The Efficacy and Cost of Alternative Strategies for Systematic Screening for Type 2 Diabetes in the U.S. Population 45–74 Years of Age

The study population for our simulation was a closed cohort representing the U.S. population 45–74 years of age without a previous diagnosis of diabetes. According to the 2000 U.S. census, 80.3 million people were 45–74 years of age (7). Based on the Third National Health and Nutrition Examination Survey, 7.7 million people 45–74 years of age were previously diagnosed with diabetes (8). Thus, 72.6 million individuals were eligible for screening.

We examined four screening tests: 1) screening with an RPG cut point of 100 mg/dl, a highly sensitive test; 2) screening with an RPG cut point of 160 mg/dl, a highly specific test; 3) screening with an RPG cut point of 130 mg/dl, a test with intermediate sensitivity and specificity; and 4) screening with a multivariate logistic equation that incorporated RPG, postprandial time (self-reported number of hours since last food or drink other than water), age, sex, and BMI (6). We also assessed three screening intervals over a 15-year period: 1) baseline and every 5 years, 2) baseline and every 3 years, and 3) baseline and every year.

We assumed that individuals with positive screening tests but RPG <200 mg/dl (94% those with positive screening tests) would undergo an OGTT for definitive diagnosis and that individuals with positive screening tests and RPG ≥200 mg/dl (6% of those with positive screening tests) would have an FPG for definitive diagnosis (6).

To estimate the sensitivity and specificity of each screening test, we applied the RPG cut points and the multivariate equation to a large dataset that included diabetes risk factors, RPG levels, and 2-h 75-g OGTTs performed on separate days. For RPG ≥100 mg/dl, sensitivity was 91% and specificity 49%; for RPG ≥130 mg/dl, sensitivity was 63% and specificity 87%; and for RPG ≥160 mg/dl, sensitivity was 44% and specificity 96%. The equation was 63% sensitive and 96% specific (6).

We estimated the prevalence of undiagnosed diabetes to be 10% at baseline and the prevalence of impaired glucose tolerance (IGT) and/or impaired fasting glucose (IFG) to be 22% (7). We assumed that the prevalence of IGT/IFG remained constant and estimated the rate of progression from IGT/IFG to diabetes to be 5.7% per year (9). Although the incidence of type 2 diabetes decreases by ∼50% in persons ≥75 years of age compared with those 64–75 years of age, the incidence increases by 50% in persons 65–74 years of age compared with those 45–64 years of age (10). For this reason, we believe that over the time frame of 15 years, the incidence remains approximately constant.

We measured the efficacy of each screening strategy by calculating the number of false-negative (screen-negative individuals with diabetes), true-positive (screen-positive individuals confirmed to have diabetes on definitive testing), and false-positive (screen-positive individuals without diabetes on definitive testing) screening tests at each screening examination and over the 15-year study period.

For each screening examination following baseline, we estimated the prevalence of undiagnosed diabetes. First, the total number of subjects eligible for screening was estimated by excluding the number of new true positives. Second, the number of new cases of undiagnosed diabetes was calculated by excluding the false negatives from the total number of eligible subjects in the population and by multiplying by 0.22 (IGT/IFG prevalence) and 0.057 (rate of progression from IGT to diabetes). Third, the total number of cases with undiagnosed diabetes was calculated by adding the number of false negatives to the number of new cases. Finally, the prevalence of undiagnosed diabetes was calculated by dividing the total number of cases with undiagnosed diabetes by the total number of eligible subjects remaining in the population. After each screening examination, the number of eligible subjects became smaller because those diagnosed with diabetes were removed from the population.

Direct medical costs included physician time ($51 per visit), RPG tests ($5.24), diagnostic FPGs ($5.24), and OGTTs ($17.22) (7). Indirect costs included the cost of patient time (1 h for an initial visit or diagnostic FPG, 2.5 h for OGTT, $8.00 per hour) and travel ($7.00 per trip) (7). In 2000, 54.4 million Americans 45–74 years of age without a diagnosis of diabetes sought medical care, and 18.2 million did not (7). For individuals who sought care, we assumed that screening was opportunistic and that the only direct medical cost was the cost of the screening test and, when required, the cost of a diagnostic FPG or OGTT. For those who had not sought medical care, we considered the direct medical cost of screening to include the cost of an outpatient visit, the screening test, and, when required, the diagnostic FPG or OGTT.

The sensitivities and specificities of various RPG cut points are plotted on a receiver operating characteristic (ROC) curve in which the 2-h 75-g OGTT served as the gold standard (Fig. 1). Each incremental improvement in sensitivity for RPG ≥130 mg/dl was associated with a substantial reduction in specificity. The multivariate equation was more sensitive than RPG alone at a given level of specificity and more specific than RPG alone at a given level of sensitivity.

Figure 2 illustrates the number of false-negative screening tests at each screening examination. The slope of the curves, particularly those with shorter screening intervals, becomes flat after several screening examinations. Except for the most specific strategy (screening with an RPG cut point of 160 mg/dl every 5 years), which does not keep pace with the number of diabetic patients entering the population each year, the number of false-negative tests falls substantially after several screening examinations. Because the sensitivities are the same, the equation generates the same number of false negatives as an RPG cut point of 130 mg/dl.

Table 1 shows the cumulative number of true-positive and false-positive screening tests for the entire 15-year screening period. The absolute difference in the number of true-positive screening tests between the most sensitive and least sensitive strategies is 4.5 million. The absolute difference in the number of false-positive screening tests between the most sensitive and least sensitive strategy is 476 million. Thus, a cut point with higher specificity minimally decreases the number of true-positive screening tests (the yield) but substantially decreases the number of false-positive tests.

Table 2 shows cost data for the entire 15-year screening period. The total cost for the most sensitive and least specific strategy, using an RPG cut point of 100 mg/dl every year, is $42.7 billion. The total cost for the least sensitive and most specific strategy, using an RPG cut point of 160 mg/dl every 5 years, is $6.9 billion. The cost per true positive identified for screening with a cut point of 100 mg/dl every 3 years is $916, with a cut point of 130 mg/dl is $642, with a cut point of 160 mg/dl is $626, and with the equation is $563. Costs are lower for opportunistic screening than for population screening. Considering a strategy using a cut point of 130 mg/dl every 3 years, the cost per true positive for opportunistic screening is $275. For population screening, the cost per true positive is $1,745.

The ADA recommends opportunistic screening for type 2 diabetes. At the same time, the ADA acknowledges that questions remain as to the optimal method and frequency of screening (2). We found that an approach that balances sensitivity and specificity—an RPG with a cut point of 130 mg/dl or a multivariate equation applied every 3 years—is optimal.

The sensitivity and specificity of the cut point used to define a positive test have a major impact on efficacy and cost. If one considers screening to be a one-time event, it is tempting to reduce the cut point in order to increase sensitivity so that no cases are missed. However, with repeated screenings, the number of false-negative individuals in the population decreases substantially regardless of the sensitivity (Fig. 2). An unfortunate consequence of using a lower and more sensitive cut point is that it decreases specificity and substantially increases the number of false-positive screening tests (Table 1). The difference in cut points does not have the same dramatic impact on the cumulative number of true positives identified (Table 1).

In addition to selecting an appropriate cut point, one must consider screening periodicity. For each screening strategy, increasing the frequency of screening from every 5 years to every year approximately quadruples the number of false-positive tests requiring definitive diagnostic testing (Table 1). Increasing the time between screenings does, however, increase the likelihood that diabetes complications may develop in the interval between screenings. The incidence of complications in type 2 diabetes is difficult to estimate because the onset and duration are unknown. In type 1 diabetes, proliferative retinopathy begins to develop 3–5 years after onset of diabetes (11) and nephropathy begins to develop 6–10 years after onset (12). Therefore, screening every 3 years should not allow complications to develop among those remaining undiagnosed. Screening every 5 years may, however, allow for the development of undiagnosed and, hence, untreated retinopathy and nephropathy. The most sensitive RPG cut point has the highest total cost, driven by the large number of false-positive screening tests. The periodicity of screening affects the total cost even more than the choice of a cut point. The total cost of screening every year is more than twice that of screening every 3 years at each RPG cut point.

Incorporating screening into ongoing medical care also reduces cost. For opportunistic screening, the cost per true-positive case identified is less than one-third that associated with population-based screening (Table 1). In opportunistic screening, a higher proportion of the total cost is incurred after a positive screening test. This is particularly true of the indirect costs of opportunistic screening, because they are incurred only with follow-up diagnostic testing. Therefore, with opportunistic screening there is a substantially higher cost associated with the most sensitive strategies. Although the absolute costs associated with population screening are less, fewer people require population screening than opportunistic screening (18.2 million vs. 54.4 million). Studies of community screening have suggested that the yield of screening may be higher among those without regular health care (13). However, even if the yield is twofold higher in population-based screening, it remains less efficient.

When evaluating the cost of strategies with intermediate sensitivity and specificity, the multivariate equation has some advantages over RPG with a cut point of 130 mg/dl. Because both screening tests have the same sensitivity, they diagnose the same number of true positives. However, because the multivariate equation is more specific than RPG with a cut point of 130 mg/dl, it generates fewer false-positives. The total cost for the screening with equation is $9.7 billion versus $11.1 billion with an RPG with a cut point of 130 mg/dl. This translates into savings of $79 per case of undiagnosed diabetes identified. The benefits of using the multivariate equation must, however, be weighed against its logistical complexities and the feasibility and cost of obtaining information on the additional risk factors included in the equation.

There are several limitations to our study. First, we have not modeled all potential screening tests or strategies. Although the ADA has recommended screening with FPG, it is not commonly performed in routine clinical practice (4), and concern has been raised that the FPG alone may not be sufficiently sensitive as a screening test (14). Studies have reported the sensitivity of an FPG cut point of 126 mg/dl to be 35–59% and the specificity to be 85–95%, comparable to the sensitivity and specificity of an RPG cut point of 160 mg/dl (15–17). Second, we cannot determine whether the costs of screening are balanced by clinical benefits of earlier diagnosis and treatment. Although recent clinical trials have demonstrated benefits associated with early treatment of IGT/IFG (18–22), prospective trials have not addressed the long-term impact of earlier diagnosis and treatment of type 2 diabetes.

In summary, we have shown that screening strategies that balance sensitivity and specificity, such as RPG with a cut point of 130 mg/dl or a multivariate equation, provide good yield and minimize false-positive tests and costs. A screening interval of 3 years is long enough to minimize false-positives, but should not allow complications to develop. Opportunistic screening is more efficient than population screening. Screening is warranted if identification of those with diabetes through screening, and their early treatment, is shown to delay or prevent complications.