During the BNDSP, a positive screening test was defined by a fasting capillary glucose ≥100 mg/dL or a casual glucose ≥140 mg/dL, and fasting was defined as absence of food ingestion for at least 4 h prior to capillary glucose test [18]. Sensitivity and specificity of capillary blood glucose screening considered in the model were 68 and 89 %, respectively [10]. These values were estimated considering the BNDSP participants reporting to be in fasting condition (47 %), for which the cut-off for positive test results was 100 mg/dL, and non-fasting condition (53 %), for which the cut-off was 140 mg/dL [19]. An additional fasting plasma glucose test and two extra physician visits were required for diagnostic confirmation of diabetes.

Universal DM2 screening as conducted in the BNDSP was compared to standard practice in Brazil at that time, that is, no organized screening. Progression of individuals with and without diabetes in the screening module is represented in Fig. 1. Individuals diagnosed with diabetes were assumed to enter the diabetes disease progression model (Fig. 2), described below.Fig. 1

Progression of individuals in the screening module, Nationwide Population Screening Program for Diabetes. Brazil, 2001

During the BNDSP, specific recommendations were given regarding treatment. Diagnosed cases were managed aiming at blood pressure target of 135/80 mmHg and fasting serum glucose level of 110 mg/dL. Therefore, we assumed that intensified glycemic and hypertension treatment were promptly initiated at diagnosis. Based on the United Kingdom Prospective Diabetes Study (UKPDS), we assumed the intensified glycemic control consisted of one or more generic drugs (metformin, glibenclamide, and/or insulin) aiming at fasting serum glucose level of 110 mg/dL [1, 20]. The effect of intensified glycemic treatment is modeled as slowing progression of microvascular complications by reducing hemoglobin A_1c, and thus lowering hazard rates for microalbuminuria, nephropathy, peripheral neuropathy, and photocoagulation [1]. No effects on macrovascular complications were assumed in the base case [1].

All persons with hypertension were assumed to receive standard hypertension treatment (targeting diastolic blood pressure of 90 mmHg) until they receive a diagnosis of diabetes, after which they receive intensified hypertension treatment (targeting diastolic blood pressure of 85 mmHg) [3] consisting of angiotensin converting enzyme inhibitor (captopril), β blocker (propranolol) or thiazides as single or combination therapy [2]. Model estimates of this approach assumed a 17–44 % relative risk reduction for stroke and 13 % risk reduction for CHD [2].

We used a modified version of the CDC/RTI type 2 diabetes cost-effectiveness simulation model to simulate the long-term health and economic consequences of the BNDSP. The CDC/RTI model is a Markov-based model developed by the Centers for Disease Control and Prevention and RTI International (CDC/RTI). The model simulates disease progression based on annual transition of multiple disease states. The outcomes include lifetime development of diabetic complications, diabetes-related health care costs, life years, and quality adjusted life years gained (QALYs) [21]. In progressing through the model from the onset of diabetes to death, people with diabetes can develop five types of complications: nephropathy, neuropathy, retinopathy, coronary heart disease (CHD), and stroke (Fig. 2). Each complication/health state has a corresponding health utility value ranging from 0 (death) to 1 (perfect health). The basic model structure and key model parameters have been previously described [21–23]. Annual transition probabilities for the health states considered in the model are presented in Table 1.Table 1

Annual transition probabilities for health states considered in the model

The model includes a screening module, and assumes that in the absence of screening, diagnosis would occur 10 years after its onset while screening would reduce this pre-diagnosis interval by 5 years [24]. Progression of persons through screening and clinical diagnosis is shown in Fig. 2.

The CDC/RTI model has been used to assess the cost-effectiveness of screening for undiagnosed type 2 diabetes and pre-diabetes in the United States [22, 23]. For this study, we adjusted epidemiological and cost parameters using Brazilian data.

Demographic and mortality data for the general population were obtained from the Brazilian National Institute for Geography and Statistics (IBGE) for 2002. Prevalence of risk factors for cardiovascular complications in the Brazilian population was based on surveys [25–27] (Table 2). Estimates of undiagnosed diabetes prevalence by age-groups and hypertension level were obtained from the BNDSP [28] (Table 2). Age and gender-specific estimates of the incidence of true diabetes in the population were obtained applying DISMOD II software [29] to data from a prevalence survey [30].Table 2

Estimated prevalence of undiagnosed diabetes, smoking, hypertension and hypercholesterolemia in the Brazilian population

Four types of costs were considered: associated with diabetes screening, diabetes treatment, diabetes-related complications, and other medical care.

Costs associated with screening and confirmatory diagnosis have been previously estimated based on actual expenses of the BNDSP, which has been previously described [19]. A total of 22.1 million capillary glucose tests were performed, of which 16.4 % were positive. Total screening and diagnostic costs were US$ 26.19 million (Int$ 104.32 million), including national level costs of diagnostic material, social mobilization and media campaign, training of healthcare workers, and management costs. The screening cost per screened individual was estimated at US$ 1.16. Diagnostic confirmation costs were US$ 2.97 per each individual screened positive during the BNDSP.

The costs of glycemic control included three resource components: drug use, physician visits, and self-testing [31]. Treatment modalities considered were oral hypoglycemic agents only, insulin only, and both oral hypoglycemic agents and insulin. Intensified glycemic control considered initial treatment with metformin. The proportion using each drug regimen varied by duration of diagnosed diabetes and was estimated from the UKPDS [1, 32]. Daily cost of medications were estimated considering average market prices assuming standard doses. Incremental cost of intensive glycemic control (relative to standard control) was estimated per year, depending on the number of years since diagnosis.

The costs of standard and intensified hypertension control were estimated to reflect clinical practice in Brazil, where standardized drug regimens include the use of generic thiazides, propranolol, and captopril. We assumed that the maximum number of drugs that would be taken at any given time was three [33]. The incremental cost of intensive hypertension control, relative to standard control, was estimated per year, depending on the number of years since diabetes diagnosis.

Cost of diabetes-related complications included the cost of nephropathy, neuropathy, retinopathy, CHD, and stroke. One-time and annual costs of complications considered in the model (Table 3) were calculated. Healthcare resource utilization was estimated based on Brazilian guidelines, considering diagnosis and annual follow-up procedures for end stage renal disease, angina and stroke; one time diagnosis of peripheral neuropathy; and one time photocoagulation procedure. Each resource was multiplied by its unit costs, which for medical procedures, exams, hospitalization and medical visits considered SUS reimbursement values [31].Table 3

Direct medical costs of diabetes complications in Brazil, 2001

One time angina and stroke costs were obtained from the National Information System on Hospitalizations in SUS (SIH), considering costs reimbursed by SUS for patients admitted to SUS in 2002 with diagnosis of the above conditions.

Direct cost of clinical nephropathy; lower extremity amputation; and death due to stroke or CHD, were obtained from follow-up data from all patients with each of these conditions admitted at the Hospital de Clínicas de Porto Alegre of the Federal University of Rio Grande do Sul (HCPA) during 2002.

Normal medical care costs that are not specific to diabetes care were estimated considering the Brazilian average government health expenditure/person (i.e., GNP per capita on health) of US$ 94 (Int$ 374) per year in 2002 [34]. Cost of death was estimated as a proportion of hospital expenditures prior to death [35], obtained from the cohort of all patients at the HCPA who progressed to death during 2002.

By estimating lifelong complications and death in a hypothetical population cohort, the model predicts the lifetime incidence of diabetes complications and QALYs for each true case of diabetes, considering utility values of the CDC/RTI model [21–23]. Per person change in life-years and QALY with screening were calculated, and the sum of all estimated costs and expected QALYs for each strategy considered in the analysis were used to calculate the incremental cost-effectiveness ratio of screening relative to no screening. All costs are presented in US dollars, considering the exchange rate in December 2001 (1 US$ = R$ 2.35). To allow for international comparison, we present results in both US$ and international dollars (Int$) considering year 2001/2002 purchasing power parity exchange rates (1 Int$ = R$ 0.59). All costs were converted to 2002 reais using the Consumer Price Index from the Brazilian Central Bank [36].

We took the perspective of the public health care system, as the costs of the screening program were paid by the publicly funded SUS and the population receiving the benefits of the strategies evaluated is covered by the SUS. A baseline 5 % discount rate was applied to future costs and QALYs [37].

We conducted one way sensitivity analyses to investigate the effect of key parameter values and assumptions in cost-effectiveness ratios (Table 4).Table 4

Sensitivity analysis

We varied the costs of screening, intensified glycemic and hypertension therapy, and diabetes complications, as well utility weights associated with diabetes and its complications by ±20 %. The time assumed between diabetes onset and screening (detection benefit from screening) varied from 4 to 6 years, and discount rates varied from 1–10 %. Risk reduction of intensive glycemic control on macrovascular complications were varied to 20 % for both myocardial infarction and stroke, assuming that 50 % of the individuals newly diagnosed with diabetes would receive metformin, and considering risk reduction estimates from UKPDS 34, which showed a relative risk reduction of 41 % for stroke and 39 % for myocardial infarction [32]. We also estimated how the cost-effectiveness of the BNDSP would change by assuming persons with diabetes would not receive intensive glucose and hypertension control under the non-screening scenario, and screening would occur only in hypertensive individuals only during the BNDSP.

This study has been carried out in accordance with the Declaration of Helsinki, and the study project was approved by the Ethics Committee of the Federal University of Rio Grande do Sul.

Screening all adults aged 40 years or older decreased the incidence of all diabetes complications considered in the model and increased survival. Cumulative incidence was reduced from 0.49 % for non-screened population to 0.28 % for the screened for end stage renal disease, from 0.76 to 0.58 % for lower extremity amputation, from 8.4 to 7.4 % for stroke, from 33.8 to 29.6 % for CHD, and from 3.6 to 2.3 % for blindness. Screening leads to a slightly longer life expectancy, adding approximately 13 weeks to the average lifespan of those detected at screening.

Compared with no screening, population screening increases lifetime costs, primarily due to increased costs of introducing treatment 5 years earlier. Modeling shows that this cost of treatment for five additional years for those diagnosed was approximately 20 times the cost of screening.

Screening adults 40 years or older would increase the life-time costs by US$ 489 but results in a gain of 0.035 life-years and 0.0157 QALYs (Table 5). In the base-case analysis, the incremental cost-effectiveness ratio was estimated to be US$ 31,147 (Int$124,060) per QALY.Table 5

Lifetime costs, life-years, QALYs and incremental cost-effectiveness per true case of diabetes diagnosed

In sensitivity analyses, cost-effectiveness ratios were sensitive to several parameters (Table 4), including discount rates, detection benefit from screening, utility rates associated with diabetes, and cost of intensified glycemic control. On the other hand, varying screening cost, complication costs, and intensified hypertension control had minimal effect.

If benefit from early glycemic control on cardiovascular outcomes was assumed, i.e., when risk reduction of intensive glycemic control on macrovascular complications was considered 20 % for both myocardial infarction and stroke, cost-effectiveness ratio fell to US$14,769/QALY (Int$ 58,825/QALY). Similarly, if screening was conducted only among hypertensive individuals, the cost-effectiveness ratio decreased to US$22,695/QALY (Int$ 90,395/QALY). Assuming that non-screened individuals did not receive intensive glucose and hypertension controls after diabetes diagnosis, incremental cost-effectiveness ratios were considerably more favorable, estimated at US$ 7505/QALY gained (Int$ 29,893/QALY).