The excess health care costs of KardioPro, an integrated care program for coronary heart disease prevention

Health Policy 119 (2015) 778–786

Contents lists available at ScienceDirect

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

The excess health care costs of KardioPro, an integrated care program for coronary heart disease prevention Christian Becker ∗ , Rolf Holle, Björn Stollenwerk Institute of Health Economics and Health Care Management, Helmholtz Zentrum München – German Research Center for Environmental Health, Neuherberg, Germany

a r t i c l e

i n f o

Article history: Received 4 June 2014 Received in revised form 17 December 2014 Accepted 19 January 2015 Keywords: Primary health care Health care reform Delivery of health care Integrated Insurance Health Health care costs

a b s t r a c t Coronary heart disease (CHD) is a major cause of death and important driver of health care costs. Recent German health care reforms have promoted integrated care contracts allowing statutory health insurance providers more room to organize health care provision. One provider offers KardioPro, an integrated primary care-based CHD prevention program. As insurance providers should be aware of the financial consequences when developing optional programs, this study aims to analyze the costs associated with KardioPro participation. 13,264 KardioPro participants were compared with a propensity score-matched control group. Post-enrollment health care costs were calculated based on routine data over a follow-up period of up to 4 years. For those people who incurred costs, KardioPro participation was significantly associated with increased physician costs (by 33%), reduced hospital costs (by 19%), and reduced pharmaceutical costs (by 16%). Overall costs were increased by 4%, but this was not significant. Total excess costs per observation year were D 131 per person (95% confidence interval: [D −36.5; D 296]). Overall, KardioPro likely affected treatment as the program increased costs of physician services and reduced costs of hospital services. Further effects of substituting potential inpatient care with increased outpatient care might become fully apparent only over a longer time horizon. © 2015 Elsevier Ireland Ltd. All rights reserved.

1. Introduction Cardiovascular disease (CVD) is the leading cause of death in industrialized countries. In addition, its direct medical costs have been estimated to sum to about 17% and 11% of overall health care expenditure in the United States and Germany, respectively, largely because of hospital stays and pharmaceuticals [1–3]. Aging populations

∗ Corresponding author at: Institute of Health Economics and Health Care Management, Helmholtz Zentrum München – German Research Center for Environmental Health, Ingolstädter Landstraße 1, 85758 Neuherberg, Germany. Tel.: +49 089 3187 3943; fax: +49 089 3187 3375. E-mail address: [email protected] (C. Becker). http://dx.doi.org/10.1016/j.healthpol.2015.01.012 0168-8510/© 2015 Elsevier Ireland Ltd. All rights reserved.

and a continually high incidence of cardiovascular risk factors are expected to contribute to a substantial increase in the prevalence of CVD in the near future [4–6]. Likewise, the economic impact of this disease is projected to triple in the United States up to 2030 [2]. Consequently, prevention and control of CVD are both recognized to be a high priority concern in health policy making [7,8]. In the German statutory health insurance system, there are several approaches to improving the prevention of CVD, including structured disease management programs and refunds for completing prevention activities [9,10]. In addition, recent health care reforms have promoted structured programs of care, based on integrated care contracts with selected service providers [11]. These integrated care contracts, which were provided with public

C. Becker et al. / Health Policy 119 (2015) 778–786

start-up funds, gave individual health insurance providers a more active role in influencing health services and the quality of care offered to their members [7,12]. One integrated care contract is KardioPro, a primary care-based prevention program that was introduced by the Siemens company health insurance (SBK) in 2006, and addressed both those with established coronary heart disease (CHD) and, unlike regular disease management programs, those with an elevated risk of CHD [13]. Although the German statutory health insurance companies are legally obliged to operate efficiently, and certain types of prevention programs, such as bonus payments and disease management programs, legally require scientific evaluation, this is not mandatory for integrated care contracts such as KardioPro [14]. Accordingly, analyses of the economic effects of CVD prevention in Germany have focused on disease management or bonus programs, and analyses of health care expenditure associated with optional prevention programs, based on integrated care contracts, remain scarce [9,10,15,16]. Yet, especially in light of the discontinuation of federal start-up financing in 2009, there needs to be awareness of the economic consequences of providing optional integrated care programs for CHD prevention in order to evaluate whether offering such a program is economically reasonable. Therefore, it is the aim of this paper to analyze the impact of the primary care-based CHD prevention program KardioPro on health care expenditure by the health insurance provider. 2. Material and methods 2.1. The cardiovascular prevention program KardioPro KardioPro is an integrated health care program for CHD prevention, which was introduced into the SBK’s standard service portfolio in 2006 [13]. Particular aims of the program were to improve early detection of CHD or CHD risk factors and to positively influence the course of disease. KardioPro addressed SBK policyholders and co-insured family members of any age who had established CHD and those who were above 45 years of age and did not have CHD. Participation in KardioPro was voluntary and written consent was obtained from participants. The program started in the Munich area and was expanded to Coburg, Berlin, Karlsruhe, Nuremberg/Erlangen, and North RhineWestphalia. The intervention was provided by office-based cardiologists and consisted of complex pathways of diagnosis and treatment that differed depending on which risk group the participants belonged to. Risk groups were determined by establishing the participants’ CHD status (known, unknown but symptomatic, unknown and asymptomatic). Within these groups, the participants were also stratified according to their individual 10-year cardiovascular risk, based on the Prospective Cardiovascular Münster study (PROCAM) score [17]. After an initial assessment, those with low coronary risk were scheduled for a 5-year follow-up visit. For the remaining participants, the disease status was intensively clarified, based on specific medical examinations, and,

779

if necessary, further diagnostics, such as computerized tomography angiography, were carried out. Depending on the results, appropriate interventions, such as percutaneous coronary intervention, were proposed or follow-up visits were scheduled, which ranged between 3 months and 5 years. All participants were offered consultation about lifestyle changes or medication. The program also included a reminder system for follow-up visits and electronic patient records in order to promote coordination among the various service providers. 2.2. Study population An intervention group and a control group were constructed from a subsample of SBK policyholders and co-insured family members, who were insured with the SBK in 2007, 2008, or 2009, were aged over 39 years (the age of the youngest KardioPro participant), and were not eligible for a nursing care level. This retrospective analysis of the costs associated with KardioPro was based on deidentified SBK routine data. The evaluation of costs did not interfere with how subjects were treated or which subjects were treated, nor did it affect data collection. According to the ethics committee of the State Chamber of Physicians of Bavaria, no ethical approval was required. First, in order to generate an intervention group, the routine data were screened for SBK policyholders or co-insured family members who had been enrolled in KardioPro in 2007, 2008, or 2009, and were insured with the SBK for the entire year prior to enrollment. If participants’ enrollment date was missing, they had likely dropped out of the program without presenting for the initial assessment. In these cases, the enrollment was assumed to have been 150 days prior to the recorded dropout date, because the SBK considered participants to be dropouts after 150 days of inactivity. In this cost analysis, however, participants who dropped out of KardioPro were retained in the intervention group, according to the intention to treat approach. Because KardioPro was designed to be part of the SBK’s regular service portfolio and potentially open to all eligible SBK members, there was no randomization of participants and control subjects. Therefore, in this retrospective analysis, a control group was constructed using propensity score matching [18]. That means, KardioPro participants and non-participants were matched, based on their probability of enrolling in the program (that is, the propensity score), which is predicted using a logistic regression model. This method allows matching on multiple characteristics, while these are included in a single score. In particular, the control group was designed by conducting one propensity score matching for each of the groups of participants enrolling in 2007, 2008, or 2009. The propensity score models were estimated based on the particular year’s participants and non-participants who were insured over the entire previous year. These models were developed using stepwise variable selection, based on the individuals’ characteristics as documented at the end of the previous year. The variables included in the propensity scores are supplied as supplemental material. These included age, sex, postal code (3 digits), annual health care

780

C. Becker et al. / Health Policy 119 (2015) 778–786

expenditure, actuarial characteristics (type of insurance and status of reduced earning capacity), and morbidity. In particular, most of the information on morbidity was based on hierarchical morbidity groups, which refer to the chronic diseases and other severe cost-intensive health conditions that are considered in the German morbiditybased risk adjustment scheme. These groups must be recorded routinely by statutory health insurance companies at the end of each calendar year, in order to enable morbidity-based risk adjustment. The method of calculating these variables is defined by the German Federal Social Insurance Authority. In particular, people are assumed to have a considered disease if, in their claims data, there were at least two validated ambulatory diagnoses in two different quarters, or one stationary diagnosis that was associated with this disease. In addition to the hierarchical morbidity groups, we included a variable that indicated the presence of obesity, which is not considered in the German risk adjustment fund. This variable was generated from ambulatory and inpatient diagnoses, where at least two ambulatory diagnoses were required. Matching was conducted using the approximate 1:1 nearest neighbor matching approach without replacement, introduced by Parsons [19]. The complete study population was created by combining all three separately matched populations. 2.3. Outcome measures The cost analysis was based on routine data from the SBK. These data included individual annual expenditure for filed claims; the costs of designing KardioPro were not provided and are not part of this analysis. Expenditure data were provided for the categories of hospital, primary care physicians, pharmaceuticals, and others (that is, aids and remedies and other service providers, dialysis, sickness payments, and optional benefits). Except for expenditure on dental care, which was not provided at the individual level for the entire study period, the cost calculation included direct medical expenditure as well as sickness payments from the SBK. Primary outcome measures were the average total costs per person per day under observation. An outcome measure with a time reference was chosen in order to exclude one of the groups to benefit from a potentially higher mortality, which would cause lower total costs. Analyses were also conducted for the costs in each of the coverage parts, as well as for costs incurred in the first, second, and third calendar years of KardioPro participation. Generally, costs were calculated by summing every person’s costs and dividing these by the number of days under observation. Although entry into KardioPro was possible at any time during the year, in the year of program entry, the average costs per person per day were calculated starting from January 1, as only the annual health care expenditures were available. Thereby, it was assumed that, due to the matching, the daily costs of participants and control subjects did not differ between January 1 and program entry. In general, the cost calculation ended at death or at December 31, 2010. Censoring occurred in cases where a matched partner left the insurance company or if a matched control

became a KardioPro participant. As costs were calculated over a period of more than a year, these were inflated to 2010 euros, using the inflation rate from German official statistics, and discounted to the year of program entry, using a discount factor of 3% [20]. Although costs per day were the dependent variable in all calculations, in the part of the results section reporting the adjusted mean costs, we expressed the value of the outcome variable in the unit “per observation year”, in order to facilitate a better understanding. 2.4. Statistical analysis The influence of KardioPro on the total costs per person per day was analyzed using a generalized linear model with gamma distribution and log-link [21,22]. This type of model recognizes the skewed distribution of cost data while avoiding the need for data retransformation. We conducted tests for distributional form and link function. According to the modified park test, the choice of gamma distribution was justified. The log-link provided the best choice for all models according to the modified Hosmer Lemeshow test, where the plot of mean residuals by deciles of the linear prediction showed no systematic pattern. To take into account observations with zero costs, which are not included in the gamma distribution, a two-part approach was selected [23,24]. In the first step, the probability of incurring positive costs was modeled using logistic regression; in the second step, the amount of positive costs was modeled for those who incurred positive costs, using generalized linear regression. As slight differences in the real distribution of the response variable and the model’s distributional assumption may possibly cause bias, especially in analyses with large numbers of observations, parameter estimates and p-values were derived by bootstrapping the original data set (1000 bootstraps) to increase robustness [25]. As one two-part model was estimated for total costs and one for each cost category, the coefficient tests for “KardioPro participation” can be seen as 10 tests according to the global hypothesis that resource utilization is the same for participants and control subjects. Therefore, the Bonferroni method was used to adjust for multiple testing [26]. Accordingly, in the tests of equality of resource consumption in the intervention and control groups, reported in Table 2, statistical significance was assumed if the p-value was below 0.05/10 = 0.005. Likewise, the equality of adjusted predictions was tested five times; statistical significance was assumed if the p-value was below 0.05/5 = 0.01. The average costs per person per day and the cost difference of KardioPro participants and matched control subjects have been adjusted for the participants’ covariate structure with regard to the variables “age”, “sex”, “year of matching”, and “health care expenditure in the year prior to program entry”. This was done using the method of “recycled predictions” [22,27]. This means that the corresponding two-part model, as described above, was estimated based on participants and control subjects. Using these parameter estimates, the costs were then predicted for two data sets: one in which every participant’s participation status was set to “yes” and one in which it was set to

C. Becker et al. / Health Policy 119 (2015) 778–786

781

Table 1 Characteristics of the combined sample.

Age (years: mean, SD) Female (n, %) Hypertension (n, %) CHD (n, %) Diabetes (n, %) Obesity (n, %) Stroke history (n, %) Congenital heart defect (n, %)

KardioPro participants (n = 13,264)

Control group (n = 13,264)

59.2 (8.7) 6080 (44.4) 3801 (28.7) 871 (6.6) 1551 (11.7) 1676 (12.6) 27 (0.2) 647 (4.9)

59.2 (8.6) 5883 (45.8) 3885 (29.3) 923 (7.0) 1665 (12.6) 1727 (13.0) 30 (0.2) 622 (4.7)

SD: standard deviation; CHD: coronary heart disease. Fig. 1. Correction for enrollment during the first calendar year. cbefore : predicted costs per day for control subjects, in the first calendar year. cafter : costs per day for participants in the period after enrollment only (unknown). caverage : predicted costs per day for participants, in the first calendar year. tbefore : number of days until enrollment. tafter : number of days after enrollment.

“no”. The subsequent difference thus isolates the marginal effect of KardioPro participation. As persons could enroll in KardioPro at any time during a calendar year, the participants’ costs per person per day during the first calendar year were likely to be underestimated (see Fig. 1). The cost differences for the first calendar year of participation were therefore calculated both with and without a correction for enrollment during the year. The correction was based on the estimates of the whole population and assumes that cbefore , the costs per insured day before enrollment, corresponded to the average cost of non-participants. As only the average costs (caverage ) were provided, the time before enrollment (tbefore ) and the time after enrollment (tafter ) were used to disaggregate the average costs, resulting in costs after enrollment (cafter ). The disaggregation was performed as follows: c after =

c average (t before + t after ) − c before t before t after

.

The input value caverage designates the KardioPro participants’ average costs as predicted through the regression model; the input value cbefore was based on the recycled prediction approach, where the status KardioPro participation was set to “no” for the same prediction population. 3. Results Between 2007 and 2009, 13,265 people enrolled in KardioPro. Matching resulted in a control group of 13,264 people overall, as there was no matching control for one of the participants. The study sample thus included 26,528 people (that is 13,264 matched pairs). During the observation period, 2356 people dropped out of KardioPro, but remained in the intervention group according to the intention to treat approach. A total of 381 people died, and 2002 observations were censored because a member of the matched pair changed insurance, and 593 members of the control group were enrolled in KardioPro in a subsequent year. In both groups, the average age was 59.2 years. Some 44.4% of the KardioPro participants were female and 6.6% had CHD. In the control group, these proportions were

45.8% and 7% respectively. Baseline characteristics regarding the comorbidities of hypertension, diabetes, obesity, and history of stroke are presented in Table 1. The results of the two-part models are presented in Table 2. According to the first parts of the models, KardioPro was significantly positively associated with incurring health care costs. Compared with the control group, the participants’ relative chance of incurring positive costs was about 16 times higher. With regard to the second part of the models, modeling the total amount of positive costs for those people who actually incurred positive costs, participation in KardioPro was significantly associated with an increase in physician costs by 33% (p < 0.001), a decrease in hospital costs by 19% (p < 0.001), a decrease in pharmaceutical costs by 16% (p = 0.002), and a decrease in other costs by 12% (p < 0.001). Finally, KardioPro was associated with an increase in total costs of 4%, although this was not significant (p = 0.176). As the five two-part models presented in Table 2 can be seen as 10 tests of equality of costs in the intervention and control groups, according to the Bonferroni correction, statistical significance was assumed if the p-value was below 0.05/10 = 0.005. Table 3 shows the mean costs per person per observation year for the five cost categories “total costs”, “primary care physicians”, “hospital care”, “pharmaceutical”, and “other”. These are based on five independently estimated two-part models, which each predict costs for the particular category for a cohort having the same distribution of the covariables “age”, “sex”, “costs incurred the year before matching”, and “year of matching” as the KardioPro participants. According to the model predicting total costs, total costs of D 3099 were incurred per observation year in the group of KardioPro participants, and D 2967 per person per observation year in the control group. Accordingly, there were excess costs of D 131 (p < 0.120, 95% confidence interval (CI): [D −36.5; D 296]). According to the model predicting costs of physician services, KardioPro participants had significant excess costs of D 215 (p < 0.001, 95% CI: [D 201; D 234]). According to the model predicting costs of hospital services, KardioPro participants had significant excess costs of D −150 (p < 0.002, 95% CI: [D −252; D −54.8]). According to the model predicting pharmaceutical costs, KardioPro participants had significant excess costs of D −110 (p < 0.002, 95% CI: [D −179; D −43.8]). As the five differences in predicted costs for the intervention and control groups can be seen as five tests of equality in

782

C. Becker et al. / Health Policy 119 (2015) 778–786

Table 2 Parameter estimates of the two-part regression model: likelihood of incurring costs (Part 1) and amount of costs for those who incur costs (Part 2). Part 1: logistic regressiona [response: positive costs per day for given cost category (yes vs. no)] Covariate

Total costs

Age (years) Sex (male vs. female) KardioPro participation (yes vs. no) Costs before participation (Q1 vs. Q4) Costs before participation (Q2 vs. Q4) Costs before participation (Q3 vs. Q4) Year of matching (2007 vs. 2006) Year of matching (2008 vs. 2006)

Physician costs

Hospital costs

Odds ratio

p-valueb

Odds ratio

pvalueb

Odds ratio

p-valueb

Pharmaceutical costs Odds ratio

p-valueb

Other costs Odds ratio

pvalueb

0.988 0.810 15.852

0.106 0.154 <0.001

0.990 0.892 9.442

0.170 0.330 <0.001

1.032 1.118 1.175

<0.001 <0.001 <0.001

1.022 0.814 1.950

<0.001 0.004 <0.001

1.004 0.696 11.220

0.124 <0.001 <0.001

0.007

<0.001

0.049

<0.001

0.238

<0.001

0.064

<0.001

0.173

<0.001

0.189

0.034

0.577

0.114

0.353

<0.001

0.420

<0.001

0.340

<0.001

0.271

0.094

1.228

0.626

0.522

<0.001

0.943

0.766

0.722

<0.001

0.758

0.154

0.698

0.028

0.722

<0.001

0.739

0.002

0.734

<0.001

0.582

<0.001

0.540

<0.001

0.518

<0.001

0.552

<0.001

0.394

<0.001

Part 2: gamma regressiona [response: amount of positive costs per day for given cost category] Covariate

Age (years) Sex (male vs. female) KardioPro participation (yes vs. no) Costs before participation (Q1 vs. Q4) Costs before participation (Q2 vs. Q4) Costs before participation (Q3 vs. Q4) Year of matching (2007 vs. 2006) Year of matching (2008 vs. 2006)

Total costs

Physician costs b

Hospital costs b

Pharmaceutical costs b

Other costs b

Exp (estimate)

p-value

Exp (estimate)

p-value

Exp (estimate)

p-value

Exp (estimate)

p-value

Exp (estimate)

p-valueb

1.018 1.068

<0.001 0.026

1.013 0.886

<0.001 <0.001

1.025 1.179

<0.001 <0.001

1.020 1.091

<0.001 0.112

0.981 1.135

<0.001 <0.001

1.040

0.176

1.326

<0.001

0.806

<0.001

0.839

0.002

0.877

<0.001

0.211

<0.001

0.310

<0.001

0.604

<0.001

0.103

<0.001

0.239

<0.001

0.311

<0.001

0.493

<0.001

0.571

<0.001

0.196

<0.001

0.295

<0.001

0.487

<0.001

0.702

<0.001

0.653

<0.001

0.377

<0.001

0.444

<0.001

0.967

0.280

0.976

0.106

1.171

<0.001

0.915

0.136

0.978

0.660

1.019

0.602

1.017

0.286

1.483

<0.001

1.072

0.302

0.928

0.090

a Parameter estimates and p-values are based on bootstrapping (1000 repetitions, percentile method), adjusted for age, sex, costs in the year before KardioPro participation, and year of matching. b Owing to multiple testing, statistical significance was assumed if the p-value was below 0.05/10 = 0.005, according to the Bonferroni method.

the predicted costs of the intervention and control groups, according to the Bonferroni correction, statistical significance was assumed if the p-value was below 0.05/5 = 0.01. After separating costs in the first, second, and third years of KardioPro, as shown in Fig. 2, the excess costs of KardioPro participants were particularly high in the first year and were negative, but not significant, in the second and third years. Excess physician costs were significantly positive in all three years. Excess hospital costs were negative in all years, although only significantly so in the second year. Excess pharmaceutical costs were significantly negative.

4. Discussion In this study, the excess health care costs associated with the integrated care program KardioPro were analyzed. Overall, the SBK’s routine data pointed toward an influence of KardioPro on health care expenditure. According to the logistic regression model, in the observation period, the chance of incurring costs was higher for KardioPro participants than for members of a matched control group. In order to calculate the excess costs associated with KardioPro, total costs were predicted for participants and control

C. Becker et al. / Health Policy 119 (2015) 778–786

783

Table 3 Costs as predicted by the generalized linear models for cohorts with a distribution of covariables as in the KardioPro participants (per person per observation year in euros). Independent variable a,c

Total costs Primary care physiciansc Hospital carec Pharmaceuticalsc Otherc

KardioPro participants (13,264) b

3095 [2989; 3208] 854 [843; 869] 1033 [971; 1099] 599 [566; 639] 540 [515; 569]

Control group (13,264)

Difference

2967 [2832; 3227] 639 [624; 653] 1183 [1102; 1263] 708 [650; 774] 507 [467; 558]

131c [−36.5; 296] 215 [201; 234] −150 [−252; −55] −110 [−179; −43.8] 32.9 [−11; 73]

a

Corresponds to the primary scenario for testing. Bootstrapped 95% confidence intervals based on 1000 repetitions (percentile method). c The predicted costs for the individual cost components do not add up to the predicted total costs, because the predicted total costs and the predicted costs in each cost category were predicted using independent prediction models. b

Fig. 2. Excess costs per person per observation year in the first 3 years of KardioPro participation. *Excess costs for the whole first calendar year with correction for enrollment during the year.

784

C. Becker et al. / Health Policy 119 (2015) 778–786

subjects according to the “recycled predictions” method using a two-part model. Whereas the results from this model suggested that KardioPro participants had higher total costs than control subjects, the difference was not significant. A possible explanation for this finding could be that KardioPro participation was associated with higher costs in some service sectors, but with lower costs in other service sectors. This was investigated by estimating the predicted costs of participants and control subjects in the individual cost categories. In particular, one independent prediction model was estimated for each of the cost categories. According to these models, there appeared to be opposing effects regarding the excess costs of the individual cost components. In particular, the KardioPro participants’ predicted physician costs were significantly higher than those of non-participating control subjects. In contrast, the participants’ predicted hospital costs and predicted pharmaceutical costs were significantly lower. These findings are also significant after adjustment for multiple testing. The fact that the sum of the predicted costs in the individual cost components is not equal to the predicted total costs is caused by the method of calculating predicted costs from two-part models rather than using crude estimates, which allowed for adjusting costs for possible confounding, but entailed random variation. Although the same parameters were included in each model, these were assigned different weights as the people incurring costs differed across cost categories. In addition, costs were predicted based on independently estimated models, which entails random error. Overall, the point estimate of the model for total costs lies within the uncertainty of the models for single cost components, as can be seen by adding the border values of the 95% confidence intervals of the models for the single cost components. Subgroup analyses of costs per year after enrollment indicated lower costs after the first year of being enrolled. Owing to different observation periods, because of the open cohort design of KardioPro, each of these subgroup analyses applies to the subsample of persons with the corresponding observation period and may not apply to the entire sample in the same way. Although integrated care is generally seen as a viable approach to improving quality of care, especially for patients with chronic conditions such as CHD, there is little knowledge about the economic consequences that health insurance companies face when introducing integrated care programs into their standard care portfolio [7,28]. Whereas the costs of integrated care contracts have been analyzed for conditions such as inflammatory bowel disease, cost analyses of CHD have focused on different service arrangements [29]. In Germany, Stock and others also evaluated a routine care program offered by a statutory health insurance company, using a matched case–control design, and found cost savings, especially for hospital and medication costs. However, they focused on a bonus program that included immunization and multiple primary prevention approaches and was not explicitly directed at CHD prevention [10,16]. In addition, Gapp and others analyzed the outcomes of German CHD disease management programs, but did not take into account health care expenditure [9].

Internationally, there have been previous studies of CHD prevention programs that were offered in standard care. However, these often did not calculate a third party payer’s actual costs, based on administrative data, but rather combined costs from several sources using economic modeling [30–32]. Several trials from the United States have investigated the impact of Medicare-sponsored coordinated care demonstrations on Medicare expenditure [33–35]. These demonstrations mainly focused on primary prevention for patients with manifest CHD and were set up as trials, which included control groups. Overall, the studies by Peikes and others and Esposito and others analyzed 16 programs and found the participants’ Medicare expenditure to be slightly lower in four of these, although not significantly. Program costs far exceeded cost savings [34,35]. Moreover, Zheng and others analyzed two programs and found significant net savings after 3 years of follow-up [33]. Although these three studies report reduced hospital utilization for some programs, only Esposito and others differentiated total expenditures and reported costs separately for Medicare Part A (which primarily refers to hospital coverage) and Part B. However, after 18 months of follow-up, there was no significant difference for both groups [34]. A further study from the United States, by Delate and colleagues, analyzed the health care costs of HMO members who were enrolled in a program of assisted long-term management after an incident coronary event, which focused on behavioral change and drug therapy. Compared with a matched population, the participants’ total expenditures were significantly lower, mostly due to lower hospital expenditure [36]. This study’s results confirm the assertion that prevention programs do not necessarily lead to cost savings, especially over a short time horizon [37]. This can be seen by testing for the difference in total expenditures over the entire observation period, which pointed toward higher expenditures for KardioPro participants, but these were not significant. However, secondary analyses of shorter time horizons and individual cost components indicate effects in different directions. As KardioPro essentially encompassed intensified primary care services provided by office-based cardiologists, such as intensified diagnostics, participants consequently incurred higher physician costs. This expenditure was particularly high in the first year of participation. During the rather short study period of up to 4 years, cost reductions could be observed in other coverage areas, especially with regard to pharmaceutical expenditure. Factors influencing such cost reductions may be the increased prescription of lower priced pharmaceuticals, better drug monitoring, and the avoidance of double prescriptions [14]. Overall cost reductions for hospital services correspond to findings of reduced hospital utilization or costs from the studies mentioned above. In multifaceted interventions such as KardioPro, changes in health care costs cannot easily be attributed to one single program component, and cost reductions cannot easily be compared across different settings. However, there are possible mechanisms explaining the lower hospital costs, pharmaceutical costs, and other costs. These could be that there was possibly earlier detection of existing health problems in KardioPro participants, who therefore

C. Becker et al. / Health Policy 119 (2015) 778–786

would have been less likely to experience fatal events or could have maintained lower doses of medication. An additional study on the effectiveness of KardioPro pointed toward the protective effect of KardioPro with regard to all-cause mortality [38]. Further analyses of medication use could not be conducted in this study because data on medication were not available. Furthermore, increased adherence to treatment guidelines was targeted by KardioPro. This is often seen as key to success in cardiovascular prevention, but has been found to be low, especially in those at risk of CHD [39,40]. Beside the secondary prevention component, KardioPro also provided elements of primary prevention, such as health education. While the costs of primary prevention may contribute to the participants’ higher excess costs in the first year, it is likely that, within the study period of up to 4 years, those health benefits that accrue in the long run could be captured to a small extent only. This might explain why, in this study, unlike in previous studies, substituting inpatient care for increased outpatient care did not offset overall health care costs [41]. In particular, as KardioPro was also directed at those without manifest CHD, the primary prevention component may positively influence modifiable risk factors and thereby facilitate positive long-term health effects. Therefore, the full potential of KardioPro may unfold in the coming years. Beyond looking at health care expenditures alone, there are further considerations that health insurance providers should take into account when deciding whether to offer additional prevention programs. Notably, providers need to determine whether, to them, the prevention program generates sufficient health improvements to justify the additional health care expenditure [42]. Although cost reductions in service areas, such as hospital and pharmaceuticals, point toward health improvements in participants, cost-effectiveness analysis would go beyond the scope of this analysis which is based on cost data only. Aside from health improvements, strategic motives could also justify additional costs. For example, integrated care contracts enable providers of health insurance to become more actively involved in shaping the delivery of health services. Selective contracting also gives health insurance providers additional information on office-based physician services. Both of these may possibly strengthen their position with regard to budget negotiations. Furthermore, prevention programs such as KardioPro may contribute to a positive image of the company and could be used as a marketing tool in order to retain customers [10,43]. Finally, there are further effects to be considered on the revenue side. Since the introduction of the German morbiditybased risk compensation scheme in 2009, all contributions are pooled and reallocated to health insurance companies, based on demographics and morbidity. As program participants receive intensified care, their health status is evaluated more often and more thoroughly than in those not participating. The participants’ health status is therefore more likely to be appropriately classified, and changes in health status are more likely to be detected earlier. Both of these could influence the financial resources allocated to the health insurance company, as this could possibly result in an allocation of more funds to the health insurance company.

785

There are limitations to this study which concern the retrospective study design. As KardioPro was not designed as a randomized controlled trial, there was no control group specified in advance, so propensity score matching was used to create a comparable control group retrospectively, based on demographics and morbidity before program entry. In this way, it could be controlled for observable characteristics; but other characteristics, not included in the available data, such as clinical parameters or behavioral risk factors, could not be taken into account, thereby allowing for potential bias [44]. In addition, income could not be considered as potential confounder, because data on the policyholders’ income were not available in this study and data on the co-insured family members are generally not routinely available. However, parts of this effect may have been covered by the place of residence, which had been controlled for. Another limitation concerns the use of health insurance routine health care expenditure data. Although these included all the SBK members’ health care expenditure, they did not include the costs of designing and controlling KardioPro, which should be taken into account when estimating the actual budget impact. 5. Conclusion The results of this study indicate that participation in the prevention program KardioPro was associated with increased physician costs but with lower costs in other areas, notably hospital services and pharmaceuticals. In the observation period, the participants’ total health care expenditures did not differ significantly from those of a matched control group. As lower costs for hospital services and pharmaceuticals may indicate health improvements for participants, similar programs may constitute one step toward improved prevention of CHD, but the entire effect may manifest fully over time with the primary prevention components becoming fully effective. Acknowledgments We thank the SBK and especially Cathleen Wenning for facilitating this project and providing the required data. We also thank the members of the KardioPro Steering Committee for their active role in designing this prevention program and their engagement in the academic interchange. Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/ j.healthpol.2015.01.012. References [1] Nichols M, Luengo-Fernandez TNR, Leal J, Gray A, Scarborough P, Rayner M. European Cardiovascular Disease Statistics 2012. European Heart Network, Brussels, European Society of Cardiology, Sophia Antipolis; 2012. [2] Heidenreich PA, Trogdon JG, Khavjou OA, Butler J, Dracup K, Ezekowitz MD, et al. Forecasting the future of cardiovascular disease

786

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

[21]

[22]

C. Becker et al. / Health Policy 119 (2015) 778–786 in the United States: a policy statement from the American Heart Association. Circulation 2011;123(8):933–44. Rayner M, Allender S, Scarborough P, British G. Cardiovascular disease in Europe. European Journal of Cardiovascular Prevention and Rehabilitation 2009;16(Suppl. 2):S43–7. Ford ES, Ajani UA, Croft JB, Critchley JA, Labarthe DR, Kottke TE, et al. Explaining the decrease in U.S. deaths from coronary disease, 1980–2000. New England Journal of Medicine 2007;356(23): 2388–98. Muller-Riemenschneider F, Andersohn F, Willich SN. Trends in agestandardised and age-specific mortality from ischaemic heart disease in Germany. Clinical Research in Cardiology 2010;99(9):545–51. Nichols M, Townsend N, Scarborough P, Rayner M. Trends in age-specific coronary heart disease mortality in the European Union over three decades: 1980–2009. European Heart Journal 2013;34(39):3007–13. Amelung V, Hildebrandt H, Wolf S. Integrated care in Germany – a stony but necessary road! International Journal of Integrated Care 2012;12:pe16. Labarthe DR, Dunbar SB. Global cardiovascular health promotion and disease prevention: 2011 and beyond. Circulation 2012;125(21): 2667–76. Gapp O, Schweikert B, Meisinger C, Holle R. Disease management programmes for patients with coronary heart disease – an empirical study of German programmes. Health Policy 2008;88(2–3):176–85. Stock S, Schmidt H, Buscher G, Gerber A, Drabik A, Graf C, et al. Financial incentives in the German Statutory Health Insurance: new findings, new questions. Health Policy 2010;96(1):51–6. Thomson S, Busse R, Crivelli L, van de Ven W, Van de Voorde C. Statutory health insurance competition in Europe: a four-country comparison. Health Policy 2013;109(3):209–25. Lisac M, Reimers L, Henke KD, Schlette S. Access and choice – competition under the roof of solidarity in German health care: an analysis of health policy reforms since 2004. Health Economics, Policy and Law 2010;5(Pt (1)):31–52. Rosenkranz S, Maier LS, Maack C, Bohm M. Hotline update of clinical trials and registries presented at the German Cardiac Society Meeting 2007: 2L-Registry, Kardio-Pro, EVER, AFFECT, VTACH, ARTS II, OPTAMI, PEPCAD I, PEPCAD II, GERSHWIN, SPICE, FIXCHF and CREDIT. Clinical Research in Cardiology 2007;96(7): 457–68. Ehlert A, Oberschachtsiek D. Does managed care reduce health care expenditure? Evidence from spatial panel data. International Journal of Health Care Finance and Economics 2014 [Epub ahead of print]. Conklin A, Nolte E, Vrijhoef H. Approaches to chronic disease management evaluation in use in Europe: a review of current methods and performance measures. International Journal of Technology Assessment in Health Care 2013;29(1):61–70. Stock S, Stollenwerk B, Klever-Deichert G, Redaelli M, Buscher G, Graf C, et al. Preliminary analysis of short-term financial implications of a prevention bonus program: first results from the German Statutory Health Insurance. International Journal of Public Health 2008;53(2):78–86. Assmann G, Cullen P, Schulte H. Simple scoring scheme for calculating the risk of acute coronary events based on the 10-year follow-up of the prospective cardiovascular Munster (PROCAM) study. Circulation 2002;105(3):310–5. D’Agostino Jr RB. Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Statistics in Medicine 1998;17(19):2265–81. Parsons L. Performing a 1: N case-control match on propensity score. In: Proceedings from the 29 h Annual SAS Users Group International Conference. 2004. Schad M, John J. Towards a social discount rate for the economic evaluation of health technologies in Germany: an exploratory analysis. European Journal of Health Economics 2012;13(2):127–44. Manning WG, Basu A, Mullahy J. Generalized modeling approaches to risk adjustment of skewed outcomes data. Journal of Health Economics 2005;24(3):465–88. Glick H, Doshi J, Sonnad S, Polsky D. Economic evaluation in clinical trials. Handbooks in health economic evaluation series, vol. x. Oxford, New York: Oxford University Press; 2007. p. 244.

[23] Afifi AA, Kotlerman JB, Ettner SL, Cowan M. Methods for improving regression analysis for skewed continuous or counted responses. Annual Review of Public Health 2007;28:95–111. [24] Buntin MB, Zaslavsky AM. Too much ado about two-part models and transformation? Comparing methods of modeling Medicare expenditures. Journal of Health Economics 2004;23(3):525–42. [25] Efron B, Tibshirani R. An introduction to the bootstrap. Monographs on statistics and applied probability., vol. xvi. New York: Chapman & Hall; 1993. p. 436. [26] Bland JM, Altman DG. Multiple significance tests: the Bonferroni method. BMJ 1995;310:170. [27] Graubard BI, Korn EL. Predictive margins with survey data. Biometrics 1999;55(2):652–9. [28] Vondeling H. Economic evaluation of integrated care: an introduction. International Journal of Integrated Care 2004;4:pe20. [29] Blumenstein I, Tacke W, Filmann N, Zosel C, Bock H, Heuzeroth V, et al. Integrated management of patients with chronic inflammatory bowel disease in the Rhine-Main Region: results of the first integrated health-care project IBD in Germany. Zeitschrift fur Gastroenterologie 2013;51(7):613–8. [30] Gillespie P, O’Shea E, Murphy AW, Byrne MC, Byrne M, Smith SM, et al. The cost-effectiveness of the SPHERE intervention for the secondary prevention of coronary heart disease. International Journal of Technology Assessment in Health Care 2010;26(3):263–71. [31] Chew DP, Carter R, Rankin B, Boyden A, Egan H. Cost effectiveness of a general practice chronic disease management plan for coronary heart disease in Australia. Australian Health Review 2010;34(2):162–9. [32] Turner DA, Paul S, Stone MA, Juarez-Garcia A, Squire I, Khunti K. Costeffectiveness of a disease management programme for secondary prevention of coronary heart disease and heart failure in primary care. Heart 2008;94(12):1601–6. [33] Zheng H, Ehrlich F, Amin J. Economic evaluation of the direct healthcare cost savings resulting from the use of walking interventions to prevent coronary heart disease in Australia. International Journal of Health Care Finance and Economics 2010;10(2):187–201. [34] Esposito D, Brown R, Chen A, Schore J, Shapiro R. Impacts of a disease management program for dually eligible beneficiaries. Health Care Financing Review 2008;30(1):27–45. [35] Peikes D, Chen A, Schore J, Brown R. Effects of care coordination on hospitalization, quality of care, and health care expenditures among medicare beneficiaries: 15 randomized trials. JAMA 2009;301(6):603–18. [36] Delate T, Olson KL, Rasmussen J, Hutka K, Sandhoff B, Hornak R, et al. Reduced health care expenditures after enrollment in a collaborative cardiac care service. Pharmacotherapy 2010;30(11):1127–35. [37] Cohen JT, Neumann PJ, Weinstein MC. Does preventive care save money? Health economics and the presidential candidates. New England Journal of Medicine 2008;358(7):661–3. [38] Witt S, Leidl R, Becker C, Holle R, Block M, Brachmann J, et al. The effectiveness of the cardiovascular disease prevention programme ‘KardioPro’ initiated by a German sickness fund: a time-to-event analysis of routine data. PLOS ONE 2014;9(12):e114720. [39] Ludt S, Wensing M, Campbell SM, Ose D, van Lieshout J, Rochon J, et al. The challenge of cardiovascular prevention in primary care: implications of a European observational study in 8928 patients at different risk levels. European Journal of Preventive Cardiology 2012;21(2):203–13. [40] Ton VK, Martin SS, Blumenthal RS, Blaha MJ. Comparing the new European cardiovascular disease prevention guideline with prior American Heart Association guidelines: an editorial review. Clinical Cardiology 2013;36(5):E1–6. [41] Vitikainen K, Linna M, Street A. Substituting inpatient for outpatient care: what is the impact on hospital costs and efficiency? European Journal of Health Economics 2010;11(4):395–404. [42] Goetzel RZ. Do prevention or treatment services save money? The wrong debate. Health Affairs (Millwood) 2009;28(1):37–41. [43] Greß S, Focke A, Hessel F, Wasem J. Financial incentives for disease management programmes and integrated care in German social health insurance. Health Policy 2006;78(2–3):295–305. [44] Austin PC. A critical appraisal of propensity-score matching in the medical literature between 1996 and 2003. Statistics in Medicine 2008;27(12):2037–49.