Details of the CDC-RTI diabetes cost-effectiveness model appear elsewhere (7,8). Briefly, it is a Markov disease progression model of type 2 diabetes. The model starts patients at the time of diagnosis of their diabetes and progresses them until death or until age 94 years. Diabetic patients in the model can develop five complications: neuropathy, nephropathy, retinopathy, CHD (i.e., cardiac arrest [CA]/myocardial infarction [MI] and angina), and stroke. Each complication has its own costs, and the model aggregates costs over a patient's lifetime. Death can result from diabetes complications or from other causes (7). The model also simulates possible side effects of aspirin therapy on gastrointestinal bleeding. The transitional probabilities between modeled disease states and intervention effectiveness parameters were derived from the Framingham Heart Study, the UK Prospective Diabetes Study (UKPDS), and other clinical trials and observational studies (9–11).

The model was validated following procedures of the American Diabetes Association Consensus Panel on Computer Modeling (12). Overall, the diabetes cost-effectiveness model accurately simulates the natural history of the diabetic population. This model has been used to evaluate the cost-effectiveness of many type 2 diabetes interventions (7,8).

We defined the demographics of our simulated population as described below. We calculated incidence rates for cohorts defined through age, sex, and race/ethnicity using data from the 2005 National Health Interview Survey, reported by the CDC. We estimated the prevalence of hypertension, high cholesterol, and smoking using National Health and Nutrition Examination Survey (NHANES) 2001–2006 data. We also obtained the initial prevalence of diabetes complications (nephropathy, neuropathy, retinopathy, CHD, and stroke) using NHANES data from 2001 to 2006. The initial A1C level was assumed to be 6.8% based on UKPDS (9,10).

We modeled daily use of 80 mg of aspirin among a cohort of U.S. residents aged 40–94 years who, in 2006, were newly diagnosed with diabetes. We defined standard care for subjects with diabetes: all patients received intensive glucose control with a treatment intensity similar to that of the intensive arm of the UKPDS (A1C 7%); and hypertensive patients were treated to meet the target diastolic blood pressure of 80 mmHg, the target blood pressure rate in the UKPDS intensive arm. We assessed the cost-effectiveness of aspirin use plus standard care compared with standard care alone. We assumed that aspirin was effective for everyone who took it. We also assumed 100% compliance. Table 1 shows the main analytic parameters' values and the sources of these parameters.

A recently published meta-analysis estimated the effect of aspirin use in primary prevention of CVD in patients with diabetes. The conclusion was that the effect of aspirin use in patients with diabetes remains unproved (13). We are aware of no data that suggest that the effect of aspirin on primary prevention of CVD in the diabetic population differs from that in the general population. Accordingly, we used the effects that were derived from a new meta-analyses of clinical trials for the general population in the base-case scenario. In the base-case analysis, for primary prevention, aspirin use reduced annual risk of coronary events by 18% and stroke by 5%; for secondary prevention, aspirin reduced the risk of major coronary events by 20% (14).

In our base-case scenario, we assumed that CVD-free patients with newly diagnosed diabetes received aspirin for primary CVD prevention. Those with a history of MI (13%) at the time of diagnosis of diabetes received aspirin for secondary CVD prevention. In addition, we assumed that the effects of aspirin use on primary or secondary prevention in the trial lasted throughout the lifetime.

Aspirin use increases risk of hemorrhagic stroke and reduces risk of ischemic stroke (14). However, we did not have reliable cost data for hemorrhagic stroke. Thus, we used the overall risk ratio (RR) of 0.95 for all stroke events, whether ischemic or hemorrhagic, in the base-case analysis. We modeled ischemic and hemorrhagic strokes separately in the sensitivity analyses, described later. Aspirin use also increases the risk of gastrointestinal bleeding. For major gastrointestinal bleeding, the RR associated with aspirin use was 1.54. We assumed no gastrointestinal bleeding for patients not using aspirin. We estimated that the gastrointestinal bleeding occurred in patients in the aspirin group at a rate of 0.03 per 100 (14). We assumed that patients who experienced aspirin side effects would stop taking aspirin.

Our analysis, conducted from a health system perspective, includes only direct medical costs. We considered the following: cost of aspirin treatment, costs of the standard care (intensive glycemic control for all patients and intensive hypertension control for patients with hypertension), costs of treating diabetes complications, and costs of treating aspirin side effects. Costs for standard care and diabetes complications were estimated using a multiplicative cost function. This cost function was based on data from adults with diabetes under different treatment regimes and with different complications and comorbidities (15,16). We obtained the costs of aspirin from Web sites of the large chain pharmacies and gastrointestinal bleeding from the literature and added them to the total costs of treatment for diabetes and its complications (Table 1) (5). We also assumed that gastrointestinal bleeding occurs only once and incurs costs only in the first year after occurrence. All costs were converted to 2006 U.S. dollars by using the Consumer Price Index for medical services.

Measures of primary outcomes for our analysis were the number of life-years (LYs) and the number of quality-adjusted LYs (QALYs), measured from time of clinical diagnosis of diabetes until the cohort reaches 94 years of age or death. We used an additive model for the utility measures (17). In this model, patients with characteristics (e.g., demographic characteristics, risk factors, complications, and BMI) differing from the baseline characteristics had those characteristics' disutility coefficients added to the intercept (17). For example, for a male subjects with BMI <30 kg/m² and without major complications, the state of having diabetes had a utility value of 0.689. If this subject had a cardiac arrest/MI, his utility score was reduced from the baseline score by 0.052. We also assumed no disutility for taking aspirin and a disutility of 0.030 for gastrointestinal bleeding (18). Costs and QALYs were discounted at 3% annually.

We conducted both one-way and probabilistic sensitivity analyses. For the former, we considered multiple scenarios. We examined the effectiveness and cost-effectiveness for several age-groups and both sexes. We used the upper and lower 95% confidence bounds of the main parameters and the parameters in diabetes subgroups reported by De Berardis et al. (13). Because the average follow-up times in the primary and secondary prevention trials for aspirin use were 5 and 3 years, respectively, we modeled interventions with benefits limited to the trial follow-up time. We also varied the cost of gastrointestinal bleeding and compliance rate and modeled cost-effectiveness of aspirin use for secondary prevention of CVD only.

Substantial proportions of diabetic patients do not receive intensive glycemic or hypertension control (19). Aspirin use might be more effective in these patients because of their increased CVD risk. We modeled the aspirin cost-effectiveness in patients with less intensive glucose or blood pressure control as the UKPDS control arms.

We also modeled ischemic and hemorrhagic strokes separately. We assumed that all strokes in the nonaspirin group were ischemic and modeled the change in costs and effectiveness of aspirin use as a result of increased risk for hemorrhagic stroke and decreased risk for ischemic stroke. We assumed an excess annual risk of hemorrhagic stroke and decreased risk of ischemic stroke as reported in the Antithrombotic Trialists' (ATT) Study (14). The multiplicative cost model did not provide costs for hemorrhagic stroke. Accordingly, we calculated that cost by adjusting the cost of ischemic stroke, using the ratios of the first year and ongoing costs between hemorrhagic stroke and ischemic stroke reported by Pignone et al. (5), respectively. This is equivalent to assuming that the cost ratios between hemorrhagic and ischemic strokes are constant as reported in the study by Pignone et al.

For our probabilistic sensitivity analysis, we randomly varied the following parameters in the model simultaneously based on distributions of their estimates: relative risk of aspirin use for primary and secondary prevention of CHD, stroke, and gastrointestinal bleeding; utilities for all health states; and cost of MI, stroke, and gastrointestinal bleeding. We assumed that efficacy and the rate of side effects followed a log-normal distribution. We assumed a log-normal distribution for MI and stroke cost, a triangular distribution for the cost of gastrointestinal bleeding, and a normal distribution for utility data with reported means and SDs. Running simulations for all age-groups was computationally prohibitive. However, the baseline effectiveness, cost, and cost-effectiveness ratios for the 55–64 years age-group were similar to those for the entire population. Accordingly, we ran 1,000 realizations using parameters on the efficacy of aspirin use reported in the ATT study (14) and the study by De Berardis et al. (13), limited to the 55–64 years age-group, and reported the results as our probabilistic sensitivity analysis.

Over the lifetime of people with newly diagnosed diabetes, aspirin use reduced the cumulative incidence of CHD events by 3.91% and the CHD mortality rate by 4.65%; however, aspirin increased the cumulative incidence of stroke by 0.51% and the stroke mortality rate by 0.28%. Overall, a person in the aspirin group gained 0.31 LYs and 0.19 QALYs (discounted at 3% annually) compared with a person who did not take aspirin, at an incremental cost of $1,700. The Incremental cost-effectiveness ratio (ICER) was $5,428 per LY gained or $8,801 per QALY gained (Table 2).

The one-way sensitivity analysis showed that the ICERs of aspirin use changed moderately if the extreme values for the effect of aspirin on primary prevention of CHD, different sex, aspirin effect on secondary prevention, and modeling different types of strokes separately hold. The ICERs did not change substantially for other variables (Table 3). Using the upper 95% of confidence limits of the effect of aspirin in primary prevention of CHD increased the ICER from $8,801/QALY at baseline to $11,289/QALY. When men and women were modeled separately using parameters in the ATT study (14) and De Berardis et al. (13) study, women always had higher CERs than men: $13,833/QALY vs. $5,752/QALY using parameters in ATT study and $22,259/QALY vs. $3,633/QALY using parameters in De Berardis et al. study. Only considering aspirin effect in secondary prevention of CVD has an ICER of $18,348/QALY. Modeling hemorrhagic stroke and ischemic stroke separately yielded an ICER of $16,484/QALY, which doubled the baseline results.

Figure 1A and B show the ICER plot from the probabilistic sensitivity analysis of the cost-effectiveness of aspirin using parameters in the general population as in the ATT study (14) and parameters in diabetes subgroup as in De Berardis et al. (13) study. Using the ATT study parameters, 100% of the realizations yielded an ICER of <$16,000/QALY gained. The 2.5th and 97.5th percentiles were $5,520 and $12,600 per QALY gained, respectively (Fig. 1A). Using parameters in De Berardis et al. (13) study, all of the 1,000 simulations yielded an ICER <$27,000 per QALY gained. The 2.5th and 97.5th percentiles were $4,916 and $16,680 per additional QALY gained, respectively (Fig. 1B).