Cost-Effectiveness Analysis of Different Screening Procedures for Type 2 Diabetes: The KORA Survey 2000

We created a cost-effectiveness model for screening a population-based sample of subjects, aged 55–74 years and who had not been previously diagnosed with diabetes, for type 2 diabetes within 1 year. We compared four strategies. 1) In fasting glucose testing only, diabetes was assumed when fasting glucose was ≥7.0 mmol/l (8). 2) In fasting glucose + OGTT, when fasting glucose was ≥6.1 and <7.0 mmol/l (impaired fasting glucose [IFG]), an OGTT was performed during a second visit. Diabetes was considered when fasting glucose was ≥7.0 mmol/l or 2-h postglucose load was ≥11.1 mmol/l (9). 3) In OGTT only, diabetes was considered when fasting glucose was ≥7.0 mmol/l or 2-h postglucose load was ≥11.1 mmol/l (9). 4) In HbA_1c + OGTT, if HbA_1c was >5.6% (10), then an OGTT was performed during a second visit.

For strategies 1–3, we assumed the necessity of separate visits, since fasting glucose tests and OGTTs require a fasting state, whereas HbA_1c measurements can be done during a regular visit. Nearly all subjects aged ≥55 years in Germany have at least one instance of contact with the health care system during 1 year.

We considered two different models (yielding a total of eight screening procedures for evaluation). In model A, all subjects of the screening population were included for the above-mentioned screening strategies. In model B, a first-step preselection (p) was performed (pfasting glucose, pfasting glucose + OGTT, pOGTT, and pHbA_1c + OGTT). Further screening was carried out only among subjects who fulfilled at least one of the following criteria: family history of type 2 diabetes, obesity (BMI 30 kg/m²), hypertension (blood pressure >140/90 mmHg), and fasting triglycerides 2 mmol/l.

Since the actual data of the selection criteria are assumed not to be available for the majority of patients, patient assessment was considered the first step of the screening program and was associated with screening costs.

The main outcomes were screening costs, the number of newly detected true-positive cases of type 2 diabetes related to the whole screening population, and incremental cost-effectiveness ratios (ICERs). Furthermore, we analyzed the percentage of identified diabetic cases in relation to all subjects with previously undiagnosed diabetes in the study population for each screening strategy.

The economic evaluations were conducted from the perspective of the statutory health insurers who cover the direct costs of the screening program (a third-party payer system) as well as from a societal viewpoint, which is the most comprehensive viewpoint, since it covers both direct and indirect costs. Economic evaluations were performed as cost-effectiveness analyses. Because of the short-term perspective of this study, discounting was not required.

We compared screening strategies using ICERs (additional costs were divided by the additional effect when a screening strategy was compared with the next less expensive or less effective one). We ruled out strategies that were less effective and more expensive than others (dominated) and those with lower effectiveness and a higher ICER (extended dominance) (17).

The clinical and epidemiological parameters are presented in Table 1. With the exception of the proportions of participation, which were derived from a population-based practice study in the U.K. (18), all data stem from the population-based KORA (Co-operative Health Research in the Region of Augsburg) Survey 2000 (3). The KORA Survey population was selected as a stratified sample from the city of Augsburg, Germany, and the surrounding districts (southern Germany, population of ∼600,000 in 1999). A total of 1,653 of 2,656 subjects aged 55–74 years (62%) could be included. After 131 subjects with known diabetes were excluded, 1,522 remained eligible, 1,353 of whom completed an OGTT. All data from the KORA Survey were calculated, accounting for the sample design.

Cost data are provided in Table 2. Direct medical costs included practitioner and laboratory testing fees. Medical costs were calculated using a price scale set by the German health care system (Einheitlicher Bewertungsmaßstab, average point value in 2002 of €0.04). When considering the societal perspective, costs were calculated as an approximation of opportunity costs. The calculation also took the productivity losses of patients (time away from work to visit the doctor) into account. We calculated a total of 1 h for each separate visit. For the OGTT, however, we considered an additional 2 h required for the test. We assessed productivity losses using the human capital approach, where the average labor cost of an employee was our approximate measure (17). We estimated the time cost of nonworking and retired subjects by applying the replacement approach (19). The proportions of retired subjects as well as fees for the general working population were derived from German statistics available through a personal communication (North Rhine-Westphalia Statistics Bureau and 20,21). All cost data were calculated for 2002 prices and given in Euros (31 December 2002: $1U.S. = €1.12347).

We varied the following input parameters (Tables 1 and 2): 1) the prevalences of disturbed glucose metabolism (type 2 diabetes, IFG, and/or elevated HbA_1c), 2) the proportions of participation for the separate visits, and 3) the labor costs included in the analyses from the societal viewpoint.

In the univariate sensitivity analyses, we decreased and increased baseline values of the input variable by 20% each.

We conducted a multivariate sensitivity analysis with simultaneous random variation of the parameters using a Monte Carlo simulation with 1,000 iterations. We entered ranges for the input data according to the KORA Survey or the U.K. practice study or estimates (for social costs) (Table 2). We assumed that distributions were either binomial or log normal (for social costs). We fitted multiplicative regression models with and without interaction to estimate an approximate multiplicative equation among costs, effects, cost-effectiveness ratios (CERs) compared with “no intervention” (average CERs) and the variation parameters. The relative relation of the four strategies to one another (in a position of dominance or extended dominance) were analyzed systematically on the simulated datasets to evaluate the proportion of the simulated populations in which relations would change. Further details of the sensitivity analyses are included in the online appendix (available at http://care.diabetesjournals.org).

We estimated the total screening costs for inhabitants aged 55–74 years in Augsburg, Germany, and surrounding districts based on the costs listed in the model (adjusted for sex and age; sample design based).

We performed all analyses using SAS for UNIX (version 8.2; SAS Institute, Cary, NC) and Stata Statistical Software (version 7.0; Stata, College Station, TX).

The proportions of subjects with undetected type 2 diabetes per screening strategy among the study population are shown in Table 3. The pfasting glucose strategy (fasting glucose testing after preselection) delivered the lowest percentage of detected cases. Using HbA_1c + OGTT (HbA_1c testing combined with an OGTT in the whole screening population without first-step preselection) as a screening strategy was the most effective in detecting cases. The age and cardiovascular risk profile (BMI, blood pressure, triglycerides, and HDL cholesterol) of the subjects identified by the selected screening strategies were very similar (data not shown).

The costs of the various screening strategies are presented in Table 3 and Fig. 1. There was a large variation in costs. The highest costs per study subject were incurred by the HbA_1c + OGTT because of the large number of subjects using this screening strategy (100% participation in the HbA_1c testing).

The strategies with a first-step preselection were all dominated and could be ruled out, both from the perspective of the statutory health insurance and from the societal viewpoint (Table 3 and Fig. 1). Fasting glucose testing and fasting glucose + OGTT were dominated by the OGTT strategy (which was more effective and incurred lower costs) and could be excluded when considered from the perspective of statutory health insurance. HbA_1c + OGTT was more effective than OGTT but incurred higher costs. ICERs considered from the societal viewpoint show that fasting glucose testing was dominated by fasting glucose + OGTT as a screening strategy and OGTT by HbA_1c + OGTT (both extended dominance). Among the remaining strategies, fasting glucose + OGTT incurred lower costs but was also less effective than using HbA_1c + OGTT (Fig. 1). ICERs (additional costs per additional detected case) from both perspectives, statutory health insurance and society, are presented in Table 3.

HbA_1c + OGTT remained the most effective and expensive strategy in all 1,000 populations generated by the Monte Carlo simulation (Table 4). The relations among the strategies and decisions about ruling out or selecting strategies remained, as they were in all of the simulated datasets from the perspective of statutory health insurance after a systematic analysis of dominances. A total of 68.3% remained from the societal perspective. The results of the regression analysis show that the variation of the parameters that were included in the sensitivity analyses had only moderate effects on the costs, effects, and CERs of the different strategies. The variation in prevalence of disturbed glucose metabolism had the largest effect on CERs. This was particularly true for the HbA_1c + OGTT strategy (each 20% decrease and increase of prevalence resulted in a 1.40-fold increase and 0.76-fold decrease of CERs from societal perspective). The results remained unchanged when we included interaction terms (data not shown). Figure 1 demonstrates the minimum and maximum costs and effects resulting from the univariate specific sensitivity analyses of the prevalence of disturbed glucose metabolism (dotted lines).

The population of 55- to 74-year-olds in Augsburg and its surrounding regions was 123,226 in the year 2000. From the KORA Survey (3), we calculated that 10,351 subjects had known diabetes and that 10,105 would have previously undiagnosed diabetes in this age-group. From the perspective of the statutory health insurance, the costs of screening (having ruled out the already diagnosed subjects) for previously undiagnosed diabetes were €503,779 using the OGTT as a screening strategy and €2,203,502 using HbA_1c + OGTT. From the societal perspective, these costs were €1,115,142 for the fasting glucose + OGTT strategy and €3,264,646 for HbA_1c + OGTT. The fasting glucose + OGTT strategy detected a total of 2,351 true cases, OGTTs 2,736 cases, and HbA_1c + OGTT 4,939 cases.

We present a decision analytic model for the evaluation of the cost-effectiveness of a number of screening procedures for type 2 diabetes, using population-based data from southern Germany. The OGTT incurred the lowest costs from the perspective of the statutory health insurance system, and a combination of the OGTT and fasting glucose testing incurred the lowest cost from the perspective of society (after dominated strategies were ruled out). However, both of these screening strategies detected only about one-third (OGTT alone) and one-fourth (fasting glucose + OGTT) of subjects with previously undiagnosed diabetes. HbA_1c testing followed by OGTT in those subjects who proved to have elevated HbA_1c yielded the highest rate of detected type 2 diabetes (more than half of all cases detected). This strategy, however, incurred the highest costs. Those strategies that were performed after a first-step preselection of subjects (considered high risk for type 2 diabetes) were all ruled out because of the additional screening costs involved in preselection.

The high effectiveness of HbA_1c testing combined with the OGTT to detect previously undiagnosed diabetes can be explained by the complete participation of all subjects in HbA_1c testing (which requires no special visit to the doctor but can be performed as an extension of another scheduled visit).

The study results remained rather stable in the sensitivity analyses. However, when a participation level of >59.5% for fasting glucose testing and >54.5% for the OGTT was achieved, an OGTT alone would be the most effective strategy and would dominate HbA_1c testing combined with an OGTT (data not shown). However, these very high participations for fasting glucose testing and the OGTT can probably only be achieved in a study setting. The participation levels in the present study, which were taken from the results of a practice study in the U.K. (30–35% participation), seem to be reasonable estimations of participation and are transferable across regional settings (18). In Germany, participation in a health check that included fasting glucose testing offered by the statutory health insurance was ∼20–25% (22).

It is difficult to compare our results with data from other studies. A U.S. study showed $758U.S. in screening costs per truly detected cases of diabetes from the societal perspective when using the fasting glucose test rather than no screening strategy (16). This is a higher cost expenditure than ours, which is €499. However, county-specific CERs largly depend on unit costs. At the same time, purchasing power parities are insufficient to adjust for differences in unit costs. Therefore, results from one country cannot be directly translated to other countries. Thus, the CERs in the present study may vary when our model is applied to other countries. OGTT and HbA_1c testing may be more or less expensive in other health care systems. Our most important result, however, is HbA_1c combined with the OGTT is the most effective, as well as most expensive, screening strategy and should not vary across health care systems. Thus, our results should also be valid in other countries.

We assumed that actual information concerning BMI, blood lipids, blood pressure, and family history of diabetes would not be available for the majority of patients. This conclusion is based on health care research data in Germany (22). We therefore considered it necessary to include costs for preselection procedures. If actual data were available, no preselection would be necessary, but the practitioner would have to select patients at high risk for diabetes screening, resulting in added cost. Generally, however, preselection reduces the effectiveness of all screening strategies.

As in the population practice study of Lawrence et al. (18), we defined successful detection of previously undiagnosed diabetes using only one fasting glucose test or OGTT. However, the American Diabetes Association recommends that a diabetes diagnosis be followed by a confirmation test. We chose a single fasting glucose and OGTT as the screening strategy, however, because we could use data on participation from a carefully designed general practice study (18). If we included confirmation testing, the costs per case detected might be higher than in the present study. Twofold testing, however, would probably reduce participation.

Several limitations of the present study must be considered. Like many other cost-effectiveness analyses of screening, we used an intermediate outcome: the number of truly positive cases detected. Including information on potential costs following the screening procedure and benefits of treatment would provide a more complete picture of the cost-effectiveness of screening for diabetes. However, no population-based data regarding the natural disease process of early detected diabetes or results describing the effectiveness of early intervention after diabetes screening are available so far (4,5).

The results of the present study are for a one-time screening situation, and they may not be applicable for ongoing screening. In the case of ongoing screening, the prevalence of undiagnosed diabetes would become lower. As can be concluded from the sensitivity analysis, a lower prevalence of undiagnosed diabetes would raise the costs per case identified and thus affect the CERs. However, the relation among strategies would probably remain the same.

The greatest strength of the present study is that it uses highly valid population-based data, whereas previous analyses used several external data resources. The sensitivity analysis showed reasonable stable results after varying the prevalence of disturbed glucose metabolism (type 2 diabetes, IFG, and elevated HbA_1c), participation, and social costs. In particular, the relation among strategies with respect to their costs and level of effectiveness remained stable in the majority of simulated populations.

In general, a cost analysis cannot determine which strategies should be implemented. The choice depends on the goal of the screening program. It may be to identify the most possible cases of previously undiagnosed diabetes or to pursue lower costs per case identified. Cost-effectiveness analyses can only indicate which strategies are dominated by others and can therefore be ruled out and show ICERs to determine which program is more effective although it incurs higher costs. A decision maker can use these information to choose the most suitable screening procedure for a program by taking into account the maximum limit to be spent per additional case detected.

Further studies are warranted in order to answer the question of which screening procedure is most appropriate. To achieve better and less costly screening, participation in screening tests needs to become more accepted by the target population. The most important issue is to evaluate the effectiveness of early intervention in diabetic subjects.

The study was supported by institutional funding (German Diabetes Research Institute) from the German Ministery of Health and by the Ministery of Science of North-Rhine Westfalia. The study was supported in part by a research grant from the German Diabetes Foundation.