- Email: [email protected]
Effects of DRG-based hospital payment in Poland on treatment of patients with stroke
Health Policy 119 (2015) 1119–1125
Contents lists available at ScienceDirect
Health Policy journal homepage: www.elsevier.com/locate/healthpol
Effects of DRG-based hospital payment in Poland on treatment of patients with stroke Victor Bystrov a,∗ , Anna Staszewska-Bystrova a , Daniel Rutkowski b , Tomasz Hermanowski c a b c
Faculty of Economics and Sociology, University of Lodz, Rewolucji 1905 r. 41, 90-214 Lodz, Poland National Health Fund, Poland Department of Pharmacoeconomics, Medical University of Warsaw, Poland
a r t i c l e
i n f o
Article history: Received 2 December 2014 Received in revised form 24 April 2015 Accepted 28 April 2015 Keywords: Hospital payment Diagnosis Related Groups Monitoring of outcomes Length of stay Stroke
a b s t r a c t A prospective payment system based on Diagnosis Related Groups (DRGs) presents strong financial incentives to healthcare providers. These incentives may have intended as well as unintended consequences for the healthcare system. In this paper we use administrative data on stroke admissions to Polish hospitals in order to demonstrate the response of hospitals to the incentives embedded in the design of stroke-related groups in Poland. The design was intended to motivate hospitals for the development of specialized stroke units by paying significantly higher tariffs for treatment of patients in these units. As a result, an extensive network of stroke units has emerged. However, as it is shown in the paper, there is no evidence that outcomes in hospitals with stroke units are significantly different from outcomes in hospitals without stroke units. It is also demonstrated that the reliance on the length of stay as a major grouping variable provides incentives for regrouping patients into more expensive groups by extending their length of stay in stroke units. The results of the study are limited by the incompleteness of the casemix data. There is a need to develop information and audit systems which would further inform a revision of the DRG system aimed to reduce the risk of regrouping and up-coding. © 2015 Elsevier Ireland Ltd. All rights reserved.
1. Introduction The Polish system of hospital payment based on DRGs was launched in mid-2008 with aims of improving resource allocation to hospitals and increasing transparency of services provision. The design of the system, including criteria for classifying patients into specific DRGs and relative sizes of DRG-based payments, was intended to provide hospitals
∗ Corresponding author. Tel.: +48 426355154. E-mail addresses: [email protected] (V. Bystrov), [email protected] (A. Staszewska-Bystrova), [email protected] (D. Rutkowski), [email protected] (T. Hermanowski). http://dx.doi.org/10.1016/j.healthpol.2015.04.017 0168-8510/© 2015 Elsevier Ireland Ltd. All rights reserved.
with incentives for containing costs without deterioration of the quality of care (see [1]). An effective payment system has to align financial incentives with the best medical practice. If a system fails to achieve such alignment, healthcare providers may have adverse incentives to exploit information asymmetry in order to increase profits. Since the payer (in this study, National Health Fund – Narodowy Fundusz Zdrowia – NFZ) has limited information about healthcare needs of an individual patient, hospitals may have incentives to underprovide healthcare services to some patients in order to reduce costs and over-provide to other patients in order to increase revenues. As the DRG-based reimbursement depends on the information coded by hospitals, there is also a risk of up-coding, i.e., exploiting the coding system
1120
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
by hospitals in order to achieve higher reimbursement. The problem of asymmetric information in the context of prospective payment system was discussed, inter alia, in [2–4]. The design of stroke DRGs varies widely across European countries (see [5]). In this paper we attempt to evaluate how Polish hospitals responded to incentives provided by the DRG system in stroke-related groups. For this purpose we use aggregate statistics on these groups as well as the patient-level data collected by the NFZ in 2010. According to the World Health Organization [6], stroke is the second leading cause of death in the world. In Poland, stroke is the third most common cause of death and the most common cause of disability in adults aged over 40 [7]. The research conducted within the International Stroke Trial and Polish National Stroke Prevention and Treatment Registry indicates that stroke mortality in Poland is relatively high (as compared to other countries) and there is large variation in stroke outcomes across Polish hospitals (see [8,9]). At the same time, Epstein et al [10] report that the costs of stroke treatment in Poland are high in relation to the national income per capita (as compared to other European countries). The development of the DRG system in Poland was accompanied by the creation of a network of specialized stroke units, which should have enabled high quality treat´ ment of stroke patients. According to Starzynska-Długosz et al. [11], the number of stroke units (eligible for thrombolysis) increased from 56 in 2007 to 150 in 2012. At the same time, the design of stroke-related DRGs provided strong financial incentives for treating patients in the stroke units. As a result, the number of patients treated in the stroke units and the number of patients classified into higherpaying stroke DRGs increased rapidly. In the years 2009–2013 the amount of reimbursement paid by the NFZ to hospitals for stroke-related groups increased by 14.93 percent (from PLN 526.11 million in 2009 to PLN 604.65 million in 2013). This growth can be mainly attributed to the redistribution of patients from lower-paying groups to higher-paying groups. While the total number of stroke hospitalizations decreased by 2.75 percent (from 94,963 in 2009 to 92,351 in 2013), the number of patients classified into higher-paying stroke DRGs increased by 37.34 percent (from 38,837 in 2009 to 53,340 in 2013). In the Polish DRG system, there are four stroke-related groups, which are defined by major diagnosis, procedures, treatment in a stroke unit, and length of stay (LoS). We use the distribution of LoS for ischemic stroke patients in order to demonstrate that the most frequent LoS is equal to a threshold value for classifying patients in higher-paying stroke DRGs. The use of LoS as a major criterion for classifying stroke patients may give hospitals an incentive to reclassify patients into a more expensive DRG. The development of the network of stroke units has coincided with a decrease of in-hospital mortality from 17.37 percent in 2009 to 15.81 percent in 2013. However, there is no evidence of significant differences in the distribution of outcomes in hospitals with stroke units and hospitals without stroke units. The outcomes are measured by in-hospital mortality rate and 30-days readmission
rate for patients with ischemic stroke. These measures are standardized by age and gender. The limitations of the data collection process implemented in Poland do not allow application of the risk-adjustment of outcomes, using comorbidities, and the resulting variation in outcome measures cannot be solely attributed to the quality of care. Comorbidities are not used as grouping variables in case of stroke-related DRGs in Poland. Though there is a possibility of reporting comorbidities to the NFZ, hospitals have no financial incentive to do so. We discuss the distribution of intensity of comorbidities coding for stroke admissions across Polish hospitals as well as the problems of data collection and quality monitoring. The paper is organized as follows. Section 2 provides a description of materials and methods. Dynamics of strokerelated DRGs, distributions of length of stay and stroke outcomes are analyzed in Section 3. Discussion of problems concerning quality monitoring is given in Section 4. Section 5 concludes the paper. 2. Materials and methods The study is based on two types of data: aggregated DRG statistics and anonymized patient-level data. The aggregated indicators are accessible on the webpage of the NFZ [12]. The database contains inter alia, the number of episodes and the average reimbursement per episode in each DRG. The patient-level data include hospital and department identifiers, age, gender, dates and modes of admission and discharge, major diagnosis, comorbidities, and DRG. The data set includes patients aged 35–100 years who were hospitalized with a diagnosis of cerebral infarction (ICD10 codes: I63.x) and discharged between January 1, 2010 and December 31, 2010. Patients transferred from another hospital are excluded from the data set. For the study of readmission rates and LoS, those patients who died before discharge are excluded from the data. The objectives of the data selection were to limit the extreme casemix variation and to use data concerning those patients who can be classified to all four strokerelated groups. 2.1. Description of Polish stroke-related DRGs The Polish system of hospital payment (JGP, Jednorodne Grupy Pacjentów) is a modification of the English system based on Healthcare Resource Groups (HRG, version 3.5). Definitions of some groups are directly adopted from the English system. Other groups, including stroke-related groups, are modified after consultations with local experts. In the Polish system, patients are included in a group on the basis of main and secondary diagnoses defined by International Classification of Diseases, version 10 (ICD10), procedures defined by the Polish modification of ICD-9, gender, age, type of admission, length of stay, and type of discharge. The size of prospective payment for a group is determined on the basis of historical data. In the absence of cost data, a system of quasi-prices and relative scores of DRG
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
groups is used (see [1,13]). All hospitals receive identical reimbursement for treating a patient in the same group. Stroke-related groups in English HRG 3.5 (A19, A22, and A23) are defined on the basis of major diagnosis, comorbidities, and age of patients. Length of stay is not used in the definitions of stroke groups in England. In the Polish system neither comorbidities nor age are used in the definitions of stroke-related groups. These groups are defined on the basis of the major diagnosis, procedures and length of stay. Two most expensive stroke-related groups in Poland (A48 and A51) require treatment in a specialized stroke unit (see Table 1). The different design of stroke groups in Poland originates from the necessity to encourage hospitals in creation of specialized stroke units. At the moment of the launch of the DRG-based system, specialized stroke units were scarce in Poland. The relatively high weights of groups requiring treatment in stroke units were intended to incentivize hospitals for treating patients in these units. Groups A48 and A51 are designed to provide patients with high-quality treatment in specialized stroke units for at least 8 days. These groups are the most expensive: the average reimbursement in group A48 is more than twice larger than in group A49 and more than four times larger than in group A50; the average reimbursement in group A51 is more than three times larger than in group A49 and more than six times larger than in group A50 (see Table 1). Large differences in reimbursement payments between groups are aimed to provide strong incentives to hospitals for quality treatment of patients in specialized stroke units. (For long stays exceeding 27 days in groups A49 and A50, and 36 days in groups A48 and A51, there is a supplementary per-diem payment). The NFZ specifies a set of requirements which should be met by hospitals to provide treatment of patients in groups A48 and A51. These requirements include qualified medical personnel, the number of beds, the type of medical equipment, and documentation. Hospitals and regional departments of the NFZ negotiate contracts, which specify the maximal volume of services in groups A48 and A51. The use of the LoS as a criterion of patient classification may provide incentives to hospitals for reclassifying patients into a more expensive group by extending their LoS. In particular, for a hospital with a stroke unit, a patient can be reclassified from group A49 into group A48 by extending her hospital stay and/or performing an additional (unnecessary) diagnostic procedure. All procedures qualifying for group A48 are also included in the list of procedures qualifying for group A49. The difference is in the minimal number of procedures: one procedure from the list is enough for qualifying a patient in A49 and two procedures are enough for qualifying a patient in A48. On the other hand, if a patient received an intensive treatment in a stroke unit, but did not survive through the first week of hospitalization, a hospital would encounter losses, because the patient would not qualify to group A48. This may put at risk high-severity patients with lower chances of survival through the first week of hospitalization, as hospitals may have a financial incentive of under-providing services to such patients. Group A51 requires expensive thrombolysis and hospitals have to
1121
treat surviving patients for at least 8 days in order to classify these patients in A51 and recuperate costs of thrombolysis. 2.2. Measurement of LoS In order to evaluate the effect of the DRG-based hospital payment on the LoS of stroke patients, we compare the distribution of LoS in hospitals with stroke units, which can classify patients in more expensive groups (A48 and A51), and hospitals without stroke units, which can only classify patients in less expensive groups (A49 and A50). For this purpose, we use data on non-deceased patients in hospitals with stroke units and hospitals without stroke units. The distributions of LoS in two types of hospitals are represented by histograms, which demonstrate differences in the empirical distributions of LoS. 2.3. Measurement of stroke outcomes To asses the performance of hospitals treating stroke patients, two outcome measures are computed: in-hospital mortality rate and 30-days all-cause readmission rate. The in-hospital mortality is computed on the basis of discharge modes and the information on readmissions is obtained by linking the NFZ data on the original stroke admissions with the data on the follow-up admissions. The outcome measures are standardized with respect to age and gender using a two-level logit model allowing for between-hospital variation in health outcomes (see, e.g., [15,16]). Standardization is carried out using all available observations. We compare the distribution of standardized mortality and readmission rates across hospitals with stroke units and the distribution of standardized rates across hospitals without stroke units. These comparisons are performed for hospitals with no fewer than 30 episodes of ischemic stroke. The reasons for such an approach are twofold. On the one hand, we tried to control sampling errors by excluding hospitals with a small number of stroke episodes. On the other hand, we aimed to retain a representative data set. The estimation of age and gender effects on mortality is carried out using data on 45,776 admissions to 360 hospitals and the comparison of standardized mortality rates uses information on 44,930 admissions to 273 hospitals. Of these 273 hospitals, there are 121 hospitals without stroke units which had 13,709 admissions, and there are 152 hospitals with stroke units which had 31,221 admissions. For the estimation of standardized readmission rates, data on 39,014 admissions are used (after the exclusion of deceased). The total number of observations used in the comparison of standardized rates is equal to 38,118 which include 26,652 admissions to 149 hospitals with stroke units and 11,466 admissions to 116 hospitals without stroke units. The central tendency, variation and the outliers for the standardized rates are estimated using statistics employed in the construction of boxplots. To this end, the central tendency is estimated using the median, and the variation is measured by the 25 percent and 75 percent quantiles of the
1122
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
Table 1 Characteristics of Polish stroke-related DRGs (2010). Code
Minimal LoS
Stroke unit
Type of stroke
Procedures
Median LoS
Average payment
A48 A49 A50 A51
8 days 4 days Not specified 8 days (if survived)
Required Not required Not required Required
Haemorrhagic or ischemic Haemorrhagic or ischemic Haemorrhagic or ischemic Only ischemic
At least 2 procedures At least 1 procedure Not specified Thrombolysis
11 9 5 11
8349.16 PLN 4023.92 PLN 1940.38 PLN 12,311.16 PLN
empirical distribution. The outliers are detected by identifying the data points which fall outside of the boxplot whiskers. These extend to the most extreme data points which are no more than 1.5 times the interquartile interval. 3. Results 3.1. Dynamics of stroke-related DRGs The stroke treatment has undergone major structural changes since the launch of the DRG system with a large increase in the share of patients treated in stroke units and a corresponding increase in the share of patients classified into high-paying DRGs (see Table 2). The fastest growing group is A51 which has grown by 546 percent in 2009–2013. However, this group is still relatively small and constitutes only 4.79 percent of all stroke admissions in 2013, because only a limited number of patients qualify to this group. The second fastest growing group is A48 with an increase of 28.19 percent in 2009–2013. As a result of this fast growth, 52.96 percent of all stroke admissions were concentrated in A48 in 2013 (as compared to 40.17 percent in 2009). This increase in groups A48 and A51 is compensated by a decrease in group A49 which does not require an extended stay in a stroke unit. The number of stroke admissions in this group decreased by 41.44 percent in 2009–2013 and the share of this group in the total number of stroke admissions decreased from 45.17 percent in 2009 to 27.20 percent in 2013. These changes result from reclassifying patients from less expensive to more expensive groups. As the number of stroke units increases, more patients can be treated in these units and classified in more expensive groups. However, as discussed in Section 3.2, the use of LoS as a classifying variable together with large differences in tariffs between groups provide incentives for gaming. Section 3.3 demonstrates that the distribution of outcomes is not significantly different between hospitals with stroke units and hospitals without stroke units.
Table 2 Number of hospitalizations in Polish stroke-related DRGs (2009–2013). Year
A48
A49
A50
A51
2009 2010 2011 2012 2013
38,151 38,687 42,319 46,944 48,908
42,892 32,417 31,434 28,088 25,117
13,234 19,695 17,194 15,310 13,894
686 1204 2285 3182 4432
Total 94,963 92,003 93,232 93,524 92,351
3.2. Analysis of the distribution of LoS The mean LoS for ischemic stroke patients, based on the data collected by the NFZ in 2010 (after the launch of the DRG-based system), was 11.6 days as compared to 14.4 days, based on the data collected by the National Stroke Prevention and Treatment Registry in 2002 (before the launch of the DRG-based system, see [9]). This number also differs from the mean LoS for ischemic stroke patients in England (HRGs A22 and A23), which was 15.3 days in 2010–2011, as reported in Hospital Episode Statistics by the Health and Social Care Information Centre [14]. The DRG-based classification of stroke hospitalizations, using LoS as a grouping variable, implies not only a reduced mean LoS but a peculiar shape of the distribution of LoS. Fig. 1 shows histograms of LoS for not deceased patients in hospitals without stroke units and hospitals with stroke units (those hospitals which had contracts with the NFZ for hospitalizations classified in groups A48 and A51). The histogram of LoS in hospitals without stroke units has no distinct mode. However, there is a distinct peak in the histogram of LoS for hospitals with stroke units: 26.5 percent of patients stay in these hospitals for 8 days. The shape of the distribution of LoS has two potential (non-medical) explanations: (1) reallocation of patients to a higher-paying DRG by extending their stay by an extra day; (2) early discharge: once a patient can be classified into a higher-paying DRG determined by the threshold LoS, there is no financial incentive to prolong the stay of the patient. It should be noted that hospitals with stroke units may not be able to treat all patients in these units. Two types of restrictions apply: a physical restriction (number of beds in a stroke unit) and a financial restriction (the total value contracted for groups A48 and A51). In result, stroke patients admitted to a hospital with a stroke unit may be treated in a neurology department or a department of internal medicine. For hospitals with a large volume of stroke hospitalizations, it implies the need to select those patients who will be admitted to a stroke unit and those who will not. This selection may be motivated by financial incentives rather than by clinical needs. In particular, there is a risk that the most severe patients with the greatest clinical needs will be treated outside stroke units because they may not qualify into groups A48 and A51.
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
0.25 0.20 0.00
0.05
0.10
0.15
Density
0.15 0.00
0.05
0.10
Density
0.20
0.25
0.30
Hospitals with stroke units
0.30
Hospitals without stroke units
1123
0 4 8
16
24
32
0 4 8
Number of days
16
24
32
Number of days
Fig. 1. Effect of stroke units on length of stay, not deceased patients.
0.4 0.3 0.2 0.1 0.0
Standardized mortality rate
Fig. 2 shows boxplots of standardized mortality rates for hospitals with and without stroke units. The median mortality rate in hospitals with stroke units is 13.43 percent as compared to 14.06 percent in hospitals without stroke units, and the interquartile intervals are respectively [10.31%,17.54%] and [10.14%,18.48%]. The differences in these descriptive statistics are not large enough to conclude that the distribution of mortality rates differs between hospitals with stroke units and hospitals without stroke units. This observation is supported by the results of Mann–Whitney U test: the hypothesis of the same distribution is not rejected (given the significance level of 10%). Boxplots for standardized readmission rates are shown in Fig. 3. The median readmission rate for hospitals with stroke units is 14.11 percent and the median readmission rate for hospitals without stroke units is 14.28 percent, and the corresponding interquantile intervals are [11.46%, 16.90%] and [10.83%, 17.80%]. As in the case of mortality rate, the differences in these descriptive statistics are not large enough to conclude that the distribution of readmission rates differs between hospitals with stroke units and hospitals without stroke units. The hypothesis of the same
0.5
3.3. Analysis of the distribution of stroke outcomes
Hospitals without stroke units
Hospitals with stroke units
Fig. 2. Boxplots of mortality rates.
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
0.3 0.2 0.0
0.1
Standardized readmission rate
0.4
1124
Hospitals without stroke units
Hospitals with stroke units
Fig. 3. Boxplots of readmission rates.
distribution is not rejected by the Mann–Whitney U test (given the significance level of 10%). The variation in stroke outcomes can be attributed either to the variation in the quality of care or the variation in the casemix. In order to separate these two factors, a risk adjustment of outcomes based on the casemix data has to be carried out. However, as it is shown in the next section, the current information system of the NFZ does not collect enough information to perform risk adjustment of outcomes. 4. Discussion: problems of quality monitoring In the Polish system, hospitals submit reimbursement claims together with statistical reports to regional branch offices of the NFZ. For each clinical episode, a statistical report has to include demographic data (age and gender), dates and modes of admission and discharge, codes of major and secondary diagnoses as well as codes of the procedures performed. The Regulation of the Minister of Health [17] allows reporting up to three comorbidities per episode. These data are stored in the Registry of Medical Services, the database of the NFZ. The NFZ controls data quality by performing plausibility checks of coded information and auditing selected hospitals. There is another system collecting information about all clinical episodes on the basis of statistical discharge cards. This system is administered by the Ministry of Health. However, information in this system is protected by the privacy law and cannot be used (under the existing legislation) for verification of the information in the database of the NFZ. The data collected by the NFZ are not protected by the privacy law. These data contain personal identifiers which allow restoring history of hospitalizations for a patient and tracking follow-up readmissions and mortality. The data could potentially be used for outcome-based
quality monitoring. However, incompleteness of the data on comorbidities makes it impossible to implement risk adjustment of outcomes. Since comorbidities are not included in the list of grouping variables for stroke-related groups in Poland, hospitals have no direct financial incentives to report them. Time pressure and the underfinancing of the coding process can contribute to the low intensity of coding as well. The average number of comorbidities per episode was computed for 273 hospitals with at least 30 stroke episodes. 16 hospitals did not report comorbidities for any patient. Other hospitals had heterogeneous coding practices. However, the total number of reported comorbidities was small with no comorbidities reported for 19,479 out of 44,924 episodes. The low intensity of coding results in low prevalence of stroke comorbidities in the administrative data as compared to the prevalence of these comorbidities in studies based on clinical data. For example, the prevalence of diabetes computed on the basis of administrative data is equal to 11.28 percent as compared to 24.1 percent reported in Niewada et al [9], and the prevalence of hypertension is equal to 37.5 percent as compared to 69.5 percent in [9]. These results are consistent with the results reported by Peltola [18], who, using administrative data from 2009, finds low values of the Charlson index of comorbidities for stroke patients in Poland as compared to other European countries. In the absence of complete casemix data, the NFZ uses raw outcome indicators. The variation in these indicators can be explained by the variation in the quality of care as well as by the variation in the casemix. The evaluation of quality based on raw outcome indicators may produce biased results. A coordinated effort by the Ministry of Health and the NFZ is needed in order to develop an integrated information system collecting comprehensive casemix data. Such system would enable the risk adjustment of outcomes and control of quality based on outcome measures. The information system should be supported by a systematic audit in order to control the quality of information submitted to the Ministry of Health and the NFZ by healthcare providers. The combination of these two systems could further inform policy decisions concerning the design of the payment system. 5. Conclusions The design of a DRG-based system embeds strong financial incentives for healthcare providers. These incentives may serve intended as well as unintended consequences for the healthcare system. In this paper, we use the data on stroke admissions in Poland in order to demonstrate how hospitals behave under the DRG-based system. The design of stroke-related groups in Poland has served the intended aim of increasing the number of patients treated in specialized stroke units. However, the use of length of stay as a grouping variable resulted in the most frequent length of stay being equal to the threshold value for reclassifying patients into more expensive groups. This
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
type of hospital behavior, which is dictated by the financial incentives embedded in the design of stroke-related groups in Poland, may have adverse consequences for patients well-being. There is an extensive network of stroke units in Poland, which can claim higher reimbursement for treatment of stroke patients. However, as demonstrated in this study, there is no evidence that the distribution of stroke outcomes in hospitals with stroke units significantly differs from the distribution of stroke outcomes in hospitals without stroke units. The study is restricted by the incompleteness of casemix data. Since outcome measures are not adjusted for casemix, their variation cannot be attributed solely to the variation in the quality of care. There is a need for further development of the information system of the NFZ which would allow systematic collection of complete casemix data. The design of stroke-related groups in Poland requires modifications in order to reduce risk of regrouping and upcoding, and achieve effective use of restricted resources which can be allocated to stroke units. The refinement of grouping criteria for DRGs and the revision of tariffs should align financial incentives of hospitals with the best medical practice. Acknowledgements This study is part of the International Research Project on Financing Quality in Healthcare (InterQuality), whose main purpose is to investigate the effect of different financing methods and incentives on quality, effectiveness and equity of access to health care. The project is funded by the European Commission under the Seventh Framework Programme (FP7) for Research and Technological Development, grant agreement number: 261369. More information about the Project can be found at http://interqualityproject.eu/. References ´ [1] Czach K, Klonowska K, Swiderek M, Wiktorzak K. Poland: The Jednorodne Grupy Pacjentów – Polish experiences with DRGs. In: Busse R, Geissler A, Quentin W, Wiley M, editors. Diagnosis-related groups in Europe. McGraw-Hill Open University Press; 2011. p. 359–80.
1125
[2] Ellis R. Creaming, skimping and dumping: provider competition on the intensive and extensive margins. Journal of Health Economics 1998;17:537–55, http://dx.doi.org/10.1016/ S0167-6296(97)00042-8. [3] Cots F, Chiarello P, Salvador X, Castells X, Quentin W. DRG-based hospital payment: intended and uninitended consequences. In: Busse R, Geissler A, Quentin W, Wiley M, editors. Diagnosis-related groups in Europe. McGraw-Hill Open University Press; 2011. p. 75–92. [4] Or Z, Häkkinen U. DRGs and quality: for better or for worse? In: Busse R, Geissler A, Quentin W, Wiley M, editors. Diagnosis-related groups in Europe. McGraw-Hill Open University Press; 2011. p. 115–29. [5] Peltola M, Quentin W. Diagnosis-related groups (DRGs) for stroke in Europe: patient classification and hospital reimbursement in 11 countries. Cerebrovascular Diseases 2013, http://dx.doi.org/10.1159/000346092. [6] WHO. Fact sheet No310, http://www.who.int/mediacentre/ factsheets/fs310/en/ [7] Udary mózgu – konsekwencje społeczne i ekonomiczne, Lazarski University, Warsaw, 2013. [8] Członkowska A, Niewada M, El-Baroni IS, Mendel T, Ryglewicz D, Sandercock P, et al. High early case fatality after ischemic stroke in Poland: exploration of possible explanations in the International Stroke Trial. Journal of the Neurological Sciences 2002;202:53–7. ´ ´ M, Ryglewicz D, Kaminski B, Członkowska [9] Niewada M, Skowronska A. Acute ischemic stroke care and outcome in centers participating in the Polish National Stroke Prevention and Treatment Registry. Stroke 2006;37:1837–43. [10] Epstein D, Mason A, Manca A. The hospital costs of care for stroke in nine European countries. Health Economics 2008;17:21–31, http://dx.doi.org/10.1002/hec.1329. ´ ´ [11] Sarzynska-Długosz I, Skowronska M, Członkowska A. Organization of acute stroke services in Poland Polish Stroke Unit Network development. Neurologia i Neurochirurgia Polska 2013;47:3–7, http://dx.doi.org/10.5114/ninp.2013.32934. [12] NFZ. Statystyka JGP, http://prog.nfz.gov.pl/APP-JGP/ ˛ [13] Kozierkiewicz A, Stamirski M, Stylo W, Trabka W. The definition of prices for inpatient care in Poland in the absence of cost data. Health Care Management Science 2006;9:281–6, http://dx.doi.org/10.1007/s10729-006-9095-z. [14] Hospital Episode Statistics. Admitted Patient Care - England, 2010–11, Health and Social Care Information Centre, http://www.hscic.gov.uk/searchcatalogue?productid=98 [15] Iezzoni LI. Risk adjustment for measuring healthcare outcomes. Health Administration Press; 1997. [16] Normand S-LT, Shahian DM. Statistical and clinical aspects of hospital outcomes profiling. Statistical Science 2007;22:206–26. ˛ Ministra Zdrowia z dnia 20 czerwca. r. w [17] Rozporzadzenie ˛ sprawie zakresu niezbednych informacji gromadzonych przez ´swiadczeniodawców, szczegółowego sposobu rejestrowania tych ˛ do informacji oraz ich przekazywania podmiotom zobowiazanym finansowania s´ wiadczen ze s´ rodków publicznych. Dziennik Ustaw Nr 123 Poz 2008;801:6636–42. [18] Peltola M. Patient classification and hospital costs of care for stroke in 10 European countries. Health Economics 2012;21(Suppl. 2):129–40, http://dx.doi.org/10.1002/hec.2841.
Contents lists available at ScienceDirect
Health Policy journal homepage: www.elsevier.com/locate/healthpol
Effects of DRG-based hospital payment in Poland on treatment of patients with stroke Victor Bystrov a,∗ , Anna Staszewska-Bystrova a , Daniel Rutkowski b , Tomasz Hermanowski c a b c
Faculty of Economics and Sociology, University of Lodz, Rewolucji 1905 r. 41, 90-214 Lodz, Poland National Health Fund, Poland Department of Pharmacoeconomics, Medical University of Warsaw, Poland
a r t i c l e
i n f o
Article history: Received 2 December 2014 Received in revised form 24 April 2015 Accepted 28 April 2015 Keywords: Hospital payment Diagnosis Related Groups Monitoring of outcomes Length of stay Stroke
a b s t r a c t A prospective payment system based on Diagnosis Related Groups (DRGs) presents strong financial incentives to healthcare providers. These incentives may have intended as well as unintended consequences for the healthcare system. In this paper we use administrative data on stroke admissions to Polish hospitals in order to demonstrate the response of hospitals to the incentives embedded in the design of stroke-related groups in Poland. The design was intended to motivate hospitals for the development of specialized stroke units by paying significantly higher tariffs for treatment of patients in these units. As a result, an extensive network of stroke units has emerged. However, as it is shown in the paper, there is no evidence that outcomes in hospitals with stroke units are significantly different from outcomes in hospitals without stroke units. It is also demonstrated that the reliance on the length of stay as a major grouping variable provides incentives for regrouping patients into more expensive groups by extending their length of stay in stroke units. The results of the study are limited by the incompleteness of the casemix data. There is a need to develop information and audit systems which would further inform a revision of the DRG system aimed to reduce the risk of regrouping and up-coding. © 2015 Elsevier Ireland Ltd. All rights reserved.
1. Introduction The Polish system of hospital payment based on DRGs was launched in mid-2008 with aims of improving resource allocation to hospitals and increasing transparency of services provision. The design of the system, including criteria for classifying patients into specific DRGs and relative sizes of DRG-based payments, was intended to provide hospitals
∗ Corresponding author. Tel.: +48 426355154. E-mail addresses: [email protected] (V. Bystrov), [email protected] (A. Staszewska-Bystrova), [email protected] (D. Rutkowski), [email protected] (T. Hermanowski). http://dx.doi.org/10.1016/j.healthpol.2015.04.017 0168-8510/© 2015 Elsevier Ireland Ltd. All rights reserved.
with incentives for containing costs without deterioration of the quality of care (see [1]). An effective payment system has to align financial incentives with the best medical practice. If a system fails to achieve such alignment, healthcare providers may have adverse incentives to exploit information asymmetry in order to increase profits. Since the payer (in this study, National Health Fund – Narodowy Fundusz Zdrowia – NFZ) has limited information about healthcare needs of an individual patient, hospitals may have incentives to underprovide healthcare services to some patients in order to reduce costs and over-provide to other patients in order to increase revenues. As the DRG-based reimbursement depends on the information coded by hospitals, there is also a risk of up-coding, i.e., exploiting the coding system
1120
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
by hospitals in order to achieve higher reimbursement. The problem of asymmetric information in the context of prospective payment system was discussed, inter alia, in [2–4]. The design of stroke DRGs varies widely across European countries (see [5]). In this paper we attempt to evaluate how Polish hospitals responded to incentives provided by the DRG system in stroke-related groups. For this purpose we use aggregate statistics on these groups as well as the patient-level data collected by the NFZ in 2010. According to the World Health Organization [6], stroke is the second leading cause of death in the world. In Poland, stroke is the third most common cause of death and the most common cause of disability in adults aged over 40 [7]. The research conducted within the International Stroke Trial and Polish National Stroke Prevention and Treatment Registry indicates that stroke mortality in Poland is relatively high (as compared to other countries) and there is large variation in stroke outcomes across Polish hospitals (see [8,9]). At the same time, Epstein et al [10] report that the costs of stroke treatment in Poland are high in relation to the national income per capita (as compared to other European countries). The development of the DRG system in Poland was accompanied by the creation of a network of specialized stroke units, which should have enabled high quality treat´ ment of stroke patients. According to Starzynska-Długosz et al. [11], the number of stroke units (eligible for thrombolysis) increased from 56 in 2007 to 150 in 2012. At the same time, the design of stroke-related DRGs provided strong financial incentives for treating patients in the stroke units. As a result, the number of patients treated in the stroke units and the number of patients classified into higherpaying stroke DRGs increased rapidly. In the years 2009–2013 the amount of reimbursement paid by the NFZ to hospitals for stroke-related groups increased by 14.93 percent (from PLN 526.11 million in 2009 to PLN 604.65 million in 2013). This growth can be mainly attributed to the redistribution of patients from lower-paying groups to higher-paying groups. While the total number of stroke hospitalizations decreased by 2.75 percent (from 94,963 in 2009 to 92,351 in 2013), the number of patients classified into higher-paying stroke DRGs increased by 37.34 percent (from 38,837 in 2009 to 53,340 in 2013). In the Polish DRG system, there are four stroke-related groups, which are defined by major diagnosis, procedures, treatment in a stroke unit, and length of stay (LoS). We use the distribution of LoS for ischemic stroke patients in order to demonstrate that the most frequent LoS is equal to a threshold value for classifying patients in higher-paying stroke DRGs. The use of LoS as a major criterion for classifying stroke patients may give hospitals an incentive to reclassify patients into a more expensive DRG. The development of the network of stroke units has coincided with a decrease of in-hospital mortality from 17.37 percent in 2009 to 15.81 percent in 2013. However, there is no evidence of significant differences in the distribution of outcomes in hospitals with stroke units and hospitals without stroke units. The outcomes are measured by in-hospital mortality rate and 30-days readmission
rate for patients with ischemic stroke. These measures are standardized by age and gender. The limitations of the data collection process implemented in Poland do not allow application of the risk-adjustment of outcomes, using comorbidities, and the resulting variation in outcome measures cannot be solely attributed to the quality of care. Comorbidities are not used as grouping variables in case of stroke-related DRGs in Poland. Though there is a possibility of reporting comorbidities to the NFZ, hospitals have no financial incentive to do so. We discuss the distribution of intensity of comorbidities coding for stroke admissions across Polish hospitals as well as the problems of data collection and quality monitoring. The paper is organized as follows. Section 2 provides a description of materials and methods. Dynamics of strokerelated DRGs, distributions of length of stay and stroke outcomes are analyzed in Section 3. Discussion of problems concerning quality monitoring is given in Section 4. Section 5 concludes the paper. 2. Materials and methods The study is based on two types of data: aggregated DRG statistics and anonymized patient-level data. The aggregated indicators are accessible on the webpage of the NFZ [12]. The database contains inter alia, the number of episodes and the average reimbursement per episode in each DRG. The patient-level data include hospital and department identifiers, age, gender, dates and modes of admission and discharge, major diagnosis, comorbidities, and DRG. The data set includes patients aged 35–100 years who were hospitalized with a diagnosis of cerebral infarction (ICD10 codes: I63.x) and discharged between January 1, 2010 and December 31, 2010. Patients transferred from another hospital are excluded from the data set. For the study of readmission rates and LoS, those patients who died before discharge are excluded from the data. The objectives of the data selection were to limit the extreme casemix variation and to use data concerning those patients who can be classified to all four strokerelated groups. 2.1. Description of Polish stroke-related DRGs The Polish system of hospital payment (JGP, Jednorodne Grupy Pacjentów) is a modification of the English system based on Healthcare Resource Groups (HRG, version 3.5). Definitions of some groups are directly adopted from the English system. Other groups, including stroke-related groups, are modified after consultations with local experts. In the Polish system, patients are included in a group on the basis of main and secondary diagnoses defined by International Classification of Diseases, version 10 (ICD10), procedures defined by the Polish modification of ICD-9, gender, age, type of admission, length of stay, and type of discharge. The size of prospective payment for a group is determined on the basis of historical data. In the absence of cost data, a system of quasi-prices and relative scores of DRG
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
groups is used (see [1,13]). All hospitals receive identical reimbursement for treating a patient in the same group. Stroke-related groups in English HRG 3.5 (A19, A22, and A23) are defined on the basis of major diagnosis, comorbidities, and age of patients. Length of stay is not used in the definitions of stroke groups in England. In the Polish system neither comorbidities nor age are used in the definitions of stroke-related groups. These groups are defined on the basis of the major diagnosis, procedures and length of stay. Two most expensive stroke-related groups in Poland (A48 and A51) require treatment in a specialized stroke unit (see Table 1). The different design of stroke groups in Poland originates from the necessity to encourage hospitals in creation of specialized stroke units. At the moment of the launch of the DRG-based system, specialized stroke units were scarce in Poland. The relatively high weights of groups requiring treatment in stroke units were intended to incentivize hospitals for treating patients in these units. Groups A48 and A51 are designed to provide patients with high-quality treatment in specialized stroke units for at least 8 days. These groups are the most expensive: the average reimbursement in group A48 is more than twice larger than in group A49 and more than four times larger than in group A50; the average reimbursement in group A51 is more than three times larger than in group A49 and more than six times larger than in group A50 (see Table 1). Large differences in reimbursement payments between groups are aimed to provide strong incentives to hospitals for quality treatment of patients in specialized stroke units. (For long stays exceeding 27 days in groups A49 and A50, and 36 days in groups A48 and A51, there is a supplementary per-diem payment). The NFZ specifies a set of requirements which should be met by hospitals to provide treatment of patients in groups A48 and A51. These requirements include qualified medical personnel, the number of beds, the type of medical equipment, and documentation. Hospitals and regional departments of the NFZ negotiate contracts, which specify the maximal volume of services in groups A48 and A51. The use of the LoS as a criterion of patient classification may provide incentives to hospitals for reclassifying patients into a more expensive group by extending their LoS. In particular, for a hospital with a stroke unit, a patient can be reclassified from group A49 into group A48 by extending her hospital stay and/or performing an additional (unnecessary) diagnostic procedure. All procedures qualifying for group A48 are also included in the list of procedures qualifying for group A49. The difference is in the minimal number of procedures: one procedure from the list is enough for qualifying a patient in A49 and two procedures are enough for qualifying a patient in A48. On the other hand, if a patient received an intensive treatment in a stroke unit, but did not survive through the first week of hospitalization, a hospital would encounter losses, because the patient would not qualify to group A48. This may put at risk high-severity patients with lower chances of survival through the first week of hospitalization, as hospitals may have a financial incentive of under-providing services to such patients. Group A51 requires expensive thrombolysis and hospitals have to
1121
treat surviving patients for at least 8 days in order to classify these patients in A51 and recuperate costs of thrombolysis. 2.2. Measurement of LoS In order to evaluate the effect of the DRG-based hospital payment on the LoS of stroke patients, we compare the distribution of LoS in hospitals with stroke units, which can classify patients in more expensive groups (A48 and A51), and hospitals without stroke units, which can only classify patients in less expensive groups (A49 and A50). For this purpose, we use data on non-deceased patients in hospitals with stroke units and hospitals without stroke units. The distributions of LoS in two types of hospitals are represented by histograms, which demonstrate differences in the empirical distributions of LoS. 2.3. Measurement of stroke outcomes To asses the performance of hospitals treating stroke patients, two outcome measures are computed: in-hospital mortality rate and 30-days all-cause readmission rate. The in-hospital mortality is computed on the basis of discharge modes and the information on readmissions is obtained by linking the NFZ data on the original stroke admissions with the data on the follow-up admissions. The outcome measures are standardized with respect to age and gender using a two-level logit model allowing for between-hospital variation in health outcomes (see, e.g., [15,16]). Standardization is carried out using all available observations. We compare the distribution of standardized mortality and readmission rates across hospitals with stroke units and the distribution of standardized rates across hospitals without stroke units. These comparisons are performed for hospitals with no fewer than 30 episodes of ischemic stroke. The reasons for such an approach are twofold. On the one hand, we tried to control sampling errors by excluding hospitals with a small number of stroke episodes. On the other hand, we aimed to retain a representative data set. The estimation of age and gender effects on mortality is carried out using data on 45,776 admissions to 360 hospitals and the comparison of standardized mortality rates uses information on 44,930 admissions to 273 hospitals. Of these 273 hospitals, there are 121 hospitals without stroke units which had 13,709 admissions, and there are 152 hospitals with stroke units which had 31,221 admissions. For the estimation of standardized readmission rates, data on 39,014 admissions are used (after the exclusion of deceased). The total number of observations used in the comparison of standardized rates is equal to 38,118 which include 26,652 admissions to 149 hospitals with stroke units and 11,466 admissions to 116 hospitals without stroke units. The central tendency, variation and the outliers for the standardized rates are estimated using statistics employed in the construction of boxplots. To this end, the central tendency is estimated using the median, and the variation is measured by the 25 percent and 75 percent quantiles of the
1122
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
Table 1 Characteristics of Polish stroke-related DRGs (2010). Code
Minimal LoS
Stroke unit
Type of stroke
Procedures
Median LoS
Average payment
A48 A49 A50 A51
8 days 4 days Not specified 8 days (if survived)
Required Not required Not required Required
Haemorrhagic or ischemic Haemorrhagic or ischemic Haemorrhagic or ischemic Only ischemic
At least 2 procedures At least 1 procedure Not specified Thrombolysis
11 9 5 11
8349.16 PLN 4023.92 PLN 1940.38 PLN 12,311.16 PLN
empirical distribution. The outliers are detected by identifying the data points which fall outside of the boxplot whiskers. These extend to the most extreme data points which are no more than 1.5 times the interquartile interval. 3. Results 3.1. Dynamics of stroke-related DRGs The stroke treatment has undergone major structural changes since the launch of the DRG system with a large increase in the share of patients treated in stroke units and a corresponding increase in the share of patients classified into high-paying DRGs (see Table 2). The fastest growing group is A51 which has grown by 546 percent in 2009–2013. However, this group is still relatively small and constitutes only 4.79 percent of all stroke admissions in 2013, because only a limited number of patients qualify to this group. The second fastest growing group is A48 with an increase of 28.19 percent in 2009–2013. As a result of this fast growth, 52.96 percent of all stroke admissions were concentrated in A48 in 2013 (as compared to 40.17 percent in 2009). This increase in groups A48 and A51 is compensated by a decrease in group A49 which does not require an extended stay in a stroke unit. The number of stroke admissions in this group decreased by 41.44 percent in 2009–2013 and the share of this group in the total number of stroke admissions decreased from 45.17 percent in 2009 to 27.20 percent in 2013. These changes result from reclassifying patients from less expensive to more expensive groups. As the number of stroke units increases, more patients can be treated in these units and classified in more expensive groups. However, as discussed in Section 3.2, the use of LoS as a classifying variable together with large differences in tariffs between groups provide incentives for gaming. Section 3.3 demonstrates that the distribution of outcomes is not significantly different between hospitals with stroke units and hospitals without stroke units.
Table 2 Number of hospitalizations in Polish stroke-related DRGs (2009–2013). Year
A48
A49
A50
A51
2009 2010 2011 2012 2013
38,151 38,687 42,319 46,944 48,908
42,892 32,417 31,434 28,088 25,117
13,234 19,695 17,194 15,310 13,894
686 1204 2285 3182 4432
Total 94,963 92,003 93,232 93,524 92,351
3.2. Analysis of the distribution of LoS The mean LoS for ischemic stroke patients, based on the data collected by the NFZ in 2010 (after the launch of the DRG-based system), was 11.6 days as compared to 14.4 days, based on the data collected by the National Stroke Prevention and Treatment Registry in 2002 (before the launch of the DRG-based system, see [9]). This number also differs from the mean LoS for ischemic stroke patients in England (HRGs A22 and A23), which was 15.3 days in 2010–2011, as reported in Hospital Episode Statistics by the Health and Social Care Information Centre [14]. The DRG-based classification of stroke hospitalizations, using LoS as a grouping variable, implies not only a reduced mean LoS but a peculiar shape of the distribution of LoS. Fig. 1 shows histograms of LoS for not deceased patients in hospitals without stroke units and hospitals with stroke units (those hospitals which had contracts with the NFZ for hospitalizations classified in groups A48 and A51). The histogram of LoS in hospitals without stroke units has no distinct mode. However, there is a distinct peak in the histogram of LoS for hospitals with stroke units: 26.5 percent of patients stay in these hospitals for 8 days. The shape of the distribution of LoS has two potential (non-medical) explanations: (1) reallocation of patients to a higher-paying DRG by extending their stay by an extra day; (2) early discharge: once a patient can be classified into a higher-paying DRG determined by the threshold LoS, there is no financial incentive to prolong the stay of the patient. It should be noted that hospitals with stroke units may not be able to treat all patients in these units. Two types of restrictions apply: a physical restriction (number of beds in a stroke unit) and a financial restriction (the total value contracted for groups A48 and A51). In result, stroke patients admitted to a hospital with a stroke unit may be treated in a neurology department or a department of internal medicine. For hospitals with a large volume of stroke hospitalizations, it implies the need to select those patients who will be admitted to a stroke unit and those who will not. This selection may be motivated by financial incentives rather than by clinical needs. In particular, there is a risk that the most severe patients with the greatest clinical needs will be treated outside stroke units because they may not qualify into groups A48 and A51.
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
0.25 0.20 0.00
0.05
0.10
0.15
Density
0.15 0.00
0.05
0.10
Density
0.20
0.25
0.30
Hospitals with stroke units
0.30
Hospitals without stroke units
1123
0 4 8
16
24
32
0 4 8
Number of days
16
24
32
Number of days
Fig. 1. Effect of stroke units on length of stay, not deceased patients.
0.4 0.3 0.2 0.1 0.0
Standardized mortality rate
Fig. 2 shows boxplots of standardized mortality rates for hospitals with and without stroke units. The median mortality rate in hospitals with stroke units is 13.43 percent as compared to 14.06 percent in hospitals without stroke units, and the interquartile intervals are respectively [10.31%,17.54%] and [10.14%,18.48%]. The differences in these descriptive statistics are not large enough to conclude that the distribution of mortality rates differs between hospitals with stroke units and hospitals without stroke units. This observation is supported by the results of Mann–Whitney U test: the hypothesis of the same distribution is not rejected (given the significance level of 10%). Boxplots for standardized readmission rates are shown in Fig. 3. The median readmission rate for hospitals with stroke units is 14.11 percent and the median readmission rate for hospitals without stroke units is 14.28 percent, and the corresponding interquantile intervals are [11.46%, 16.90%] and [10.83%, 17.80%]. As in the case of mortality rate, the differences in these descriptive statistics are not large enough to conclude that the distribution of readmission rates differs between hospitals with stroke units and hospitals without stroke units. The hypothesis of the same
0.5
3.3. Analysis of the distribution of stroke outcomes
Hospitals without stroke units
Hospitals with stroke units
Fig. 2. Boxplots of mortality rates.
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
0.3 0.2 0.0
0.1
Standardized readmission rate
0.4
1124
Hospitals without stroke units
Hospitals with stroke units
Fig. 3. Boxplots of readmission rates.
distribution is not rejected by the Mann–Whitney U test (given the significance level of 10%). The variation in stroke outcomes can be attributed either to the variation in the quality of care or the variation in the casemix. In order to separate these two factors, a risk adjustment of outcomes based on the casemix data has to be carried out. However, as it is shown in the next section, the current information system of the NFZ does not collect enough information to perform risk adjustment of outcomes. 4. Discussion: problems of quality monitoring In the Polish system, hospitals submit reimbursement claims together with statistical reports to regional branch offices of the NFZ. For each clinical episode, a statistical report has to include demographic data (age and gender), dates and modes of admission and discharge, codes of major and secondary diagnoses as well as codes of the procedures performed. The Regulation of the Minister of Health [17] allows reporting up to three comorbidities per episode. These data are stored in the Registry of Medical Services, the database of the NFZ. The NFZ controls data quality by performing plausibility checks of coded information and auditing selected hospitals. There is another system collecting information about all clinical episodes on the basis of statistical discharge cards. This system is administered by the Ministry of Health. However, information in this system is protected by the privacy law and cannot be used (under the existing legislation) for verification of the information in the database of the NFZ. The data collected by the NFZ are not protected by the privacy law. These data contain personal identifiers which allow restoring history of hospitalizations for a patient and tracking follow-up readmissions and mortality. The data could potentially be used for outcome-based
quality monitoring. However, incompleteness of the data on comorbidities makes it impossible to implement risk adjustment of outcomes. Since comorbidities are not included in the list of grouping variables for stroke-related groups in Poland, hospitals have no direct financial incentives to report them. Time pressure and the underfinancing of the coding process can contribute to the low intensity of coding as well. The average number of comorbidities per episode was computed for 273 hospitals with at least 30 stroke episodes. 16 hospitals did not report comorbidities for any patient. Other hospitals had heterogeneous coding practices. However, the total number of reported comorbidities was small with no comorbidities reported for 19,479 out of 44,924 episodes. The low intensity of coding results in low prevalence of stroke comorbidities in the administrative data as compared to the prevalence of these comorbidities in studies based on clinical data. For example, the prevalence of diabetes computed on the basis of administrative data is equal to 11.28 percent as compared to 24.1 percent reported in Niewada et al [9], and the prevalence of hypertension is equal to 37.5 percent as compared to 69.5 percent in [9]. These results are consistent with the results reported by Peltola [18], who, using administrative data from 2009, finds low values of the Charlson index of comorbidities for stroke patients in Poland as compared to other European countries. In the absence of complete casemix data, the NFZ uses raw outcome indicators. The variation in these indicators can be explained by the variation in the quality of care as well as by the variation in the casemix. The evaluation of quality based on raw outcome indicators may produce biased results. A coordinated effort by the Ministry of Health and the NFZ is needed in order to develop an integrated information system collecting comprehensive casemix data. Such system would enable the risk adjustment of outcomes and control of quality based on outcome measures. The information system should be supported by a systematic audit in order to control the quality of information submitted to the Ministry of Health and the NFZ by healthcare providers. The combination of these two systems could further inform policy decisions concerning the design of the payment system. 5. Conclusions The design of a DRG-based system embeds strong financial incentives for healthcare providers. These incentives may serve intended as well as unintended consequences for the healthcare system. In this paper, we use the data on stroke admissions in Poland in order to demonstrate how hospitals behave under the DRG-based system. The design of stroke-related groups in Poland has served the intended aim of increasing the number of patients treated in specialized stroke units. However, the use of length of stay as a grouping variable resulted in the most frequent length of stay being equal to the threshold value for reclassifying patients into more expensive groups. This
V. Bystrov et al. / Health Policy 119 (2015) 1119–1125
type of hospital behavior, which is dictated by the financial incentives embedded in the design of stroke-related groups in Poland, may have adverse consequences for patients well-being. There is an extensive network of stroke units in Poland, which can claim higher reimbursement for treatment of stroke patients. However, as demonstrated in this study, there is no evidence that the distribution of stroke outcomes in hospitals with stroke units significantly differs from the distribution of stroke outcomes in hospitals without stroke units. The study is restricted by the incompleteness of casemix data. Since outcome measures are not adjusted for casemix, their variation cannot be attributed solely to the variation in the quality of care. There is a need for further development of the information system of the NFZ which would allow systematic collection of complete casemix data. The design of stroke-related groups in Poland requires modifications in order to reduce risk of regrouping and upcoding, and achieve effective use of restricted resources which can be allocated to stroke units. The refinement of grouping criteria for DRGs and the revision of tariffs should align financial incentives of hospitals with the best medical practice. Acknowledgements This study is part of the International Research Project on Financing Quality in Healthcare (InterQuality), whose main purpose is to investigate the effect of different financing methods and incentives on quality, effectiveness and equity of access to health care. The project is funded by the European Commission under the Seventh Framework Programme (FP7) for Research and Technological Development, grant agreement number: 261369. More information about the Project can be found at http://interqualityproject.eu/. References ´ [1] Czach K, Klonowska K, Swiderek M, Wiktorzak K. Poland: The Jednorodne Grupy Pacjentów – Polish experiences with DRGs. In: Busse R, Geissler A, Quentin W, Wiley M, editors. Diagnosis-related groups in Europe. McGraw-Hill Open University Press; 2011. p. 359–80.
1125
[2] Ellis R. Creaming, skimping and dumping: provider competition on the intensive and extensive margins. Journal of Health Economics 1998;17:537–55, http://dx.doi.org/10.1016/ S0167-6296(97)00042-8. [3] Cots F, Chiarello P, Salvador X, Castells X, Quentin W. DRG-based hospital payment: intended and uninitended consequences. In: Busse R, Geissler A, Quentin W, Wiley M, editors. Diagnosis-related groups in Europe. McGraw-Hill Open University Press; 2011. p. 75–92. [4] Or Z, Häkkinen U. DRGs and quality: for better or for worse? In: Busse R, Geissler A, Quentin W, Wiley M, editors. Diagnosis-related groups in Europe. McGraw-Hill Open University Press; 2011. p. 115–29. [5] Peltola M, Quentin W. Diagnosis-related groups (DRGs) for stroke in Europe: patient classification and hospital reimbursement in 11 countries. Cerebrovascular Diseases 2013, http://dx.doi.org/10.1159/000346092. [6] WHO. Fact sheet No310, http://www.who.int/mediacentre/ factsheets/fs310/en/ [7] Udary mózgu – konsekwencje społeczne i ekonomiczne, Lazarski University, Warsaw, 2013. [8] Członkowska A, Niewada M, El-Baroni IS, Mendel T, Ryglewicz D, Sandercock P, et al. High early case fatality after ischemic stroke in Poland: exploration of possible explanations in the International Stroke Trial. Journal of the Neurological Sciences 2002;202:53–7. ´ ´ M, Ryglewicz D, Kaminski B, Członkowska [9] Niewada M, Skowronska A. Acute ischemic stroke care and outcome in centers participating in the Polish National Stroke Prevention and Treatment Registry. Stroke 2006;37:1837–43. [10] Epstein D, Mason A, Manca A. The hospital costs of care for stroke in nine European countries. Health Economics 2008;17:21–31, http://dx.doi.org/10.1002/hec.1329. ´ ´ [11] Sarzynska-Długosz I, Skowronska M, Członkowska A. Organization of acute stroke services in Poland Polish Stroke Unit Network development. Neurologia i Neurochirurgia Polska 2013;47:3–7, http://dx.doi.org/10.5114/ninp.2013.32934. [12] NFZ. Statystyka JGP, http://prog.nfz.gov.pl/APP-JGP/ ˛ [13] Kozierkiewicz A, Stamirski M, Stylo W, Trabka W. The definition of prices for inpatient care in Poland in the absence of cost data. Health Care Management Science 2006;9:281–6, http://dx.doi.org/10.1007/s10729-006-9095-z. [14] Hospital Episode Statistics. Admitted Patient Care - England, 2010–11, Health and Social Care Information Centre, http://www.hscic.gov.uk/searchcatalogue?productid=98 [15] Iezzoni LI. Risk adjustment for measuring healthcare outcomes. Health Administration Press; 1997. [16] Normand S-LT, Shahian DM. Statistical and clinical aspects of hospital outcomes profiling. Statistical Science 2007;22:206–26. ˛ Ministra Zdrowia z dnia 20 czerwca. r. w [17] Rozporzadzenie ˛ sprawie zakresu niezbednych informacji gromadzonych przez ´swiadczeniodawców, szczegółowego sposobu rejestrowania tych ˛ do informacji oraz ich przekazywania podmiotom zobowiazanym finansowania s´ wiadczen ze s´ rodków publicznych. Dziennik Ustaw Nr 123 Poz 2008;801:6636–42. [18] Peltola M. Patient classification and hospital costs of care for stroke in 10 European countries. Health Economics 2012;21(Suppl. 2):129–40, http://dx.doi.org/10.1002/hec.2841.