics, University of Bergen, Bergen, Norway. E-mail: [email protected]

Copyright © 2015 John Wiley & Sons, Ltd.

HOW DO HOSPITALS RESPOND TO PRICE CHANGES?

621

in many Organisation for Economic Co-operation and Development (OECD) countries including Australia, Denmark, England, France, Germany, Norway and Sweden (Street et al., 2007). Under ABF, based on DRG classification system, a hospital’s income depends on the price of each DRG and the number of patients discharged within each DRG. DRG prices are fixed in advance and based on cost data collected from a representative group of hospitals in previous years. Hence, the price per DRG is meant to reflect the average costs of providing treatment for typical patients within the DRG. Furthermore, the price per DRG is independent of the costs incurred by any particular hospital. The DRG system is often referred to as a ‘prospective’ payment system. This is indeed the case if the price is set for a given patient’s diagnosis that is outside of hospitals’ control. In practice, a significant number of DRGs are based on treatment and also depend on patients’ complications and whether the patient is treated as day case or is admitted to hospital for at least one night. It has been argued that this generates an implicit cost sharing with higher costs being associated with higher revenues, for a given diagnosis (McClellan, 1997). The strengths and weaknesses of ABF systems are illustrated in several theoretical models (Chalkley and Malcomson, 1998a,1998b; Ma, 1994; Ellis and McGuire, 1986; Ellis, 1998, and Shleifer, 1985). The main benefits of this type of hospital financing are that it stimulates efficiency, because ABF is a form of yardstick competition (Shleifer, 1985) and that payments are patient-based (money follows the patient), which generates an incentive for hospitals to treat a larger number of patients. The main drawback is that within each category of service (DRG), there may be substantial variation in the appropriate provision of services. This creates selection problems because hospitals will have (financial) incentives to attract profitable patients (cream-skimming) and to skimp and/or dump unprofitable patients within each DRG. As long as hospitals care about profits to some extent, higher prices imply higher marginal revenues, which give incentives to increase activity (e.g. Ellis, 1998; Kaarboe and Siciliani, 2011; see, e.g. Section 4, in Chalkley and Malcomson (2000), for a review). We estimate how price changes in a given DRG affect the number of patients within the same DRG.1 We also investigate whether changes in the difference in DRG price between day case and overnight admission, for a given diagnosis, change the proportion of patients admitted as day cases. Similarly, we test whether changes in the difference in DRG price for patients with complications and without complications change the proportion of patients admitted with complications (commonly referred to as ‘upcoding’). The DRG price is defined as the national average treatment cost (NATC), which is paid to hospitals for each DRG in each year. We employ a large administrative database from Norway that records patient-level information of all hospitals over the five-year period 2003-2007. These data are aggregated at DRG level. Our main dependent variable is the number of patients treated within a DRG, in each hospital, each year. National average treatment costs are revised regularly in Norway. Major revisions took place in 2004, 2006 and 2007, and minor ones in 2005. The NATCs in the three years where major revisions took place are based on the average costs of a subsample of around 25 Norwegian hospitals in 2002, 2004 and 2005, respectively. There is therefore a time lag of 2 years for the changes in costs to be reflected in price changes. Price changes generated by DRG variations can, therefore, be considered to be predetermined. We estimate a range of empirical models to examine the effects of DRG price changes on the volume of patients treated. We include hospital fixed effects and DRG fixed effects to control for unobservable heterogeneity across hospitals and DRGs. We include year dummies to capture the time trend in a flexible way. In our preferred specification, we also include a linear time trend interacted with each DRG to control for different dynamics in volume across different DRGs over time. The results suggest that price increases of a given DRG lead to increases in the number of patients treated for the same DRG: a 10% rise in price leads to about 0.8–1.3% increase in the number of patients treated for DRGs

1

The aforementioned theoretical papers rest on the assumption that the price per DRG is not provider specific. Hence, the price (typically) has to be combined with a lump-sum transfer in order to ensure that the providers stay in the market. Miraldo et al. (2011) consider the case where lump-sum transfers are not allowed but the fixed prices can be adjusted to reflect exogenous cost differences between providers. This resembles the institutional settings in England and in the USA, where DRG-based prices are allowed to vary among providers to reflect exogenous cost differences. In Norway, all hospitals face the same DRG-based prices in the period studies.


Health Econ. 25: 620–636 (2016) DOI: 10.1002/hec

622

J. JANULEVICIUTE ET AL.

that are medical. For medical DRGs, we find a price response for both DRGs that refer to emergency and elective patients. In contrast, we find no price effect for DRGs that are surgical (for both emergency and elective patients). Moreover, we find that changes in the difference in the price between DRGs with and without complications for a given diagnosis increase the proportion of patients with complication. That is, we provide evidence of upcoding. The effect is detected for three out of four subgroups of DRGs considered (medicalelective, surgical-emergency and surgical-elective). A 10% increase in the ratio in prices for patients with and without complications increases the proportion of patients with complications by 0.3–0.4 percentage points. For medical DRGs, the estimated price effects are not only due to changes in the proportion of patients with complications: when we replicate the analysis for only the DRGs that are not differentiated by the presence of complications, we still find that higher prices increase the number of patients for a given DRG. 1.1. Relation to literature There are a few studies that investigate how changes in prices affect the number of patients treated. Dafny (2005) analysed how hospitals responded to changes in DRG prices for 43% of Medicare admissions. The study identified DRGs, based on a 1998 policy reform that had changed the descriptions of the DRGs to which patients were to be assigned. More specifically, the change involved eliminating the distinction between patients ‘aged over 69 years’ and ‘aged under 70 years’, so that the only qualifiers were ‘with complications’ or ‘without complications’. This resulted in substantial changes in the DRG prices for the two groups of patients. There was little evidence that US hospitals increased the volume of admissions for diagnoses that were subject to the price increases. The hospitals primarily responded by upcoding patients into diagnoses with the largest price increases (which is consistent with previous findings by Silverman and Skinner, 2004). The effect was highest among for-profit hospitals. Gilman (2000) investigates how the introduction of procedure-based DRGs for HIV-treatments affected the intensity of services provided for (new) the high paying procedural DRGs and for the lower paying non-procedural DRGs. The author finds that hospitals responded to the new incentives by increasing (decreasing) the intensity of services for higher (lower) paying DRGs (Hafsteinsdottir and Siciliani (2010) for a theoretical analysis of refinements of DRG-based prospective payment systems). Seshamani et al. (2006) look at whether different financial constraints for hospitals affect 30-day mortality rates and whether some patient groups are affected more than others. Their study investigates the reductions in Medicare payments that result from the Balanced Budget Act (BBA) of 1997. The authors found no significant changes in mortality rates between high- and low-BBA-impact hospitals. Lindrooth et al. (2007) examine whether exogenous change (cuts) in the average reimbursements for patient admissions affects how intensely patients are treated at the hospitals. They estimate the effect of the BBA by comparing treatment intensities in pre-BBA and post-BBA periods. The study uses the data from 11 states, for 16 disease categories for Medicare patients. The authors found that the intensity of treatment was reduced for the diagnoses that were more generously reimbursed before the cuts associated with the BBA, while there were no changes in treatment activity for unprofitable diagnoses. Farrar et al. (2009) examine the effects of the introduction of ‘payment by results’, a DRG-type-based hospital payment system based on fixed prices in England. They found some evidence that hospital activity grew more quickly when the system moved from a payment based on fixed budget to a DRG pricing system. Allen et al. (2015) analyse a policy change in England under which the price for one form of treatment was set proactively rather than reactively based on historically reported levels of cost. They found that hospitals responded strongly to this proactive price change. Liang (2015) finds that hospital caseload in orthopaedic departments increases with the profitability of DRGs in Taiwan. Mikkola et al. (2001) find that during the first 2 years after introduction of DRG-based financing in one Swedish county, a 20% increase in productivity was realized, compared with other counties. However, the effect was temporary. Some early studies from Norway (Biørn et al. (2003); Kjerstad (2003)) find that the introduction of ABF in Norway had a significantly positive effect on the number of patients treated and DRG points produced. Copyright © 2015 John Wiley & Sons, Ltd.



623

Most of the existing evidence on the effect of changes in hospitals’ prices relates to the USA. Health systems differ to a great extent across the OECD countries in terms of health insurance design and institutional arrangements. There is more limited evidence on the effect of hospital price changes within national health systems. These are typically characterised by a stronger public component with little presence of private providers and a strong presence of publicly owned hospitals. Hospitals remain complex organisations with different types of internal allocation rules. Whether financial incentives as induced by price changes bite within such public organisations is not clear a priori. This is even more the case within the Norwegian context where the DRG tariff only contributes to 40–60% of the funding. The nonprofit motive may be so pervasive that financial incentives may prove rather weak. On the other hand, policymakers in Norway and other OECD countries introduced DRG systems precisely to stimulate such public organisations to behave more efficiently. Our study fills this gap. The two early studies mentioned previously investigated the effect of the introduction of Norwegian DRG system in the early years. Now that the system has settled and the quality of the data has improved, it is time to explore whether significant changes in prices lead to further changes in hospital behaviour. Overall, we find evidence of incentive effects. Our analysis therefore suggests that price effects may be present even in hospitals that are publicly owned and when DRG prices are used to allocate resources for only about half of the hospital funding. The paper is organised as follows. Section 2 presents the institutional settings. Section 3 describes the data and descriptive statistics. Section 4 presents the methods. Section 5 provides and discusses the estimation results. Section 6 contains the concluding remarks. In the Appendix, we outline a simple theoretical model of a hospital respond to changes in the DRG weight.

2. INSTITUTIONAL SETTINGS Specialised healthcare providers in Norway are predominantly publicly owned and, since 2002, have been organised as state-owned enterprises within five regional health authorities (north, mid, west, south and east).2 A regional health authority has the responsibility for providing specialist health care to all patients within its region.3 The provision of health care is organised through health enterprises that are owned and governed by the regional health authorities. The regional health authorities can also contract with private suppliers to provide treatment. However, this outsourcing is quite small compared with the overall treatment activity and is also confined to only a few diagnoses. Every individual in Norway is entitled to care under the publicly funded health system. Health care is allocated on the basis of need or expected benefit, and not ability to pay. The public health system covers both elective and emergency treatments. Emergency patients need to be stabilised before discharge. Patient access to specialised health care is either through a referral system (elective care) or through emergency care. The main difference between the two forms of care is that emergency care is provided for the sudden onset of a medical condition of such nature that failure to render immediate care would reasonably result in deterioration of the injured person’s medical condition. Elective care is meant to capture medical conditions where the medical care can be planned and booked days, weeks or months in advance (Prop. 91 L, 2010-2011, Ministry of Health and Care Services, 2011). Another important feature is the patient’s right to free choice of hospital, which came into effect on a national level in 2001. However, relatively few patients have opted to receive treatment outside of their hospital catchment areas (Vrangbæk et al., 2007). The regional health authorities are financed by a mixture of block grants and a DRG activity-based system. The block grant is based on a risk-adjusted capitation formula. Together, the two types of funding are meant to 2 3

The number of regions is four since June 2007, when south and east were merged. See the study of Hagen and Kaarboe (2006) and Magnussen et al. (2007) for the descriptions of Norwegian hospital sector and the 2002 reform where hospital ownership was transferred from county councils to the central government.



624


cover the running cost of providing somatic hospital services (mental health is instead financed by block grants). The central government determines the relative shares of the two sources of funding (the activity-based and the capitation-based), on a yearly basis. To calculate the treatment costs within DRGs, the total running costs accrued for somatic care are collected from a sample of Norwegian hospitals. Some of the costs are directly related to specific DRGs, for example, the cost of pacemakers and dialysis. Other costs are directly related to the departments (specialities), but not to specific DRGs. One such example is nursing time at a speciality. These costs are typically allocated to the different DRGs based on each DRGs share of the activity (e.g. number of bed days) within the speciality. Other costs are only indirectly related to specific departments. Examples of such costs are the cost of the top management, the costs of the blood bank and the cost of the laundry departments. In 2004, the average relative cost share of the indirect costs was 14% (Anthun and Torvik, 2006). The indirect costs are first allocated to the different specialities based on standardized rules. The rules differ depending on which costs are being allocated, but the main allocation mechanisms are the specialities relative share of either the hospital’s total man year, activity measured as the number of bed days or floor area. The indirect costs are then allocated to the different DRGs based on how large share of the speciality’s activity of the different DRGs amounts to. When all costs are allocated to the different DRGs, the hospital’s DRG cost is determined as the sum of the costs for that DRG divided by the number of patients treated in the DRG. The national average cost for a specific DRG is calculated as the average costs for that DRG, where the average is taken over all hospitals that are part of the cost allocation project. The final part of the cost allocation project is to relate the national average cost of a DRG to the average cost of all hospital stays. This cost ratio will determine the DRG’s DRG weight. The DRG weights form the basis for the prices that are to be paid for future hospital treatments. More specifically, hospital prices are determined by the product of the average cost of all hospital stays, the DRG weights and the share of the ABF grant. There is a two-year lag between the collection of the cost data and the year in which they are used to set prices for the respective DRGs. DRG weights are not adjusted in relation to access of care or excess demand. In other words, DRG weights are not lifted upwards for treatments where access of care is perceived as poor or for treatments where waiting lists are long. Because hospitals are public, they cannot distribute any potential profits. Doctors within public hospitals tend to be salaried. Despite these institutional features, doctors within hospital may still have strong incentives to expand volume. This is because each department within the hospital has allocated the resources based on DRG points. In 2005, 80% of somatic specialities received resources based on the DRG points they produced (Harsvik and Kjekshus (2007, table 287)). Departments with low volumes are bound to be downscaled. Higher volumes imply that departments can expand. Senior doctors need to make sure that their department is adequately funded, generating an incentive to treat more generously reimbursed patients. Moreover, at the more aggregated hospital level, hospitals’ managers also have incentives to generate surpluses because these can be invested in higher hospital capacity. The Norwegian DRG system introduced an adjustment for outlier costs and payments in 2006. However, this applies only to a small proportion of patients within each DRG.4 Therefore, despite this adjustment, the payment system remains mainly prospective. However, like other DRG systems across the OECD countries, including Medicare in the USA, there is a retrospective element built into the Norwegian system. This is due to DRGs being based on ‘treatments’ as opposed to ‘diagnosis’ for a significant portion of DRGs and DRGs being split according to complications (in about half of DRGs). Both treatments and complications are choice

4

To qualify for outlier payments, a patient’s treatment has to belong to a DRG where the average length of stay plus 1.65 times the standard deviation of the length of stay is longer than 20 days. In addition, the patient’s length of stay has to be at least 10 days longer than the average length of stay plus 1.65 times the standard deviation of the length of stay (Directorate of Health, Ministry of Health and Care Services, 2006).



625


Table I. Changes in reimbursement system I

II

III

IV

V

Year

Cost year

Cost-based reimbursement per DRG point (in NOK)

Activity-based funding share

Cost-based price per DRG point (in NOK)

2003 2004 2005 2006 2007

2000 2002 2002 2004 2005

28 863 27 496 27 484 27 630 27 356

60% 40% 60% 40% 40%

17 318 10 998 16 490 11 052 10 942

DRG, diagnosis-related group; NOK, Norwegian krone. In column (III) real prices; deflator at municipality and regional level. See http://www.regjeringen.no/Upload/KRD/Vedlegg/KOMM/TBU/H-2218mOmslag.pdf

variables of the hospitals. Variations in prices can lead to change the composition of treatments, and the incentive to classify the patient as a complicated case.5 Table I summarises the changes in the financing system for the years from 2003 to 2007. Column II shows which year is used to calculate the cost-based DRG weights. These weights were updated every second year until 2006; from 2007 on, the weights were updated each year. Column III shows the reimbursement level for a hospital stay with a DRG weight of one (i.e. a hospital stay with the average cost of all hospital stays) for each year, in real terms. The cost-based reimbursement dropped in the first year but remained relatively stable thereafter. Column IV shows the ABF shares. These dropped from 60% to 40% between 2003 and 2004 rose to 60% in 2005 and then returned to 40% in 2006 and 2007. Column V in Table I shows the actual price for a hospital stay with a DRG weight of one. This cost-based price is the product of the ABF share (column IV) and the reimbursement per stay (column III). Changes in ABF shares lead to substantial changes in costbased prices per hospital stay. Indeed, in some years, both the cost calculations and the ABF share change, while in other years, either the cost calculations are updated or the ABF share is changed. We argue that the 40–60 shares are exogenously determined and external to the health system. For the period in question, they are the result of political processes where centre–right and centre–left coalitions respectively have determined the outcomes. Typically centre–left coalitions favour a low DRG share. More important than the 40–60 split for total activity levels are the yearly steering documents stipulating DRG production objectives for each regional health authority. Table II shows the changes in activity over time, where activity is expressed in DRG points and in the number of patients treated.6 We can see that the number of DRG points increases every year. Also, the increases in DRG points are slightly higher when the ABF share is high: the increase in DRG points is 2.1% when the ABF share decreased from 60% to 40% (2003–2004). As the ABF share increased from 40% in 2004 to 60% in 2005, the number of DRG points increases by 3.8%. Furthermore, the table shows that when the ABF share fell to 40% in 2006, the increase in total DRG points produced is at least 1% lower than the year before. The same trend is observed for the number of patients treated. When the ABF share was kept at 40% (2006–2007), the increase in both DRG points and the number of patients treated is lower. These trends suggest that hospital output in aggregate may be responding to changes in prices. However, these changes cannot be

Outlier adjustments have been referred to as ‘prospective’ cost sharing under which higher costs lead to higher reimbursements. Different reimbursements that arise from different treatments for a given diagnosis have instead being referred to as ‘retrospective’ cost sharing (McClellan, 1997; Gilman, 1999; Hafsteinsdottir and Siciliani, 2012). DRGs that differ based on complications are ‘prospective’ if whether a patient is assigned to a DRG with or without complication is exogenous. Recent studies however suggest that hospitals may upcode in response to financial incentives (Dafny, 2005). In such cases, DRGs with and without complications may account for ‘retrospective’ cost sharing. 6 A hospital stay’s DRG point is equal to the DRG weight of the stay. Hence, total hospital activity is measured as the sum of the DRG weights from all hospital stays. 5



626


Table II. Changes in activity: diagnosis-related group points and patients treated 2003–2004

2004–2005

2005–2006

2006–2007

From 60% to 40%

From 40% to 60%

From 60% to 40%

Kept at 40%

2.08% 3.04%

3.79% 4.02%

2.64% 4.12%

3.48% 1.52%

Years Change in ABF share: Change in total DRG points* Change in total patients treated

ABF, activity-based financing; DRG, diagnosis-related group. *DRG point is the sum of the DRG weights from all hospital stays.

distinguished from other factors that may be affecting aggregate hospital production. To identify whether hospitals are responding to price changes, we need to examine the changes by DRG. Based on the existing theory, we can identify different mechanisms through which higher prices can influence the number of patients treated. First, as suggested by Chalkley and Malcomson (1998a and 1998b), in the presence of excess demand and long waiting times (a key feature of publicly funded services), a higher price can induce providers to admit more patients by reducing the severity threshold over which the patient is recommended for treatment. In our empirical analysis, we will compare results when additional control for severity measures (such as changes in age, gender and co-morbidities). Second, when more than one DRG payment is available within the same diagnosis and the price markup is higher for the more expensive DRG, the provider may have an incentive to provide marginal patients with the more intensive treatment (Hafsteinsdottir and Siciliani, 2010). Within the Norwegian context, DRG prices are systematically different for patients treated as day case or with an overnight stay. We exploit this margin in our empirical analysis to test whether larger differences in prices between overnight-stay and day-case patients lead to a smaller proportion of patients admitted as day cases. Third, when DRG prices are differentiated for patients with and without complications, providers may have an incentive to classify more patients as with complications, and such incentive may be stronger the larger the difference in prices (Kuhn and Siciliani, 2013).

3. DATA The data are taken from Norwegian Patient Register, which covers the entire population of patients receiving hospital treatment during the period from 2003 to 2007. Hospitals that admitted fewer than 50 patients each year are excluded. Our final sample includes 63 hospitals. There are about 500 DRGs in each year. DRGs can be medical or surgical and elective or emergency (see Section 2 for definitions) and refer to patients admitted as day case or overnight.7 Our key dependent variable is the number of patients treated within each DRG, hospital and year. Rehabilitations and other specific treatments, where additional reimbursements apply, are excluded from the sample. These procedures are partly financed on the basis of the patient’s length of stay in hospital, not exclusively on a DRG basis. Such procedures amount to about 9% of the volume and apply mostly to elective treatments. Some observations are also excluded because of missing information about the DRG, the DRG weight or the hospital where the treatment was provided (0.4%). Our final sample consists of 6 008 674 patients. Descriptive statistics are presented in Table III. Forty-seven per cent of patients are male. The average patient age is about 50–51 years. The average number of co-morbidities is 1.14 in 2003 and increases slowly over time to 1.23 in 2007. The average DRG weight is relatively stable over time, around 0.81. The changes over time in average prices paid per admission are mainly driven by the changes in the ABF share. The number of patients treated has increased over time: there were 1 122 860 patients in 2003 and 1 272 144 patients in 2007. 7

Day cases are usually elective treatments, either as a separate treatment or as part of a treatment series. It is defined as treatments where duration at the hospital lasts for more than 5 h but the patient does not stay overnight. Outpatients are not included in the analysis because the DRG system did not apply for the outpatients during the analysis period.



627


Table IIIa. Descriptive statistics. Sample means* Year

Age (years)

Male

Number of co-morbidities

Length of stay (days)

DRG weight

Price

2003 2004 2005 2006 2007

50.02 50.35 50.84 51.10 51.53

0.47 0.47 0.47 0.47 0.47

1.14 1.15 1.18 1.23 1.23

3.40 3.23 3.13 3.01 2.88

0.81 0.81 0.81 0.79 0.81

14 642 9299 13 984 9352 9371

Number of patients 1 122 1 157 1 203 1 253 1 272

860 039 530 101 144

DRG, diagnosis-related group. *Hospitals that admit less than 50 patients are dropped. All statistics are weighted by the number of patients except the total number of patients. Number of DRGs in each year: 484 (2003), 492 (2004), 504 (2005), 513 (2006) and 517 (2007).

Table III b. Volume of patients by method of admission and type of care Type of patients Elective Emergency Births

Medical

Surgical

1 727 442 2 195 393 225 950

1 482 920 376 578 391

The numbers of patients by method of admission and the type of treatment are shown in Table IIIb. Elective patients amount to 53% of the total volume. Emergency patients account for 43%, and about 4% are births. Sixty-nine per cent of patients receive medical treatment, and 31% receive surgical treatment. Note however that while 53% of patients with medical treatments are emergency patients, only 20% of surgical treatments are for emergency patients (with therefore 80% being elective ones). Therefore, while medical treatments are relatively equally split between emergency and elective patients, surgical treatments are dominated by elective patients. Table IV provides more details about the changes over time in the DRG weights, across the whole distribution. Although the average DRG did not vary much from year to year, there were significant changes within each year. For example, between 2003 and 2004, the DRG weight reduced by at least 0.64 in 1% of the DRGs, by at least 0.15 in 10% of the DRGs and by at least 0.07 in 25% of the DRGs. Recall that the average DRG weight is 0.81. Therefore, a change of 0.15 for a specific DRG is substantial. Similarly, the DRG weight increased by at least 0.36 in 10% of the DRGs, by at least 0.99 in 5% of the DRGs and by at least 2.96 in 1% of the DRGs. Similar changes in DRG weights also occurred between 2005 and 2006 and between 2006 and 2007. The changes were much smaller between 2004 and 2005 as the costs were not updated between these years and only minor changes were made by the Ministry of Health on the DRG grouping system (Ministry of Health and Social Services, 2005).

4. METHODS We estimate a range of empirical models to examine the effect of DRG price changes on the volume of patients treated. The main explanatory variable is the price pjt paid to hospitals for each DRG j in each year t. The price is constructed in the following way. Define gjt as the weight attached to DRG j in year t, zt as the share of ABF (which varies from 0.4 to 0.6, see column IV in Table I) and Mt as the cost-based price per DRG point for that particular year (see column V in Table I).8 The price pjt is simply given by the product of the three items and can be written as pjt ¼ gjt zt M t : 8

Mt is based on the national average cost for producing a DRG point.



628


Table IV. Distributions of changes in diagnosis-related group weights between consecutive years Year 2003–2004 2004–2005 2005–2006 2006–2007

1%

5%

10%

25%

50%

75%

90%

95%

99%

Mean

0.64 0.02 1.56 1.08

0.25 0.01 0.38 0.36

0.15 0.00 0.21 0.17

0.07 0.00 0.08 0.06

0.00 0.00 0.01 0.01

0.08 0.00 0.06 0.05

0.36 0.00 0.27 0.18

0.99 0.00 0.63 0.49

2.96 0.23 1.83 1.40

0.11 0.01 0.04 0.01

There are three sources of variations in the price, pjt. The first (gjt) is the result of relative changes in DRG weights over time (which is different for different DRGs). The second (zt) is due to the national changes in the ABF component that apply to all treatments. The final one (Mt) is the cost-based price (i.e. the monetary converter) per DRG point. The last two components are absolute changes that affect all DRGs and vary only over time, while the first source of variation causes relative prices to change. As already shown in Table IV and discussed at the end of Section 3, the changes in DRG weights are frequent and substantial. We exploit this variation in order to separate the effect of price changes on the volume of activity from other changes over time. The models are estimated in log–log form, and the results are interpreted as elasticities. Initial Box–Cox tests supported the use of the log–log form.9 Our first model (Model 1) is ln yjkt ¼ β0 þ β1 ln pjt þ αj þ αk þ αt þ εjkt ;

(1)

where yjkt is the number of patients treated within DRG j at hospital k in year t and pjt is the price for DRG j at time t (and pjt = gjtztMt). To control for unobserved heterogeneity (for example, the differences in management style in hospitals, as well as different levels of activity across DRGs), we include hospital fixed effects, αj, and DRG fixed effects, αk. Year dummies are included to capture the time trend: αt is the year dummy for year t. εjkt is the error term. The effect of zt and Mt is captured by the year dummies, because they are perfectly correlated (note that 1n (pjt) = 1n(gjt) + 1n(zt) + 1n(Mt)). The only variation in price that we explore to estimate β1 is, therefore, the one that is due to the variation in the DRG weights gjt. Model 1 is also estimated for pairs of consecutive years, to see whether the price effect is different from 1 year to another. One potential concern is the endogeneity between the activity and the changes in the NATC that results from technological developments. For example, a new more effective technology may make a treatment cheaper and more easily administered, which could result in more patients treated and a lower NATC (a negative correlation between activity and NATC). In such cases, our estimates may be biased downwards. On the other hand, a new more expensive and effective technology might arise, which increases the demand for care and the cost, resulting in a higher volume and a higher NATC (a positive correlation between activity and NATC). In such cases, our estimates may be biased upwards. We tackle these potential biases in two different ways. First, NATCs are updated with a delay of 2 years; therefore, current prices are predetermined and do not reflect current costs. In addition, and perhaps more importantly, we allow for differential time trends across different DRGs, which allow us to control for differential trends in technological developments. We include additional interactions between each DRG and a linear time trend (αj * t) in Model 2 (our preferred specification): (2) ln yjkt ¼ β0 þ β1 ln pjt þ β2j ln αj * t þ αj þ αk þ αt þ εjkt : In Model 3, we add additional controls to Model 2 to allow for changes in patients’ characteristics (age, gender and number of co-morbidities). 9

The fit of Model 1 (Table VI) was maximised with a lambda value of 0.19. As the fit was practically equivalent with the log transformation (0.4431 vs 0.4540), we use the log transformation for ease of interpretation.




629

Then, we estimate Model 2 for four subgroups of patients: medical-emergency, medical-elective, surgicalemergency and surgical-elective patients. We also test whether the changes in activity driven by the variations in the DRG weights were driven by the variations in the patients’ characteristics by testing directly whether the changes in price affect patients’ age, gender and number of co-morbidities. We run models that are analogous to Model 1: (3) mjkt ¼ β0 þ β1 ln pjt þ αj þ αk þ αt þ εjkt ; where mjkt is either the following: (i) the mean patient age; (ii) the mean number of co-morbidities; and (iii) the proportion of male patients within DRG j, hospital k and year t. Moreover, we test whether the differences in prices between DRGs with complications and without complications for a given diagnosis affect the proportion of patients with complications. We identify 129 DRG pairs. We run the following model: h i nc (4) þ β2j ln αj * t þ αj þ αk þ αt þ εjkt ; cjkt ¼ β0 þ β1 ln pcc jt =pjt where cjkt is the proportion of patients with complications within DRG j, hospital k and year t; pcc jt is the DRG is the DRG price for patients without complications. price for patients with complications and pnc jt Analogously, we also test whether the differences in prices between DRGs that involve an overnight stay and a day case affect the proportion of patients treated as day cases. All DRGs have a differential payment along this dimension. Therefore, the sample size remains the same to the one in Equation (1). Finally, as a falsification test, we run the analysis using a sample of women who gave birth.10 As in the study of Dranove and Wehner (1994), we do not expect to find behaviour responses on the number of births.11

5. RESULTS 5.1. Price effect on the number of patients treated We expect to find a positive effect of the price on the number of patients treated, because increases in price for a given DRG generate stronger incentives to treat patients. The first column in Table VI shows the results for Model 1 that controls for hospitals’ fixed effects and DRG fixed effects (Equation 1). Changes in prices generated by the variations in the DRG weights have a positive and significant effect (at 5% level) on the number of patients treated. The overall price elasticity is 0.042: when the price increases by 10%, the volume increases by 0.42%. We use 2003 as the reference year. The year dummies over the period 2004–2007 are positive and suggest that activity increased steadily over time (by 10.6% within a period of 5 years). The following four columns in Table V look at the changes from 1 year to another (i.e. between 2003 and 2004, 2004 and 2005, 2005 and 2006 and 2006 and 2007). The effect of the price generated by DRG variations on number of patients treated is in the range of 0.058–0.78 in all years except for years 2004–2005, where the coefficient is negative. When positive, the price effect is statistically significant at 5% level except for years 2005–2006. There was very little change in DRG weights between 2004 and 2005, and these few changes drive the negative counter-intuitive coefficient between these 2 years. The last two columns of Table VI show the results for Models 2 and 3, where we add a linear trend interacted with time and patients’ characteristics. The price elasticity is stable and equal respectively to 0.043 and 0.049. Overall, the results in Table VI support the hypothesis that increases in prices generated by the variations in DRGs have a positive effect on the number of patients treated within each DRG. Hospitals could encourage or discourage home births, but it would still be less likely to find a significant effect for this sample. The mean DRG weight for this sample has increased from 2.23 in 2003 to 3.56 in 2007. 11 Childbirths were used to test the validity of the two-stage least square approach when estimating physician-induced demand (Dranove and Wehner, 1994). 10



630


Table V. Regression results of price effect on overall activity-dependent variable, ln(yjkt), log of number of patients treated within diagnosis-related group j, hospital k and year t Model 1 ln (price)

0.042* (2.091) Ref 0.005*** (4.187) 0.049*** (3.946) 0.101*** (5.868) 0.106*** (5.801)

2003 2004 2005 2006 2007 2

Adjusted R Observations

0.441 146 258

2003–2004

2004–2005

2005–2006

2006–2007

0.070* (2.081)

0.333* (2.482)

0.058 (1.547)

0.078* (2.177)

Ref

—

—

Ref

—

—

0.150** (2.759) —

Ref

—

—

0.059** (3.201) —

0.064*** (3.660) —

0.440 58 298

0.443 58 582

0.443 58 679

0.005 (0.678) 0.443 58 590

Model 2

Model 3

0.043* (2.051) Ref 0.116 (1.715) 0.178 (1.305) 0.296 (1.451) 0.366 (1.344) 0.442 146 258

0.049* (2.260) — 0.111 (1.691) 0.034 (0.010) 0.054 (0.005) 0.068 (0.005) 0.445 101 995

Robust t-statistics clustered by hospital in parentheses. Number of observations: the total number of diagnosis-related group (DRG) hospital and year combinations. Model 1 contains fixed effects for DRGs and hospital fixed effects (not reported). Model 2 contains fixed effects for DRGs and hospital fixed effects, and linear time trends interacted with DRG (not reported). Model 3 adds other covariates to Model 1: age, gender and number of co-morbidities. *p < 0.05; **p < 0.01; ***p < 0.001.

We also investigate whether the price effect is uniform across different types of patients. We estimate Model 2 (our preferred specification) for four subgroups of patients: medical-elective, medical-emergency, surgicalelective and surgical-emergency patients. The results are presented in Table VI. It shows that higher DRG prices increase the number of patients for medical DRGs and not for surgical DRGs. Moreover, the price effect is larger (0.13) for elective patients than emergency patients (0.08). Therefore, a 10% increase in price increases the number of patients with medical treatment by 0.8–1.3%. The result that the price effect impacts on medical DRGs seems plausible. A hospital’s decision about whether to admit a patient for medical treatment is more open to discretion compared with surgical treatment.12 Moreover, the price effect on medical DRGs is higher for elective patients than for emergency patients where the latter should be less sensitive to price. It may be surprising that medical emergency patients also respond to price.13 However, a hospital’s decision about whether to admit a patient for emergency treatment is open to discretion in some instances. For example, a patient may be sent to the hospital by a GP who has doubts about appropriate treatment. These cases will vary in seriousness, and it may be that when prices increase, the hospitals are willing to admit more emergency patients, instead of sending them back to their GPs.

5.2. Differential price effect on proportion of patients admitted as day cases and with complications The price effect found on the number of patients treated could be due to different mechanisms. It could be explained by the changes in severity thresholds that would allow treating more patients at the margin who would otherwise not be treated at all (or not treated in hospital; for example, the family doctor of the patient could prescribe a drug treatment), or it could be due to a reallocation of patients across DRGs: more precisely, (a) patients may be allocated to a more generous DRG that involves an overnight stay, as opposed to day-case DRG, and/or (b) patients could be allocated to a DRG with complications, as opposed to a DRG without complications. We explore whether these two mechanisms have some merits. 12 13

Overuse of supply sensitive care is also reported in Darthmouth Atlas Project (darthmouthatlas.org). Analysing a cross section of southeastern hospitals in the USA, Hegji (2007) also finds that emergency room visits generate profits for the hospitals.



631


Table VIa. Price effect on number of patients. Subgroup analysis

ln (price) 2004 2005 2006 2007 2


Medical-elective

Medical-emergency

Surgical-elective

Surgical-emergency

0.129* (2.545) 0.200* (1.995) 0.217 (1.104) 0.405 (1.351) 0.494 (1.233) 0.591 35 953

0.079** (3.005) 0.011 (0.176) 0.092 (0.804) 0.088 (0.505) 0.170 (0.738) 0.779 53 933

0.075 (1.517) 0.178 (1.956) 0.435* (2.573) 0.595* (2.305) 0.810* (2.362) 0.684 30 246

0.039 (0.775) 0.013 . (0 127) 0.031 (0.172) 0.056 (0.204) 0.066 (0.182) 0.690 24 727

Robust t-statistics clustered by hospital in parentheses. Includes fixed effects for diagnosis-related groups, fixed effects for hospitals and linear trend interacted with each diagnosis-related group. *p < 0.05; **p < 0.01; ***p < 0.001.

Table VIb. Effect of price differences on proportion of day cases o

d

ln (price /price ) 2004 2005 2006 2007 2


Medical-elective

Medical-emergency

Surgical-elective

Surgical-emergency

0.047** (2.718) 0.023* (2.072) 0.006 (0.524) 0.034 (1.598) 0.035 (1.612) 0.814 35 953

0.005 (0.912) 0.014 (1.797) 0.014 (1.035) 0.023 (1.092) 0.028 (1.027) 0.737 53 933

0.017 (0.780) 0.070** (2.851) 0.114* (2.440) 0.176* (2.576) 0.243* (2.518) 0.829 30 246

0.001 (0.074) 0.051* (2.060) * 0.097 (2.009) 0.146* (2.009) 0.193* (2.018) 0.491 24 727

Robust t-statistics clustered by hospital in parentheses. o d o d o d ln (price /price ) = ln (price ) – ln(price ), where price is the price of diagnosis-related group (DRG) with overnight stay and price is the price of DRG for day cases. Weighted by the number of patients. Includes fixed effects for DRGs, fixed effects for hospitals, and linear trend interacted with each DRG. *p < 0.05; **p < 0.01; ***p < 0.001.

Table VIb shows the results of the effect of changes in prices on the proportion of patients admitted as day cases for the four subgroups of patients. We never find that a higher difference in price between DRGs with an overnight stay and without (i.e. day case) for a given diagnosis reduces the proportion of patients admitted as day cases, as we would intuitively expect (in one occasion, i.e. medical-elective patients, we find the opposite). Within the Norwegian context, the changes in price differentials are due mainly to the changes in DRGs that involve an overnight stay because DRGs for day cases have few exceptions stayed unchanged. Note also that each DRG in Norway is split between day cases and overnight stay, and therefore, the sample size in Tables VIa and VIb is identical. To conclude, the price effects on the number of patients treated described in Section 5.1 are not due to the changes in the allocation of patients within a given diagnosis between DRGs, which involve a day case and an overnight stay. Analogously, in Table VII, we explore whether the differences in price, for a given diagnosis, between DRGs with complications and without complication affects the proportion of patients with complications. Not all diagnoses or treatments have DRGs split by complications. Within the Norwegian context, we identify 129 pairs of DRGs. Table IX suggests that a 10% increase in the ratio in prices for patients with and without complications increases the proportion of patients with complications by 0.3–0.4 percentage points for three Copyright © 2015 John Wiley & Sons, Ltd.


632


Table VII. Effect of price differences on proportion of patients with complications (129 diagnosis-related group pairs) cc

nc

ln (price /price ) 2004 2005 2006 2007 2


Medical-elective

Medical-emergency

Surgical-elective

Surgical-emergency

0.032* (2.134) 0.002 (0.093) 0.015 (0.474) 0.002 (0.053) 0.023 (0.383) 0.632 12 051

0.015 (1.194) 0.017 (1.869) 0.032 (1.910) 0.046 (1.801) 0.055 (1.747) 0.861 15 208

0.031** (2.651) 0.032 (1.483) 0.073 (1.708) 0.117 (1.827) 0.158 (1.808) 0.806 8433

0.039* (2.188) 0.069 (1.074) 0.154 (1.213) 0.241 (1.268) 0.306 (1.203) 0.721 7167

Robust t-statistics clustered by hospital in parentheses. cc nc cc nc cc nc ln (price /price ) = ln (price ) – ln(price ), where price is the price of diagnosis-related group (DRG) with complications and price is the price of DRG without complications. Weighted by the number of patients. Includes fixed effects for DRGs, fixed effects for hospitals, and linear trend interacted with each DRG. *p < 0.05, **p < 0.01, ***p < 0.001

out of four subgroups of patients (medical-elective, surgical-elective and surgical-emergency). The coefficient is positive but smaller and not significant for medical-emergency patients. Although the number of surgical patients tends to be unresponsive to price changes (as suggested in Table VII) because of the demand for surgery being relatively inelastic, providers still respond to the changes in price through a different margin, that is, the allocation of patients to DRGs with and without complications. Finally, we test whether the changes in prices for DRGs that are not split by complications exhibit different effects on the number of patients compared with the main analysis that includes all DRGs in Table VI. Table VIII suggests that the results for such DRGs are very similar to those presented in Table VI. An increase in the DRG price by 10% increases the number of patients with medical DRGs by 0.8–1.6%. We therefore conclude that the price effects in Table VI cannot be explained by the changes in the allocations of patients between DRGs with and without complications (and note moreover that Table IX suggests that ‘upcoding’ is present in particular among surgical DRGs where the price effects outlined in Table VI are absent).

Table VIII. Price effect on number of patients for diagnosis-related groups that are not split by complications

ln (price) 2004 2005 2006 2007 2


Medical-elective

Medical-emergency

Surgical-elective

Surgical-emergency

0.166* (2.566) 0.114 (1.957) 0.030 (0.280) 0.145 (0.987) 0.163 (0.816) 0.622 16 314

0.081* (2.450) 0.094*** (3.600) 0.087** (2.620) 0.175*** (3.675) 0.177** (2.790) 0.762 25 083

0.178* (2.126) 0.120 (1.322) 0.438** (3.287) 0.556** (2.588) 0.768** (2.731) 0.676 15 087

0.061 (0.727) 0.068 (1.140) 0.044 (0.607) 0.079 (0.705) 0.106 (0.716) 0.631 11 877

Robust t-statistics clustered by hospital in parentheses. Includes fixed effects for diagnosis-related groups, fixed effects for hospitals, and linear trend interacted with each diagnosis-related group. *p < 0.05; **p < 0.01; ***p < 0.001. Copyright © 2015 John Wiley & Sons, Ltd.


633


Table IX. Price effect on patients’ characteristics and sensitivity analysis

agejkt co-morbiditiesjkt malejkt Sensitivity analysis (Model 1) Births only

Coefficient on ln (price)

Number of observations

0.041 (1.6745) 0.1428*** (5.4564) 0.005 (0.0870)

146 258 146 258 146 258

0.071 (0.490)

1399

agejkt: mean age within diagnosis-related group (DRG) j, hospital k and year t. co-morbiditiesjkt: mean number of co-morbidities within DRG j, hospital k and year t. malejkt: proportion of males within DRG j, hospital k and year t. Robust t-statistics clustered by hospital in parentheses. Number of observations: the total number of DRG hospital and year combinations. All models contain fixed effects for DRG and hospital groups and year dummies (not shown). *p < 0.05; **p < 0.01; ***p < 0.001.

5.3. Sensitivity analysis Finally, we test whether the price changes induced by the variations in DRG weights also affect the characteristics of the patients who are treated. Table 9 shows the results when we estimate the effect of price changes on the average patients’ age, the number of co-morbidities and the proportion of male patients. We find that higher prices increase co-morbidities but have no effect on age and proportion of male patients. This suggests that providers may be admitting more severe patients as a result of the increase in price. We then estimate the model on the sample of births, which we expect to be exogenous and, therefore, not influenced by the changes in prices. The estimation was in line with our expectations, and we found that the price effect was not significant.

6. CONCLUDING REMARKS We investigate how relative changes in DRG prices affect the number of patients treated within the same DRG in Norwegian hospitals. In most of the countries where healthcare services are publicly provided, the demand for health care exceeds their ability to supply it. It is therefore important to explore whether price mechanisms might be used as a way to increase activity in the hospitals. As predicted by an economic theory (Chalkley and Malcomson, 1998a and 1998b), increased prices per treatment with a prospective payment system give incentives to increase the number of those treatments. On the other hand, because of the restriction of capacity in hospitals and weak profit motives within publicly owned hospitals, the price could have little or no effect on the number of patients treated. The existing studies (Kjerstad, 2003; Biørn et al., 2003) show that the introduction of ABF had a positive effect on an activity in Norway. In this paper, we investigate whether the changes in DRG prices have any impact on the number of patients treated. The changes in prices are created by the changes in DRG weights and the ABF share. If hospitals increase activity when prices increase, either through the better use of their existing resources or by adjusting to the increased demand for some specific procedures, this finding might provide important information for policymakers. The results suggest that a 10% rise in DRG price leads to about 0.8–1.3% increase in the number of patients treated for DRGs that are medical. In contrast, we find no price effect for DRGs that are surgical. The result is plausible because whether a patient is admitted for medical treatment is more open to discretion compared with surgical treatment. We also investigate whether the changes in price change the composition of patients. We find clear evidence of upcoding, that is, the incentive to classify a patient with complications (which is line with the study of Dafny, 2005). The effect is detected for different subgroups of DRGs: medical-elective, surgical-emergency and surgical-elective. A 10% increase in the ratio in prices for patients with and without complications increases Copyright © 2015 John Wiley & Sons, Ltd.


634


the proportion of patients with complications by 0.3–0.4 percentage points. Therefore, even though the number of surgical patients is unresponsive to the changes in price, its composition is. Overall, our analysis suggests that the changes in DRG prices influence hospital behaviour. Some limitations and venues for future research should be mentioned. First, we do not have available information on costs at specific hospitals. Average reimbursements for some treatments might be more profitable for some hospitals than for others. Second, we have no data on outpatient visits, and we could not explore possible volume effects in this dimension or possible substitution effects between inpatient care and outpatient visits. Third, we have not examined the effect on the quality of the treatments provided. More research with better data is needed on these issues. A final important issue that could to be addressed in the future is the price effect on waiting times. Using the information on waiting times in Norway, Martinussen and Hagen (2009) found evidence of cream skimming during the first period after the introduction of ABF. In Norway, patients may choose a hospital where treatment is to be provided, and it would be useful to determine whether hospitals compete for patients. The main qualitative indicator for attracting patients is the waiting time for the treatment. Thus, if hospitals attempt to attract more patients within specific diagnosis, they will try to have shorter waiting times for those treatments. If increases in price lead to shorter waiting time for a treatment, they may be used as a mechanism to influence patients’ prioritisation.

APPENDIX A SIMPLE MODEL OF HOSPITAL BEHAVIOUR In this appendix, we put forward a simple theoretical model of how a hospital will respond to the changes in the DRG weight. Consider a hospital that treat i = 1, 2, …, I different patient groups with DRG weight D1, D2, …, DI. The hospital (might) care about patients health and obtain (altruistic) utility β∑Ii¼1 ni . The hospital’s profit is given by T+ pα∑Dini ϕ(n1, n2, …, nI), where T > 0 is a block grant, p > 0 is the price per DRG point, α > 0 is the DRG share, ϕ(n1, n2, …, nI) is the cost function and ni is the number of patients of type i the hospital treat. For simplicity, we assume p = 1 and that φðn1 ; ; n2 ; …; ; nI Þ ¼ ∑Ii¼1 ϕ i ðni Þ, that is, the cost independence between the patient groups. The hospital maximizes I

max

n1 ;n2 ;…;nI

U¼β ∑ ni þ T þ pα∑Di ni ϕ ðn1 ; ; n2 ; …; ; nI Þ: i¼1

The first-order conditions are given by αDi ϕ ′ ðni Þ ¼ 0; i¼1; 2; …; I: The second-order condition is given by I

∏ ϕ ″ ðni Þ: i¼1

Obviously, the second-order condition is satisfied if φ() is strictly convex that we assume. By differentiating the first-order conditions and using the implicit function theorem, we obtain I

α∏ ϕ} ðni Þ dn1 ¼

i¼2 I

∐ ϕ} ðni Þ

dD1 ⇒

dn1 α > 0: ¼ dD1 ϕ} ðn1 Þ

i¼1

From the aforementioned equation, we obtain the following results: Copyright © 2015 John Wiley & Sons, Ltd.



• •

635

The curvature of the cost function and the DRG weight determine how much the hospital will respond to the increases in the DRG share. This is quite obvious as the DRG weight determines the marginal income and the curvature determines how much marginal cost changes when production changes. The higher the DRG share, the larger the increase in the number of patients treated for a given increase in the DRG weight. How much production is increased will of course also depend on the cost changes; more specifically, how much marginal costs changes.

REFERENCES

Allen T, Fichera E, Sutton M. 2015. Can payers use prices to improve quality? Evidence from English hospitals. Health Economics doi: 10.1002/hec.3121 Anthun KS, Torvik LH. 2006. Fordeling av felleskostnader i helseforetak». SINTEF Helse. (Allocation of indirect costs in health enterprises). Biørn E, Hagen TP, Iversen T, Magnussen J. 2003. The effect of activity-based financing on hospital efficiency: A panel data analysis of DEA efficiency scores 1992– 2000. Health Care Management Science 6: 271–83. Chalkley M, Malcomson JM. 1998a. Contracting for Health Services with Unmonitored Quality. The Economic Journal 108: 1093–1110. Chalkley M, Malcomson JM. 1998b. Contracting for Health Services when Patient Demand does not reflect Quality. Journal of Health Economics 17: 1–19. Chalkley M, Malcomson JM. 2000. Government Purchasing of Health Services In (ed.) Culyer AJ, Newhouse JP. Handbook of Health Economics, Chapter 15, Volume 1, Part A, 847–890. Elsevier. Dafny LS. 2005. How Do Hospitals Respond to Price Changes? American Economic Review 95(5): 1525–1547. Dranove D, Wehner P. 1994. Physician-Induced Demand for Childbirths. Journal of Health Economics 13: 61–73. Ellis R, McGuire TG. 1986. Provider Behaviour under Prospective Reimbursement: Cost Sharing and Supply. Journal of Health Economics 5: 129–151. Ellis R. 1998. Creaming, Skimping and Dumping: Provider Competition on the Intensive and Extensive Margins. Journal of Health Economics 17: 537–556. Farrar S et al. 2009. Has payment by results affected the way that English hospitals provide care? Difference-in-differences analysis. British Medical Journal 339: b3047. Gilman BH. 2000. Hospital Response to DRG Refinements: the Impact of Multiple Reimbursement Incentives on Inpatient Length of Stay. Health Economics 9: 277–294. Gilman BH. 1999. Measuring hospital cost-sharing incentives under refined prospective payment. Journal of Economics and Management Strategy 8(3): 433–452. Hafsteinsdottir E, Siciliani L. 2012. Hospital cost sharing incentives: evidence from Iceland. Empirical Economics 42(2): 539–561. Hagen TP, Kaarboe O. 2006. The Norwegian hospital reform of 2002: central government takes over ownership of public hospitals. Health Policy 76: 320–333. Hafsteinsdottir EJG, Siciliani L. 2010. DRG Prospective Payment Systems: Refine or not Refine? Health Economics 19: 1226–1239. Harsvik T, Kjekshus LE. 2007. "INTORG – Organisasjon og ledelsesstrukturer ved norske sykehus". SINTEF Helse. (INTORG – Organisational- and managementstructures at Norwegian hospitals). Hegji CE. 2007. Hospital Activity and Hospital Profits. Journal of Hospital Marketing & Public Relations 17(2): 3–11. Kaarboe O, Siciliani L. 2011. Multi-Tasking, Quality and Pay for Performance. Health Economics 20: 225–238. Kjerstad E. 2003. Prospective Funding of General Hospitals in Norway: Incentives for Higher Production? International Journal of Health Care Finance and Economics 3: 231–251. Kuhn M, Siciliani L. 2013. Manipulation and Auditing of Public Sector Contracts. European Journal of Political Economy 32: 251–267. Liang LL. 2015. Do Diagnosis-Related Group-Based Payments Incentivise Hospitals to Adjust Output Mix? Health Economics 24: 454–469. Lindrooth RC, Bazzoli GJ, Clement J. 2007. The Effect of Reimbursement on the Intensity of Hospital Services. Southern Economic Journal 73(3): 575–587. Ma AC. 1994. Health Care Payment Systems: Cost and Quality Incentives. Journal of Economics and Management Strategy 3(1): 93–112. Magnussen J, Hagen TP, Kaarboe O. 2007. Centralized or Decentralized? A Case Study of Norwegian Hospital Reform. Social Science & Medicine 64: 2129–2137. Copyright © 2015 John Wiley & Sons, Ltd.


636


Martinussen PE, Hagen TP. 2009. Reimbursement Systems, Organisational Forms and Patient Selection: Evidence from Day Surgery in Norway. Health Economics, Policy, and Law 4: 139–158. McClellan M. 1997. Hospital reimbursement incentives: an empirical analysis. Journal of Economics and Management Strategy 6(1): 91–128. Mikkola H, Keskimaki I, Hakkinen U. 2001. DRG-Related Prices Applied in a Public Health Care System- can Finland learn from Norway and Sweden? Health Policy 59: 37–51. Ministry of Health and Social Services 2005. "Informasjonshefte om Innsatsstyrt Finansiering 2005." ["Information on Activity-based financing 2005."] Ministry of Health and Care Services 2006. "Innsatsstyrt finansiering 2006". (Activity-based financing rule book 2006). Ministry of Health and Care Services 2011. "Lov om pasientrettigheter". Proposition 91 L. (Act on Municipal health and care services). Miraldo M, Siciliani L, Street A. 2011. Price Adjustment in the Hospital Sector. Journal of Health Economics 30(1): 112–125. Shleifer A. 1985. A Theory of Yardstick Competition. Rand Journal of Economics 16(3): 319–327. Seshamani M, Schwartz JS, Volpp KG. 2006. The Effect of Cuts in Medicare Reimbursement on Hospital Mortality. Health Service Research 41(3): Part I, 683–700. Silverman E, Skinner J. 2004. Medicare upcoding and hospital ownership. Journal of Health Economics 23: 369–389. Street A, Vitikainen K, Bjorvatn A, Hvenegaard A. 2007. Introducing Activity-Based Financing: A Review of Experience in Australia, Denmark, Norway and Sweden. CHE Research paper 30. Vrangbæk K, Østergren K, Birk HO, Winblad U. 2007. Patient Reactions to Hospital Choice in Norway, Denmark, and Sweden. Health Economics, Policy, and Law 2: 125–152.



HOW DO SMOKERS RESPOND TO CIGARETTE TAXES? EVIDENCE FROM CHINA'S CIGARETTE INDUSTRY.

Do consumers of manufactured cigarettes respond differently to price changes compared with their Roll-Your-Own counterparts? Evidence from New Zealand.

How do cells respond to ultrafine surface contours?

How do biting disease vectors behaviourally respond to host availability?

How do older adults respond to active Australia physical activity questions? Lessons from cognitive interviews.

How to do it. Give evidence.

Do not attempt cardiopulmonary resuscitation documentation: do doctors respond to national changes?

How to respond at interview.

Do wheezy infants recovering from bronchiolitis respond to inhaled salbutamol?

price laws work? Evidence from retail scanner data and TUS-CPS.

Weight conversations in romantic relationships: What do they sound like and how do partners respond?

How do cattle respond to sloped floors? An investigation using behavior and electromyograms.

A narrative inquiry: how do nurses respond to patients' use of humour?

Q&A: How do plants sense and respond to UV-B radiation?

Q&A: How do plants respond to ethylene and what is its importance?

Q&A: How do plants respond to cytokinins and what is their importance?

How do people respond to health news? The role of personality traits.

How Do Payers Respond to Regulatory Actions? The Case of Bevacizumab.

How do patients respond when confronted with telephone access barriers to care?

Do thymic malignancies respond to target therapies?

How to do a quick search for evidence.

How to respond at interview: 2.

How tree roots respond to drought.

Daylight and absenteeism--evidence from Norway.