We built a state transition model of screening results and type 2 diabetes disease progression to simulate lifetime diabetes-related health care costs and QALYs. Our target population comprised individuals who had no history of T2DM or cardiovascular events. We stratified the population into age categories of 30–44, 45–59 and 60–74 years old. For each age category, we also stratified patients into BMI status as follows: underweight, BMI < 18.5; normal weight, BMI 18.6–24.9; overweight, BMI 25–29.9; and obese, BMI ≥ 30 [7, 8]. In total, 12 stratifications were used to create the state transition models. We used the same tree structure for all 12 models with different parameter values. The tree structures in the state transition model comprise three main arms (Fig. 1), “screening results by HbA1c test”, “no screening year” and “T2DM progression”. Screening result arms consist of parameters: incidence of T2DM and sensitivity and specificity of HbA1c testing to determine how many individuals from the population go to the T2DM progression arm. In the no screening year arm, people are classified as the condition that was determined in the screening results arm. For example, if a person was categorized as false negative, they had a higher complication rate while they were in the no screening year because they technically missed a chance for early detection and treatment. In the T2DM progression arm, people die or may experience complications based on each relative risk. A first-order Monte Carlo simulation (microsimulation) of a hypothetical cohort of 50,000 people was performed to estimate the lifetime expected costs and expected QALY. The cycle length of the model was set to 1 year. A willingness-to-pay (WTP) threshold of 50,000 USD per QALY gained was used as the acceptable level for ICER. An annual discount rate of 2% was applied to both cost and benefit [9]. The incremental cost-effectiveness ratio (ICER) was estimated as an indicator of cost effectiveness using the following formula: ICER = (Cost interval_a – Cost interval_b)/(QALYd interval_a – QALY interval_b). The model was developed and analyzed using TreeAge Pro 2016 (TreeAge Software, Williamstown, MA, USA).
Parameters in the state transition models
Screening results by HbA1c test
The sensitivity and specificity of the HbA1c test were calculated in each population using real data from St. Luke’s International Hospital, Tokyo. We used the same data set and methodology to calculate sensitivity and specificity when we recommended different intervals based on risk stratification [6] (Table 2). The methodology has been described in detail elsewhere [10, 11]. Briefly, linear random effect models with random intercept and random slope, adjusted for sex, age and BMI at first measurement of HbA1c as continuous value, were used to generate predicted HbA1c. Sensitivity and specificity were calculated by comparing the observed HbA1c value and the generated predicted HbA1c value as the gold standard.
Table 2
Sensitivity and specificity rates applied to the state transition models.
Age category | BMI level | Interval | HbA1c sensitivity (%) | HbA1c specificity (%) |
30–44 | Underweight (< 18.5 kg/m2) | Annual | ≒0.0 | 100.0 |
3 years | 50.0 | 100.0 |
10 years | 81.8 | 99.9 |
Normal (18.5–25 kg/m2) | Annual | ≒0.0 | 100.0 |
3 years | 47.5 | 99.9 |
6 years | 68.1 | 99.9 |
Overweight (25–30 kg/m2) | Annual | ≒0.0 | 100.0 |
3 years | 54.5 | 99.7 |
4 years | 57.8 | 99.6 |
Obese (≥ 30 kg/m2) | Annual | ≒0.0 | 100.0 |
2 years | 37.5 | 99.1 |
3 years | 62.2 | 98.6 |
45–59 | Underweight (< 18.5 kg/m2) | Annual | ≒0.0 | 100.0 |
3 years | 66.7 | 99.8 |
10 years | 70.0 | 99.7 |
Normal (18.5–25 kg/m2) | Annual | ≒0.0 | 100.0 |
3 years | 56.0 | 99.7 |
6 years | 73.5 | 99.5 |
Overweight (25–30 kg/m2) | Annual | 2.5 | 100.0 |
3 years | 61.1 | 99.1 |
4 years | 66.8 | 99.0 |
Obese (≥ 30 kg/m2) | Annual | ≒0.0 | 100.0 |
3 years | 62.0 | 98.0 |
4 years | 71.4 | 98.0 |
60–74 | Underweight (< 18.5 kg/m2) | Annual | ≒0.0 | 100.0 |
3 years | 50.0 | 99.5 |
6 years | 87.5 | 99.3 |
Normal (18.5–25 kg/m2) | Annual | ≒0.0 | 100.0 |
3 years | 50.3 | 99.3 |
7 years | 72.7 | 99.1 |
Overweight (25–30 kg/m2) | Annual | ≒0.0 | 100.0 |
3 years | 52.3 | 99.1 |
5 years | 64.5 | 98.7 |
Obese (≥ 30 kg/m2) | Annual | ≒0.0 | 100.0 |
3 years | 60.0 | 97.5 |
4 years | 70.0 | 97.5 |
Cost
Direct costs estimated in this study include screening cost and type 2 diabetes treatment. Indirect costs were not considered in this study. We estimated the unit costs of screening with the HbA1c test (including consumables, staff time and laboratory processing costs) as USD 80.00. We had to assume the fee for type 2 diabetes screening with HbA1c because medical cost for prevention is not covered in Japan; thus, there was no official price list for the type 2 diabetes screening test in Japan. We estimated it by summing the cost for a T2DM patient with a stable glycemic condition who received a routine HbA1c test followed by a doctor’s consultation.
For the treatment fee, we used published cost data for T2DM. Fukuda et al. [12] reported detailed treatment costs for T2DM as well as proportion rates for each T2DM-related complication. In our model, we aggregated to one treatment cost for any complication based on the proportions of Japanese people experiencing complications. We also estimated annual treatment fees for false positive patients by summing nutrition education and physical exercise education and assuming no drug prescription fees.
Utility
We assumed the utility value to be that of full health and set at 1. We assigned 0.785 utility for those with T2DM without any complications based on a previous study [13]. We calculated a single utility value for those with T2DM with any complication based on multiple studies. Fukuda et al. [12] thoroughly reported treatment costs for patients with T2DM and the proportion of T2DM-related complications using the Japan Medical Data Center Claims Database. We first retrieved utility values for each T2DM-related complication from previous studies and then weighted each utility value based on the proportion reported to aggregate into one utility, which represents the average utility value for patients with T2DM with any complication (Table 1).
Table 1
Parameters used in the tree model
| Items | Type of distribution of PSA | Point estimate | Distribution parameters for PSA | Ref |
Cost (USD) | Screening cost | Gamma | 100 | α = 27.3, λ = 1/292 | |
Annual treatment fee among T2DM patients without complication | Gamma | 3,500 | α = 84146, λ = 1/3 | 12) |
Annual treatment fee among T2DM patients with complication | Gamma | 8,000 | α = 26540, λ = 1/22 | 12) |
Annual treatment fee among false positive patients with no medication | Gamma | 1,400 | α = 1752, λ = 1/60 | 12) |
Discount rate | - | 0.02 | - | 23) |
Utility | Utility for healthy population | - | 1 | | 13) |
Utility for T2DM patients without complication | - | 0.785 | - | 13) |
Utility for T2DM patients with complication | - | 0.638 | - | 13) |
Risk | Mortality rate among healthy population | Life Table | - | - | |
Relative risk towards mortality rate among T2DM patients with complications | LogNormal | 5.61 | Log (mean) = 1.65, SE = 0.08 | 16) |
Relative risk towards mortality rate among T2DM patients without complications | LogNormal | 2.61 | Log (mean) = 0.95, SE = 0.14 | 16) |
Annual complication rate | Beta | 0.014 | α = 2.27, β = 160.3 | 18) |
Relative risk towards complication rate among T2DM patients with treatment | LogNormal | 0.79 | Log (mean) = -0.23, SE = 0.09 | 19) |
Table 4
Suggested screening strategies based on cost-effectiveness analysis, stratified by age and BMI group
| Age groups |
30–44 | 45–59 | 60–74 |
BMI group | Underweight | 10-year | 10-year | 3-year |
Normal | 6-year | 6-year | 3-year |
Overweight | 4-year | 3-year | 3-year |
Obese | 3-year | 3-year | 3-year |
Risks from type 2 diabetes
The age-dependent mortality rate for people without type 2 diabetes was obtained from the life table reported by the Ministry of Health, Labour, and Welfare in Japan [14]. We assumed that people with type 2 diabetes receiving the appropriate treatment would achieve the same mortality rate as people without type 2 diabetes based on a recent study [15]. The relative risks of mortality for patients with type 2 diabetes with and without complications were set to 5.22 and 2.61, respectively [16, 17].
The annual incidence of T2DM complications was set to 0.014 based on a previous study [18]. Furthermore, we assumed that patients receiving appropriate treatment would experience fewer complications than those receiving no treatment. There are no published complication rate data for patients with no treatment; thus, we decided to retrieve data from the report, which compared metformin therapy versus conventional therapy. We treated conventional therapy as no treatment, so patients receiving appropriate treatment with metformin therapy had a 0.78-fold lower complication rate [19].
Probabilistic sensitivity analysis (PSA)
The robustness of the model results was explored by considering the impact of the model assumptions and the uncertainties in the model input parameters. This was done using deterministic and probabilistic sensitivity analyses (PSA), using the distribution for each parameter shown in Table 1. The PSA explored the uncertainties in the model parameters by randomly sampling 1,000 people with 1,000 iterations on each parameter distribution. We calculated the cost, QALYs and ICERs from this sample.