Effectiveness of the German disease management programs: quasi-experimental analyses assessing the population-level health impact

Diabetes and cardiovascular disease are two highly prevalent diseases that pose a large health and cost burden on patients and society. In Germany around 40% of deaths and around 18% of annual health care expenditures are attributable to these two conditions, with coronary heart disease (CHD) being the major contributor to the burden of cardiovascular disease [1, 2]. The adequate management of diabetes and CHD is highly complex and challenging for health care providers, patients and the health care system as a whole. It requires regular check-ups, close monitoring of blood pressure, cholesterol and blood glucose, active patient involvement in managing the disease between check-ups and good coordination of care between general practitioners, nurses, diabetes educators, specialists, and hospitals [3, 4]. Various disease management programs (DMPs) have been developed to address these aspects, and many of them have been shown to be effective in improving intermediate outcomes, such as control of blood glucose, hemoglobin and lipid levels [5, 6].

In 2002/2003 DMPs for type 2 diabetes and CHD were rolled out nationwide in Germany in the statutory health insurance (SHI) system, with the aim of reinforcing guideline care and improving quality of care for the chronically ill [7]. Since their implementation the proportion of patients enrolled in the German DMPs for type 2 diabetes and CHD rose to more than 50% of patients with each respective indication, and today around 6 million people in total are enrolled in the two DMPs [8]. However, despite its high enrollment and estimated annual program costs of around 870€ million (assuming 6 million enrollees and annual DMP administration costs of 145€), the effectiveness of the German DMPs for diabetes and CHD has remained uncertain. A systematic review of the effectiveness of the German DMP for type 2 diabetes, by Fuchs et al., found a positive effect across most studies on process parameters, such as the proportion of patients receiving annual check-ups and indicated medication. Furthermore, the few studies that analyzed mortality and survival also found large positive effects for the type 2 diabetes DMP [9‐12]. Authors further discussed that, due to spillover-effects that might occur when physicians also treat non-enrollees according to DMP standards, the net effect of the intervention might even be underestimated in studies that compare DMP enrollees and controls [13]. However, all of these studies are based on non-randomized comparisons using data from insurance claims, cohort studies or registries and use regression, matching or propensity score methods to control for selection bias. As those approaches do not balance unobservable factors it is possible that the reported effects on health outcomes and mortality might be explained by differences in unobserved characteristics between DMP enrollees and non-enrollees [14, 15]. Those methodological limitations in evaluations of the DMPs have been discussed previously by other authors [16, 17].

In this study, we use a different approach, adopting a population-level health perspective, in evaluating the effectiveness of German DMPs for type 2 Diabetes and CHD in reducing mortality. We apply a quasi-experimental study design, specifically a synthetic control (SC) study, in assessing whether the 2002/2003 introduction of the German type 2 diabetes and CHD DMPs led to reduced population-level mortality relative to other European countries. This approach allowed us to estimate the net effect of the German DMPs on mortality, while accounting for potential spill-over effects and avoiding selection bias.

German DMPs for type 2 diabetes and CHD belong to the biggest DMPs worldwide and administrative costs of those programs are high; however, their effectiveness remains unclear. In contrast to previous studies that analyzed the effect of DMPs using regression, matching or propensity score methods in German datasets, we applied a quasi-experimental study design to compare mortality trends in the German elderly population and mortality trends in other European countries before and after the introduction of the DMPs for DM2 and CHD. This allowed us to estimate not only the net effect of the German DMPs on mortality, but also effects from any potential spill-over of the interventions. Consistently across analyses and outcomes, we observed no clear effect associated with the DMPs in Germany, when compared to a synthetic control group comprising several European countries.

With the introduction of DMPs for type 2 diabetes and CHD in 2002 and 2003 every statutory insured patient with one of these conditions was eligible for participating in those programs. Given this ad-hoc introduction without a scientific testing phase, the short-term and long-term effectiveness and cost-effectiveness of the programs could not be tested in randomized study designs. Since then, several studies aimed to analyze the impact of DMPs on process and health outcomes using observational data. The majority of these focused on diabetes, and most are based on data from insurance claims, cohort studies or registries. Generally, the evidence from these studies suggests a positive impact of DMPs on process and health outcomes. A recent systematic review on the German DMPs for type 2 diabetes identified nine studies, which observed large improvements in mortality and morbidity associated with the DMPs. Stock et al. found a mortality reduction of 49%, Miksch et al. a reduction of 22% and Drabnik et al. a reduction of 49% [10‐12]. However, each of the analyzed outcomes was only assessed by a single or a limited number of studies [9]. Since the publication of this review, further research based on observational data has pointed towards benefits associated with mortality and morbidity in the DMPs for COPD [19]. Most of the analyses in those studies, however, share methodological limitations which may have led to bias in the observed results. Each uses some method to account for selection bias, i.e. to account for differences in individuals enrolled in DMPs and individuals not enrolled in the DMPs. The methods applied to do so differ widely; some use propensity score matching based on several potentially relevant confounders [12, 25], while others more simple matching techniques [11, 26] or adjust for a small selection of variables, such as age and gender in regression analyses [27‐29]. Across all of these studies, however, it is possible that further unobserved variables led to confounding in the observed effect estimates - a limitation also described by other authors commenting on methods for evaluating German DMPs [16, 17]. Furthermore, those studies based exclusively on German data within the SHI system were not able to assess the full effect of the DMP introduction, including potential spill-over effects due to general changes in the care patterns of general practitioners after the DMP implementation. Such effects may also have improved quality of care in DMP non-enrollees.

Our quasi-experimental approach allowed us to avoid the difficulties surrounding the observable and unobservable confounders at the individual level, and to estimate the full net effect of the DMP implementation at the population-level, including potential spill-over effects. These potential spill-over effects are important and should be included when evaluating a health policy from a comprehensive healthsystem perspective. We analyzed circulatory and all-cause mortality, two very distal endpoints in a potential logic model, but could not analyze process outcomes, intermediate clinical outcomes or patient relevant outcomes such as quality of life and patient satisfaction. However, long-term improvements in those process outcomes are expected to prolong life expectancy via improved control of blood sugar levels, hypertension and lipids.

Due to large confidence intervals in our effect estimates, our results cannot reject the possibility that DMPs have a positive effect, but make large positive effects on mortality as reported in the other studies very unlikely [10‐12]. Given the results of this study on the effect of DMPs on circulatory and all-cause mortality, the investments for operating the two DMPs of approximately 870€ million may be questionable. Previous RCTs showed generally a positive effect of DMPs for diabetes on glycemic control, and process outcomes, but the success of these programs were often dependent on the design of the programs and the settings in which they were applied [5, 30]. Furthermore, the efficacy observed in these trials is likely to differ from the effectiveness observed in the real–world setting [31]. Many things that occur in the translation and implementation process of interventions into real world practice, including the rule of halves, could result into reduced or negligible effectiveness [1, 2]. Given all of this, future research that clarifies the impact of German DMPs on both short and long-term type 2 diabetes and CHD-related outcomes is highly warranted.

Our study has several strengths: We included best available data from several European countries in our analysis, which provided rich geographic variability in the construction of the synthetic control group. We considered multiple population-level aspects that may have differed between Germany and the control countries, including the prevalence of obesity, hypertension, smoking and alcohol consumption, the population age-structure, the gross domestic product (GDP) and the unemployment rate. This ensured that these potentially important confounders were accounted for in creating the synthetic control group. The use of time-series data from seventeen years, as opposed to a simple pre-, post-intervention analysis, also ensured that unobserved time-varying confounders were accounted for (Boutell 2018). Our results also show that the applied approach provided a well-matched synthetic control group with respect to the pre-intervention outcome trends. In addition, our various sensitivity analyses considering a later intervention start and comparing results in various exposed and non-exposed age groups show that our methods and results are robust towards changes in key assumptions.

Our approach was not without limitations, and it is important to consider these in interpreting the results. As described above, only 5–6 million out of the 25 million Germans over the age of 55 years are enrolled in the DMPs for type 2 diabetes or CHD, which is around 20% of this demographic strata. Our analysis, however, does not distinguish between enrollees and non-enrollees. This means that if, for example, the DMPs reduced mortality relatively by 20% without the existence of any spillover effects the maximal effect we could detect with our QE approach is a reduction of 4% in mortality. The large confidence intervals around our ATT estimate suggest that even mortality reductions of 20–50% as reported by previous studies could not be rejected with this approach [10, 11]. The relatively low power is related to the limited number of comparison countries and years of data. Furthermore, in our analysis where we took 2003 as intervention start, we included only four pre-intervention data points; conducting a synthetic control analysis based on this limited number of data points may lead a synthetic control group that is not well-comparable to the intervention group. However, our results suggest that the pre-intervention outcome trend in Germany was quite stable, and the synthetic control group matched this well. We have included a selection of covariates that are critical population-level determinants of circulatory and all-cause mortality; however, it is possible that this is not a comprehensive list, and that other variables should have been included in the analysis. Finally, mortality outcomes lie at the distal position of the causal pathway of the DMPs; shorter-term outcomes such has improved management, quality-of-life and patient satisfaction are also important for individuals and in aggregate from a public health perspective, yet are not captured in this study.

In this quasi-experimental study applying a synthetic control design and analysis, we observed no effect of the German DMPs for type 2 diabetes and CHD on all-cause or circulatory mortality at the population-level. This pattern was consistent across all main, subgroup and sensitivity analyses. Confidence intervals were wide, and we could not reject the possibility of a positive effect. However, according to our analysis, the large mortality reductions as reported in previous studies are unlikely. Given the substantial costs for administration and operation of the programs, further comparative effectiveness research is needed to clarify the role of German DMPs for type 2 diabetes and CHD.