Efficient Cutoff Points for Three Screening Tests for Detecting Undiagnosed Diabetes and Pre-Diabetes: An economic analysis

We evaluated three screening tests based on the ADA recommendations and available data: fasting plasma glucose (FPG), HbA_1c, and random capillary blood glucose (CBG). We examined eight cutoff values for each screening test for the two goals (Table 1). For the first screening test, all individuals would receive an FPG test. Subjects with an FPG value exceeding the cutoff value but less than the value used to define impaired fasting glucose or undiagnosed diabetes would get an oral glucose tolerance test (OGTT). For the other two screening tests, all individuals would receive an HbA_1c test or a random capillary blood test, and those who had test values above the cutoff point would get an OGTT.

By dividing total cost of screening by the number of positive cases identified, we estimated the cost per positive case identified by each cutoff value, screening test, and the goal of screening. We conducted the analysis from two perspectives: a single-payer health care system perspective (including only the direct medical costs) and a societal perspective (including both the direct medical and nonmedical costs).

We defined undiagnosed diabetes and pre-diabetes using current ADA criteria (11). If the goal of screening was to identify pre-diabetes, we also included undiagnosed diabetes in the total number of positive cases identified because identification of undiagnosed diabetes is an unintentional but beneficial by-product of screening for pre-diabetes. Because the ADA recommends opportunistic screening, we applied each screening test to individuals who visited a health care provider at least once in the past year and did not report a previous diagnosis of diabetes. We restricted our analysis to individuals who were aged 45–74 years because there are no data on both the sensitivity and specificity of FPG and HbA_1c tests for other age-groups. Applying these criteria resulted in 54.5 million individuals in the U.S. who would be eligible for tests for both screening goals in year 2000. We obtained the total number of civilian people aged 45–74 from the U.S. 2000 Census (12).

The total number of positive cases identified was calculated by multiplying the proportion of cases identified by the number of cases in the screening population. The former, by definition, is equal to the sensitivity of the screening test whereas the latter is equal to the total number of individuals eligible for screening multiplied by the prevalence of undiagnosed diabetes alone or both pre-diabetes and undiagnosed diabetes. We estimated the sensitivity and specificity of the FPG and HbA_1c tests by using results from the third U.S. National Health and Nutrition Examination Survey (13) and those from a capillary blood test using data collected in a large community study (14) (Table 1). We estimated that the prevalences of undiagnosed diabetes and pre-diabetes were 8.8 and 36.8%, respectively, based on data from the third U.S. National Health and Nutrition Examination Survey (13).

We included both medical and nonmedical costs to estimate the direct cost of the screening test. Direct medical costs included laboratory tests, personnel time, and other material costs (e.g., costs of copying and mailing). Nonmedical costs included transportation to a health care provider and the patient’s time spent traveling and receiving tests. The total direct cost associated with each cutoff value of the screening test was calculated as a sum of the various resources used (e.g., physician time, laboratory tests) (15). The cost of each resource was the product of the following three components: number of physical units used to screen one person, the unit value of the resource, and the number of individuals screened (15). We obtained the cost data from various sources: laboratory test costs were from Medicare, physician fees were from HealthCare Consultants of America (16), transportation costs to a health provider were from the literature (17,18), and the cost of time for medical office staff other than physicians and patients was obtained from the Bureau of Labor Statistics (19). All costs were expressed in 2000 U.S. dollars.

The cost of identifying a case of pre-diabetes or undiagnosed diabetes varied by study perspective, purpose of screening, and cutoff value of the screening test. From a single-payer perspective, if the purpose of screening was to identify both pre-diabetes and undiagnosed diabetes, the costs per case identified by cutoff value ranged from $125 to $321 for the CBG test, from $114 to $476 for the FPG test, and from $153 to $536 for the HbA_1c test. From the societal perspective and for the same purpose of screening, the costs per case identified ranged from $175 to $389 for the CBG test, from $172 to $674 for the FPG test, and from $215 to $605 for the HbA_1c test. If the screening was to identify individuals with diabetes alone, from the single-payer perspective the costs per case identified by cutoff value ranged from $392 to $671 for the CBG test, from $556 to $717 for the FPG test, and from $590 to $817 for the HbA_1c test. For the same screening purpose, from the societal perspective, the costs per case identified ranged from $504 to $990 for the CBG test, from $816 to $1,177 for the FPG test, and from $728 to $1,165 for the HbA_1c test.

For all three screening tests, costs per case identified first decreased and then increased as the cutoff value of a screening test increased (Fig. 1). This pattern was independent of the goal of screening. Also, the efficient cutoff value of a screening test was lower for screening for pre-diabetes and undiagnosed diabetes together than for undiagnosed diabetes alone. For detecting both pre-diabetes and undiagnosed diabetes, the optimal cutoff point was 100 mg/dl for the FPG test, 5.0% for the HbA_1c test, and 100 mg/dl for the CBG test. In comparison, the most efficient cutoff points for detecting undiagnosed diabetes alone were 110 mg/dl, 5.7%, and 120 mg/dl, respectively.

The estimated cost associated with the most efficient cutoff point also differed by screening tests. For screening for both pre-diabetes and undiagnosed diabetes, the FPG test had the lowest cost, followed by the CBG test and the HbA_1c test. In comparison, for screening undiagnosed diabetes alone, the CBG test had the lowest cost, followed by the FPG test and the HbA_1c test. This pattern was independent of the study perspective.

The ADA recommends screening men and women ≥45 years at 3-year intervals for pre-diabetes and undiagnosed diabetes during a health care office visit (3). However, it has been unclear what cutoff value of a screening test should be different for each of the two screening goals. Previous studies evaluated different cutoff points of FPG tests, HbA_1c tests, and CBG tests without considering resource uses (14,20). Previously, we combined both the number of individuals identified and the resources used by five screening strategies to determine the strategies used for screening for pre-diabetes and their cutoff values (15). We were not able, however, to examine the full spectrum of cutoff values of the various biochemical screening tests evaluated in the study because of a lack of data. In addition, we did not evaluate the appropriate cutoff value if the goal of screening was to screen for undiagnosed diabetes alone. Furthermore, the definition of pre-diabetes has changed since our previous study (9). In our current study, we reanalyzed the data used previously to obtain all possible and logical cutoff values that can be used for screening for undiagnosed diabetes alone and for both undiagnosed diabetes and pre-diabetes based on the new definition of pre-diabetes.

Our results indicate that a lower cutoff value should be used for identifying individuals with pre-diabetes and undiagnosed diabetes than for identifying individuals with undiagnosed diabetes alone. The cost per positive case identified was determined by two factors: the number of cases identified and the total cost of screening. At a low cutoff value (e.g., 80 mg/dl for the random capillary blood glucose test), the screening test has a high sensitivity but low specificity. Although the test can identify more positive cases, the cost of the screening is also high, which results in a high cost per case identified. As the cutoff value of a screening test increases, the sensitivity of the test and the number of cases identified would decrease. This decrease, however, is offset by the lower total costs of the screening test, which yields a lower cost per case identified. After the lowest cost cutoff value point, further increasing the cutoff value would lead to a higher cost per case identified because of the reversed effect.

The most efficient cutoff value used for screening undiagnosed diabetes alone was too high for screening for both pre-diabetes and undiagnosed diabetes because the screening cost accompanied by a higher specificity was outweighed by a reduction in the number of cases identified. The reverse argument can be made that the most efficient cutoff point for screening for both pre-diabetes and undiagnosed diabetes was too low for screening for undiagnosed diabetes alone because the higher screening cost outweighed the benefit of identifying more cases.

For all three tests, our analysis indicated lower cutoff values than previously reported (14,20,21). Two factors may explain the difference between our results and those of previous studies. First, our cutoff values were based on costs per case identified, whereas previous studies were based only on the sensitivity and specificity of the screening test. Second, it is not clear what criteria were used for selecting the recommended or used cutoff values in previous studies. The best cutoff value is impossible to determine based on the sensitivity and specificity of the test alone because of the tradeoff between the two measures, as discussed earlier. If a subjective judgment were used to set the “appropriate” cutoff value, these studies might have given too much weight to the specificity of the screening test. In screening for undiagnosed diabetes, true positive cases would ultimately be determined by either the OGTT or FPG test. A highly specific cutoff value can reduce the total cost of screening but result in a high unit cost per case identified because fewer positive cases would be identified.

Our study has some limitations. First, we defined our cases of pre-diabetes or diabetes with a single FPG or OGTT test because of a lack of data on the specificity and specificity of another confirmation test at a different moment. This may lead to some misclassification between individuals with pre-diabetes and undiagnosed diabetes. For example, a person with pre-diabetes based on the FPG value of the first test may actually be found to have diabetes if he or she undergoes a test to confirm OGTT results. Also, some people would be “mislabeled” as having pre-diabetes if their results for the confirmation test were negative. Although this misclassification may have important clinical implications, it would have little impact on our study conclusion. For example, we performed a sensitivity analysis on how the efficient cutoff point would change if an OGTT confirmation test was added for all patients with identified cases of pre-diabetes or diabetes, assuming a sensitivity and specificity of 90%. Although the cost per case identified would go up because of the additional cost and decreased number of confirmed cases, the most efficient cutoff point for each test remained the same.

Second, our data on the sensitivity and specificity of FPG and HbA_1c tests were derived from a nationally representative sample but excluded individuals >74 years or <45 years of age. This population, however, is recommended for pre-diabetes screening by the ADA. Our data on the sensitivity and specificity of the CBG test were derived from a predominately low-income population. A previous study showed that the sensitivity and specificity of the FPG test could vary by ethnic groups or level of obesity in the European population (10). If the performance of the biochemical test varied across subpopulation groups, our most efficient cutoff values may not be applicable for all other subpopulation groups. In addition, our data on the sensitivity and specificities of the screening test were based on the scenario of prevalence screening and the performance of the tests may be different under the scenario of incidence screening. Our conclusion, however, that a lower cutoff value should be used when screening for both pre-diabetes and undiagnosed diabetes rather than only undiagnosed diabetes is likely to remain the same.

Third, we did not impose any costs for either missed cases of pre-diabetes or undiagnosed diabetes cases or for the potential negative benefits of detecting and sequentially treating those with abnormal results associated with different cutoff values. For clinicians, the cost of a case missed may have different implications because of concerns about liability. For some people with pre-diabetes or undiagnosed diabetes but with no diabetic symptoms, detecting and treating the condition may lower their quality of life.

Finally, our study focused on the costs of detection as opposed to the overall costs of a screening program and subsequent treatment. The cost of case detection is only a small part of the overall cost of implementing any screening program, either screening for pre-diabetes and undiagnosed diabetes together or for undiagnosed diabetes alone. Screening for asymptomatic chronic conditions such as pre-diabetes or undiagnosed diabetes may affect subsequent use of health services, subsequent work income, and subsequent sick leave from work. Moreover, a more important and costly task for any screening program would be ensuring appropriate treatment for the identified individuals. In addition, regarding the question of whether screening for pre-diabetes or undiagnosed diabetes is a worthwhile public health investment, the cost of detection and appropriate treatment is only part of the equation. To answer this question, one also needs to know the long-term health benefits of treating pre-diabetes and undiagnosed diabetes. Unfortunately, we currently have limited information on this issue. Because different cutoff values would yield different numbers of individuals who are labeled as having pre-diabetes or diabetes, we do not know if the most efficient cutoff value for identifying positive cases would remain the same when the long-term benefits and all the costs produced by the screening (including the cost of follow-up treatment) are considered.

Because of the different recommendations from professional organizations for screening for undiagnosed diabetes and pre-diabetes (3,4,9) and their implications in terms of health care resources, it is likely that physicians in different regions and communities will differ in implementing screening for pre-diabetes and undiagnosed diabetes. We have shown that the most efficient cutoff values for three biochemical screening tests depend on the goal of screening. Physicians should use lower cutoff values for detecting pre-diabetes and undiagnosed diabetes at the same time than for detecting diagnosed diabetes alone.