The Cost-effectiveness of Screening for Type 2 Diabetes FREE

TYPE 2 DIABETES mellitus is a common and serious disease in the United States, but one third of those affected are unaware they have it.^1- 2 Onset may occur 9 to 12 years before diagnosis is made with current clinical practice.³ At clinical diagnosis of diabetes, microvascular complications and several risk factors for macrovascular complications are frequently found.^4- 6 Early detection of type 2 diabetes through screening may therefore be an appropriate public health strategy.⁷ However, the health benefits of early detection and treatment of type 2 diabetes have never been firmly demonstrated.⁸ Additionally, screening may lead to misdiagnosis, inappropriate investigation and treatment, avoidable adverse effects, and unnecessary psychosocial and economic costs.^9- 10 With the current increase in costs of health care, consideration must be given to the cost and benefit of screening for any disease.

If early treatment of type 2 diabetes reduces the incidence or slows the progression of major complications, it might sufficiently reduce the costs of treatment during later years to offset the costs associated with screening and early treatment. Even if lifetime costs are higher with screening and early treatment, the costs might still be offset by the benefits derived from additional years of life gained or improved quality of life. We estimate the lifetime costs and benefits of opportunistic screening (ie, during routine contact with the medical care system) for type 2 diabetes. We assume that opportunistic screening will reduce the prediagnosis interval by 5 years from 10.5 years (under current clinical practice) to 5.5 years. The lifetime costs and benefits of opportunistic screening were compared with current clinical practice using recommended cost-effectiveness methods.¹¹

We developed a semi-Markov Monte Carlo simulation model (@Risk, Version 3.5 c for Windows NT, Palisades Inc, Newfield, NJ, and Excel, Version 7, Microsoft, Redmond, Wash) using assumptions similar to those used in previous diabetes cost-effectiveness studies.^12- 14 First, a hypothetical population without clinically diagnosed diabetes is selected and assigned to either opportunistic screening or current clinical practice. Second, for a cohort of 10,000 individuals with diabetes, the model simulates the development and progression of the major complications of the disease under each assigned alternative. This cohort is followed from onset of diabetes until death or age 95 years. The model parameters (eg, health states, transition probabilities, and costs) relied on major population surveys, epidemiologic studies, clinical trials, and other clinical studies.^14- 22 Interpretation and selection of the most appropriate parameters were accomplished by an expert panel convened for the study. The model adopts the perspective of a single payer for all direct medical costs. We did not consider direct nonmedical or indirect costs. We also did not address the issue or the costs of providing access to medical care for persons who lack access. Costs and benefits are discounted at 3% and costs are expressed in 1995 US dollars. The model structure, parameter assumptions, and relevant references are described below.

The model consists of a screening module and a disease progression module (Figure 1). The screening module assumes that each person is screened only once. The module determines the prevalence of undiagnosed diabetes, costs associated with screening, the likelihood that screening results in a diagnosis of diabetes, and the timing of diagnosis and treatment. A fasting plasma glucose test (FPGT) is used as the screening test (cutoff value, 6.1 mmol/L [110 mg/dL]; sensitivity, 80%; specificity, 96%).⁸ For persons testing positive to the first test, an oral glucose tolerance test (OGTT) is used to confirm diabetes. All persons with diabetes (true-positives and false-negatives) enter the disease progression module. Persons with a true-positive test result begin treatment when they are diagnosed through screening, and those with a false-negative test result begin treatment at clinical diagnosis. Conversely, all individuals who do not have diabetes (including false-positives) are excluded from the disease progression module.

The disease progression module^13- 15 models the natural history of diabetes and calculates the average costs of treatment. Unlike previous studies, this module begins at onset of diabetes rather than at clinical diagnosis. The module allows people to develop disease complications before their condition is diagnosed, as a function of the prevalence of complications at clinical diagnosis.^3- 5 At initial onset of diabetes, all persons are assumed to have no complications.

The disease states modeled for each of the major microvascular complications of diabetes are shown in Table 1. The probability of developing these complications varies with duration of diabetes, race, ethnicity, and level of glycemic control. Monte Carlo techniques are used to progress individuals through the model. At each step, a random number is drawn. Irreversible transition to the next health state occurs if the random number is less than or equal to the transition probability for progression from the current health state to the subsequent state. The transition probabilities (Table 2) are based on studies by Eastman et al¹⁴ and recalibrated to correspond to number of years from onset (rather than from diagnosis) and were thus shifted by 10 years to reflect the average length of the interval from disease onset to diagnosis. For example, if the transition probability for a complication at years 1 through 10 after clinical diagnosis was x, we used a transition probability of x for years 11 through 20 after onset. The relative risk adjustments for race are specific to each complication. Persons who have proliferative retinopathy or macular edema and who receive eye examinations are assumed to be treated with appropriate photocoagulation treatment. Only 1 lower-extremity amputation (LEA) is permitted per individual in a lifetime.

Mortality is modeled as a competing risk for each of the major complications of diabetes. Increased mortality rates among persons with diabetes are attributed to increased mortality due to cardiovascular disease, end-stage renal disease (ESRD), LEA, and other causes. The mortality rate is determined sequentially. For persons who undergo an LEA, the mortality rate varies by anatomic level of amputation.²⁹ If a patient does not have an LEA during the year or successfully survives the operation, the model assigns that person the mortality rate for ESRD, provided this condition is present.³⁰ If ESRD is not present, the mortality rate is determined from a combination of the cardiovascular and noncardiovascular mortality rates. After adjustment for age, sex, and race, the annual noncardiovascular mortality among persons with diabetes is 2.75 times that of persons without diabetes.¹⁴ Cardiovascular mortality rate, estimated from the Framingham Heart Study,¹⁹ is a function of age, sex, systolic blood pressure, total cholesterol level, high-density lipid levels, smoking behavior, and left ventricular hypertrophy.

An estimated 9 to 12 years (mean, 10.5 years) elapse between onset and clinical diagnosis of type 2 diabetes.³ Because variation is considerable across individuals, we assume that this distribution is approximately normal and that the range represents a 70% confidence interval. With screening, both the mean interval from onset to diagnosis and the distribution around the mean will be altered. We assume that screening lessens the mean prediagnosis interval by 5 years from 10.5 to 5.5 years. The likelihood that screening will identify an individual with diabetes is uniformly distributed across the prediagnosis phase. The prevalence of undiagnosed diabetes is estimated at 3.2% of the US population between the ages of 20 and 74 years.³¹ However, this prevalence varies across the population.⁴ Thus, we modeled the likelihood of undiagnosed diabetes as a logit function of age, race, ethnicity, obesity, and hypertension using data from the second National Health and Nutritional Examination Survey.³² We then applied this model to the US population and estimated the prevalence of undiagnosed diabetes for alternative screening cohorts.

The model assumes 1-time screening because this scenario allowed us to use currently available epidemiologic and clinical data or their direct extrapolation for most parameters. Repeat screening scenarios require assumptions for which data are not available. These include the incidence of diabetes onset and its relationship to the prevalence of diabetes risk factors in the population and screening test performance parameters adjusted to a previously screened population.

The progression of diabetes and its complications are modeled as a function of glycemic levels.¹⁴ While total glycemic exposure is the dominant factor affecting the risk of microvascular complications among persons with type 1 diabetes,³³ studies are investigating this relationship among persons with type 2 diabetes.^21,34- 35 We assumed that glycemic level affects the incidence of microvascular complications in type 2 diabetes, as many have espoused,^36- 40 and used the same relative benefits from treatment per unit difference in glycosylated hemoglobin (HbA_1c) level as was found in the Diabetes Control and Complications Trial (DCCT) of type 1 diabetes.³³ Because the link between glycemic level and cardiovascular disease is uncertain,^41- 43 our base case model restricts the causal links to the microvascular complications, which may understate some of the benefits of early detection and treatment.

The HbA_1c value is assumed to be 6.8% at onset of diabetes with an annual increase of 0.2 percentage points, resulting in an estimated 8.9% at clinical diagnosis,²² a value comparable with that found in the United Kingdom Prospective Diabetes Study (UKPDS) (9.0%).³⁴ Under screening, the HbA_1c level is on average 7.8% at diagnosis (roughly the midpoint from onset to clinical diagnosis). We estimated the SD at clinical diagnosis as 20% to 26% of the mean HbA_1c value,^22,34 and assumed that at screening diagnosis this SD would be roughly half (13%) because true duration is constant across individuals. We also assumed that the coefficient of variation remains constant as the mean glycemic level varies. On the basis of UKPDS results (HbA_1c, 9.0% at randomization and 6.9% at 1 year), we estimated that the reduction in HbA_1c level after 1 year is 2.1 percentage points with treatment by diet only. Thed initial HbA_1c reduction was not sustained over time in the UKPDS study.²⁰ Thus, our model assumes a rate of increase in HbA_1c levels after treatment is initiated that is comparable with the UKPDS diet-only group rate at 0.156 percentage points per year ([7.5%−6.8%]/4.5 years). The average HbA_1c values for the cohort are not permitted to decrease below 6.0% or exceed 11%.

We assume that screening will occur during an already scheduled physician visit. Therefore, cost will arise for all subjects from the screening test and additional physician time (estimated by the expert panel as a quarter of a physician visit). Persons with positive test results will consume additional resources by way of a confirmatory test (OGTT) and a full physician visit. Routine costs not specific to diabetes are estimated to average $1939 per year.^12- 13 In addition, costs associated with diabetes occur in 3 categories: outpatient and case management services, self-monitoring, and drugs.¹² Four treatment modalities were developed: diet and exercise only, oral hypoglycemic agents only, insulin only, and both oral hypoglycemic agents and insulin. The proportion using each modality varied by duration of diagnosed diabetes (estimated from the 1989 National Health Interview Survey Diabetes Supplement [NHIS]¹⁶). Also, based on data from NHIS, we estimated that compared with people without diabetes, people with diabetes not receiving insulin use 4 additional physician visits per year, and those receiving insulin use 5 additional visits per year. The cost of diet and exercise treatment at each visit was based on the level of care provided in the conventional treatment arm of the DCCT. For costing oral hypoglycemic treatment in the base case, glyburide at half the maximum dose was used, as was done by Eastman et al.^14- 15 The cost of insulin treatment was based on the experiences of the control group in the Veterans Affairs trial.²¹ The costs of routine treatment specific to diabetes and the costs of treating diabetes complications (Table 3) are expressed in US 1995 dollars and adjusted for inflation. The model assumes that treatment costs vary by duration from onset regardless of when diagnosis occurred. This assumption may underestimate the benefits of screening to the extent that earlier implementation of diet may delay the need for drugs.

Primary outcome measures were additional life-years and quality-adjusted life-years (QALYs), as measured from onset of diabetes. A utility value of 1.0 is used for each year of life lived without major complications and less than 1 for each year with a major complication (0.69 for blindness,⁴⁷ 0.61 for ESRD,¹³ and 0.8 for LEA²⁹). When multiple complications are present, the adjustment uses the lowest of the values.

We first modeled the effect of screening using a set of baseline assumptions for all adults 25 years or older. We then separately modeled age- and race-specific groups to examine the effects of screening subpopulations at relatively high risk of developing diabetes and its complications. Because of their relatively favorable cost-effectiveness ratio, the cohort aged 25 to 34 years is selected as the reference case. Sensitivity analysis was based on examining a change in only 1 parameter at a time and was performed for the following parameters: the screening method used, sensitivity and specificity of the FPGT, length of the prediagnosis interval, prevalence of undiagnosed diabetes, alternative assumptions about treatment (intensive glycemic level control and low-cost treatment through diet and exercise only during the early treatment period for persons diagnosed at screening), cost of physician's time for the screening test, and discount rate. Sensitivity to mortality rates was not assessed because of previously reported⁴⁸ insensitivity to this parameter.

Screening of all adults aged 25 years or older decreases the average age at diagnosis by nearly 6 years (Table 4). The lifetime cumulative incidence of ESRD, blindness, and LEA are reduced by 26%, 35%, and 22%, respectively, and years of life without major complications are increased (0.08 years, 0.27 years, and 0.15 years). The lifetime incidence of cardiovascular disease is not substantially affected (relative increase, 0.2%) because of increased survival. Screening increases the lifetime costs of treatment by $3388 but results in a gain in life-years of only 0.02 years (1 week). The incremental cost of screening over current clinical practice per additional life-year is estimated at $236,449, and the cost per QALY is estimated at $56,649.

It is more cost-effective to screen younger cohorts who can potentially gain more life-years free from major complications (Table 4). The greatest gain from screening is observed for blindness, for which the cumulative incidence is reduced by 7.5 percentage points for adults aged 25 to 34 years and by 0.5 percentage points for persons aged 65 years or older. Among adults aged 25 to 34 years, the lifetime incidence of cardiovascular disease actually increases by 0.8 percentage points, attributed to their small gain in longevity. In this age cohort, lifetime cost of treatment decreases by $1275 with screening, for an average gain in life-years of 0.13 (approximately 7 weeks). The cost per life-year for adults aged 25 to 34 years is $35,768 ($13,376/QALY), which is about one sixth the ratio obtained for all adults. Decreasing gains in both life-years and QALYs are found with increasing age. Compared with persons aged 25 to 34 years, the cost per life-year is nearly twice as high for adults aged 35 to 44 years ($64,878 [$18,707/QALY]), and 19 times higher for those aged 55 to 64 years ($681,989 [$116,908/QALY]. For adults aged 65 years or older, there is no gain in life-years.

Compared with all adults, screening results in greater reductions in the cumulative incidence of major complications among African Americans, resulting in larger increases in life-years and QALYs (Table 5). The model estimates savings of $5539 in the lifetime costs of treatment with screening among African Americans aged 25 to 34 years compared with $1275 among all races of the same age. The cost-effectiveness ratios for African Americans aged 25 to 34 years are $2219 per life-year and $822 per QALY, after screening costs are incorporated.

If HbA_1c levels of 7.0% or more⁴⁹ are used as the screening test, the cost-effectiveness ratio is $46,948 per life-year ($18,790/QALY) (Figure 2). Thus, using HbA_1c as the screening test appears slightly more cost-effective because treatment is directed at patients at higher risk (HbA_1c, ≥7.0% at screening) of developing complications. The model is moderately sensitive to the performance characteristics of the screening test (FPGT). If sensitivity and specificity each were 10% lower, the cost per case detected increases, and the incremental cost per life-year rises to $56,274 ($21,044/QALY).

To test the sensitivity to shorter prediagnosis intervals,⁵⁰ we halved this interval from 10.5 years to 5.5 years and found the health benefits of screening to be decreased and the cost per life-year to be increased nearly 3-fold to $137,148 ($38,983/QALY). We found screening more cost-effective at higher prevalence of undiagnosed diabetes, all else being constant, because of the lower cost per case detected. For example, a prevalence of 1.00% results in a cost per life-year of $21,937 ($8,203/QALY); a prevalence of 0.35% results in a cost per life-year of $79,434 ($29,704/QALY).

If patients receive intensive glycemic treatment rather than standard treatment, the health benefits of screening are reduced. Depending on duration of disease, the annual cost of routine care under intensive treatment ranges from $3111 to $3355, more than double the cost under standard treatment. With screening, these higher treatment costs accrue for an additional 5.9 years, and increase the cost per life-year to $132,549 ($46,760/QALY), which is more than 3-fold higher than under baseline assumptions. In contrast, if treatment consists only of diet and exercise during the lead time, the cost per life-year is reduced to $18,449 ($6979/QALY). This is because the treatment cost for diet and exercise would be only $476 instead of $709 in the base case, reflecting fewer physician visits (2 instead of 4) and lower case management costs.

The cost of the screening test is small compared with the cost of the physician's time. Therefore, we left the former unchanged and eliminated the latter and found that screening would be more cost-effective at $1058 per life-year ($396/QALY). Raising the discount rate from 3% to 5% increases treatment costs to $767 while leaving screening costs unaffected. At the same time, life-years decrease from 0.13 years to 0.07 years. Thus, cost per life-year more than doubles to $98,305 ($32,844/QALY).

Opportunistic screening of all adults aged 25 years or older for type 2 diabetes would cost $236,449 per life-year gained and $56,649 per QALY gained. In comparison, screening mammography for women aged 50 years or older costs from $3400 to $83,830 per life-year gained, annual screening for cervical cancer for women aged 21 years or older costs $50,000 per life-year gained, and hypertension screening for asymptomatic men and women 20 years old costs $48,000 and $87,000, respectively.^51- 53 We found that the benefits of early detection and treatment accrue more from postponement of complications and the resulting improvement in quality of life than from additional life-years. The cost-effectiveness of screening may be enhanced by targeting groups with higher lifetime incidence of major diabetes complications. Although the cost per case detected is higher among younger (because of lower prevalence) adults, these extra costs are more than offset by the reduced costs from lifetime complications prevented. Thus, it will be more cost-effective to screen younger than older adults. Among groups, such as African Americans, with both higher prevalence of diabetes and higher lifetime incidence of complications, screening may be even more cost-effective. The benefits of screening are greater the sooner after onset that detection occurs because the opportunity to reduce the development of major complications is enhanced. For example, in a hypothetical cohort of adults whose diabetes begins at age 30 years, the sooner after onset that screening and early treatment begins, the greater the reduction in the cumulative incidence of ESRD and the more favorable the cost-effectiveness ratio (Table 6).

We did not assess the effect of repeated screening or the influence of noncompliance with screening and treatment. Both these situations may reduce the cost-effectiveness of screening. Repeat screening scenarios might decrease the cost-effectiveness because the prevalence of undiagnosed diabetes would likely be lowered in subsequent screening efforts and result in a higher cost per case detected. The sensitivity analyses found the model moderately sensitive to the prevalence of undiagnosed diabetes. Noncompliance results in some expenditures without gain in benefit. On the other hand, several of our assumptions may be conservative. A screening program will also identify persons with impaired glucose tolerance who may benefit from early intervention. Our model also does not take into account the potential benefit of early initiation of glycemic level control on the incidence of cardiovascular disease. There is emerging evidence that glycemic level and cardiovascular disease may be associated.^41- 43 Early detection may offer the opportunity to influence macrovascular risk factors other than hyperglycemia. However, we did not incorporate this benefit in our model because of a paucity of empirical data. Finally, our model assumes that microvascular complications are affected only by current glycemic levels. Had we used a less conservative approach of modeling cumulative glycemic exposure (eg, HbA_1c multiplied by duration),⁵⁴ the benefits of screening would be enhanced.

Our model was not sensitive to the choice of screening test but was moderately sensitive to the assumptions concerning the performance characteristics of the FPGT. As recently recommended by the American Diabetes Association (ADA),⁷ if a second FPGT was used for confirmation in place of an OGTT, we estimate that the total costs would decline by only 2%. The results are thus not particularly sensitive to the choice of the confirmatory test because most of the medical cost associated with the confirmation test is due to the additional physician visit rather than the test itself. For the same reason, using an HbA_1c value of 7.0% or more to confirm diabetes, rather than an OGTT, will not dramatically change the cost-effectiveness of screening. The model was sensitive to the assumptions concerning the length of the prediagnosis interval, the prevalence of undiagnosed diabetes, and the intensity of glycemic control therapy. If intensive glycemic control became the standard of care for type 2 diabetes, screening may be less cost-effective. Average glycemic levels with intensive treatment may be sufficiently low that the lifetime incidence of complications is also low, thus allowing little room for further improvement through early diagnosis and treatment. Our estimates of cost-effectiveness are also sensitive to the discount rate and are higher (less cost-effective) at higher discount rates. We considered the possibility that diagnosis of diabetes may have a transient or more long-lasting effect on quality of life. However, because of lack of empirical data we did not assess this issue.

Our study suggests that it is more cost-effective to opportunistically screen young adults for type 2 diabetes, contrary to the current recommendations of the ADA to screen only persons aged 45 years or older.⁷ Age-specific cost-effectiveness ratios from this study offer one means of assessing the relative opportunity costs of screening various subpopulations and for framing decisions concerning age thresholds for screening within the constraints of finite financial resources. Our results suggest that the selection of appropriate age groups should not ignore younger adults. However, factors other than the cost-effectiveness of a single intervention (eg, number of people needing the intervention) may influence the optimal decision.⁵⁵ Thus, when selecting a constellation of interventions aimed at improving overall public health, each choice may not necessarily be the most cost-effective for an individual patient.⁵⁵

Our analysis takes a single-payer perspective. Thus, the results of our study may be more directly applicable to Medicare and some Medicaid populations. Although targeting young adults for screening may have the lowest opportunity cost, a multipayer system may not have the financial incentive (eg, reimbursement) to readily adopt such a policy. Implementation of the findings of our study will require the acceptance and advocacy of influential groups like state Medicaid agencies, the Health Care Financing Administration, the American Association of Health Plans, and the National Center for Quality Assurance, as well as the ADA.