Introduction

Results from clinical trials and translation studies in real-world settings show that type 2 diabetes can be prevented or delayed by lifestyle interventions in individuals with increased risk of type 2 diabetes.1 Those lifestyle interventions were also found to be cost-effective or even cost saving2 from a healthcare system perspective in many countries. A recent systematic review of trial-based studies demonstrated that various interventions to prevent type 2 diabetes in high-risk populations were both cost-effective and feasible in diverse settings.3 The gap between gold-standard results shown in trials and effectiveness in everyday practice is usually addressed in pragmatic studies designed to be practical and applicable in real-world clinical settings. Such pragmatic lifestyle programs have been successfully established in the routine practice in some Western countries4–7 and there is robust real-world evidence that these programs are effective in lowering weight and hemoglobin A1c (HbA1c) if implemented in practice.8 The systematic review on the effectiveness of pragmatic lifestyle interventions for preventing type 2 diabetes found that these interventions led to weight loss, and adherence to guidelines was significantly associated with greater weight loss.9 A quasi-experimental evaluation of a nationwide diabetes prevention program, utilizing electronic health records from over 2 million patients, provided causal evidence of improved glycemic control and reduced risk factors in the behavior change program for pre-diabetes.8 A systematic review found that lifestyle interventions targeting high-risk individuals were generally cost-effective for preventing type 2 diabetes from a healthcare system or societal perspective.2

In the USA, the National Diabetes Prevention Program (US-NDPP) was authorized by the US Congress in 2010 and since then has been delivered by over 1500 organizations, reaching almost 300 000 people.4 The British National Institute for Health and Care Excellence (NICE) recommends a lifestyle intervention for individuals with intermediate categories of hyperglycemia for preventing type 2 diabetes,5 6 and the National Health Service (NHS) England launched the NDPP (NHS-NDPP) in 20166 10 into significant health benefits for participants.8 In Germany, several programs have been recommended.11 However, to date, no such program was put into action.

Whether a national program like those in the USA and the UK would be a good use of healthcare resources in Germany is unknown. Evidence from the USA or the UK may not be directly applicable to Germany due to two main reasons. First, the population characteristics of people at high risk of developing type 2 diabetes might differ. Second, the costs for treating diabetes and its complications are highly context-specific and country-specific. In German setting, the lifestyle programs of diabetes prevention were evaluated in a handful of studies.12–15 These studies either overlooked diabetes complications12 13 or primarily focused on weight management, without directly estimating the impact of lifestyle interventions on diabetes incidence.14 15 Furthermore, previous studies often ignored productivity losses and evaluated those programs from a narrow healthcare system perspective and did not consider potential heterogeneity in the cost-effectiveness across the age range.2 Without that, both the societal value of these programs as well as the target populations with the most favorable cost-effectiveness ratio remain unknown.

In this study, we aim to examine the long-term health and economic impact of a pragmatic lifestyle program, as successfully implemented in large scale in other countries, for Germany from a healthcare payer and societal perspective. We also examine how the effectiveness and cost-effectiveness of the lifestyle program to prevent type 2 diabetes vary by age group at both individual and population levels.

Methods and materialsEvaluation approach

We used an established and validated diabetes simulation-model,16 adapted it to the German context, and simulated the lifetime cost-effectiveness of a lifestyle diabetes prevention program from a healthcare system and societal perspective. The healthcare system perspective included only direct medical costs. The societal perspective also included patient-time costs associated with diabetes self-management (direct non-medical costs) and productivity losses (indirect costs). We also projected the population-level impact of a nationwide implementation of the program, under different theoretical levels of participation in the eligible population. The workflow of the analysis is shown in online supplemental file 1, figure 1 .

Intervention

The lifestyle intervention programs evaluated were based on the recommendation of NICE public health guidelines (PH38) and programs being implemented in the NHS-NDPP program.6 The program targeted individuals with a high risk of developing type 2 diabetes (HbA1c at range 6.0%–6.4% or 42–46 mmol/mol) who were identified during a routine check-up. The program includes a core component of 13 group education sessions in the first year followed by seven maintenance sessions over the following 2 years. The program covers topics like diabetes risk assessment and risk identification, physical activity, weight management, dietary advice, and medications, with an annual review by a general practitioner and blood tests.6 7

Simulation model

We used the Centers for Disease Control and Prevention (CDC) RTI Diabetes Cost-Effectiveness Model (CDC-RTI DM), an established and validated Markov cohort model designed to simulate the development and progression of type 2 diabetes and to assess the cost-effectiveness of various prevention and treatment interventions including diabetes and intermediate hyperglycemia screening. The model is described in detail elsewhere.16 In brief, the model simulates disease progression based on annual transitions of multiple disease states. It follows cohorts of patients to either death or 95 years of age, simulating the development of diabetes-related complications on three microvascular disease paths (nephropathy, neuropathy, and retinopathy) and two macrovascular disease paths (coronary heart disease and stroke). This model is closed, namely starts with a fixed population that is followed over time without any new subjects that can enter or leave the study at different times. The model outcomes include lifetime development of diabetic complications (nephropathy, neuropathy, retinopathy, coronary heart disease and stroke), diabetes-related healthcare costs, life years, and quality-adjusted life years (QALYs) gained. The model was internally and externally validated17 and has been used extensively in different cost-effectiveness analyses in the USA context.18–20 Details of the simulation model are described in a technical report in online supplemental file 2.

For this analysis, the CDC-RTI DM was modified to fit the German setting. We parametrized the model using the best available German-specific data, or international data if no German source was found. We relied on prospective German studies to assess the epidemiologic parameters as well as statutory health insurance (SHI) claims data to assess the cost parameters. After adaptation, we validated the model in parts (all-cause cumulative mortality, cumulative mortality of coronary heart disease, and stroke) as recommended in the principles of good practice for modeling in healthcare evaluation.21 Extensive information on parameters, data sources, assumptions, and limitations is provided in online supplemental file 1 and in online supplemental file 4.

Key parameters of the simulation model

The key model parameters required in the analysis are presented in online supplemental file 1, table 1.

Effectiveness of the lifestyle intervention program: the effectiveness of the program on reducing type 2 diabetes incidence (the relative risk of diabetes in participants undertaking a pragmatic lifestyle program) was based on data from a recent meta-analysis.22 We conservatively assumed that the program causes an effect only on type 2 diabetes incidence and do not affect a disease progression, the effect of the program is the same for all subgroups and that the effect held only during the program and would be marginalized after the program stopped. Furthermore, we assumed that adherence of participants was equivalent to that seen in the pragmatic real-world intervention studies from which relative risk reduction was derived.22

Costs of the lifestyle intervention program: the cost of the program was taken from Roberts et al7 and converted to Euros by applying a purchasing power parity (PPP) conversion based on 2015 values. This covered staff costs, including a social security contribution, and blood test costs. For validation, we also calculated the program costs bottom-up using German prices per service. As both approaches returned similar results (not shown), we utilized only the PPP-converted costs in this analysis.

Health utilities

Health utility decrements associated with diabetes and diabetes complications were taken from a large population-based German study (see online supplemental file 1, table 2).23

Cost parametersDirect medical cost of diabetes and complications

The yearly baseline medical cost for a patient with and without diabetes as well as excess costs related to risk factors or diabetes complications were taken from Kähm et al who used a panel data approach based on German claims data of an SHI to estimate the excess costs of diabetes complications.24 Annual costs of medical care for a patient with intermediate hyperglycemia were calculated using the age-adjusted and sex-adjusted cost ratio of ‘intermediate hyperglycemia’ versus ‘no diabetes’ taken from a population-based cohort study in southern Germany.25 The multiplicative cost approach was used in the simulation.26

Direct non-medical costs for diabetes care

The patient time costs due to diabetes self-management were taken from Icks et al27 where time spent on self-management activities was assessed using a validated questionnaire, categorized as lost leisure time, and valued with a mean adjusted net wage per hour in Germany. The intervention-related direct non-medical costs (cost for exercise classes and equipment, food and food preparation items, and transportation) were not included in the analysis because no reliable information on such costs is available.

Indirect societal costs (productivity losses)

The costs of productivity loss according to the human capital approach were taken from Ulrich et al.28 The costs were due to losses in productivity such as sick days and long-term incapacity to work. These indirect societal costs were restricted to the age group <65 years old. The costs due to unpaid work, presenteeism or premature death were not considered.

In the analysis, the cost-effectiveness was calculated from two perspectives: (1) a healthcare system perspective that only included direct medical costs, (2) a societal perspective that included direct medical, non-medical costs (the patient time costs), and indirect societal costs.

Simulation population

The baseline population consisted of two cohorts of persons aged 35–54 and 55–74 with newly diagnosed intermediate hyperglycemia (HbA1c at range 6.0%–6.4% or 42–46 mmol/mol). The characteristics and the distribution of the initial cohort by age, sex, smoking habits, hypertension, and high cholesterol status were estimated from the Cooperative Health Research in the Region of Augsburg (KORA) S4 study, a population-based study from southern Germany covering individuals 25–74 years of age.25

Cost-effectiveness of the program and thresholds for adoption

We estimated the incremental cost-effectiveness ratios (ICER) in €/QALY by dividing the discounted incremental cumulative costs by the discounted incremental QALYs. All costs were adjusted to 2020 values and future costs and QALYs were discounted by an annual discount rate of 3.5%, which is recommended by NICE.29

In Germany, no explicit willingness-to-pay (WTP) is defined30; we, therefore, assessed the probability that the intervention is cost-effectiveness at the following WTP thresholds: of 20 000 €/QALY corresponding to the NICE recommendations,31 50 000 €/QALY corresponding to Neumann et al,12 and 100 000 €/QALY as a maximum value. We also estimated the WTP needed to achieve a probability of cost-effectiveness more than 50%, 75%, and 95% based on cost-effectiveness acceptability curves (CEACs), which show the probability that an intervention is cost-effective at different WTP values.

We also compared the cumulative incidence of type 2 diabetes, the average number of years lived with type 2 diabetes in both intervention and routine care groups, and the number of avoided type 2 diabetes cases.

Sensitivity analysis

In the univariate sensitivity analysis, we checked the robustness of the results by broadly considering a range of percentages for the costs of the intervention (50% to 150%), the baseline medical costs (80% to 120%), and the program effect on type 2 diabetes incidence (50% to 150%). In the probabilistic sensitivity analysis (PSA), we simultaneously varied intervention costs and effects on type 2 diabetes incidence of the intervention. This approach accounts for decision uncertainty by considering a multiplier for intervention costs (uniformly distributed between 0.5 and 1.5) and assuming Gamma and log-normal distributions for baseline medical costs and effect relative ratios, respectively. We performed 1000 iterations to compute cost-effectiveness and analyzed the distribution of cost-effectiveness ratios including the mean, median, credible intervals, and probabilities to achieve different WTP thresholds.

The PSA was performed only from the healthcare system perspective because of the software limitations.

The statistical analysis and graphs were performed in R V.4.4.1.

To report the results, we followed the recommendations of the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) (see online supplemental file 3).

Population-level analysis

In view of the high public health significance of diabetes prevention and the relevance of informing decision-making, we analyzed the effect of a nationwide diabetes prevention program also at population level in Germany. We focused on the affected German population, namely persons with intermediate hyperglycemia, as a target group of the program. We assumed that the program would be offered to a patient at risk by a general practitioner. In Germany, 45% of patients participated in a disease-related training after a clinician’s recommendation.32 In this analysis, we assumed five scenarios where population proportions of participation in the nationwide program were 5%, 10%, 25%, 50% or 100% of all individuals with intermediate hyperglycemia to reflect a full range of the theoretical and expected participation. We estimated a reduction in type 2 diabetes incidence, average number of years lived with diabetes, and total costs of the program over lifetime. The simulation was based on two closed cohorts (35–54 and 55–74 years old) followed from starting age to the death.

ResultsCost-effectiveness of the program at individual level

The lifetime costs and QALYs for both intervention and non-intervention groups as well as resulting ICERs are shown in table 1. The cost-effectiveness planes are shown in figure 1. In all analyses, the lifestyle program led to a gain in QALYs and an increase in costs. The ICERs for the lifestyle intervention program were 14,690€ per QALY gained for the 35–54 years olds and 14,372€ per QALY gained for the 55–74 years olds from a payer’s perspective over lifetime. From a societal perspective (including indirect societal costs and direct non-medical cost) the intervention costed −3,805€ per QALY gained for 35–54 year olds and 4,579€ for 55–74 year olds, and dominated the comparator in the younger cohort (was more effective and cost saving). The main driver for cost savings from a societal perspective was the patient time costs (online supplemental file 1, table 7).

Table 1

Lifetime incremental effectiveness, incremental cost and cost-effectiveness of implementing a lifestyle type 2 diabetes prevention program, by age group and study perspective

Figure 1

Cost-effectiveness plane: incremental cost and QALYs relative to no intervention in the probabilistic sensitivity analysis. QALYs, quality-adjusted life years.

Diabetes-related outcomes

The intervention led to a lower cumulative incidence of type 2 diabetes (20.61% vs 21.05%) at the end of the simulation for the 35–54 years old cohort and a lower number of years lived with type 2 diabetes (0.03 years reduction per person in the same scenario) (online supplemental file 1, table 3). The effect was more pronounced after 10 years than at the end of the simulation. The relative risk reduction (the proportion of prevented type 2 diabetes cases) was higher at 10 years then at the end of the simulation. For the older participants, the health benefit in reducing diabetes incidence is greater. We found no significant changes in the incidence of type 2 diabetes complications or any direct effect on all-cause mortality. However, the onset of complications and mortality was slightly postponed to older ages (not shown).

Sensitivity analysis

The distribution of incremental costs, incremental effects, and cost-effectiveness ratios from PSA is shown in online supplemental file 1, table 4. The CEACs are shown in figure 2, and the probabilities of the intervention to be cost-effective at different WTP thresholds are shown in online supplemental file 1, table 5. At a WTP threshold of 20 000 €/QALY, 50 000€/QALY and 100 000€/QALY, the probability of the intervention to be cost-effective is 65.4%, 98.0%, and 99.8% in 35–54 year olds and 74.0%, 99.2% and 99.9% in 55–74 year olds from a healthcare system perspective over lifetime.

Figure 2

Lifetime cost-effectiveness acceptability curves from a healthcare system perspective. QALYs, quality-adjusted life years.

The WTP thresholds needed to achieve a probability ≥50% of cost-effectiveness based on CEACs ranged from 14 730 to 15 603 €/QALY. To achieve a probability of cost-effectiveness ≥95%, a WTP of 40 390 to 31 745 €/QALY would be necessary.

Key factors impacting cost-effectiveness calculations in the sensitivity analysis were relative risk reduction of type 2 diabetes, costs of the interventions, and baseline medical costs (online supplemental file 1, table 6 and figure 2).

Population-level analysis

The absolute number of prevented or delayed cases, accumulated QALYs, and costs would depend on the fraction of the population participating in the program (table 2). The absolute numbers of diabetes cases are shown in online supplemental file 1, table 8.

Table 2

Estimated aggregated health outcomes and costs of implementing a nation-wide diabetes prevention program in Germany

If the program was offered to everyone with diagnosed intermediate hyperglycemia in both age groups in Germany (~8.5 million people) and if 25% of those participated in the program, around 10 500 incident cases of type 2 diabetes (2.5% in relative terms) would be avoided and around 35 200 QALYs would be gained over a lifetime perspective. The discounted lifetime cost of the program would be around 1 08 517 million € from a healthcare system perspective and 1 13 324 million € from a societal perspective. The upfront budget impact of delivering the intervention in the first 2 years would be 678 million €.

Discussion

This study is the first to comprehensively examine the long-term cost-effectiveness of introducing a program to prevent type 2 diabetes targeting individuals with intermediate hyperglycemia in Germany. Furthermore, it adds to previous economic evaluations by quantifying the potential impact in two age groups and by two different perspectives (healthcare system and societal) as well as by estimating the potential costs and consequences at the level of a Germany-wide program.

Neumann et al estimated the long-term cost-effectiveness of a hypothetical lifestyle intervention program for the prevention of type 2 diabetes using a four-state Markov model in Germany.12 However, that model did not account for diabetes complications and their costs and did not allow extrapolating the results on a population level.

Only a few studies compared younger and older cohorts in cost-effectiveness analysis, mostly focusing on diabetes screening.33 Furthermore, many other cost-effectiveness analyses ignored indirect societal costs and direct non-medical costs, although it is known from cost of illness studies that those are substantial.27 28

Our main results are comparable with previous international studies, for example, in the UK and the USA.4–6 A recent systematic review evaluating programs to prevent type 2 diabetes reported a median ICER of 13 761 $/QALY (12 653 €/QALY), which is in the same range as our results. While ICERs in each varied between studies, all but one showed ICERs far below the commonly used WTP threshold (US$ 50,000).34

In our study, the intervention in the older group had a higher probability of being cost-effective from the healthcare system perspective compared with the younger group, which agrees with Neumann et al.35 However, from a societal perspective, the intervention in the younger cohort led to lower ICERs and even became cost saving, as the extra productivity losses were only attributed to the under-65 group in our method, so the costs of the diabetes case were higher for them. In the younger group, the effect is more pronounced since this cohort lives longer with diabetes. We showed ICERs for different cost structures, highlighting that it is essential to include direct non-medical and indirect societal costs in the evaluation of population-wide programs to understand their full impact and to avoid underestimation of benefits. In our simulation study, we adapted the CDC-RTI DM to the German setting and relied on the best available German-specific and international data. We also showed the potential cost-effectiveness of a pragmatic lifestyle intervention scaled up to the total population at risk of developing diabetes, which had been done for the USA36 and the UK,7 but not for Germany. This population-level analysis shows that if 25% of people with a HbA1c of 6.0%–6.5% (~2 million people) would participate in the program, one could prevent 10 500 incident cases and gain 35 200 QALYs from a societal perspective over their lifetime. However, such a scenario would come with substantial upfront investments of around 678 million €, which is almost 1.5% of the annual healthcare expenditure for diabetes in Germany.37 In light of around 0.5 million incident cases of diabetes that occur in Germany annually, this highlights that high-risk individual-level prevention strategies as analyzed in this study need to be complemented with population-level policies and interventions that address the upstream determinates of obesity and type 2 diabetes.38–41 Moreover, patient participation in the lifestyle interventions for individuals at risk of diabetes varied from 15% to 80%,42 so our assumption might be considered conservative.

There are strengths and weaknesses of our study. Our strengths included that we used a well-established simulation model for modeling prevention of type 2 diabetes. We parameterized it with epidemiological and economic inputs from German studies that were based on large population-based cohorts or population-representative health insurance claim data. Also, we performed PSA. This analysis, though limited to the healthcare perspective due to software constraints, provides a more comprehensive picture of cost-effectiveness uncertainty compared with univariate analysis.

Our first limitation was that assumptions were required for our simulation analysis. However, we purposely made our assumptions conservative. For example, it is unknown if and how lifestyle programs affect the probability of developing type 2 diabetes beyond the active program phase. We assumed no preventative effect beyond the active program phase. Other assumptions include no reversion from intermediate hyperglycemia to normoglycemia, the same utilities in both intervention and routine care scenarios, adherence equivalent to those in real-world setting studies, etc. A detailed discussion of these assumptions is given in online supplemental file 1.

Our study is also subjected to typical limitations of all simulation studies: dependence on a model structure and parameters. Moreover, not all primary clinical data used in the model are available for Germany; some had to be taken from other countries. The baseline population in the model was parameterized based on the KORA S4 study, a large population-based cohort. KORA was conducted specifically in the Augsburg region, which may not fully represent the entire German population and may not capture all demographic variations across Germany. Regional differences in health outcomes have been observed between northeast and south Germany. Also, the study covers a wide adult age range but does not include very young or very old populations.

We did not include individuals aged 75 and older in our analysis, even though this demographic is growing and increasingly plays a crucial role in national health systems. Assessing the potential effect of the pragmatic lifestyle intervention on this population would indeed be a valuable contribution. However, we chose not to include individuals over 75 years at baseline for two reasons. First, most trial evidence on the effectiveness of lifestyle interventions focuses on individuals younger than 75. Second, our simulation was based on the baseline cohort characteristics from the KORA S4 study, which only includes individuals up to 75 years old. We lacked population-based data on the distribution of risk factors (such as smoking habits, hypertension, and high cholesterol) for those older than 75, which is why they were excluded from our analysis.

Another limitation is that our study did not include program implementation expenses. Moreover, we adopted the costs of the diabetes prevention program directly from Roberts et al,7 which might pose some questions on the transferability as medical care systems and budgeting differ between the UK and Germany. The costs of pragmatic lifestyle interventions varied widely depending on what was included in the program; however, we appointed the program recommended by NICE,6 7 suggesting that the total costs of such a program depend only on the costs of its components in a particular healthcare system. Alternatively, the costs could be taken from another intervention implemented in Germany. Due to the limitations of a Markov model structure, we also could not calculate the loss of productivity due to premature death or identify the largest driver in indirect societal diabetes costs. Unfortunately, no cost structure for any such intervention was found in the literature. However, considering the robustness of our results towards variations in our crucial model assumptions, these limitations are unlikely to change the general conclusions of our study.

In this health-economic analysis, we could not identify which component of the intervention plays the most significant role because this objective was out of scope of this study. As was shown, several key elements contribute to the effectiveness of diabetes prevention programs. The original DPP was a 3-year randomized trial followed by 7 years of modified intervention follow-up, suggesting longer term interventions are beneficial.43 Moreover, flexible interventions delivered by skilled lifestyle coaches that accommodate individual preferences and reflect local community and cultural contexts may achieve better outcomes.44 However, this type of investigation aligns more with the primary objectives of experimental studies.