Projecting the incidence and costs of major cardiovascular and kidney complications of type 2 diabetes with widespread SGLT2i and GLP-1 RA use: a cost-effectiveness analysis

Aims/hypothesis Whether sodium–glucose co-transporter 2 inhibitors (SGLT2is) or glucagon-like peptide-1 receptor agonists (GLP-1 RAs) are cost-effective based solely on their cardiovascular and kidney benefits is unknown. We projected the health and economic outcomes due to myocardial infarction (MI), stroke, heart failure (HF) and end-stage kidney disease (ESKD) among people with type 2 diabetes, with and without CVD, under scenarios of widespread use of these drugs. Methods We designed a microsimulation model using real-world data that captured CVD and ESKD morbidity and mortality from 2020 to 2040. The populations and transition probabilities were derived by linking the Australian Diabetes Registry (1.1 million people with type 2 diabetes) to hospital admissions databases, the National Death Index and the ESKD Registry using data from 2010 to 2019. We modelled four interventions: increase in use of SGLT2is or GLP-1 RAs to 75% of the total population with type 2 diabetes, and increase in use of SGLT2is or GLP-1 RAs to 75% of the secondary prevention population (i.e. people with type 2 diabetes and prior CVD). All interventions were compared with current use of SGLT2is (20% of the total population) and GLP-1 RAs (5% of the total population). Outcomes of interest included quality-adjusted life years (QALYs), total costs (from the Australian public healthcare perspective) and the incremental cost-effectiveness ratio (ICER). We applied 5% annual discounting for health economic outcomes. The willingness-to-pay threshold was set at AU$28,000 per QALY gained. Results The numbers of QALYs gained from 2020 to 2040 with increased SGLT2i and GLP-1 RA use in the total population (n=1.1 million in 2020; n=1.5 million in 2040) were 176,446 and 200,932, respectively, compared with current use. Net cost differences were AU$4.2 billion for SGLT2is and AU$20.2 billion for GLP-1 RAs, and the ICERs were AU$23,717 and AU$100,705 per QALY gained, respectively. In the secondary prevention population, the ICERs were AU$8878 for SGLT2is and AU$79,742 for GLP-1 RAs. Conclusions/interpretation At current prices, use of SGLT2is, but not GLP-1 RAs, would be cost-effective when considering only their cardiovascular and kidney disease benefits for people with type 2 diabetes. Graphical abstract Supplementary Information The online version contains peer-reviewed but unedited supplementary material available at 10.1007/s00125-022-05832-0.


Transition probabilities for the model.
We used a linked dataset derived from the National Diabetes Services Scheme (NDSS) to estimate the incidence of hospitalization for myocardial infarction (MI), stroke, and heart failure (HF), the incidence of end-stage kidney disease (ESKD), and all-cause mortality. MI, stroke, and HF admission was defined from hospital admissions data when one of the following International Classification of Diseases (ICD)-10 codes was listed as the primary diagnosis for an admission: MI (ICD-10 codes: I21-I22), stroke (I60-I64), and HF (I110, I130, I132, or I50). We included admissions to public hospitals only. ESKD was defined from the Australia and New Zealand Dialysis and Transplant Registry as initiation of kidney replacement therapy (receipt of dialysis or a kidney transplant). Allcause mortality was derived from the National Death Index.
Incidence rates were estimated for the period covering 1 July 2016 to 31 May 2019. We included a look-back period from 1 July 2010 to 30 June 2016 to split the NDSS population into their appropriate health state (i.e., whether they had had a prior admission for MI, stroke, or HF, or developed ESKD). This was used to split subsequent follow-up time by the presence or absence of prior MI, stroke, HF, and/or ESKD. If an individual experienced any of these outcomes during follow-up, their risk time was split into periods representing time before and after this event. This way, we could estimate the effect of each prior event on subsequent events. To partition follow-up time, individuals were followed from 1 July 2016, date of NDSS registration, or migration into one of the four states in this study, until an event (MI, stroke, HF, ESKD, or death), migration out of one of the four states, or end of follow-up. Except for death, once an event occurred, individuals were followed subsequently until another event, migration out, or end of follow-up. Because admissions are only recorded in hospital admitted datasets at discharge, we terminated follow-up on 31 May 2019 to avoid underestimating admission rates.
Follow-up time was then split into 6-month intervals by attained age (10-100 years), duration of diabetes (0-50 years), and calendar time, and event counts and risk time were then tabulated in these intervals, with each tabulation assigned the midpoint of the interval. Age at diagnosis of diabetes was added to the tabulation as attained age minus duration of diabetes. Incidence and mortality rates were then analyzed via a Poisson model (one for each outcome) with spline effects of attained age, duration of diabetes, age at diagnosis of diabetes, a linear effect of calendar time, and binary effects of prior MI, stroke, HF, and ESKD, using log-person-time as the offset. Models were fit for men and women separately. The models were then used to predict the incidence of each complication by any combination of age, duration of diabetes, and prior MI, stroke, HF, and ESKD, with the prediction year set at 2019; however, the effect of calendar time was partially carried forward from 2020-2040. While the use of SGLT2is and GLP-1 RAs was relatively low during the estimation period, we nevertheless corrected these incidence rates for use of SGLT2is and GLP-1 RAs during the estimation period.

Validation of transition probabilities and model structure
We validated the transition probabilities and model structure using two tests. First, we compared the actual vs. predicted number of events using the actual population structure among people with type 2 diabetes. Events were predicted using the transition probabilities described above and actual demographic structure of the population with type 2 diabetes for three financial years (we selected 3 so that the first 6 years could be used to define prior events). This test (ESM Fig. 4-8) demonstrated that the modelled transition probabilities could recapitulate reality reasonably well in the absence of any other model inputs.
Second, we compared the actual vs. predicted number of events using the actual population structure in the first year, then a modelled population structure in the second 2 years. I.e., we tested our full model over the years financial years 2016-17 to 2018-19 to see if the model could recapitulate reality for years in which we had data. This test (ESM Fig. 9) demonstrated that our full model could recapitulate reality reasonably well.  200,159,565 $201,678,615 $1,519,050 $201,864,236 $1,704,577,697,920 $6,610,658,368 $32,960,448 $6,617,250,016 $39,552,694,475,256 $1,592,768,706,810 $1,580,359,115,926 Total productivity costs $8,551,923,232 $8,481,164,758,422 $8,473,169,753,542 Total societal costs $25,174,623,995 $25,107,320,303,067 $26,556,767,334 $1,382,143 47,405 All costs are presented in 2020 Australian dollars. All health economic outcomes have been subject to 5% annual discounting.

Background and objectives
3 Give the context for the study, the study question, and its practical relevance for decision making in policy or practice. 3

Methods
Health economic analysis plan 4 Indicate whether a health economic analysis plan was developed and where available.

Study population 5
Describe characteristics of the study population (such as age range, demographics, socioeconomic, or clinical characteristics).
Appendix -ESM Table 1 Setting and location 6 Provide relevant contextual information that may influence findings.

5, 6
Comparators 7 Describe the interventions or strategies being compared and why chosen.

3, 7
Perspective 8 State the perspective(s) adopted by the study and why chosen.

4, 5
Time horizon 9 State the time horizon for the study and why appropriate. 5

Discount rate
10 Report the discount rate(s) and reason chosen. 5

Selection of outcomes 11
Describe what outcomes were used as the measure(s) of benefit(s) and harm(s).

6
Measurement of outcomes 12 Describe how outcomes used to capture benefit(s) and harm(s) were measured.

6
Valuation of outcomes 13 Describe the population and methods used to measure and value outcomes.

Measurement and valuation of resources and costs
14 Describe how costs were valued. Table 1 Currency, price date, and conversion 15 Report the dates of the estimated resource quantities and unit costs, plus the currency and year of conversion. Table 1 Rationale and description of model 16 If modelling is used, describe in detail and why used. Report if the model is publicly available and where it can be accessed.

Characterising distributional effects
19 Describe how impacts are distributed across different individuals or adjustments made to reflect priority populations.

N/A
Characterising uncertainty 20 Describe methods to characterise any sources of uncertainty in the analysis.

7, 8
Approach to engagement with patients and others affected by the study 21 Describe any approaches to engage patients or service recipients, the general public, communities, or stakeholders (such as clinicians or payers) in the design of the study.

Study parameters
22 Report all analytic inputs (such as values, ranges, references) including uncertainty or distributional assumptions. Table 1 Summary of main results 23 Report the mean values for the main categories of costs and outcomes of interest and summarise them in the most appropriate overall measure. Table 2 Effect of uncertainty 24 Describe how uncertainty about analytic judgments, inputs, or projections affect findings. Report the effect of choice of discount rate and time horizon, if applicable.