Health Policy 113 (2013) 206–215

Contents lists available at ScienceDirect

Health Policy journal homepage: www.elsevier.com/locate/healthpol

Resources allocation and health care needs in diabetes care in Danish GP clinics Troels Kristensen a,b,∗ , Kim Rose Olsen a,b , Camilla Sortsø c , Charlotte Ejersted d , Janus Laust Thomsen b,e , Anders Halling b a Institute of Public Health, Centre of Health Economics Research, Faculty of Health Sciences, University of Southern Denmark, J.B. Winsløws Vej 9B, DK-5000 Odense C, Denmark b Institute of Public Health, Research Unit of General Practice, Faculty of Health Sciences, University of Southern Denmark, J.B. Winsløws Vej 9A, DK-5000 Odense C, Denmark c Department of Business and Economics, Centre of Health Economics Research, Faculty of Social Sciences, University of Southern Denmark, Campusvej 55, DK-5230 Odense M, Denmark d Department of Endocrinology, Odense University Hospital, Kløvervænget 6, DK-5000 Odense C, Denmark e DAK-E Danish Quality Unit of General Practice, J.B. Winsløws Vej 9A, DK-5000 Odense C, Denmark

a r t i c l e

i n f o

Article history: Received 11 February 2013 Received in revised form 9 September 2013 Accepted 15 September 2013 Keywords: General practice Expenditure variation Resource utilisation band Fee-for-service The Johns Hopkins Adjusted Clinical Groups (ACG) System

a b s t r a c t Background: In several countries, morbidity burdens have prompted authorities to change the system for allocating resources among patients from a demographic-based to a morbidity-based casemix system. In Danish general practice clinics, there is no morbiditybased casemix adjustment system. Aim: The aim of this paper was to assess what proportions of the variation in fee-for-service (FFS) expenditures are explained by type 2 diabetes mellitus (T2DM) patients’ co-morbidity burden and illness characteristics. Methods and data: We use patient morbidity characteristics such as diagnostic markers and co-morbidity casemix adjustments based on resource utilisation bands and FFS expenditures for a sample of 6706 T2DM patients in 59 general practices for the year 2010. We applied a fixed-effect approach. Results: Average annual FFS expenditures were approximately 398 euro per T2DM patient. Expenditures increased progressively with the patients’ degree of co-morbidity and were higher for patients who suffered from diagnostic markers. A total of 17–25% of the expenditure variation was explained by age, gender and patients’ morbidity patterns. Conclusion: T2DM patient morbidity characteristics are significant patient related FFS expenditure drivers in diabetes care. To address the specific health care needs of T2DM patients in GP clinics, our study indicates that it may be advisable to introduce a morbidity based casemix adjustment system. © 2013 Elsevier Ireland Ltd. All rights reserved.

1. Introduction

∗ Corresponding author at: University of Southern Denmark, Institute of Public Health, Centre of Health Economics Research, J.B. Winsløws Vej 9B, Odense C, Denmark. Tel.: +45 65503877; fax: +45 6550 3880. E-mail addresses: [email protected] (T. Kristensen), [email protected] (K. Rose Olsen), [email protected] (C. Sortsø), [email protected] (C. Ejersted), [email protected] (J.L. Thomsen), [email protected] (A. Halling). 0168-8510/$ – see front matter © 2013 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.healthpol.2013.09.006

Caring for patients with multi-morbidity – the presence of several chronic diseases in one individual – is an important challenge facing health services in developed countries [1]. Currently, the management of patients with multi-morbidity is the norm rather than the exception and presents a challenge to the single-disease and fragmented focus in the health system [1]. Due to the ageing of

T. Kristensen et al. / Health Policy 113 (2013) 206–215

populations and new technologies, this challenge is expected to continue and grow in the future [2]. Primary care is considered a more cost-effective place to begin addressing multi-morbidity rather than waiting for patients to appear in secondary care [3,4]. Multi-morbidity may not be adequately addressed in the current structures of primary care systems [4,5]. Primary care is often organised around single diseases with limited time allocated to a consultation rather than systems supporting multi-morbidity-based guidance. Primary care health professionals often have to apply the evidence from single disease guidelines when treating multi-morbidity patients. For instance in Denmark, more could be done to develop clinical guidelines and indicators that fully address the realities of patients with multi-morbidity conditions [6]. The OECD has reported that Danish general practice has not been reorganised to deliver a new set of functions in care co-ordination and integration that it ought to perform. This implies a need to redesign general practice consultation and conceptualisation of multi-morbidity in the organisational structures of the health care system [7,8]. Inappropriate coordination and integration may be driven by inappropriate economic incentives in remuneration systems to encourage the treatment of more chronic diseases in a single consultation. Several countries with publicly funded general practice (GP) clinics have reoriented their remuneration systems towards a morbidity-based casemix adjustment system [9]. Denmark has yet to reorient its resource allocation system in the general practice sector towards morbiditybased casemix systems [3]. Prior to potential reforms of the remuneration system in Danish GP clinics, it is relevant to investigate the extent to which public resources are allocated according to morbidity status. A recent study focusing on all types of GP patients concluded that morbidity measures were significant patient-related fee-for-service (FFS) expenditure drivers [10]. Morbidity characteristics explained 18-31%, age and gender 13% and volume of activity explained about 35% of the resource allocation through FFS. However, that study did not explore the association between morbidity and FFS within a specific chronic disease area in primary care. Because diabetes patients have one of the highest co-morbidity rates with other chronic diseases and represent an increasing economic burden to the health care system, this study limits its focus to diabetes patients in primary care [11]. Almost all of these patients are type 2 diabetes mellitus (T2DM) patients. Type 1 diabetes patients are treated in the hospital sector. In addition to missing morbidity adjustment, the FFS component may be too dominant in Denmark [12,13]. We anticipate that lack of morbidity-adjusted remuneration and the dominant FFS component lead to short GP visits focused on one problem. The average T2DM patient with co-morbidities may not come back to the GP over and over again in a way that reflects their health-care need. This means that the present resource allocation for T2DM patients may not reflect health care needs. The aim of this study was to describe and analyse the extent to which negotiated patient-level fee-for-service (FFS) expenditures are associated with T2DM patients’ co-morbidity burden.

207

1.1. The Danish general practice sector and the allocation of resources for diabetes patients Danish GPs are self-employed professionals who contract with one of five regions that have the overall operational and planning responsibilities for the health care system [12]. GP services are financed by taxes, and there are no user fees for GP services. GPs are gatekeepers, and the main coordinators of care for patients. The regions compensate GPs through a combination of per capita fees (30%) and FFS (70%), and nearly all citizens are registered with a specific GP [12,13]. This mix of FFS and capitated payment systems is expected to balance the incentives for over-provision of services inherent in FFS against the incentives for under-provision of services inherent in capitation [14,15]. The Danish Regions and the Danish Organisation of GPs (PLO) negotiate per capita and FFS compensation for GPs. Fees are used strategically to form incentives for specific services such as preventive care [13]. The present remuneration system that lacks morbidity adjustments (except concerning fees for supplementary services) incentivises the GP to follow a one disease or procedure approach per visit. There is no additional payment for diagnosis coding; the only incentive for GPs to code diagnoses is to encourage better organisation and quality improvements. Nevertheless, many Danish GPs have used the International Classification of Primary Care code (ICPC-2) for several years [16,17]. In 2006, Danish GPs were encouraged to implement a Data Capture module called “Sentinel Data Capture”. Data collected include, e.g. all diagnoses of patient contacts and all disbursement codes [18]. These new data offer an opportunity to examine the association between the present allocation of resources and diabetic patients’ morbidity burden. Actual cost data on Danish GP clinics are unavailable [10]. 2. Methods Descriptive statistics and a regression approach were employed to explore the association between FFS expenditures per capita in diabetes care and the patients’ morbidity burden in general practice. Due to the nested nature of the data and the results of a Hausmann test, we applied a fixed-effects data model that recognises the stratification of patients within GP clinics [19,20]. The model takes the following form: FFSEij = ˛ + Xij ˇ + uj + vij ,

i = 1, . . .I, j = 1, . . .N

(1)

where FFSEij represents patient level FFS expenditures for T2DM patient i registered with general practice clinic j, and FFSEi is the sum of GP services (sik ) weighted by politically  negotiated fees for each service (pk ). Thus, FFSEi = pk sik , where k = 1, . . ., M. T2DM patients were identified using the patient’s personal identification number and ICPC code T90 listed in chapter T (Endocrine/Metabolic and Nutritional). The parameter Xij is a row vector of explanatory variables containing the characteristics of patient i in clinic j. The parameter uj is the clinic-specific effect referring to the conditional mean of the annual expenditures per individual treated by clinic j. This GP clinic fixed effect

208

T. Kristensen et al. / Health Policy 113 (2013) 206–215

captures clinic-specific relative expenditures (fixed effects) after allowing for differences in patient characteristics. vij is the unexplained noise term, which was assumed to be normally distributed with zero mean and a constant variance. The morbidity-based grouping of The Johns Hopkins Adjusted Clinical Groups (ACGs) into resource utilisation bands (RUBs) was used to measure the patients’ co-morbidity burden in addition to T2DM. This system has been demonstrated to be able to robustly measure morbidity in an individual and is considered useful in European national health care systems [9,21,22]. The six RUBs are constructed by combining the about 100 unique mutually exclusive combinations of conditions designated ACGs in six simplified mutually exclusive RUBS that measure the overall morbidity burden. The reason why grouping of ACGs into RUBs is useful is that less divisions means less administrative complexity. The grouping of ACGs into six groups of diagnoses with equal resource utilisation (RUB0-RUB5) is sufficient to group the patients according to morbidity burden. More groups create more room for gaming and make the system more complicated to understand, implement and monitor. The ACG system software was used to assign a six level co-morbidity measure, where RUB0 indicates little or no co-morbidity and RUB5 a very high level of co-morbidity [23]. To determine the sensitivity of the results to different co-morbidity and diagnostic characteristics, we specified five fixed effects models (1-5) with different combinations of four subsets of covariates: the age and gender of the patient, the RUB groupings and two sets of diagnostic markers based on the International Classification of Primary Care (ICPC2) developed by WONCA International Classification Committee (WICC) [24]. All diagnostic markers were entered as dummy variables. The role of the diagnostic markers was to describe and capture patient co-morbidity characteristics according to the structure of the present ICPC-2 coding system. The applied diagnostic markers should be confined to those actually used for coding diagnoses over and above the RUBs based on the ACG casemix system. One set of diagnostic markers is based on 17 types of body systems/problem areas (i.e. symptoms and/or diseases) according to the ICPC-2 classification (see Fig. 2). This set of ICPC2 chapter markers captures the coding of diagnoses across specific body systems and two non-body system areas: general and unspecified and social problems. Another alternative set was based on the classification of ICPC-2 diagnoses into 7 chapter components (i.e. components that distinguish symptoms, diagnoses, work, and care processes – see Fig. 3). The latter set of diagnostic markers included process codes, symptoms/complaints, infections, neoplasms, injuries, congenital anomalies, and other diagnoses. This set of chapter components captures the reasons for visits and the focal components of work and care processes. To measure the prevalence of diagnostic markers per patient in each of the ICPC-2 classifications and measure the variation explained by co-morbidity and morbidity rather than volume, we limited our analysis to the range of different diagnoses per patient. Thus, we excluded the volume of diagnosis to be able to explore the explanatory power of multiple diagnosis/co-morbidity characteristics.

Gender and age have been widely applied to explain expenditure variation [25]. Similar to the rest of the female population [26,27], female diabetes patients have been found to report more disease experiences [28]. Younger and elderly patients are expected to be more expensive [10]. We included dummy variables identifying whether the patient received care for other diseases. FFS expenditures were expected to increase progressively in the degree of multi-morbidity and/or the number of diabetes comorbidities. In the regression, we excluded RUB0 as a reference group for RUBs 1-5. The likelihood ratio test statistic and Akike’ Information Criterion (AIC) were used to compare the fit of the five models (1-5) [29]. In line with a related study, no log transformation of FFSE was used [10]. Finally, the rho (or interclass correlation coefficient) = u2 /(v2 +u2 ) based on the variance components of the composite error term v2 and u2 was used to quantify the ratio of unexplained variation between the patient and GP clinic level. 3. Data The present sample is restricted to T2DM patients in Danish sentinel clinics for the year 2010. Sentinel clinics are clinics that coded ICPC-2 diagnoses for episodes of care for on average more than 70% of their patients in at least 6 months and continued to do so in the year [15]. The Danish Quality Unit of General Practice (DAK-E) collect these data to promote improved organisation and quality in the GPs’ clinics as well as research and monitoring. In 2010, only a limited number of GPs (3-4%) had started to code sufficiently to qualify as sentinel clinics. We analysed the FFS expenditures of all T2DM patients (6706 patients) from a sample of 139,527 GP patients, who were representative of Danish GP clinics and received care in 59 sentinel clinics in 2010 [16]. Our analysis combines several data sets: (i) the Danish General Practice Database (DAMD), which includes data linked to the patient’s personal ID; (ii) ACGs with similar expenditures grouped in RUBs; (iii) the tariff agreement on GP services between PLO and the association of Danish regions. The tariff agreement on GP services and the National Health Service disbursement codes were used to calculate and map service expenditure data for each patient in 2010. Thus, we were able to identify the FFS expenditures for T2DM patients in 2010. 3.1. Descriptive patient characteristics Table 1 reports descriptive patient characteristics regarding annual FFSE, non-diagnostic markers and morbidity markers based on RUBs using means and percentiles (p5; p50; p95) for all covariates. Table 1 shows that the average annual FFSE per patient was D 398.5 (p5 = 95.7; p95 = 841.0). Fifty-six percent of these T2DM patients were male, and the average patient was 65 years old (p5 = 44; p95 = 85). On average, patients visited their GP 7.84 times (p5 = 1; p95 = 20), and 10.9 diagnoses were recorded during the year. Within the RUB-range from 0 to 5, the average patient had a multi-morbidity case mix index value of 2.2 (p5 = 0; p95 = 3). Data on the patients’

T. Kristensen et al. / Health Policy 113 (2013) 206–215

209

Fig. 1. Box plot of annual FFS expenditures per diabetes patient by co-morbidity burden (RUB).

Table 1 Descriptive diabetes patient characteristics.

Expenditures in euro (D ) Non-diagnostic markers Number of diagnoses during the year Age (years) Sex (male = 0, female = 1) Number of visits Morbidity markers based on RUBs Resource utilisation band 0-5

Mean

SD

P5

P50

P95

398.5

236.1

96.7

353.3

841.0

10.87 65.84 0.44 7.84

9.19 12.59 0.49 6.72

2 44 0 1

8 67 0 6

28 85 1 20

2.22

1.11

0

3

3

SD standard deviations, 5% percentile (p5); 50% percentile (p50); 95% percentile (p95).

morbidity based on RUB and ICPC-2 chapters are presented in the descriptive results section.

4. Results

4.1.1. T2DM patients’ co-morbidity burden characteristics Fig. 2 depicts the distribution of T2DM patients by additional morbidity burden based on mutually exclusive RUBs. The distribution indicates how often additional combinations of chronic diseases occur in the present sample

This section reports descriptive results on the association between FFSE per T2DM patient and morbidity burden.

4.1. Descriptive results Fig. 1 reveals that patients with T2DM accounted for high expenditure levels, which increased according to the level of co-morbidity (RUB). More T2DM co-morbidities and other independent diseases imply higher annual expenditures. This evidence is consistent with the evidence from the RUBs [9,30]. Fig. 1 shows that the mean FFS expenditure per T2DM patient increases less than proportionally from RUB0 to RUB2 and begins increasing progressively from RUB2 to RUB5. The mean (median) annual FFS expenditures per T2DM patient within each RUB level were as follows: RUB0: D 256.6 (245.2), RUB1: D 318.6 (292.30), RUB2: D 339.1 (312.6) RUB3: D 464.0 (417.6), RUB4: D 604.9 (536.4) and RUB5: D 753.1 (735.2). The variability of the FFS expenditures also increased with the morbidity burden.

Fig. 2. Distribution of diabetes patients by additional illness burdens.

210

T. Kristensen et al. / Health Policy 113 (2013) 206–215

of T2DM patients. These data confirm that the majority of T2DM patients have a high degree of multi-morbidity and expected health care utilisation [31]. A majority of the patients (78.6%) were grouped in RUBs with additional chronic diseases (RUB2-RUB5). Only 13.3% were grouped in RUB0. 4.1.2. Prevalence and pattern of co-morbidity characteristics within morbidity levels Fig. 3 reveals the prevalence and patterns of different diagnostic markers based on ICPC-2 chapters per patient in addition to T2DM by morbidity burden (RUB2-RUB5). The diagnostic markers are defined as follows: body systems (B-Y), general and unspecified (A) and social problems (Z). The shape of the lines in the diagram reflects how often combinations of diseases occur depending on the level of multi-morbidity (RUB2-RUB5). For instance, more than 85% in RUB4 have received a cardiovascular diagnosis, and approximately 70% have a musculoskeletal diagnosis. Thus, it can be calculated that 0.85 × 0.70 = 0.60% of T2DM patients in RUB4 are likely to have both co-morbidities. Fig. 3 shows that the patterns of co-morbidities are similar across RUB-levels (RUB2-RUB5). As expected, the RUB-levels appear to imply an increasing prevalence of co-morbidities in terms of other diagnoses among T2DM patients. For instance, RUB5 patients have the highest prevalence of diagnostic markers across all body systems, symptoms and problem areas. The most prevalent diagnostic marker across all RUB-levels is cardiovascular disease. The following are ‘general and unspecified’, ‘digestive’, ‘skin’, ‘respiratory’ and ‘musculoskeletal’. The alternative ICPC-2 chapter component classification of diagnostic markers is plotted in Fig. 4. T2DM patients at RUB-levels (RUB2-RUB5) have received a mixture of chapter component markers in addition to a low prevalence of congenital anomalies. The most prevalent T2DM diagnostic markers are “other diagnoses”, “symptoms/complaints” and “infections”. Fig. 4 shows that classifications at higher RUB-levels imply simultaneously higher levels of both “injuries, symptoms/complaints”, “infections”, “injuries” and “other diagnoses”. T2DM patients at level RUB0 primarily exhibit injuries and process codes. 4.2. Association between FFSE per T2DM patient and morbidity burden Table 2 displays the results of equation 1, applying the five specifications that include different sets of T2DM patient markers (demographic and multi-morbidity case mix and two sets of diagnostic characteristics). Models 1-5 are able to explain 3.6-25.0% of the variation in the FFSE for primary care patients in sentinel clinics, as revealed by the overall R2 statistic. Nearly all covariates were significant. 4.2.1. Demographic markers Model 1 indicates that age and gender explained 3.6% of the variation in FFSE. Women had higher FFSE than men. Models 1-5 all revealed that FFSE were u-shaped functions of age, with younger (p5 = 44) and older (p95 = 85)

patients having higher expenditures. When Models 3-5 were adjusted for diagnostic markers, the gender effect found in Models 1 and 2 was transferred to the markers. This indicates that the gender effect was driven by genderspecific diagnostic markers. 4.2.2. RUB markers Compared to Model 1, the inclusion of RUBs (Model 2) reveals that a significant share of the variation in expenditures was driven by co-morbidity characteristics. Co-morbidity measures (RUBs) increased the explanatory power from 3.6% to 17%. The numerical values and significance of the RUB-coefficients indicate that FFSE increased progressively in the degree of multi-morbidity. The coefficients of the RUBs were all estimated related to the reference group (RUB0). For instance, the results of Model 2 show that a patient in RUB3 was D 201.9 more expensive than a patient in RUB0 due to morbidity characteristics. In Models 3-5, the significance of the RUB-coefficients diminished because of the introduction of more detailed and symptom-/disease-specific markers in terms of ICPC-2 chapters (body system, symptoms and social problem markers) and ICPC-2 component markers. 4.2.3. Diagnostic markers Models 3-5 show that both ICPC-2 chapter markers and chapter component markers improve the overall explanatory power of the model. The R2 (overall) statistic increased from 17.0% (Model 2) to 23.9% (Model 3), 21.9% (Model 4) and 25.0% (Model 5). The introduction of diagnostic markers in Models 3-5 shifts some of the significance and magnitude of coefficients from the RUB-coefficients to the individual body system diagnostic markers and the chapter component. The most expensive diagnostic markers were the “psychological” and “urological” diabetes patient markers. After controlling for patient characteristics, the rho statistic in Table 2 indicates that GP clinic characteristics explained 14-18% of the remaining variation in patient expenditures. This means that the unexplained variation in FFSE is more related to patient characteristics than to GP clinic characteristics. However, the rho statistic also indicates that GP clinic characteristics are important for explaining politically negotiated FFSE for T2DM patients. Likelihood ratio and AIC test statistics showed that Model 5 fitted the data in the best way. 5. Discussion This paper contributes to the literature in several ways. First, we applied simplified mutually exclusive RUB categorisations based on the ACG system to describe the distribution of politically negotiated FFS expenditures as a function of co-morbidity among T2DM patients in Danish sentinel GP clinics. Second, we used these RUBs to describe the prevalence of T2DM patients’ “ICPC-2 chapter markers” and “ICPC-2 chapter component markers” by co-morbidity burden. Third, we estimated the association between politically negotiated FFSE and comorbidity burden in T2DM patients, unlike previous

T. Kristensen et al. / Health Policy 113 (2013) 206–215

211

Fig. 3. Prevalence of diabetes patient co-morbidity characteristics by co-morbidity burden. Note: The ACG system software (version 9.0) was used to assign a six level co-morbidity burden measure (RUB0-RUB5). RUB 2 (···: dot) indicates low co-morbidity, RUB3 (–·–: dash dot) moderate co-morbidity, RUB4 (-·-·-: short dash) high co-morbidity and RUB5 (—: solid) very high co-morbidity. Due to little or no co-morbidity RUB0 and RUB1 were excluded from the figure.

studies that estimate the association between measures of “actual” GP clinic costs and a morbidity measure.

5.1. FFSE per T2DM patient and co-morbidity burden Overall, patient level FFSE increased progressively as a function of increasing co-morbidity, and the variation in FFSE per patient could be explained by patient co-morbidity characteristics. Comorbidity measures explained more than age and gender. This suggests that resource allocation rules based on both age, gender and co-morbidity characteristics are more appropriate than only based on age and gender [10,32]. In line with related literature, combinations of age, gender and RUBs and diagnostic markers explained 17-25% of the variance in FFSE per T2DM patient in Danish GP clinics [33]. A previous Danish study of all GP patients in the same sentinel clinics explained 32–44% of the variance in FFSE [10]. This difference in explanatory power is not surprising, because the present study is limited to analysing the relationship between an index condition (T2DM) and co-morbidities, rather than the relationship between all patients and all diagnoses. We found that 14-18% of the unexplained variation in FFSE was explained by the GP clinic characteristics captured by provider-specific fixed effects. This indicates that the way GP clinics manage their T2DM patients has some impact on the FFSE. Nonetheless, the majority of this variation seems to be due to unobserved variation in GP clinics’ T2DM populations.

5.2. Randomly distributed T2DM patients? If type T2DM patients with many co-morbidities are randomly distributed across GP clinics, no systematic difference in remuneration may occur. It is, however, expected that chronic diseases cluster in practices in deprived areas [34,35]. Our data confirm that T2DM patients are systematically distributed across GP clinics. First, a non-parametric Kruskal-Wallis test and an equivalent ANOVA test for the differences in the mean RUB-scores across the GP clinics were rejected (p < 0.0002). Second, we tested the joint significance of a row of clinic characteristics against the average RUB-score per GP clinic: (1) the geographic location of the GP clinic by region, (2) the number of physicians per clinic, (3) average physician age, (4) the proportion of female physicians, (5) list size, and (6) the number of T2DM patients. The F-test of the joint explanatory power rejected the hypothesis of no systematic differences (p < 0.0032). Third, a new Danish study of the same GP clinics also confirms that there are systematic differences across sentinel clinics in their multi-morbidity patient profiles [10]. 5.3. RUBs and prevalence of diagnostic markers As expected, RUBs seem to be consistent and robust measures of prevalence of diagnostic markers and comorbidity burden. The most prevalent diagnostic marker in all RUBs is cardiovascular disease, representing high

212

T. Kristensen et al. / Health Policy 113 (2013) 206–215

Fig. 4. Prevalence of diabetes patient component characteristics by co-morbidity burden. Note: The ACG system software was used to assign a six level co-morbidity burden measure (RUB0-RUB5). RUB1 (—: dash) indicates little co-morbidity, RUB2 (···: dot) low co-morbidity, RUB3 (–·–: dash dot) moderate co-morbidity, RUB4 (-·-·-: short dash) high co-morbidity and RUB5 (—: solid) very high co-morbidity. Due to little or no co-morbidity RUB0 was excluded from the figure.

numbers of patients with hypertension and ischaemic heart disease [36]. Blood pressure is measured at least once a year in T2DM patients, and the treatment of hypertension is considered as important as maintaining high blood glucose to prevent complications [19]. The large number of skin diseases in RUB4 and RUB5 may represent, for instance, diabetic foot ulcers. Diabetic foot ulcers are known to be expensive to treat [19]. T2DM does not directly affect the respiratory system, and hence the high prevalence of the respiratory marker in RUB4 and RUB5 likely represents a common co-morbidity. 5.4. Room for improvement of morbidity adjustment In general, our results indicate that there is room for improvement of the association between negotiated FFSE per T2DM patient and measures of co-morbidity characteristics in Danish GP clinics. However, our analysis gives no answer as to why. Potential explanations could be one or more of the following: (a) GPs handle more diseases in one consultation and thereby treat patients with and without co-morbidities at the same “price”;

(b) T2DM patients with many co-morbidities are underserved and/or receive treatment for their co-morbidities in secondary care; (c) T2DM patients’ needs for GP services are more related to e.g. socioeconomics than observed morbidity.

In case of (a) patients will receive the treatment they need, but GPs with many T2DM patients with comorbidities will face either a lower remuneration (if they focus their work on more complex and time-consuming patients) or work longer hours (because treating more diseases in one consultation may be more time-consuming). This connection has been shown to hold in a slightly different analytical framework in [37]. In case of (a) or (b), there is a need to increase incentives to serve patients with co-morbidities if they do not receive the required treatment in secondary care. For instance, it might be appropriate to adopt a morbidity-based casemix system as part of the GP remuneration system and increase incentives to address more co-morbidities in one consultation rather than short GP visits focused on one problem. In case of (c) it is considered more relevant to adjust the remuneration system according to e.g. socioeconomics [38].

T. Kristensen et al. / Health Policy 113 (2013) 206–215

213

Table 2 Type-2 diabetes patient expenditure function estimates, 2010. Model 1 Demographic makers Age (years) Age squared Sex RUB markers RUB1 RUB2 RUB3 RUB4 RUB5 ICPC-2 chapters markers General and unsp. A Blood, blood for. B Digestive D Eye F Ear H Cardiovascular K Musculoskeletal L Neurological N Psychological P Respiratory R Skin S Endocrine/Metab. T Urological U Pregnancy, childb. W Female genital X Male genital Y Social problems Z Chapter components markers PROCESS CODES SYMPTOMS/COMP. INFECTIONS NEOPLASMS INJURIES CONGENITAL AN. OTHER DIAG. Constant N u v Rho R2 overall AIC Likelihood-ratio test Number of GP clinics

6.126*** −0.0237* 22.46***

Model 2 7.167*** −0.0399*** 10.23* 57.21*** 79.30*** 201.9*** 347.3*** 485.7***

Model 3 8.527*** −0.0515*** −2.340 −6.591 −7.327 37.09*** 86.42*** 146.8***

Model 4 8.303*** −0.0489*** 1.682 −20.68 −14.25 64.24*** 165.8*** 270.4***

33.57*** 71.11*** 51.74*** 14.85 29.89** 64.09*** 39.82*** 54.83*** 76.33*** 64.49*** 64.14*** 31.05*** 76.90*** 74.35* 17.59 35.18** 58.35**

91.68* 6706 79.57 215.8 0.120 0.0364 91056.53 1138.43*** 59

−36.67

−86.05*

6706 81.07 202.3 0.146 0.170 89989.1 790.19*** 59

6706 87.40 187.8 0.181 0.239 89171.91 354.57*** 59

Model 5 8.612*** −0.0520*** −3.306 −17.40 −22.20* 20.66 69.00*** 128.2** 4.698 69.12*** 47.55*** 6.945 24.10* 59.25*** 31.34*** 50.69*** 70.83*** 53.94*** 48.65*** 28.44*** 68.35*** 60.58 11.53 32.46** 59.19**

54.44*** 64.03*** 68.43*** -3.255 58.90*** 52.21* 64.40*** −84.48* 6706 81.90 195.8 0.162 0.219 89506.48 412.92*** 59

45.21*** 18.05** 18.45** -15.85 32.18*** 33.05 18.83* −91.88* 6706 86.30 187.2 0.180 0.250 89127.56 – 59

Rho = u2 /v2 +u2 where v2 and u2 are the variance components of the composite error term. All estimates of ␤-coefficientsin the table represent the quantification of the parameters associated with patient-level characteristics. Akike’ information criterion (AIC) was used to compare the fit of the models (1-5). The pairwise Likelihood ratio-tests show the results of testing Model 1 versus Model 2, Model 2 versus Model 3 etc., respectively. * p < 0.05. ** p < 0.01. *** p < 0.001.

5.5. Markers for morbidity adjustment

5.6. Considerations for a blended remuneration system

We find that the RUB-markers are the most appropriate for casemix adjustment among the three sets of markers. The non-mutually exclusive ICPC-2 chapter markers and ICPC-2 chapter component markers describe the prevalence of morbidity and diagnoses rather than co-morbidity burden/casemix. It could be useful to reward GPs with complex patients according to RUB classes 0-4 and merge RUB4 and RUB5 because RUB5 is sparsely populated. Patients with a higher morbidity burden seem to have more fluctuations in FFSE in GP clinics. The explanation may be that some of these T2DM patients are treated in hospitals rather than GP clinics.

Unblended strategies to payment reforms may result in distortions in practice patterns [39,40]. Blended systems seem to be more appropriate [40]. A change towards increased blending of systems could be a system, where fees are differentiated according to a morbidity burden measure such as RUBs, either through morbidity adjustment of the variable FFS payments and/or fixed capitation payments. Due to the prospective nature of FFS and capitation fees, morbidity-adjusted remuneration requires prospective diagnostic information based on assumptions. However, adjustments of fees such as capitation fees can in practice be made both retrospectively and prospectively

214

T. Kristensen et al. / Health Policy 113 (2013) 206–215

[14,15]. Retrospective calibrations are somewhat more difficult to implement (e.g. they involve an “after the fact” settling-up component), but they have the advantage of doing a better job of accounting for differences among risk pools. Another advantage is a higher predictive capability and that prospective calibration can include acute medical conditions (including injuries/accidents) [14]. In contrast, retrospective calibrations may establish poorer incentives for diagnostic coding and appropriate medical care than prospective calibration [14,15]. Whether a possible morbidity-based remuneration system in Danish primary care should be retrospective or prospective and based on adjustments of FFS or capitation fees is a political decision. Proponents for morbidity adjustments of capitation payment may argue that GPs are less reluctant to undertake strategic coding of chronic diseases for adjustment of capitation payment than strategic coding of morbidity-adjusted FFS. The justification is that it is more problematic for the GP to change a general categorisation of patients for permanent capitation payments than to code a single service for the same patient. Morbidity-adjusted FFS may, for instance, incentivise a GP to code a T2DM diabetes patient’ visit due to pain in the throat as T2DM rather than pain in the throat. In contrast, proponents for adjustment of FFS fees may argue that strategic coding could be minimised if all FFS fees were calibrated proportionally according to a morbidity index such as RUB0-5. Thus, a patient with at higher morbidity burden will release a proportionally higher fee for all service. 5.7. Policy and research implications This study indicates that there may be gaps between the resource allocation actually observed and the observed morbidity burden for T2DM in Danish primary care. Thus, it may be useful to implement a morbidity-based case mix system. This could potentially contribute to improved quality of care and equity without sacrificing efficiency [15,40,41]. In Denmark, it is considered technically feasible to implement a casemix system, because the GPs have started to code diagnoses and Danish authorities have started to pilot-test the ACG system in the hospital sector [9]. Simplified morbidity categories such as RUBs are relatively easy to understand and implement (they only require demographic and diagnostic information [14]). However, the scale and extent to which morbidity adjustment (i.e. risk adjustment) of Danish GPs’ remuneration systems are needed as a critical element of reforming is not well explored. More research than this pilot study is needed. Finally, it should be recognised that the effect of morbidity adjustment is dependent on the context and the way this adjustment is implemented [40]. For instance, the balance between FFS and capitation fees may influence the effect of morbidity-adjusted payments and incentives for cost savings [14,42,43]. 5.8. Limitations of this study The limitations of this type of study have been described [10]. The present study is based on a subsample of all T2DM diabetes patients from a dataset of 139,527 patients that is

representative of GP patients but not GP clinics [10]. Due to this method of collecting the sample, the results may not be representative of all T2DM patients and sentinel clinics. The findings may also be confounded by difficulties with proper coding and registration such as miscoding, misclassification and misdiagnosis of T2DM patients [44]. However, the data from sentinel GP clinics are considered the best available for research purposes [18]. The systematic distribution of T2DM patients across sentinel clinics is a result of the way GP patients are linked to GPs. On the one hand, we do not believe that GPs are cream-skimming T2DM patients with fewer morbidities, as patients choose the GP and not vice versa. GP patients are able to choose the age and sex of the GP, but no quality indicators are available. On the other hand, we have reason to believe that the GPs select their patient populations, when they decide to buy a clinic. 6. Conclusion Large variations exist in the FFSE per T2DM patient in Danish primary care. T2DM patients have a high degree of co-morbidity, and patient co-morbidity characteristics are significant FFSE drivers. An increasing prevalence of co-morbidity implies significantly higher and more variability in FFSE. Denmark has yet to reorient the resource allocation system for GP clinics towards morbidity-based remuneration. T2DM patients’ age, gender and co-morbidity characteristics explained 17-25% of the variation in negotiated FFSE. We find that there may be room for improvement of the association between morbidity burden and resource allocation given a standard mixed remuneration system. The Danish authorities have undertaken active piloting of the ACG system [8]. Therefore, it may be both technically feasible and policy-relevant to introduce a morbidity-based casemix adjustment system or at least advisable to conduct further research in this area. Acknowledgment We are grateful for comments from two anonymous referees and The Mt Hood 2012 Challenge, Baltimore Maryland, USA. References [1] Salisbury C. Multimorbidity: redesigning health care for people who use it. Lancet 2012;380(9836):7–9. [2] Fortin M, Lapointe L, Hudon C, Vanasse A. Multimorbidity is common to family practice: is it commonly researched? Canadian Family Physician 2005;51:244–5. [3] Starfield B. Primary care. Participants or gatekeepers? Diabetes Care 1994;17(Suppl. 1):12–7. [4] Starfield B, Lemke KW, Bernhardt T, Foldes SS, Forrest CB, Weiner JP. Comorbidity: implications for the importance of primary care in ‘case’ management. The Annals of Family Medicine 2003;1(1):8–14. [5] Dawes M. Co-morbidity: we need a guideline for each patient not a guideline for each disease. Family Practice 2010;27(1):1–2. [6] OECD Health Division. OECD reviews of health care quality: Denmark – executive summary; assessment and recommendations. OECD Better policies for better lives; 2013. p. 1–34. [7] Kadam U. Redesigning the general practice consultation to improve care for patients with multimorbidity. BMJ 2012;345:e6202.

T. Kristensen et al. / Health Policy 113 (2013) 206–215 [8] Boult C, Counsell SR, Leipzig RM, Berenson RA. The urgency of preparing primary care physicians to care for older people with chronic illnesses. Health Affairs (Millwood) 2010;29(5):811–8. [9] Starfield B, Kinder K. Multimorbidity and its measurement. Health Policy 2011;103(1):3–8. [10] Kristensen T, Rose Olsen K, Schroll H, Thomsen JL, Halling A. Association between fee-for-service expenditure and morbidity burden in primary care. The European Journal of Health Economics 2013;(July), http://dx.doi.org/10 1007/s10198-013-0499-7 [Epub ahead of print]. [11] van Baal PH, Engelfriet PM, Boshuizen HC, van de Kassteele J, Schellevis FG, Hoogenveen RT. Co-occurrence of diabetes, myocardial infarction, stroke, and cancer: quantifying age patterns in the Dutch population using health survey data. Population Health Metrics 2011;9:51, http://dx.doi.org/10.1186/1478-7954-9-51. [12] Olejaz M, Juul Nielsen A, Rudkjøbing A, Okkels Birk H, Kransik A, Hernández-Ouevedo C. Denmark: health system review. Health Systems in Transition 2012;14(2):1–192. [13] Pedersen K, Andersen J, Søndergaard J. General practice and primary health care in Denmark. Journal of American Board of Family Medicine 2012;25(March–April):S34–8. [14] Ellis RP, Pope GC. Diagnoses-based risk adjustment for medicare capitation payments. Health Care Financing Review 1996;17(3):101–28. [15] Newhouse JP, Butin MB, Chapman JD. Risk adjustment and medicare: taking a closer look. Health Affairs 1997;16(5):26–43. [16] Schroll H, Christensen B, Andersen JS, Sondergaard J. Danish General Medicine Database – future tool! The Danish Society of General Medicine. Ugeskrift for Laeger 2008;170(12):1013. [17] Schroll H. Data collection perspectives from patient care in general practice. Ugeskrift for Laeger 2009;171(20):1681–90. [18] Schroll H, Christensen RD, Thomsen JL, Andersen M, Friborg S, Sondergaard J. The Danish model for improvement of diabetes care in general practice: impact of automated collection and feedback of patient data. International Journal of Family Medicine 2012 [volume: 208123]. [19] Kristensen T, Laudicella M, Ejersted C, Street A. Cost variation in diabetes care delivered in English hospitals. Diabetic Medicine 2010;27(8):949–57. [20] Laudicella M, Olsen KR, Street A. Examining cost variation across hospital departments – a two-stage multi-level approach using patient-level data. Social Science & Medicine 2010;71(10):1872–80. [21] Orueta JF, Lopez-De-Munain J, Baez K, Aiarzaguena JM, Aranguren JI, Pedrero E. Application of the ambulatory care groups in the primary care of a European national health care system: does it work? Medical Care 1999;37(3):238–48. [22] Weiner JP, Starfield BH, Steinwachs DM, Mumford LM. Development and application of a population-oriented measure of ambulatory care case-mix. Medical Care 1991;29(5):452–72. [23] The Health Services Research & Development Center at The Johns Hopkins University. The Johns Hopkins ACG system, Technical Reference Guide Version 9.0. Baltimore, MD, USA: Bloomberg School of Public Health; 2009. [24] WONCA. International classification of primary care ICPC-2-R, 2nd ed. New York: Oxford University Press; 2005. [25] Schokkaert E, Van de Voorde C. Direct versus indirect standardization in risk adjustment. Journal of Health Economics 2009;28(2):361–74. [26] Sayer GP, Britt H. Sex differences in morbidity: a case of discrimination in general practice. Social Science & Medicine 1996;42(2):257–64.

215

[27] Gijsbers van Wijk CM, Kolk AM, van den Bosch WJ, van den Hoogen HJ. Male and female morbidity in general practice: the nature of sex differences. Social Science & Medicine 1992;35(5): 665–78. [28] Kautzky-Willer A, Kamyar MR, Gerhat D, Handisurya A, Stemer G, Hudson S, et al. Sex-specific differences in metabolic control, cardiovascular risk, and interventions in patients with type 2 diabetes mellitus. Gender Medicine 2010;7(6):571–83. [29] Akaike H. A new look at the statistical model identification. IEEE Transactions on Automatic Control 1974;19(6):716–23. [30] Zielinski A, Kronogard M, Lenhoff H, Halling A. Validation of ACG case-mix for equitable resource allocation in Swedish primary health care. BMC Public Health 2009;9:347, http://dx.doi.org/10.1186/1471-2458-9-347. [31] Charlton J, Latinovic R, Gulliford MC. Explaining the decline in early mortality in men and women with type 2 diabetes: a populationbased cohort study. Diabetes Care 2008;31(9):1761–70. [32] Schokkaert E, Van de Voorde C. Risk selection and the specification of the conventional risk adjustment formula. Journal of Health Economics 2004;23:1237–40. [33] Lehnert T, Heider D, Leicht H, Heinrich S, Corrieri S, Luppa MH, et al. Health care utilization and costs of elderly persons with multiple chronic conditions. Medical Care Research and Review 2011;68(4):389–420. [34] Hossain MP, Palmer D, Goyder E, Nashas EL. Social deprivation and prevalence of chronic kidney disease in the UK: workload implications for primary care. Quarterly Journal of Medicine 2012;105:167–75. [35] Williams ED, Maliano DJ, Zimmet PZ, Kavanagh AM, Stevenson CE, Oldenburg BF, et al. Area-level socioeconomic status and incidence of abnormal glucose metabolism. Diabetes Care 2012;35:1455–60. [36] Gaede P, Vedel P, Larsen N, Jensen GV, Parving HH, Pedersen O. Multifactorial Intervention and cardiovascular disease in patients with type 2 diabetes. New England Journal of Medicine 2003;348(5):383–93. [37] Olsen KR. Patient complexity and GPs’ income under mixed remuneration. Health Economics 2012;21(6):619–32. [38] Borgquist L, Andersson D, Engström S, Magnusson H:. Primärvårdsersätning efter socioekonomi eller sjukdomsbörda. Registerstuide från Östergötland visar på olika svar beroende på analyseniveå. Läkartidningen 2010;37(107):2158–60. [39] Krasnik A, Groenewegen PP, Pedersen PA, von SP, Mooney G, Gottschau A, et al. Changing remuneration systems: effects on activity in general practice. Britsh Medical Journal 1990;300(6741):1698–700. [40] Goroll AH. Payment reform to support lasting practice reform in primary care. Journal of ambulatory care management 2011;34(1):33–7. [41] Newhouse JP. Reimbursing health plans and health providers: efficiency in production versus selection. Journal of Economic Literature 1996;XXXIV:1236–63. [42] Ash AS, Ellis RP. Risk-adjusted payment and performance assessment for primary care. Medical Care 2012;50(8):643–53. [43] McClellan. Reforming payments to healthcare providers: the key to slowing healthcare cost growth while improving quality? Journal of Economic Perspectives 2011;25(2):69–92. [44] de Lusignan S, Sadek N, Mulnier H, Tahir A, Russell-Jones D, Khunti K. Miscoding, misclassification and misdiagnosis of diabetes in primary care. Diabetic Medicine 2012;29(2):181–9.