A Systematic Review of Cost-Effectiveness Studies of Newer Non-Insulin Antidiabetic Drugs: Trends in Decision-Analytical Models for Modelling of Type 2 Diabetes Mellitus

Abstract

Background

We performed a systematic overview of the cost-effectiveness analyses (CEAs) comparing Non-insulin antidiabetic drugs (NIADs) with other NIADs for the treatment of type 2 diabetes mellitus (T2DM), using decision-analytical modelling (DAM), focusing on both the economic results and the underlying methodological choices.

Methods

Eligible studies were CEAs using DAM to compare NIADs within the glucagon-like peptide-1 (GLP1) receptor agonists, sodium-glucose cotransporter-2 (SGLT2) inhibitors, or dipeptidyl peptidase-4 (DPP4) inhibitor classes with other NIADs within those classes for the treatment of T2DM. The PubMed, Embase and Econlit databases were searched from 1 January 2018 to 15 November 2022. Two reviewers screened the studies for relevance by titles and abstracts and then for eligibility via full-text screening, extracted the data from the full texts and appendices, and then stored the data in a spreadsheet.

Results

The search yielded 890 records and 50 studies were eligible for inclusion. The studies were mainly based on a European setting (60%). Industry sponsorship was found in 82% of studies. The CORE diabetes model was used in 48% of the studies. GLP1 and SGLT2 products were the main comparators in 31 and 16 studies, respectively, while one study had DPP4 and two had no easily discernible main comparator. Direct comparison between SGLT2 and GLP1 occurred in 19 studies. At a class level, SGLT2 dominated GLP1 in six studies and was cost effective against GLP1 once as part of a treatment pathway. GLP1 was cost effective in nine studies and not cost effective against SGLT2 in three studies. At a product level, oral and injectable semaglutide, and empagliflozin, were cost effective against other within-class products. Injectable and oral semaglutide were more frequently found cost effective in these comparisons, with some conflicting results. Most of the modelled cohorts and treatment effects were sourced from randomised controlled trials. The following model assumptions varied depending on the class of the main comparator: choice of and reasoning behind risk equations, the time until the treatment switch, and how often the comparators were discontinued. Diabetes-related complications were emphasised on par with quality-adjusted life-years as model outputs. The main quality issues were regarding the description of alternatives, the perspective of analysis, the measurement of costs and consequences, and patient subgroups.

Conclusion

The included CEAs using DAMs have limitations that hinder their ability to inform decision makers on the cost-effective choice: lack of updated reasoning behind the choice of key model assumptions, over-reliance on risk equations based on older treatment practices, and sponsorship bias. The question of which NIAD is cost effective for the treatment of which T2DM patient is a pressing one and the answer remains unclear.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40273-023-01268-5.

Key Points for Decision Makers

Compared with their older counterparts, newer Non-insulin antidiabetic drugs are cost effective for treating type 2 diabetes mellitus and show great promise in treating diabetes and its complications through indirect and direct effects. An overview of the recent findings in the cost-effectiveness literature and the underlying methodological choices in the decision-analytical models could aid decision makers in prioritisation.

This review found that as comparators, glucagon-like peptide-1 receptor agonists and sodium-glucose cotransporter-2 inhibitors were the favoured classes. Within these classes, semaglutide (injectable or oral) and empagliflozin were the favoured products, respectively. It is challenging to provide conclusions on the cost-effective option among these products due to different underlying methodological choices, sponsorship bias, and outdated information populating the model.

Decision makers face several difficulties when prioritising between the newer Non-insulin antidiabetic drug. The field of cost-effectiveness analyses in type 2 diabetes mellitus could benefit from using modelling practices, mainly treatment switch assumptions and risk equations, that better align with real-world practice and contemporary follow-up data for modelling treatment effects over time.

Introduction

Type 2 diabetes mellitus (T2DM) is a chronic progressive condition that poses a growing public health concern worldwide [1]. The estimated global total diabetes-related health expenditure for adults with diabetes will reach US dollars (US$) 1.03 trillion in 2030 and US$1.05 trillion in 2045 [1]. A UK model estimates that around 90% of the total healthcare costs for treating diabetes can be attributed to T2DM [2]. Furthermore, the costliest aspect of diabetes is the diabetes-related complications (DRCs), which carry great direct and indirect economic costs and a massive impact on health-related quality of life [3–5].

Several Non-insulin antidiabetic drugs (NIADs) are used for the pharmacological treatment of T2DM, where the management of hyperglycaemia is central. The main NIAD classes are commonly divided into older NIADs (oNIADs; e.g. metformin [MET], sulphonylurea [SU], and thiazolidinediones), and newer NIADs (nNIADs; e.g. dipeptidyl peptidase-4 [DPP4] inhibitors, glucagon-like peptide-1 [GLP1] receptor agonists, sodium-glucose cotransporter-2 [SGLT2] inhibitors) [6].

In 2018, a paradigm shift occurred where treatment guidelines in the consensus reports from the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD) [7–16] increasingly emphasise using products from the GLP1 and SGLT2 classes of pharmaceuticals. This emphasis was based on cardiovascular outcome trials, which were shown to have a protective effect against cardiorenal DRCs while treating hyperglycaemia and promoting weight loss [10]. Later updates have led to the recommendation of even earlier use of GLP1 or SGLT2 products, independent of HbA1c or first-line medication, if the patient is at high risk for atherosclerotic cardiovascular disease (ASCVD), chronic kidney disease (CKD), or heart failure (HF) [14].

Prevention or reduction of the occurrence of DRCs could lead to substantial improvements in the quality of life for patients with T2DM and a reduction in the increasing economic burden of the T2DM pandemic [3]. It thus becomes increasingly important to evaluate the value for money of the nNIADs [17]. One of the most commonly used methods for evaluating the cost effectiveness of diabetes treatment is the decision analytical model (DAM). Many factors complicate predicting how diabetes progresses over time, and DAMs can incorporate multiple sources of evidence to estimate how interventions differ over a long time horizon. Using many sources also allows DAMs to compensate for the short durations of clinical trials [18].

The ADA issued guidelines for diabetes modelling in 2004, highlighting seven different factors that complicate diabetes modelling: the long time horizon of disease progression, the involvement of multiple organ systems, the use of several types of medications that affect different outcomes, the vast array of complications that differ in terms of costs and how they affect quality of life, and the difficulties related to diagnosis [19]. These guidelines suggest that confidence in the models and their reliability can increase if transparency, validation, and the inclusion of different types of uncertainty are considered when constructing them. Adherence to these guidelines was investigated in a review from 2015, where it was found to be lacking but improving [20]. Methodological aspects of the models have also been reviewed [21–23], and Asche et al. have commented that clinically inconsequential changes in clinical parameters are given too much weight in DAMs and that time horizons in the models are too long [24].

One significant barrier to using nNIADs more than oNIADs is that the price has consistently been much higher [6, 10]. Despite the acquisition cost being much higher [11, 13–15], the literature on cost effectiveness has favoured the nNIADs [25–31] for second-line treatment. However, it remains unclear which of the nNIADs is the cost-effective choice, for which patients, and under which model assumptions [25]. An overview of the underlying methodological choices in the DAMs used to compare these nNIADs can provide decision makers with a better understanding of the basis of the economic results on which they base their decisions on choosing the appropriate nNIAD. We therefore aimed to provide an overview of the economic outcomes and differences in methodological choices, trends, and model assumptions by conducting a systematic review of the literature on cost-effectiveness analyses (CEAs) using DAM to compare nNIADs against other nNIADs, for the treatment of patients with T2DM.

Methods

This systematic review was conducted in 2021, updated in 2022, and adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [32].

Eligibility Criteria

Types of Studies

Studies were eligible to be included if they were CEA or cost-utility analyses that used any DAM to compare one or several nNIADs as interventions for the treatment of T2DM. The studies must have compared at least one nNIAD with another, either comparing different classes or products or the same classes and products within those classes (e.g., GLP1 class vs. SGLT2 class, or GLP1 product vs. SGLT2 product, such as semaglutide vs. empagliflozin).

Editorials, comments, conference abstracts, protocols, reviews, and studies that were not original research articles published in English were excluded.

Types of Intervention

The primary aim of the interventions in the CEAs must be managing the hyperglycaemic aspect of T2DM. The therapy in the intervention can be nNIADs monotherapy, combination therapy of nNIADs with oNIADs, nNIADs, or insulin (INS).

Types of Comparators

Eligible comparators include nNIADs of the SGLT2, GLP1, or DPP4 classes on the product level or aggregated class-level representations based on the pooling of data from the treatment effect of several products.

Information Sources and Literature Search Strategy

Data were collected by searching the PubMed, EMBASE, and EconLit databases. The initial search period was from 1 January 2018 to 8 October 2020. The search was updated twice and the final search date was 15 November 2022. The search was limited to start from 2018 to include studies from the year the 2018 ADA/EASD consensus report was published [10]. The literature search was first performed in PubMed using a combination of Medical Subject Heading (MeSH) terms and free-text keywords, and subsequently adapted to EMBASE and EconLit. In addition, reference lists in eligible studies were scanned to identify additional relevant articles. The entire search strategy is available in Online Resource Appendix A1.

Selection Process

Screening Process

First, titles and abstracts were screened for relevance by two authors independently (HVBL and EPJ), erring to the side of inclusion and blinding the decision process using the browser application Rayyan [33]. Second, HVBL and EPJ accessed the full text and reviewed the studies in-depth for final inclusion based on the predefined eligibility criteria.

Data Collected and the Collection Process

A spreadsheet for collecting data from the studies was developed and tested on all studies to ensure the validity of the extraction sheet. After an iterative process, the testing resulted in a standardised sheet for data collection. Data were extracted under three main categories: (1) basic study characteristics, including country of study, funding source, analysis perspective, time horizon, model type and name, subgroups analysed, and background medication and combination therapy of the modelled cohort; (2) main outcomes of the economic evaluation and sensitivity analysis (SA), such as cost, effect measure, incremental cost-effectiveness ratio (ICER) and currency, the key drivers of results, and types and results of SA; and (3) modelling parameters and assumptions, including the source of baseline characteristics of the modelled cohort, patient subgroup, treatment effect, adverse events (AEs), DRCs, risk equations, treatment switch, and treatment switch type. The term ‘main comparator’ was used when a comparator in the study was easily discernible as the one against which all other drugs were compared.

Quality Assessment of the Included Studies

The overall quality of the included studies was assessed using Drummond’s 10-point checklist for assessing economic evaluations [17], which consists of 10 essential questions with supporting sub-questions. HVBL assessed all studies and EPJ assessed a random sample of 10 studies. Any disagreement was resolved through consensus, while major disagreement resulted in EPJ assessing more studies until consensus was achieved. The answers to the 10 main questions are presented in Table 4, along with a summary of the main results. Each question was answered using one of the following four responses: ‘Yes’ (adequate), ‘Partial yes’ (partially adequate), ‘Unclear’ (cannot tell), ‘No’ (not adequate), and ‘Not applicable’. For the checklist questions that were subjective in nature, the ADA treatment guidelines [10, 11, 13–15] were used as a reference regarding the adequacy of the answers.

Table 4.

Quality assessment via Drummonds 10-point checklist

Study, year	Research aim	Alternatives described	Effectiveness established	Identification of costs and consequences	Measurement of costs and consequences	Valuation of costs and consequences	Extrapolation and discounting	Incremental analysis	Sensitivity analysis presentation	Discussion of study results
Barnett, 2018 [37]	Yes	Partial yes	Yes	Partial yes	Unclear	Unclear	Yes	Yes	Partial yes	Partial yes
Basson, 2018 [36]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Ericsson, 2018 [35]	Yes	Partial yes	Yes	Yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Tzanetakos, 2018 [34]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Ericsson, 2019 [44]	Yes	Partial yes	Yes	Yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Gæde, 2019 [52]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Hunt, 2019 [43]	Yes	Partial yes	Yes	Yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Johansen, 2019 [68]	Yes	Partial yes	Yes	Yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Malkin, 2019a [41]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Yes	Partial yes
Malkin, 2019b[42]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Yes	Partial yes
Raya, 2019 [39]	Yes	Partial yes	Partial yes	Partial yes	Unclear	Yes	Yes	Yes	Yes	Partial yes
Viljoen, 2019 [38]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Yes	Partial yes
Bain, 2020 [51]	Yes	Yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Capel, 2020 [50]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Gorgojo-MartíNez, 2020 [49]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Johansen, 2020 [48]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Martín, 2020 [47]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Capehorn, 2021 [57]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Guzauskas, 2021 [64]	Yes	Yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Malkin, 2021 [53]	Yes	Yes	Yes	Partial yes	Unclear	Partial yes	Yes	Yes	Yes	Yes
Risebrough, 2021 [70]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Ehlers, 2022a [58]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Ehlers, 2022b [59]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Ekhlasi, 2022 [83]	Partial yes	Partial yes	Yes	Partial yes	Partial yes	Partial yes	Yes	Yes	Partial yes	Yes
Eliasson, 2022 [60]	Yes	Yes	Yes	Yes	Unclear	Yes	Yes	Yes	Partial yes	Unclear
Franch-Nadal, 2022 [63]	Yes	Yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Hu, 2022 [79]	Yes	Partial yes	Yes	Partial yes	Unclear	Unclear	Yes	Yes	Partial yes	Partial yes
Malkin, 2022 [61]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Ruan, 2022 [81]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Stafford, 2022 [73]	Yes	Partial yes	Yes	Yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Viljoen, 2022 [62]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Chien, 2020 [75]	Yes	Yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Yes	Partial yes
Hu, 2021 [76]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Lin, 2021 [78]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Zupa, 2021 [71]	Yes	Yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Yes	Partial yes
Chakravarty, 2018 [65]	Yes	Yes	Yes	No	No	Partial yes	NA	Yes	Partial yes	Partial yes
Neslusan, 2018 [66]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Hou, 2019 [74]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Ramos, 2019 [40]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Ramos, 2020a [45]	Yes	Yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	No	Partial yes
Ramos, 2020b [46]	Yes	Yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Reifsnider, 2020 [69]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Yes	Partial yes
Van der Linden, 2020 [54]	Yes	Partial yes	Yes	Yes	Unclear	Yes	Yes	Yes	Partial yes	Yes
Ehlers, 2021 [56]	Yes	Yes	Yes	Partial yes	Partial yes	Yes	Yes	Yes	Partial yes	Yes
Gourzoulidis, 2021 [55]	Yes	Partial yes	Yes	Partial yes	Partial yes	Yes	Yes	Yes	Partial yes	Yes
Ramos, 2021 [77]	Yes	Yes	Yes	Partial yes	Partial yes	Yes	No	Yes	Partial yes	Partial yes
Reifsnider, 2021 [67]	Yes	Yes	Yes	Partial yes	Partial yes	Yes	Yes	Yes	Partial yes	Yes
Lasalvia, 2022 [82]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Partial yes
Peng, 2022 [80]	Yes	Partial yes	Yes	Unclear	Unclear	Yes	Yes	Yes	Yes	Partial yes
Reifsnider, 2022 [72]	Yes	Partial yes	Yes	Partial yes	Unclear	Yes	Yes	Yes	Partial yes	Yes

NA not applicable, Partial yes only a partial answer to the question

Results

Identified Studies

From the initial search, 890 unique studies were identified, of which 807 were removed after title and abstract screening, leaving 83 full-text articles. After reviewing the full texts, 50 studies met the inclusion criteria and were included for data extraction (see Fig. 1).

Fig. 1.

Flow-chart for study selection, adapted from the PRISMA-guidelines for the reporting systematic reviews [32]. DAM decision-analytical model, nNIAD Non-insulin antidiabetic drugs

Basic Characteristics

The results of extracted data from included studies can be found in Table 1, which summarises the general characteristics of the studies. The majority of studies (n = 30) were conducted in Europe [34–63], of which many were conducted in the UK (n = 9) [37, 38, 40, 45, 46, 48, 51, 57, 62], and one-third of the studies were conducted in North America (n = 10) [64–73] and Asia (n = 8) [74–81]. Two studies were conducted outside these regions: one each from Colombia [82] and Iran [83].

Table 1.

General characteristics of the studies

Study, year	Country	Comparison	Funding	Perspective	Time horizon, years	Model type	Model name	Subgroups analysed	Background medication	Combination therapy
Barnett, 2018 [37]	UK	Switch from sitagliptin 100 mg 1d to liraglutide 1.8 mg 1w within the first year of analysis vs. no switch	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 8.5	None stated	Dual therapy	MET
Basson, 2018 [36]	France	Dulaglutide 1.5 mg 1w vs. exenatide QW ?? mg 1w	Eli Lilly	Statutory health insurance	40	Markov with Monte Carlo simulation	QuintilesIMS CORE Diabetes Model	None stated	Triple therapy or above	MET+SU
Ericsson, 2018 [35]	Sweden	Liraglutide 1.8 mg 1d + basal vs. lixisenatide 20 mg 1d + basal and IDegLira vs. lixisenatide 20 mg + basal	Novo Nordisk	Societal	40	Markov chains	IHECM-T2D	None stated	Dual therapy	MET
Tzanetakos, 2018 [34]	Greece	Exenatide QW2 mg 1w vs. INS glargine, liraglutide 1.2 mg 1d	AstraZeneca	Third-party payer	40	Discrete event stochastic simulation	Cardiff	None stated	Dual and triple therapy	MET, MET+SU
Ericsson, 2019 [44]	Sweden	Semaglutide-I 1 mg 1w vs. dulaglutide 1.5 mg 1w, Lixisenatide ?? mg 1d	Novo Nordisk	Societal	40	Markov cohort with risk equations	IHE-DCM	None stated	Dual and triple therapy	MET, MET+INS
Gæde, 2019 [52]	Denmark	Semaglutide-I 0.5 and 1 mg 1w vs. dulaglutide 1.5 mg 1w and semaglutide-I 0.5 and 1 mg 1w vs. exenatide QW ?? mg 1w, liraglutide 1.2 and 1.8 mg 1d, Lixisenatide ?? mg 1d	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual therapy	MET
Hunt, 2019 [43]	Netherlands	Semaglutide-I 0.5 and 1 mg 1w vs. INS glargine U100 1d and semaglutide-I 0.5 and 1 mg 1w vs. dulaglutide 0.75 and 1.5 mg 1w	Novo Nordisk	Societal	Lifetime	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	A subgroup analysis on BMI 30+, 35+	Dual and triple therapy	MET, MET+SU
Johansen, 2019 [68]	Canada	Semaglutide-I 0.5 and 1 mg 1w vs. dulaglutide 0.75 and 1.5 mg 1w	Novo Nordisk	Societal	40	Markov cohort with risk equations	IHECM-T2D	None stated	Dual and triple therapy	MET
Malkin, 2019a [41]	Estonia	Semaglutide-I 1 mg 1w vs. liraglutide 1.2 mg 1d	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	Whole analysis on BMI >35	Dual and triple therapy	MET, SU, TZD
Malkin, 2019b [42]	Slovakia	Semaglutide-I 0.5 and 1 mg 1w vs. dulaglutide 1.5 mg 1w	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	Whole analysis on BMI >35	Dual therapy	MET
Raya, 2019 [39]	Spain	IDegLira vs. GLP1, INS + GLP1, basal INS, MDI	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 8.5	Four subgroups: MDI, GLP1+INS, basal INS, GLP1	Dual and triple therapy	MET, GLP1, INS
Viljoen, 2019 [38]	UK	Semaglutide-I 0.5 and 1 mg 1w vs. dulaglutide 1.5 mg 1w	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	THIN cohort used in SA	Dual therapy	MET
Bain, 2020 [51]	UK	Semaglutide-O 14 mg 1d vs. empagliflozin 25 mg 1d, liraglutide 1.8 mg 1d, sitagliptin 100 mg 1d	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual and triple therapy	MET, SU, SGLT2
Capel, 2020 [50]	Spain	Exenatide QW 2 mg 1w vs. dulaglutide 1.5 mg 1w, liraglutide 1.2 and 1.8 mg 1d, lixisenatide 20 mg 1d	AstraZeneca	Healthcare payer	40	Discrete event stochastic simulation	Cardiff	None stated	Dual therapy	MET
Gorgojo-MartíNez, 2020 [49]	Spain	Semaglutide-I 0.5 and 1 mg 1w vs. empagliflozin 10 and 25 mg	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual and triple therapy	MET, TZD, MET+TZD
Johansen, 2020 [48]	UK	Semaglutide-I 1 mg 1w vs. liraglutide 1.2 mg 1d	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual and triple therapy	MET, SU, S, COMBOS
Martín, 2020 [47]	Spain	Semaglutide-I 0.5 and 1 mg 1w vs. dulaglutide 1.5 mg 1w, sitagliptin 100 mg 1d	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual and triple therapy	Vs. dulaglutide: MET vs. sitagliptin: MET, SU, TZD
Capehorn, 2021 [57]	UK	Semaglutide-I 1 mg 1w vs. empagliflozin 25 mg 1d	Novo Nordisk	Healthcare payer	Lifetime	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual therapy	MET
Guzauskas, 2021 [64]	USA	Semaglutide-O 14 mg 1d vs. empagliflozin 10 and 25 mg 1d, liraglutide 1.8 mg 1d, sitagliptin 100 mg 1d, background (MET + SU)	Not industry	Healthcare payer	Lifetime	Individual patient-level Monte Carlo microsimulation	UKPDS OM2	None stated	Dual and triple therapy	MET+SU
Malkin, 2021 [53]	Netherlands	Semaglutide-O 14 mg 1d vs. empagliflozin 25 mg 1d, sitagliptin 100 mg 1d, liraglutide 1.8 mg 1d	Novo Nordisk	Societal	Lifetime	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	A subgroup analysis on BMI 30+	Dual and triple therapy	MET, SU, SGLT2
Risebrough, 2021 [70]	USA	Semaglutide-O 14 mg 1d vs. dulaglutide 1.5 mg 1w, liraglutide 1.8 mg 1d, Semaglutide-I 1 mg 1w	Novo Nordisk	Payer perspective	Lifetime	State transition cohort model with competing risk approach	Not stated	None stated	Dual and triple therapy	1–2 OADs
Ehlers, 2022a [58]	Denmark	Semaglutide-O ?? mg 1d vs. empagliflozin ?? mg 1d	Boehringer Ingelheim	Health sector	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9.5	None stated	Dual therapy	MET
Ehlers, 2022b [59]	Denmark	Semaglutide-I 1 mg 1w vs. empagliflozin 25 mg 1d	Boehringer Ingelheim	Payers’ perspective	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9.5	None stated	Triple therapy or above	MET
Ekhlasi, 2022 [83]	Iran	Dulaglutide 1.5 mg 1w vs. liraglutide 1.8 mg 1d	Not industry	Health system	10	Markov state transition model	Not stated	None stated	Triple therapy or above	≥2 OADs
Eliasson, 2022 [60]	Sweden	Semaglutide-O 14 mg 1d vs. empagliflozin 25 mg 1d, sitagliptin 100 mg 1d	Novo Nordisk	Societal and payers	40	Markov state transition model	IHE-DCM	None stated	Dual and triple therapy	MET, MET+SU
Franch-Nadal, 2022 [63]	Spain	Semaglutide-O 14 or 7 mg 1d vs. empagliflozin 25 mg 1d, sitagliptin 100 mg 1d, liraglutide 1.8 mg 1w	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9.0	None stated	Dual and triple therapy	MET, SU, SGLT2
Hu, 2022 [79]	China	Semaglutide-I 1 mg 1w vs. dulaglutide 1.5 mg 1w	Not industry	Healthcare providers	40	Individual patient-level Monte Carlo microsimulation	UKPDS OM2	None stated	Dual therapy	MET
Malkin, 2022 [61]	Portugal	Semaglutide-O 14 mg 1d vs. empagliflozin 25 mg 1d, dulaglutide 1.5 mg 1w	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9.0	None stated	Dual and triple therapy	1–2 OADs
Ruan, 2022 [81]	China	Semaglutide-I 0.5 or 1 mg 1w vs. dulaglutide 1.5 mg 1w	Novo Nordisk	Healthcare system	40	Markov state transition model	IHE-DCM	None stated	Dual therapy	MET
Stafford, 2022 [73]	Canada	Semaglutide-I 1 mg 1w vs. canagliflozin 300 mg 1d	Novo Nordisk	Healthcare payer and societal	40	Markov state transition model and individual patient simulation	IHE-DCM and ECHO-T2DM	None stated	Dual therapy	MET
Viljoen, 2022 [62]	UK	Semaglutide-I 1 mg vs. dulaglutide 3 mg 1w, dulaglutide 4.5 mg 1w	Novo Nordisk	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9.0	None stated	Dual therapy	MET
Chien, 2020 [75]	Taiwan	Comparison of 8 different strategies with 5 second-line classes (SGLT2, DPP4, GLP1, SU, ins)	Not industry	National health insurance	40	Patient-level fixed-time increment, Monte Carlo microsimulation model	Cardiff	Scenario analysis had Taiwanese T2DM individuals	Triple therapy or above	MET
Hu, 2021 [76]	China	Dapagliflozin 10 mg 1d + saxagliptin 5 mg 1d vs. dapagliflozin 10 mg 1d vs. saxagliptin 5 mg 1d	Not industry	Healthcare service providers	50	Individual patient-level Monte Carlo microsimulation	UKPDS OM2	None stated	Triple therapy or above	MET
Lin, 2021 [78]	China	Five different strategies for DPP4 inhibitors were compared: linagliptin 5 mg, saxagliptin 5 mg, alogliptin 25 mg, sitagliptin 100 mg, and vildagliptin 50 mg	Not industry	Healthcare service providers	Lifetime	Risk equation model/submodels/unclear terms	COMT	None stated	Dual therapy	MET
Zupa, 2021 [71]	USA	Empagliflozin 25 mg 1d vs. semaglutide-I 1 mg 1w	Not industry	Healthcare system	3	Markov state transition model	Not stated	Alternate case where patients started with complications consistent with EMPA-REG OUTCOME and SUSTAIN 6 populations	Dual and triple therapy	SoC (none stated)
Chakravarty, 2018 [65]	USA	Dapagliflozin vs. liraglutide, SU, DPP4, pioglitazone	AstraZeneca	Third-party payer	1	Decision tree	Not stated	None stated	Dual therapy	MET
Neslusan, 2018 [66]	USA	Canagliflozin 300 mg 1d vs. dapagliflozin 10 mg 1d	Janssen Scientific Affairs	Third-party payer	30	Markov with microsimulation at the patient level	ECHO-T2DM	None stated	Dual therapy	MET
Hou, 2019 [74]	China	Canagliflozin 100 mg 1d vs. dapagliflozin 10 mg 1d	Not industry	Healthcare service providers	Lifetime	Risk equation model/submodels/unclear terms	COMT	None stated	Dual therapy	MET
Ramos, 2019 [40]	UK	Empagliflozin ?? mg 1d + SoC vs. saxagliptin ?? mg 1d + SoC, sitagliptin ?? mg 1d + SoC	Boehringer Ingelheim	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual and triple therapy	MET, SU, GLP1, TZD, MEG, INS
Ramos, 2020a [45]	UK	Empagliflozin ?? mg 1d + SoC vs. SoC, liraglutide ?? mg 1d + SoC	Boehringer Ingelheim	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual and triple therapy	MET, SU, TZD, MEG, INS added as per SoC
Ramos, 2020b [46]	UK	Empagliflozin 25 mg 1d vs. semaglutide-O 14 mg 1d	Boehringer Ingelheim	Healthcare payer	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual therapy	MET
Reifsnider, 2020 [69]	USA	Comparison of two different second-line strategies, empagliflozin or sitagliptin	Boehringer Ingelheim	Healthcare payer	Lifetime	Individual patient-level Monte Carlo microsimulation	DICE platform	No subgroup, but analysis divided into T2DM individuals with or without CVD	Triple therapy or above	MET
Van der Linden, 2020 [54]	Netherlands	Dapagliflozin ?? mg 1d vs. DPP4 (represented by sitagliptin mainly)	AstraZeneca	Societal	40	Fixed-time increment stochastic simulation model	Cardiff	None stated	Triple therapy or above	MET+SU
Ehlers, 2021 [56]	Denmark	Empagliflozin ?? mg 1d vs. liraglutide 1.8 mg 1d	Boehringer Ingelheim	Healthcare sector	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual therapy	MET
Gourzoulidis, 2021 [55]	Greece	Empagliflozin ?? mg 1d vs. dapagliflozin ?? mg 1d	Boehringer Ingelheim	Public payer	Lifetime	Individual patient-level discrete-event simulation	Not stated	None stated	Dual and triple therapy	SoC (MET, SU, INS)
Ramos, 2021 [77]	China	Empagliflozin ?? 1d vs. liraglutide ?? 1w, sitagliptin ?? mg1d	Boehringer Ingelheim	Healthcare payer providers	50	Markov with Monte Carlo simulation	IQVIA CORE Diabetes Model v. 9	None stated	Dual and triple therapy	SoC (MET, SU, INS)
Reifsnider, 2021 [67]	USA	Empagliflozin ?? 1d vs. dapagliflozin ?? mg 1d, canagliflozin ?? mg 1d	Boehringer Ingelheim	Third-party payer	Lifetime	Individual patient-level discrete-event simulation	Not stated	None stated	Dual and triple therapy	SoC (MET, SU, INS)
Lasalvia, 2022 [82]	Colombia	Dapagliflozin ?? mg 1d vs. DPP4	AstraZeneca	Unclear, third-party payer or healthcare system	5	Patient-level fixed-time increment, Monte Carlo microsimulation model	Cardiff	None stated	Dual therapy	MET
Peng, 2022 [80]	Taiwan	SGLT2 vs. DPP4	Not industry	Healthcare sector	10	Markov state transition model	Not stated	With and without CVD history	Unclear	Unclear
Reifsnider, 2022 [72]	USA	Empagliflozin ?? mg 1d vs. liraglutide ?? mg 1d	Boehringer Ingelheim	Payers’ perspective	Lifetime	Individual patient simulation	Not stated	With and without CVD history	Triple therapy or above	MET

When no dosage or ‘??’ is written, the dosage was unclear. Comparisons list the main comparators and then the alternatives separated by a comma, while ‘and’ is written to show that several different comparisons were made. When one medication is combined with another in the comparison, it is explicitly stated or symbolised with a ‘+’ symbol

1d once per day, 1w once per week, BMI body mass index, CVD cardiovascular disease, COMT Chinese Outcomes Model for T2DM, CORE Centre for Outcomes Research, Cardiff Cardiff Diabetes Model, DICE discretely integrated condition event, DPP4 dipeptidylpeptidase-4 inhibitors, ECHO-T2DM Health Outcomes Model of Type 2 Diabetes Mellitus, GLP1 glucagon-like peptide-1 receptor agonist, IDegLira combination of liraglutide and INS degludec, IHECM-T2D Swedish Institute for Health Economics Cohort Model for T2DM, INS insulin, MDI multiple daily injections, MEG meglitinide, MET metformin, SA sensitivity analysis, SGLT2 sodium-glucose cotransporter-2 inhibitors, SU sulfonylurea, semaglutide-I or O semaglutide injection or oral, SoC standard of care, T2DM type 2 diabetes mellitus, TZD thiazolidinedione, UKPDS UK Prospective Diabetes Study, UKPDS OM UK Prospective Diabetes Study outcomes model

Only 9 of 50 studies were not sponsored by the industry [64, 71, 74–76, 78–80, 83]. Novo Nordisk funded the most studies (n = 23) [35, 37–39, 41–44, 47–49, 51–53, 57, 60–63, 68, 70, 73, 81], followed by Boehringer Ingelheim (n = 11) [40, 45, 46, 55, 56, 58, 59, 67, 69, 72, 77]. The remaining five studies were funded by AstraZeneca [34, 50, 54, 65, 82] and one each by Eli Lilly [36] and Janssen Scientific Affairs [66].

The vast majority of modelling studies (n = 24) [36–43, 45–49, 51–53, 56–59, 61–63, 77] used IQVIA's CORE Diabetes Model (CDM) [84, 85], followed by the Cardiff Diabetes Model [86] (n = 5) [34, 50, 54, 75, 82], the Swedish Institute for Health Economics Cohort Model for T2DM (IHE-DCM) [87] (n = 5) [35, 44, 60, 68, 81], the UK Prospective Diabetes Study (UKPDS) Outcomes Model 2 (UKPDS-OM2) [88] (n = 3) [64, 76, 79], the Chinese Outcomes Model for T2DM (COMT) [89] (n = 2) [74, 78], and one study each with the Economic and Health Outcomes Model of T2DM (ECHO-T2DM) [90] and the Discretely Integrated Condition Event platform [91]. One study used both the IHE-DCM and the ECHO-T2DM to produce and compare their results [73]. Eight studies did not state the model name, of which three studies [71, 80, 83] used Markov state transition models, three used individual patient simulations [55, 67, 72], and one each used a state transition cohort model with a competing risk approach [70] and a decision tree [65].

The majority of studies conducted analyses using the payer’s perspective (n = 31) [34, 36–42, 45–52, 55, 57, 59, 61–67, 69, 70, 72, 75, 77]. Less frequently, a healthcare sector perspective (n = 10) [56, 58, 59, 71, 74, 76, 78–81] and a societal perspective were used in six studies [35, 43, 44, 53, 54, 68]. Two studies used both a payer’s perspective and a societal perspective [60, 73], while the perspective employed by one study was unclear [82].

Few studies (n = 3) had a time horizon of 5 years or lower [65, 71, 82]. The majority had a time horizon of either 40 years (n = 12) [34–36, 44, 50, 54, 60, 68, 73, 75, 79, 81] or 50 years (n = 21) [37–42, 45–49, 51, 52, 56, 58, 59, 61–63, 76, 77], while some studies (n = 11) reported a lifetime horizon [43, 53, 55, 57, 64, 67, 69, 70, 72, 74, 78]. The remainder had a time horizon of 30 years [66] or 10 years [80, 83].

The most frequently used background medication for the modelled cohorts at baseline was MET (n = 25) [35, 37, 38, 42, 46, 50, 52, 56–59, 62, 65, 66, 68, 69, 72–76, 78, 79, 81, 82], while most of the remaining studies (n = 23) reported MET combined with a range of other classes of medications [34, 36, 39–41, 43–45, 47, 48, 51, 53–55, 60, 61, 63, 64, 67, 70, 71, 77, 80, 83]. Sometimes the background medication was described as ‘Standard of Care’ (SoC) or ‘1–2 oral antidiabetic drugs’. Two studies had unclear reporting on the background medication [71, 80]. With regard to combination therapy, most studies (n = 22) involved dual or triple therapy, while some studies (n = 19) had dual therapy and the remaining eight studies had triple therapy or above [36, 54, 59, 69, 72, 75, 76, 83]. In one study, the background medication was unclear [80].

Cost-Effectiveness Results and Uncertainty

This section first describes how the comparisons between the classes are presented and then the methodological choices regarding the CEAs. An overview can be found in Table 2.

Table 2.

Cost-effectiveness and uncertainty results

Study, year	Comparison	Cost, year	Effect measure	Which is cost effective	ICER (currency)	Threshold	Quadrant (location in source)	Key driver	Sensitivity analysis	First sensitive factor	Second sensitive factor	Third sensitive factor
Barnett, 2018 [37]	Liraglutide vs. sitagliptin	Direct, 2016	QALY, LY, incidence of DRC, onset of DRC	Liraglutide	15,423/QALY, (GBP)	GBP20,000/QALY	QNE (Fig. 2)	Lower rates and delayed onset of DRCs offset the higher treatment cost	DSA generally confirmed findings. Except lipid or hypoglycaemia difference only led to the same effect but more cost, and BMI or SBP difference only led to 150k+ and 279k+ ICER; PSA likewise	Lipid difference only	Hypoglycaemia difference only	BMI difference only
Basson, 2018 [36]	Dulaglutide vs. exenatide QW	Direct, 2014	QALY, LY, incidence of DRC, onset of DRC	Dulaglutide	Dominant (EUR)	EUR30,000/QALY	QSE (Fig. 2)	Lower rates and delayed onset of DRCs offset the higher treatment cost	DSA generally confirmed findings; PSA likewise	Variation in treatment efficacy	Time on treatment	Shorter time horizon (10 years)
Ericsson, 2018 [35]	Liraglutide + basal INS vs. lixisenatide + basal INS, IDegLira	Direct, 2016; indirect, 2015	QALY, LY, incidence of DRC	Liraglutide and IDegLira	Vs. lixisenatide: 30.802/QALY Scenarios: IDegLira vs. lixisenatide: 34.800/QALY | 23.984/QALY | dominant (SEK)	SEK100,000–1,000,000/QALY	QNE for liraglutide vs. lixisenatide (Fig. 1, Table 4); QNE for IDegLira vs. lixisenatide = QNE or QSE dependent on dosage (Table 5)	More complications avoided	DSA generally confirmed findings. Except liraglutide was not CE when HbA1c changes were assumed equal; PSA likewise	Hba1c reduction of liraglutide critical in achieving CE	Liraglutide 1.8 mg replaced with 1.2 mg led to liraglutide being dominant	Shorter time horizon (10 years)
Tzanetakos, 2018 [34]	Exenatide QW vs. INS glargine, liraglutide	Direct, 2016	QALY, LY, incidence of DRC	Exenatide QW	Vs. INS glargine: 4499/QALY vs. liraglutide: 2827/QALY (EUR)	EUR36,000/QALY	QNE for exenatide once-weekly vs. IG, QNE for exenatide once-weekly vs. liraglutide (Table 3)	Vs. INS glargine: lower cumulative incidence of DRCs (IHD, MI, hypoglycaemia) offset higher acquisition cost vs. liraglutide: lower cumulative incidence of DRCs (CHF, stroke) and fatality offset higher acquisition cost	DSA generally confirmed findings; PSA likewise	Lower utility to BMI weights	Switching HbA1c threshold	Hba1c effect of exenatide QW 1w set to upper limit
Ericsson, 2019 [44]	Semaglutide-I vs. dulaglutide, lixisenatide	Direct, 2018; indirect, 2016	QALY, LY, incidence of DRC	Semaglutide-I	Vs. dulaglutide: dominant, vs. lixisenatide: dominant (SEK)	SEK500,000/QALY	QSE for semaglutide-I vs. dulaglutide, QSE for semaglutide-I vs. lixisenatide (Fig. 1)	For both comparisons: longer time to onset of DRCs, reduction of DRC incidence (particularly retinopathy) led to reduced costs and higher quality of life with semaglutide-I	DSA confirmed findings. Except with intensification at lower HbA1c threshold for semaglutide-I vs. dulaglutide; PSA confirmed base case	No drift in last intensification	Intensification at HbA1c 7.5%	Shorter time horizon (20 years)
Gæde, 2019 [52]	Semaglutide-I vs. dulaglutide, exenatide QW, liraglutide, lixisenatide	Direct, 2017	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	Primary analysis: dominant Secondary analysis: semaglutide-I 0.5 mg dominant, semaglutide-I 1 mg dominant (DKK)	DKK250,000/QALY (based on UK threshold)	QSE for semaglutide-I 0.5 mg vs. dulaglutide, QSE for semaglutide-I 1 mg vs. dulaglutide (Table 2)	Delay to treatment intensification, lower rates, and delayed onset of DRCs	DSA generally confirmed findings. Including only statistically significant differences made semaglutide-I 0.5 mg non-CE. With different treatment intensification, semaglutide-I was non-dominant but still CE; PSA confirmed base case	Only statistically significant differences	Treatment switch at 3 or 5 years made semaglutide-I 1 and 0.5 mg not dominant	Shorter time horizon (10 years)
Hunt, 2019 [43]	Semaglutide-I vs. INS glargine, dulaglutide	Direct + indirect, 2017	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	Semaglutide-I 0.5 mg vs. dulaglutide – direct: 4671/QALY direct + indirect: dominant Semaglutide-I 1 mg vs. dulaglutide – direct: 2861/QALY direct + indirect: dominant Semaglutide-I 0.5 mg vs. INS glargine – direct: 11,310/QALY direct + indirect: 4988/QALY Semaglutide-I 1 mg vs. INS glargine – direct: 7515/QALY direct + indirect: 495/QALY (EUR)	EUR12,900/QALY	Direct costs: QNE for semaglutide-I 0.5 mg vs. glargine, QNE for semaglutide-I 1 mg vs. glargine, QNE for semaglutide-I 0.5 mg vs. dulaglutide, QNE for semaglutide-I 1 mg vs. dulaglutide with indirect costs: QNE, QNE, QSE, and QSE respectively (Table 4)	Semaglutide-I vs. INS glargine: reduction in cumulative incidence of DRCs and increased time to their onset. Largest cost saving from avoided CVD complications. This offset higher pharmacy costs of semaglutide-I. Combining direct + indirect reduced ICER further. Semaglutide-I vs. dulaglutide: same, but avoided ophthalmological complications most notable. Combining direct + indirect led to semaglutide-I dominating	DSA confirmed findings; PSA likewise	Hypoglycaemia difference only	SBP difference only	Lipid difference only
Johansen, 2019 [68]	Semaglutide-I vs. dulaglutide	Direct + indirect, 2017	QALY, LY, incidence of DRC	Semaglutide-I	Semaglutide-I 0.5 mg vs. dulaglutide 0.75 mg: dominant, semaglutide-I 1 mg vs. dulaglutide 1.5 mg: dominant (CAD)	CAD50,000/QALY	QSE for semaglutide-I 0.5 mg vs. dulaglutide 0.5 mg, QSE for semaglutide-I 1 mg vs. dulaglutide 1.5 mg (Fig. 3, Table 3)	Higher cost of semaglutide-I due to longer treatment period (than dulaglutide) before switch to INS, offset by reduced complications, and longer onset until them. Same concept led to higher QALY for semaglutide-I	DSA confirmed findings; PSA: low dose, 66% probability of being CE at 50k/QALY; high dose 73% for PSA scenario analyses: 98% CE for both dose comparisons	Scenario: HbA1c drift at 0.14%, discontinuing GLP1 at 8% HbA1c threshold, and replacing it with INS	Scenario: No INS effect, HbA1c remain at 8% when reached	Shorter time horizon (10 years)
Malkin, 2019 [41]	Semaglutide-I vs. liraglutide	Direct, drug, unknown year, DRC 2017	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	523/QALY (EUR)	EUR52,390/QALY (3 times Estonian GDP/capita)	QNE (Table 4)	Reduced incidence and delayed time to onset of DRC and avoidance of ulcer, amputation, neuropathy, hypoglycaemia, CVD	DSA confirmed findings. In some cases, semaglutide-I dominated; PSA likewise	Shorter time horizon (10 years)	Use of UKPDS 82	0% discount rate
Malkin, 2019 [42]	Semaglutide-I vs. dulaglutide	Direct, year unknown	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	Semaglutide-I 0.5 vs. dulaglutide 1.5: dominant, semaglutide-I 1 vs. dulaglutide 1.5: dominant (EUR)	EUR25,536/QALY (28 times average monthly wage in Slovakia)	QSE for semaglutide-I 0.5 mg vs. dulaglutide 1.5 mg, QSE for semaglutide-I 1 mg vs. dulaglutide 1.5 mg (Table 5)	Reduced incidence and delayed time to onset of DRC and avoidance of ulcer, amputation, neuropathy, hypoglycaemia, CVD	DSA confirmed findings. In DSA it is dominant all the time; PSA is 57–72% likely to be CE	Treatment switch at 7.5% HbA1c	Statistically significant differences only	0% discount rate
Raya, 2019 [39]	IDegLira vs. GLP1, INS mix + GLP1, basal INS, MDI	Direct, 2016	QALY, LY, incidence of DRC, onset of DRC	IDegLira	Vs. MDI: 3013/QALY, vs. basal: 6890/QALY, vs. GLP1: dominant, vs. GLP1 + INS: dominant (EUR)	EUR30,000/QALY	QNE for IDegLira vs. MDI, QNE for IDegLira vs. basal, QSE for IDegLira vs. INS+GLP1, QSE for IDegLira vs. GLP1 (Table 4, Fig. 2)	Vs. MDI: improved glycemic control led to fewer DRCs, higher cost offset by reduced cost of DRC vs. GLP1 + INS: cost saving because of fewer DRCs vs. basal: higher cost offset by reduced DRCs and delayed onset vs. GLP1: cost saving in general, fewer DRCs	DSA generally confirmed findings. Highly sensitive to (leads to non-CE) abolishment of HbA1c difference though (vs. MDI); PSA likewise	HbA1c difference abolishment vs. MDI OR basal INS	NPH INS cost applied	Statistically significant differences only
Viljoen, 2019 [38]	Semaglutide-I vs. dulaglutide	Direct, 2016	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	Semaglutide-I 0.5 vs. dulaglutide 1.5: dominant, semaglutide-I 1 vs. Dulaglutide 1.5: dominant (GBP)	GBP20,000/QALY	QSE for semaglutide-I 0.5 mg vs. dulaglutide 1.5 mg, QSE for semaglutide-I 1 mg vs. dulaglutide 1.5 mg (Tables 2 and 4, ESM Fig. 2)	Reduced incidence and delayed time to onset of DRC, higher cost due to increased survival offset by fewer DRCs	DSA generally confirmed findings; PSA likewise	Only including statistically significant differences between semaglutide-I 0.5 mg and dulaglutide 1.5 mg	Treatment switch at 7.5%	Dominant in all other scenarios
Bain 2020, [51]	Semaglutide-O vs. empagliflozin, sitagliptin, liraglutide	Direct (2019 drug; 2018 DRC)	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-O 14 mg	Vs. empagliflozin: 11,006/QALY vs. sitagliptin: 4930/QALY vs. liraglutide: dominant (GBP)	GBP20,000–30,000/QALY	QNE for semaglutide-O 14 mg vs. empagliflozin 25 mg, QNE for semaglutide-O 14 mg vs. sitaglipin100 mg, QSE for semaglutide-O 14 mg vs. liraglutide 1.8 mg (Table 3, Fig. 2)	Benefit from reduced incidence of DRC and longer mean time to onset of any DRC with semaglutide-O in all comparisons. Increased cost of semaglutide-O from higher acquisition price and longer time to treatment intensification	DSA confirmed findings. Dominance over liraglutide and CE against others unchanged; PSA confirmed base case	Shorter time horizon (10 years)	Treatment switch at 8.0% HbA1c	26-week treatment effects applied
Capel, 2020 [50]	Exenatide QW vs. dulaglutide, liraglutide, lixisenatide	Direct, 2018	QALY	Exenatide QW 2 mg 1w	Vs. dulaglutide: dominant, vs. liraglutide 1.2 mg: dominant, vs. liraglutide 1.8 mg: dominant, vs. lixisenatide: dominant (EUR)	EUR20,000/QALY gained	QSE for exenatide vs. dulaglutide, QSE for exenatide vs. liraglutide 1.2 mg, QSE for exenatide vs. liraglutide 1.8 mg, QSE for exenatide vs. lixisenatide, (Table 3, Fig. 1)	None stated	DSA completely confirmed findings in base case; PSA likewise	Dominant conclusion not changed	Dominant conclusion not changed	Dominant conclusion not changed
Gorgojo-MartíNez, 2020 [49]	Semaglutide-I vs. empagliflozin	Direct, 2018	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	Semaglutide-I 1 mg vs. empagliflozin 10 mg: 161/QALY Semaglutide-I 1 mg vs. empagliflozin 25 mg: 625/QALY (EUR)	EUR30,000/QALY	QNE for semaglutide-I 0.5 mg vs. empagliflozin 10 mg, QNE for semaglutide-I 1 mg vs. empagliflozin 25 mg (Table 3)	Lower rates and delayed onset of DRCs led to higher clinical benefit for semaglutide-I. Higher acquisition cost and longer survival led to higher cost for semaglutide-I. Greater Hba1c reduction biggest contributor to superiority over empagliflozin	DSA generally confirmed findings, except some changes made semaglutide-I CE instead of dominant; PSA likewise	Shorter time horizon (10 years)	Lower 95% CI of HbA1c treatment difference	5% discount rates
Johansen, 2020 [48]	Semaglutide-I vs. liraglutide	Direct, 2018	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	Semaglutide-I dominant (GBP)	GBP20,000/QALY	QSE (Table 2, Fig. 4)	Longer time to onset of DRCs, time to intensification, higher survival, greater ‘avoidance’ of DRCs	DSA completely confirmed findings, except under a statistically deterministic scenario, where it was still CE; PSA likewise	Shorter time horizon (10 years)	SGLT2 and SU discontinued at treatment intensification	Only statistically significant differences
Martín, 2020 [47]	Semaglutide-I vs. dulaglutide, sitagliptin	Direct (2019 drug; 2018 DRC)	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	0.5 mg semaglutide-I vs. dulaglutide: dominant, 1 mg semaglutide-I vs. dulaglutide: dominant, 0.5 mg semaglutide-I vs. sitagliptin: dominant, 1 mg semaglutide-I vs. sitagliptin: dominant (EUR)	EUR30,000/QALY	QSE for semaglutide-I 1 mg vs. dulaglutide, QSE for semaglutide-I 1 mg vs. sitagliptin, QSE for semaglutide-I 0.5 mg vs. dulaglutide, QSE for semaglutide-I 0.5 mg vs. sitagliptin (Table 2)	Longer time to onset of DRC, time to intensification, higher survival, greater ‘avoidance’ of DRCs	DSA completely confirmed findings; PSA likewise	Dominant conclusion not changed	Dominant conclusion not changed	Dominant conclusion not changed
Capehorn, 2021 [57]	Semaglutide-I vs. empagliflozin	Direct, 2019	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	4439/QALY (GBP)	GBP20,000/QALY	QNE (Table 2, Fig. 4)	Reduced incidence of DRCs, delayed onset of DRCs, delayed treatment intensification, driven by greater HbA1c reduction	DSA confirmed findings; PSA likewise	Fixed HbA1c over time and treatment intensification at 3 years	Shorter time horizon (10 years)	Different hypoglycaemia disutilities
Guzauskas, 2021 [64]	Semaglutide-O vs. empagliflozin, liraglutide, sitagliptin, MET + SU	Direct, 2020	QALY, LY, incidence of DRC	Semaglutide-O, except vs. empagliflozin	Vs. empagliflozin: 458.400/QALY vs. liraglutide: 40.100/QALY vs. sitagliptin: 145.200/QALY vs. background (MET+SU): 117.500/QALY (US$)	US$100,000–250,000/QALY	QNE for semaglutide-I vs. liraglutide, QNE for semaglutide-I vs. empagliflozin, QNE for semaglutide-I vs. sitagliptin, QNE for semaglutide-I vs. background medication (Table 2)	Semaglutide-O had fewer MACE and cardiovascular deaths	No mention of robustness to change; but costs/QALYs very sensitive to changes in treatment effect	Change in MACE reduction of semaglutide-O	Changes in Hba1c reduction of semaglutide-O	Change in heart failure and nephropathy HR of semaglutide-O
Malkin, 2021 [53]	Semaglutide-O vs. empagliflozin, sitagliptin, liraglutide	Direct + indirect, 2019	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-O	Semaglutide-O vs. empagliflozin: 13,770/QALY, w indirect = 7061/QALY Semaglutide-O vs. sitagliptin: 5938/QALY, w indirect = 516/QALY Semaglutide-O vs. liraglutide: dominate, w indirect cost = dominate (EUR)	EUR20,000/QALY	QNE for semaglutide-O vs. empagliflozin, QNE for semaglutide-O vs. sitagliptin, QSE for semaglutide-O vs. liraglutide. with indirect cost: QNE, QNE, and QSE, respectively (Table 3, Fig. 1)	Greater Hba1c reductions were the key driver of clinical benefits in all comparisons. BMI made smaller contributions. Other RF made small or no contributions	DSA generally confirmed findings; PSA showed 52.7, 70.8%, 68.3% chance of semaglutide-O being CE vs. empagliflozin, sitagliptin, liraglutide, respectively	Treatment switch at 8.0% HbA1c	Use of UKPDS 82	Shorter time horizon (30 years)
Risebrough, 2021 [70]	Semaglutide-O vs. dulaglutide, liraglutide, semaglutide-I	Direct, 2019	QALY, LY, incidence of DRC, event-free survival	Semaglutide-O	Vs. semaglutide-I: 163,737/QALY, vs. dulaglutide: dominate, vs. liraglutide: dominate (US$)	US$20,000/QALY	QNE for semaglutide-O vs. injectable semaglutide, QSE for semaglutide-O vs. dulaglutide, QSE for semaglutide-O vs. liraglutide (Table 5)	Small differences in AE estimates, HbA1c benefits and event-free survival led to cost savings vs. dulaglutide and liraglutide	DSA confirmed findings; PSA focuses on semaglutide-O vs. semaglutide-I and shows that with increasing WTP, semaglutide-I is more likely to be CE	Daily cost of semaglutide-O	Weight treatment effect of semaglutide-O	Daily cost of semaglutide-O
Ehlers, 2022 [58]	Semaglutide-O vs. empagliflozin	Direct, 2020	QALY, LY, incidence of DRC	Semaglutide-O not cost effective	Semaglutide-O vs. empagliflozin: 1,930,548/QAL (DKK)	DKK357,100/QALY (1 time GDP/capita)	QNE (Fig. 1)	The cost-effectiveness result was driven by a major difference in treatment costs, reflecting the large unit cost difference of Semaglutide-O vs. empagliflozin	DSA confirmed findings; PSA likewise, semaglutide-O was CE in 16% of simulations at set WTP	Discontinuation of semaglutide-O and empagliflozin, and switch to higher-dose, long-acting INS alogliptin in third-line	Shorter time horizon (5 years)	Using trial product estimand
Ehlers, 2022 [59]	Semaglutide-I vs. empagliflozin	Direct, 2020	QALY, LY	Semaglutide-I not cost effective	Semaglutide-I vs. empagliflozin: 745,561/QALY, (DKK and EUR)	DKK357,100/QALY (1 time GDP/capita)	QNE (Fig. 1)	None stated	DSA confirmed findings; PSA likewise	Third-line treatment assumption (comparators replaced with INS)	Third-line treatment assumption (comparators replaced with INS) + third-line occurring at 8% HbA1c instead of 7.5%	Shorter time horizon (5 years)
Ekhlasi, 2022 [83]	Dulaglutide vs. liraglutide	Direct, 2018	QALY	Dulaglutide	Dominant (US$)	US$3598.483/QALY (1 time GDP/capita in Iran 2018)	QSE (Fig. 2)	None stated	DSA confirmed findings; PSA likewise	Cost of liraglutide	Cost of dulaglutide	Hba1c reduction of liraglutide
Eliasson, 2022 [60]	Semaglutide-O vs. empagliflozin, sitagliptin	Direct + indirect, 2019	QALY, incidence of DRC	Semaglutide-O	Direct: vs. empagliflozin = 239,001/QALY, vs. sitagliptin = 120,848/QALY, indirect: vs. empagliflozin = 191,721/QALY, vs. sitagliptin 95,234/QALY (SEK)	SEK500,000/QALY	Direct costs: QNE for semaglutide-O vs. empagliflozin, QNE for semaglutide-O vs. sitagliptin, with indirect costs: QNE and QNE (Fig. 2, Table 1)	Greater reductions in HbA1c led to fewer DRCs, longer time to INS initiation, fewer hypo events, and, lastly, better projected survival	DSA confirmed findings; PSA likewise	3-year fixed duration until INS initiation	No QOL impact of BMI change	0% discount rate
Franch-Nadal, 2022 [63]	Semaglutide-O vs. empagliflozin, sitagliptin, liraglutide	Direct, 2020	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-O	Vs. empagliflozin: 1339/QALY, vs. sitagliptin: dominant, vs. liraglutide: dominant, 7 mg semaglutide-O vs. sitagliptin: 2011/QALY (EUR)	EUR30,000/QALY	QNE for semaglutide-O vs. empagliflozin, QSE for semaglutide-O vs. sitagliptin, QSE for semaglutide-O vs. liraglutide, QNE for 7 mg semaglutide-O vs. sitagliptin (Table 1)	Extra clinical benefit from using semaglutide-O was due to reduced cumulative incidence and later onset of DRCs. Higher treatment cost of semaglutide-O was offset by the lower DRCs	DSA confirmed findings; PSA likewise	Shorter time horizon (10 years)	UKPDS HbA1c progression with no changes in treatment intensification	Lower 95% CI of HbA1c estimated treatment difference applied
Hu, 2022 [76]	Semaglutide-I vs. dulaglutide	Direct, 2021	QALY, LY, incidence of DRC	Semaglutide-I	26,957.44/QALY(US$)	US$12,551.5–37,654.50/QALY (1–3 times GDP/capita)	QNE (Fig. 2)	None stated	DSA showed high sensitivity to time-related factors, reversing conclusions of base-case; PSA showed 30.2%, 48.2% and 2.8% chance of being CE at above 3 times GDP/capita, between 1 and 3 times GDP/capita, and below 1 times GDP/capita, respectively	Discounting factor	Shorter time horizon	MI disutility score
Malkin, 2022 [61]	Semaglutide-O vs. empagliflozin, dulaglutide	Direct, 2021	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-O	Vs. empagliflozin: 23,571/QALY, vs. dulaglutide: 23,927/QALY, (EUR)	EUR30,000/QALY	QNE for semaglutide-O vs. empagliflozin, QNE for semaglutide-O vs. dulaglutide (Fig. 2)	Reduced incidence and time to onset of DRCs, the higher costs of semaglutide-O were offset by this. Higher HbA1c and weight reduction were the biggest drivers when comparing with empagliflozin and dulaglutide, respectively	DSA confirmed findings; PSA likewise	Shorter time horizon (10 years)	Discount factor of 0%	Only statistically significant differences from NMA
Ruan, 2022 [81]	Semaglutide-I vs. dulaglutide	Direct, 2021	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	Semaglutide-I dominant (CNY)	CNY80,976/QALY (1 times GDP/capita)	QSE (Fig. 2)	Semaglutide-I reduced and delayed the occurrence of DRCs, and reduced mortality. This offset the increase treatment cost	DSA confirmed findings; PSA likewise	Shorter time horizon (5 years)	Shorter time horizon (10 years)	HbA1c threshold at 7.0%
Stafford, 2022 [73]	Semaglutide-I vs. canagliflozin	Direct + indirect, 2019	QALY, LY, incidence of DRC	Semaglutide-I	Using the IHE-DCM model: 14,127/QALY; using the ECHO-T2DM model: 13,188/QALY (CAD)	CAD50,000/QALY	QNE and QNE using both models (Fig. 1)	QALY gains for semaglutide-I were mainly driven by later use of INS caused by higher HbA1c reduction, and greater initial weight loss. Fewer DRCs offset some of the higher treatment costs of semaglutide-I	DSA confirmed findings; PSA likewise	10% worse HbA1c effect for semaglutide-I	HbA1c threshold set to 7.5%	Discount rate at 3.5%
Viljoen, 2022 [62]	Semaglutide-I vs. dulaglutide	Direct, 2020	QALY, LY, incidence of DRC, onset of DRC	Semaglutide-I	Vs. dulaglutide 3 mg: dominant, vs. dulaglutide 4.5 mg: 228/QALY (GBP)	GBP20,000/QALY	QSE for semaglutide-I vs. dulaglutide 3 mg, QNE for semaglutide-I vs. Dulaglutide 4.5 mg (Fig. 3)	Extra clinical benefit from using semaglutide-I was due to reduced cumulative incidence and later onset of DRCs. Higher treatment cost of semaglutide-I was offset by the lower DRCs	DSA confirmed findings; PSA likewise	HbA1c threshold set to 7.5%	Lower 95% CI of HbA1c treatment difference	Addition of basal INS, then basal bolus
Chien, 2020 [75]	No main vs. classes: MET, SU, DPP4, SGLT2, GLP1, unspecified INS	Direct, 2019	QALY, LY, incidence of DRC	Arm7 (SGLT2)	Arm 6: – Arm 7 dominates Arm 3 and Arm 8 is cost effective against Arm 1 and Arm 2 Arm 4 and Arm 5 extended dominated Arm 1: MET+SU -> +DPP4 Arm 2: MET +SGLT2 -> +DPP4 Arm 3: MET+DPP4 -> +SU Arm 4: MET+DPP4 -> +SGLT2 Arm 5: MET+GLP1 -> +SU Arm 6: MET+SU -> +DPP4 Arm 7: MET+SU -> +SGLT2 Arm 8: MET+INS -> +SU (NT)	NT770,770/QALY (forecasted GDP/capita in Taiwan in 2019)	Arm 7 vs. Arm 3: QSE Arm 7 vs. Arm 8: QSE Arm 7 vs. Arm 1: QNE Arm 7 vs. Arm 2: QNE Arm 4 and Arm 5: extended dominated (Table 3, text)	None stated	DSA confirmed findings; PSA likewise	Baseline HbA1c	Baseline age	Hba1c threshold
Hu, 2021 [76]	No main vs. dapagliflozin + saxagliptin, dapagliflozin, saxagliptin	Direct, 2019	QALY, LY	Dapagliflozin	Dapagliflozin + saxagliptin vs. saxagliptin: 217,530/QALY dapagliflozin vs. dapagliflozin + saxagliptin: dominate dapagliflozin vs. saxagliptin: 12,191/QALY (US$)	US$10,425.29–31,275.88/QALY (1–3 times GDP/capita in 2019)	QSE for dapagliflozin + MET vs. dapagliflozin + saxagliptin + MET, QNE for dapagliflozin + MET vs. saxagliptin + MET, QNE for dapagliflozin + saxagliptin + MET vs. saxagliptin (Table 4)	None stated	DSA confirmed findings; PSA likewise	Discount rate	Saxagliptin acquisition cost	Dapagliflozin acquisition cost
Lin, 2021 [78]	No main vs. linagliptin 5 mg, saxagliptin 5 mg, alogliptin 25 mg, sitagliptin 100 mg, and vildagliptin 50 g	Direct, 2019	QALY, LY, incidence of DRC	Alogliptin	Alogliptin vs. linagliptin: extended dominate, vs. saxagliptin: dominate, vs. vildagliptin: dominate (US$)	US$10,276/QALY (China capita per GDP in 2019)	QSE for alogliptin vs. linagliptin, QSE for alogliptin vs. saxagliptin: QSE for alogliptin vs. vildagliptin	Reduced cumulative incidence of DRCs	DSA confirmed findings; PSA likewise	Cost of alogliptin and sitagliptin	Reduction of HbA1c for sitagliptin and alogliptin	Discount rate
Zupa, 2021 [71]	Empagliflozin vs. semaglutide-I	Direct, 2020	QALY, LY	Semaglutide-I	19,964/QALY (US$)	US$50,000–100,000/QALY	QNE (text)	None stated	DSA confirmed findings; PSA likewise	Daily cost of semaglutide-I	Heart failure risk of semaglutide-I	Stroke risk of semaglutide-I
Chakravarty, 2018 [65]	Dapagliflozin vs. liraglutide, SU, DPP4, pioglitazone	Direct, 2016	QALY	Dapagliflozin	Vs. liraglutide: dominant, vs. DPP4: dominant, vs. TZD: 25.835/QALY, vs. SU: 19.005/QALY (US$)	US$50,000/QALY	QSE for dapagliflozin vs. GLP1, QNE for dapagliflozin vs. SU, QSE for dapagliflozin vs. DPP4, QNE for dapagliflozin vs. TZD (Fig. 4, Table 5)	Change in body weight	DSA generally confirmed findings; PSA likewise	Change in weight impact of treatment	Change in Hba1c	Change in SBP
Neslusan, 2018 [66]	Canagliflozin vs. dapagliflozin	Direct, 2016	QALY, LY, incidence of DRC, onset of DRC	Canagliflozin	Dominant (US$)	US$100,000/QALY	QSE (Fig. 2)	Cost offsets from higher acquisition cost, and QALY gains were driven by better HbA1c lowering, which also led to lower event rates from complications, longer time to INS, less INS use	DSA completely confirmed findings; PSA likewise	Shorter time horizon (5 years)	Later treatment intensification	Real-world patient characteristics
Hou, 2019 [74]	Canagliflozin vs. dapagliflozin	Direct 2017	QALY, LY, incidence of DRC	Canagliflozin	Canagliflozin100mg dominant (US$)	US$9117/QALY (GDP/capita of China in 2017)	QSE (Table 3)	Driven by the reduced cumulative incidence of macrovascular and microvascular complications	DSA big impact from cost of drugs, moderate/small impact of disutility/costs of complications; PSA confirmed base case	Cost of canagliflozin and dapagliflozin	Disutility	Cost of complications
Ramos, 2019 [40]	Empagliflozin vs. sitagliptin, saxagliptin	Direct, 2018	QALY, LY, incidence of DRC	Empagliflozin	Vs. sitagliptin: 6464/QALY, vs. saxagliptin: 3878/QALY (GBP)	GBP20,000/QALY	QNE for empagliflozin vs. sitagliptin, QNE for empagliflozin vs. saxagliptin, (Table 5, Fig. 3)	Higher initial cost of SGLT2 offset by higher QALYs and LYs. Higher cost amd more DRCs for empagliflozin because of increased survival. Lower renal complication costs	DSA confirmed findings; PSA likewise	Shorter time horizon (5 years)	Hba1c threshold for treatment switch at 9%	Cardiovascular outcomes up to 3 years
Ramos, 2020 [45]	Empagliflozin vs. SoC and liraglutide	Direct, 2018	QALY, LY, incidence of DRC	Empagliflozin	Empagliflozin + SoC dominant vs. liraglutide + SoC, 6428/QALY vs. SoC alone (GBP)	GBP20,000–30,000/QALY	QSE for empagliflozin vs. liraglutide, QNE for empagliflozin vs. SoC (Table 5, Fig. 3)	Treatment costs, survival, lower CV mortality	DSA confirmed findings; PSA likewise	Shorter time horizon (5 years)	Treatment switch threshold at 9%	CVOT outcome benefits applied for full treatment duration
Ramos, 2020 [46]	Empagliflozin vs. Semaglutide-O	Direct, year unknown	QALY, LY, heart failure	Empagliflozin	With hHF: empagliflozin dominant, without hHF: ICER = 186,690/QALY (GBP)	GBP20,000–30,000/QALY	QSE for empagliflozin (with hHF effect) vs. oral semaglutide, QSW for empagliflozin (without hHF effect) vs. oral semaglutide (Table 4, Fig. 1)	Inclusion of hHF effect of empagliflozin. Lower cost of empagliflozin	DSA confirmed findings; but some scenarios very sensitive; PSA likewise	Excluding the treatment effect on hHF	Treatment intensification at different HbA1c thresholds	BMI polynomial utility approach
Reifsnider, 2020 [69]	Empagliflozin vs. sitagliptin	Direct, 2018	QALY, LY, CVD-free LY, incidence of DRC	Empagliflozin	Base case: 6967/QALY in CVD: 3589/QALY in non-CVD: 12,577/QALY (US$)	US$50,000–150,000/QALY	QNE for base case, QNE in CVD population, QNE in non-CVD population (Table 1, ESM Fig. SA3)	Base case: least complications with empagliflozin in the CVD pop: longer CVD-free survival and less cardiovascular death, fewer rates of DRCs in general in the non-CVD pop: lower or similar rates of DRCs	DSA confirmed findings, and showed empagliflozin to be either CE or dominant, depending on parameter. More likely to be dominant in the CVD pop.; PSA confirmed findings	Rebate percentage applied to the wholesale acquisition cost	Shorter time horizon (1 year)	Adherence to empagliflozin (80%*) or commercial perspective
Van der Linden, 2020 [54]	Dapagliflozin vs. DPP4	Direct + indirect, 2018	QALY, LY, incidence of DRC	Dapagliflozin	Vs. DPP4 class: dominant (EUR)	EUR20,000/QALY	QSE (Table 7, Fig. 2)	Dapagliflozin reduced the incidence of micro- and macrovascular complications, in exchange for more urinary tract infections and gastrointestinal infections, which increased quality of life. Dapagliflozin was cost saving due to lower treatment costs and reduced DRCs	DSA confirmed findings. No change from being dominant; PSA confirmed base case	Dominant conclusion not changed	Dominant conclusion not changed	Dominant conclusion not changed
Ehlers, 2021 [56]	Empagliflozin vs. liraglutide	Direct, 2019	QALY, LY, incidence of DRC	Empagliflozin	Empagliflozin dominant (DKK)	DKK357,100/QALY (1 times the GDP/capita)	QSE (Fig. 2, Table 4)	Longer survival of empagliflozin, higher total cost of liraglutide	DSA confirmed findings. In scenario where liraglutide effects were extended to 13 years (Hba1c 8.5%), liraglutide would have extreme ICER (>3 mill/QALY); PSA likewise	Dominant conclusion not changed	Dominant conclusion not changed	Dominant conclusion not changed
Gourzoulidis, 2021 [55]	Empagliflozin vs. dapagliflozin	Direct, 2020	QALY, LY, incidence of DRC	Empagliflozin	Vs. dapagliflozin: 965/QALY (EUR)	EUR36,000/QALY	QNE (Table 3, ESM Fig. 1)	Empagliflozin had reduced many DRCs, with longer life as a result	DSA confirmed findings, but empagliflozin showed dominance in 3-year model, not shown in tornado diagram; PSA confirmed findings	Discount rate of costs	HR: dapagliflozin vs. empagliflozin	Discount rate health
Ramos, 2021 [77]	Empagliflozin vs. liraglutide, sitagliptin	Direct, 2019	QALY, LY, incidence of DRC	Empagliflozin	Vs. liraglutide: dominant vs. sitagliptin: 75,349/QALY (RMB)	RMB212,676/QALY (3 times GDP/capita)	QSE for empagliflozin vs. liraglutide, QNE for empagliflozin vs. sitagliptin, (Table 4, Fig. 3)	Longer survival of empagliflozin, lower cost from less HF and renal complications, despite higher costs from longer survival	DSA confirmed findings, except when changing treatment switch for liraglutide to 13 years, at HbA1c 8.5%, which puts empagliflozin in QSW; PSA confirmed base case	Effects of liraglutide extended to 13 years /HbA1c threshold of 8.5%	Shorter time horizon (5 years)	CV outcomes only used for 3 years
Reifsnider, 2021 [67]	Empagliflozin vs. dapagliflozin, canagliflozin	Direct, 2020	QALY, LY, incidence of DRC	Empagliflozin	Empagliflozin vs. canagliflozin: dominate empagliflozin vs. dapagliflozin: 3054/QALY Empagliflozin vs. SoC: 32,848/QALY (US$)	US$50,000–150,000/QALY	QSE for empagliflozin vs. canagliflozin, QNE for empagliflozin vs. dapagliflozin, (Table 1, ESM Fig. OS1)	Longer overall survival and reduced rates of clinical events	DSA generally confirmed findings. Using treatment effects that favoured the comparators, made comparators dominant; PSA confirmed base case	Reducing HR for comparator vs. empagliflozin	Shorter time horizon (1, 3, 5, and 10 years)	Commercial perspective
Lasalvia, 2022 [82]	Dapagliflozin vs. DPP4	Direct, 2020	QALY, LY, incidence of DRC	Dapagliflozin	1964.80/QALY (US$)	US$5710–17,129.9/QALY (1–3 times GDP/capita)	QNE (Table 4)	None stated	DSA confirmed findings; PSA likewise	Change of time horizon	Hba1c threshold for treatment switch at 9%	Weight reduction effect maintenance
Peng, 2022 [80]	SGLT2 vs. DPP4	Direct, 2020	QALY, incidence of DRC	SGLT2	With CVD history: 3244.07/QALY; without CVD history: 4185.64/QALY (US$)	US$30,038–90,114/QALY (1–3 times GDP/capita)	QNE and QNE for comparisons with and without CVD history (Table 1)	None stated	DSA confirmed findings; PSA likewise	Cost of DPP4	HR of SGLT2 vs. DPP4 on all-cause death	HR of SGLT2 vs. DPP4 on stroke
Reifsnider, 2022 [72]	Empagliflozin vs. liraglutide	Direct, 2019	QALY, LY, incidence of DRC	Empagliflozin	Empagliflozin dominant (US$)	None stated	QSE (Table 1)	Fewer DRCs over time and longer survival due to empagliflozin's effect on patients with CVD	DSA completely confirmed findings; PSA likewise	Disutility of injectable treatment	Drug acquisition cost	Treatment effect for patients with CVD (HR of empagliflozin vs. liraglutide)

Comparisons list the main comparators and then the alternatives separated by a comma, while ‘and’ is written to show that several different comparisons were made. When one medication is combined with another in the comparison, it is explicitly stated or symbolised with a ‘+’ symbol

1w once per week, AE adverse event, BMI body mass index, CAD Canadian dollars, CE cost-effective, CHF chronic heart failure, CI confidence interval, CNY Chinese Yuan, CV cardiovascular, CVD cardiovascular disease, CVOT Cardiovascular Outcome Trial, DKK Danish kroner, DPP4 dipeptidylpeptidase-4 inhibitors, DRCs diabetes-related complications, DSA deterministic sensitivity analysis, ECHO-T2DM Health Outcomes Model of Type 2 Diabetes Mellitus, ESM electronic supplementary, material, EUR Euro, GBP British pound sterling, GDP gross domestic product, GLP1 glucagon-like peptide-1 recepter agonist, HR hazard ratio, hHF hospitalisation for heart failure, ICER incremental cost-effectiveness ratio, IDegLira combination of liraglutide and INS degludec, IHD ischaemic heart disease, IHE-DCM Institute for Health Economics Cohort Model for T2DM, INS insulin, LY life years, MACE major adverse cardiac events, MET metformin, MDI multiple daily injections, MI myocardial infarction, NMA network meta-analysis, NT Taiwan new dollar, pop population, PSA probabilistic sensitivity analysis, QALY quality-adjusted life-years, QOL quality of life, QSE south-east quadrant, QNE north-east quadrant, QSW south-west quadrant, QW every week, RMB Renminbi, SBP systolic blood pressure, SEK Swedish kroner, SoC standard of care, US$ United States dollar, SGLT2 sodium-glucose cotransporter-2 inhibitors, semaglutide-I/O semaglutide injectable/oral, SU sulfonylurea, TZD thiazolidinedione, UKPDS UK Prospective Diabetes Study, WTP willingness to pay

Overview of Comparisons between the Newer Non-insulin Antidiabetic Drugs

In the included studies, the GLP1 and SGLT2 classes were the main comparators in 31 and 16 studies, respectively, while one study used DPP4 exclusively [78]. The remaining three studies compared alternatives where it was difficult to discern which was the main comparator [75, 76]. No studies evaluated the cost effectiveness of combining the two most frequently compared classes—SGLT2 and GLP1. The dosage of one or more comparators was mentioned in 35 of 50 studies. When a single product was represented as two different comparators with different dosages, the comparator with the higher dosage was always favoured. The results of the comparisons are presented in three sections based on the most frequent comparisons between classes of nNIADs: (1) GLP1 compared with GLP1 or another non-SGLT2; (2) SGLT2 compared with SGLT2 or another non-GLP1; and (3) GLP1 compared with SGLT2. The study, which compared different products from the DPP4 class, found alogliptin to be cost effective [78]. To simplify the descriptions of the comparators in Sects. 3.3.2–3.3.4, the nNIAD mentioned first is considered the main comparator unless otherwise stated. The results of the comparisons are briefly summarised below. Table 2 provides further details on the comparisons and the economic results in general.

Glucagon-Like Peptide-1 (GLP1) Compared with GLP1 or Another Non-Sodium-Glucose Cotransporter-2 (SGLT2)

Comparisons of products within the GLP1 class occurred in 14 studies, and comparisons of the GLP1 class of products against non-SGLT2 products occurred in six studies.

In 10 studies, injectable semaglutide was found to be primarily dominant but cost effective against other GLP1 products [39, 42, 43, 45, 49, 53, 63, 69, 80, 82]. In the one study where oral semaglutide was compared with injectable semaglutide, oral semaglutide was dominant against it and all other comparators [70]. The three remaining comparisons were between other GLP1 products and gave conflicting results [36, 50, 83]. For the six comparisons with GLP1 and other classes as comparators, injectable semaglutide was dominant compared with dulaglutide and sitagliptin [47], and cost-effective compared with INS glargine [43]. Exenatide was found to be cost effective against INS glargine and liraglutide [34], while liraglutide was cost effective against sitagliptin [37]. IDegLira, a combination of liraglutide and INS degludec, was categorised as an nNIAD of the GLP1 class if compared with nNIADs or similar combinations of nNIADs and INS. IDegLira and liraglutide, combined with INS, were found to be cost effective or dominant compared with other GLP1 products combined with INS, GLP1 as a class, and multiple forms of INS treatment [35, 39].

SGLT2 Compared with SGLT2 or Another Non-GLP1

Comparisons of products within the SGLT2 class occurred in four studies, and in six studies, SGLT2 products were compared with SGLT2 or non-GLP1 products.

When only SGLT2 products were compared, empagliflozin was found to be cost effective against dapagliflozin [55, 67] and dominant against canagliflozin [67], while canagliflozin was dominant against dapagliflozin in two studies [66, 74]. For the six comparisons with SGLT2 and other classes, empagliflozin was found to be cost effective against sitagliptin [69] and saxagliptin [40], while dapagliflozin was found to be dominant in one study against the DPP4 class and cost effective in another [54, 82]. Finally, dapagliflozin in combination with MET was the dominant strategy in a study comparing combinations of dapagliflozin and saxagliptin [76], and one class-level comparison found SGLT2 to be cost effective against DPP4 [80].

GLP1 Compared with SGLT2

Direct comparison of GLP1 and SGLT2 occurred in 19 studies. Aggregating the results to a class-level perspective, GLP1 was cost effective against SGLT2 in nine comparisons and not cost effective in three, while SGLT2 dominated GLP1 in six comparisons. Additionally, one study found SGLT2 as a third-line choice to be the cost-effective option in a set of comparisons where introducing GLP1 as a third-line choice was dominated.

From the product-level perspective, semaglutide (oral or injectable) and empagliflozin were the most frequently used main comparators. Oral semaglutide was found to be cost effective in five of seven comparisons with empagliflozin [51, 53, 58, 60, 61, 63, 64] but was not cost effective in two comparisons [58, 64]. Injectable semaglutide was found to be cost effective in all comparisons with empagliflozin [49, 57, 59] except one [59]. Injectable semaglutide was also found to be cost effective against canagliflozin [73]. Empagliflozin dominated liraglutide in all their comparisons [45, 56, 72, 77]. When empagliflozin was compared with injectable semaglutide and oral semaglutide, it dominated the latter [71] but was not cost effective against the former [46]. Dapagliflozin was compared with liraglutide once and dominated [65]. One study compared injectable semaglutide and empagliflozin [71] with no easily discernible main comparator and found the former to be cost effective.

Willingness to Pay and Incremental Cost-Effectiveness Ratio (ICER) Estimates

In the majority of studies (n = 21), a willingness-to-pay (WTP) threshold similar to that used in the UK (GBP20–30,000/quality-adjusted life-years [QALYs] or equivalent value in Euros or local currency) was utilised [34, 36–40, 45–55, 57, 61–63], followed by the World Health Organisation (WHO)-recommended WTP threshold of one to three times the gross domestic product per capita (n = 14) [41, 56, 58, 59, 74–83]. The remaining studies used a North American WTP threshold (n = 9) [64–71, 73] or a Swedish threshold (n = 3) [35, 60], and one study each used thresholds specific to The Netherlands [43] and Slovakia [42], while the last study did not state a threshold [72].

A total of 116 ICER estimates for base cases and reported scenarios were extracted from the included studies. Most of these ICERs were in the northeast quadrant of an incremental cost-effectiveness plane (n = 65; higher costs and effects), while the remaining were located in the southeast quadrant (n = 54; lower costs, higher effects), except one that was found in the southwest quadrant (lower costs, lower effects). When comparing the base case and the scatterplot resulting from the probabilistic sensitivity analysis (PSA), there was a clear consensus about the cost effectiveness and location of the estimates, indicating ICER estimates that were robust to changes in parameters chosen in the respective analyses.

Effect Measures and Key Drivers

All studies used QALYs and life-years (LYs) as effect measures, except for five studies [50, 60, 65, 80, 83] that only used QALYs. The majority of the included studies emphasised presenting the DRC output of the models and how this output affected the results, while six studies [50, 65, 76] did not report DRC output. Among the studies that reported DRC output, 19 reported both DRC incidence and the time until onset [36–39, 41–43, 47–49, 51–53, 57, 61–63, 66, 81], while 24 only reported incidence [34, 35, 40, 44, 45, 54–56, 58, 60, 64, 67–70, 72–75, 77–80, 82]. Two studies focused explicitly on the effect of the treatments on CVD-related DRCs, with one including hospitalisation from HF [46] and included CVD-free LYs [69]. One study included event-free survival as an effect measure [70].

The reporting of key drivers was based on an explicit description of the key drivers of the results of the economic analysis, taken from a dedicated section. If not described explicitly, the authors interpreted the key drivers as the results reported as most important for their results. The authors observed a general tendency to emphasise the effect of the nNIADs on time until the onset of DRCs and reduction of the cumulative incidence of DRCs, or both, as being key in achieving cost effectiveness over the comparators. It was not possible to discern whether key drivers of the results were reported in nine studies [50, 59, 71, 75, 76, 79, 80, 82, 83].

Sensitivity Analysis

All included studies conducted both deterministic sensitivity analysis (DSA) and PSA. The included studies either directly reported parameters that had the greatest impact on the ICER or supplied graphs or tables from which it was possible to extract the information. We extracted information on the parameters that affected the results most, second most, and third most, according to the degree to which the change in the parameter affected the ICER. These parameters were further grouped into broad categories. Among these categories, the most frequently occurring first-, second-, and third-most sensitive parameters were changes to the time horizon of the analysis, changes to the treatment effect of the comparators, and differing assumptions regarding the cost of the comparators, respectively. Four studies [47, 50, 54, 56] reported that the conclusion of dominance remained unchanged (see Table 2 for more details).

Model Parameters and Assumptions

This section presents the key model parameters and assumptions from the included studies, along with notable differences between the studies. A detailed presentation can be found in Table 3.

Table 3.

Model parameters

Study, year	Comparison	Baseline table	Treatment effect	Adverse events	Diabetes-related complications	Risk equations	Treatment switch	Treatment switch type
Barnett, 2018 [37]	Liraglutide vs. sitagliptin	Yes	HbA1c, systolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, severe hypoglycaemic event, non-severe hypoglycaemic event	Severe hypoglycaemic event, non-severe hypoglycaemic event	Ophthalmic, diabetic foot, renal complications, CVD	Base case: UKPDS 68; SA: UKPDS 82	After 3 years = comparators discontinued, all progress to basal INS NPH	Comparators discontinued
Basson, 2018 [36]	Dulaglutide vs. exenatide QW	Yes	HbA1c, systolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, nausea, severe hypoglycaemic event, non-severe hypoglycaemic event, injection site reaction	Severe hypoglycaemic event, non-severe hypoglycaemic event, nausea, injection site reaction	Myocardial infarction, ischaemic heart disease, heart failure, stroke, severe vision loss, amputation, peripheral vascular disease, gross proteinuria, neuropathy, ulcer, hemodialysis, peritoneal dialysis, renal transplant, cataract, retinopathy, macular edema, vision-threatening retinopathy	Base case: No specific UKPDS RE mentioned; SA: none stated	After 2 years = comparators discontinued, basal INS glargine IU40	Comparators discontinued
Ericsson, 2018 [35]	Liraglutide + basal INS vs. lixisenatide + basal INS, IDegLira	Yes	HbA1c, BMI, systolic blood pressure, total cholesterol, LDL, HDL, triglycerides, severe hypoglycaemic event, non-severe hypoglycaemic event	Severe hypoglycaemic event, non-severe hypoglycaemic event	Proliferative retinopathy, retinopathy, macular edema, severe vision loss, neuropathy, peripheral vascular disease, amputation, microalbuminuria, gross proteinuria, end-stage renal disease, ischaemic heart disease, myocardial infarction, stroke, heart failure	Base case: SweNDR, UKPDS OM2; SA: none stated	After 8.8% HbA1c threshold = comparators discontinued, patients switched to basal-bolus	Comparators discontinued
Tzanetakos, 2018 [34]	Exenatide QW vs. INS glargine, liraglutide	Yes	HbA1c, weight, total cholesterol, HDL, systolic blood pressure, discontinuation in general, non-severe hypoglycaemic event, severe hypoglycaemic event, nausea	Discontinuation in general, non-severe hypoglycaemic event, severe hypoglycaemic event, nausea	Myocardial infarction, congestive heart failure, stroke, amputation, ischaemic heart disease, blindness, end-stage renal disease, non-severe hypoglycaemic event, severe hypoglycaemic event	Base case: UKPDS 68; SA: none stated	After 8% HbA1c threshold = comparators discontinued, replace with basal INS rescue regimen	Comparators discontinued
Ericsson, 2019 [44]	Semaglutide-I vs. dulaglutide, lixisenatide	Yes	HbA1c, BMI, systolic blood pressure, total cholesterol, LDL, HDL, triglycerides, severe hypoglycaemic event, non-severe hypoglycaemic event	Severe hypoglycaemic event, non-severe hypoglycaemic event	Proliferative retinopathy, non-proliferative retinopathy, macular edema, severe vision loss, neuropathy, peripheral vascular disease, amputation, microalbuminuria, macroalbuminuria, end-stage renal disease, ischaemic heart disease, myocardial infarction, stroke, heart failure	Base case: SweNDR; SA: none stated	Uncontrolled on MET: HbA1c 8.22% exceeded = basal INS added, GLP1 stopped. Next step not described Uncontrolled on INS: HbA1c 8.31% exceeded = bolus added, GLP1 stopped	Comparators discontinued
Gæde, 2019 [52]	Semaglutide-I vs. dulaglutide, exenatide QW, liraglutide, lixisenatide	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, estimated glomerular filtration rate, non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, microalbuminuria, gross proteinuria, hemodialysis, peritoneal dialysis, renal transplant, background retinopathy, proliferative retinopathy, macular edema, cataract, blindness, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe hypoglycaemic event, severe nocturnal hypoglycaemic event, diabetic ketoacidosis, lactic acid event, laser treatment, neuropathy, amputation, gangrene, ulcer	Base case: UKPDS 68; SA: none stated	After HbA1c 7.5% exceeded = add basal INS, 8.0 exceeded = add basal-bolus + stop GLP1	Comparators continued, then discontinued
Hunt, 2019 [43]	Semaglutide-I vs. INS glargine, dulaglutide	Yes	HbA1c, systolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Severe hypoglycaemic event, non-severe hypoglycaemic event, severe nocturnal hypoglycaemic event, non-severe nocturnal hypoglycaemic event	myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, hemodialysis, peritoneal dialysis, renal transplant, non-severe hypoglycaemic event, severe hypoglycaemic event, laser treatment, cataract, blindness, neuropathy, amputation, gangrene, ulcer	Base case: none stated; SA: UKPDS 68 and 82	After 3 years = comparators discontinued and all get same dosage of basal INS glargine U100	Comparators discontinued
Johansen, 2019 [68]	Semaglutide-I vs. dulaglutide	Yes	HbA1c, systolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, estimated glomerular filtration rate, non-severe hypoglycaemic event, severe hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event	Background retinopathy, macular edema, proliferative retinopathy, severe vision loss, neuropathy, amputation, microalbuminuria, macroalbuminuria, end-stage renal disease, ischaemic heart disease, myocardial infarction, stroke, congestive heart failure, non-severe hypoglycaemic event, severe hypoglycaemic event	Base case: UKPDS 82 and 68, CDC, WESDR, REP; SA: UKPDS 68	After 3 years = comparators discontinued, basal INS Hagedorn for all	Comparators discontinued
Malkin, 2019 [41]	Semaglutide-I vs. liraglutide	Yes	HbA1c, systolic blood pressure, BMI	None stated	Any, stroke, myocardial infarction, angina, congestive heart failure, peripheral vascular disease, neuropathy, amputation, ulcer, end-stage renal disease, gross proteinuria, microalbuminuria, macular edema, cataract, proliferative retinopathy, background retinopathy, severe vision loss	Base case: UKPDS 68; SA: UKPDS 82	After 5 years = comparators discontinued, all get same dosage of basal INS glargine U100	Comparators discontinued
Malkin, 2019 [42]	Semaglutide-I vs. dulaglutide	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, estimated glomerular filtration rate, non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Any, stroke, myocardial infarction, angina, congestive heart failure, peripheral vascular disease, neuropathy, amputation, ulcer, end-stage renal disease, gross proteinuria, microalbuminuria, macular edema, cataract, proliferative retinopathy, background retinopathy, severe vision loss	Base case: UKPDS 68; SA: none stated	After 3 years = comparators discontinued, all get same dosage of basal INS glargine U100	Comparators discontinued
Raya, 2019 [39]	IDegLira vs. GLP1, INS mix + GLP1, basal INS, MDI	No	HbA1c, systolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI	Hypoglycaemia only in SA	Myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, microalbuminuria, gross proteinuria, haemodialysis, peritoneal dialysis, renal transplant, background retinopathy, proliferative retinopathy, macular edema, severe vision loss, cataract, neuropathy, ulcer, amputation, non-severe hypoglycaemic event, severe hypoglycaemic event, gangrene, laser treatment	Base case: none stated; SA: UKPDS OM2, UKPDS OM	After 5 years = comparators discontinued, all progress to basal bolus	Comparators discontinued
Viljoen, 2019 [38]	Semaglutide-I vs. dulaglutide	No	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, estimated glomerular filtration rate, severe hypoglycaemic event, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Severe hypoglycaemic event, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Myocardial infarction, angina, peripheral vascular disease, stroke, congestive heart failure, microalbuminuria, gross proteinuria, end-stage renal disease, hemodialysis, peritoneal dialysis, renal transplant, background retinopathy, proliferative retinopathy, macular edema, severe vision loss, cataract, neuropathy, ulcer, amputation, severe hypoglycaemic event, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Base case: UKPDS 68; SA: UKPDS 82, UKPDS HbA1c	After 3 years = comparators discontinued, all progress to basal INS	Comparators discontinued
Bain, 2020 [51]	Semaglutide-O vs. empagliflozin, sitagliptin, liraglutide	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, BMI, severe hypoglycaemic event, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, microalbuminuria, gross proteinuria, hemodialysis, peritoneal dialysis, renal transplant, background retinopathy, macular edema, severe vision loss, cataract, neuropathy, ulcer, amputation, non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event, laser treatment, blindness, gangrene	Base case: UKPDS 68; SA: UKPDS 82	After HbA1c 7.5% exceeded = stop semaglutide, basal INS started	Comparators discontinued
Capel, 2020 [50]	Exenatide QW vs. dulaglutide, liraglutide, lixisenatide	No	HbA1c, weight, discontinuation associated with AEs, nausea, non-severe hypoglycaemic event, severe hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event, nausea	Ischaemic heart disease, congestive heart failure, myocardial infarction, stroke, amputation, blindness, end-stage renal disease, severe hypoglycaemic event, non-severe hypoglycaemic event	Base case: UKPDS 68; SA: none stated	After HbA1c 7.5% exceeded = comparators discontinued, switch to basal INS, 8.0 exceeded = switch to basal-bolus + stop GLP1	Comparators discontinued
Gorgojo-Martínez, 2020 [49]	Semaglutide-I vs. empagliflozin	No	HbA1c, systolic blood pressure, BMI	None stated	Any, stroke, myocardial infarction, angina, congestive heart failure, peripheral vascular disease, neuropathy, amputation, ulcer, end-stage renal disease, gross proteinuria, microalbuminuria, macular edema, cataract, proliferative retinopathy, background retinopathy, severe vision loss, blindness, hemodialysis, non-severe hypoglycaemic event, severe hypoglycaemic event, gangrene	Base case: none stated; SA: UKPDS 68 and 82	Both treatments switched to basal INS glargine U100, after 3 years	Comparators discontinued
Johansen, 2020 [48]	Semaglutide-I vs. liraglutide	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, estimated glomerular filtration rate, non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, microalbuminuria, gross proteinuria, hemodialysis, peritoneal dialysis, renal transplant, background retinopathy, proliferative retinopathy, macular edema, severe vision loss, cataract, neuropathy, ulcer, amputation, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe hypoglycaemic event, severe nocturnal hypoglycaemic event	Base case: UKPDS 68; SA: UKPDS 82	After HbA1c 7.5% exceeded = stop semaglutide, basal INS started	Comparators discontinued
Martín, 2020 [47]	Semaglutide-I vs. dulaglutide, sitagliptin	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, estimated glomerular filtration rate, non-severe hypoglycaemic event, severe hypoglycaemic event	Hypoglycaemia	Myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, microalbuminuria, gross proteinuria, hemodialysis, peritoneal dialysis, renal transplant, background retinopathy, macular edema, severe vision loss, cataract, neuropathy, ulcer, amputation, non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event, laser treatment, blindness, gangrene, severe vision loss, end-stage renal disease	Base case: UKPDS 68; SA: UKPDS 82	After 3 years = basal INS glargine U100	Comparators discontinued
Capehorn, 2021 [57]	Semaglutide-I vs. empagliflozin	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, estimated glomerular filtration rate	Non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe hypoglycaemic event, severe nocturnal hypoglycaemic event	Myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, microalbuminuria, gross proteinuria, hemodialysis, peritoneal dialysis, renal transplant, background retinopathy, proliferative retinopathy, macular edema, severe vision loss, cataract, neuropathy, ulcer, amputation, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe hypoglycaemic event, severe nocturnal hypoglycaemic event, laser treatment, blindness	Base case: UKPDS 68; SA: UKPDS 82	After exceeding threshold: HbA1c 7.5% = replace therapy with basal INS	Comparators discontinued
Guzauskas, 2021 [64]	Semaglutide-O vs. empagliflozin, liraglutide, sitagliptin, MET+SU	Yes	HbA1c, weight, severe hypoglycaemic event, discontinuation associated with AEs, MACE, heart failure, nephropathy	Severe hypoglycaemic event, discontinuation associated with AEs	Heart failure, ischaemic heart disease, myocardial infarction, stroke, blindness, ulcer, amputation, nephropathy, severe hypoglycaemic event, cardiovascular death, MACE	Base case: UKPDS OM2; SA: none stated	After HbA1c 8.5% exceeded = add INS therapy to comparators, except for sitagliptin, which is also discontinued	Comparators continued
Malkin, 2021 [53]	Semaglutide-O vs. empagliflozin, sitagliptin, liraglutide	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, LDL, BMI, non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	None stated	Base case: UKPDS 68; SA: UKPDS 82	After HbA1c 7.5% exceeded = discontinue drug/comparator, replace with basal INS	Comparators discontinued
Risebrough, 2021 [70]	Semaglutide-O vs. dulaglutide,liraglutide,semaglutide-I	Yes	HbA1c, weight, systolic blood pressure, discontinuation associated with AEs, nausea, vomiting, hypoglycaemia, diarrhoea, ischaemic heart disease, myocardial infarction, congestive heart failure, stroke, revascularisation, proteinuria, serum creatinine, neuropathy, ulcer, amputation	Discontinuation associated with AEs, non-severe hypoglycaemic event, severe hypoglycaemic event, nausea, vomiting, diarrhoea	Non-severe hypoglycaemic event, severe hypoglycaemic event, ischaemic heart disease, myocardial infarction, congestive heart failure, stroke, amputation, blindness, renal failure	Base case: UKPDS OM1; SA: none stated	After 3 years = switch to INS	Comparators discontinued
Ehlers, 2022 [58]	Semaglutide-O vs. empagliflozin	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, BMI	Non-severe hypoglycaemic event, severe hypoglycaemic event, urinary tract infection, gastrointestinal	Myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, hemodialysis, peritoneal dialysis, renal transplant, non-severe hypoglycaemic event, severe hypoglycaemic event, laser treatment, cataract, blindness, neuropathy, ulcer, amputation, macular edema, urinary tract infection, proliferative retinopathy, background retinopathy	Base case: UKPDS 82 and 68; SA: none stated	After HbA1c 7.5% exceeded = add basal INS	Comparators continued
Ehlers, 2022 [59]	Semaglutide-I vs. empagliflozin	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, BMI, non-severe hypoglycaemic event, severe hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event, urinary tract infection, gastrointestinal	Myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, hemodialysis, peritoneal dialysis, renal transplant, non-severe hypoglycaemic event, severe hypoglycaemic event, laser treatment, cataract, blindness, neuropathy, ulcer, amputation, macular edema, urinary tract infection, proliferative retinopathy, background retinopathy	Base case: UKPDS 68 and 82; SA: none stated	After HbA1c 7.5% exceeded = add basal INS	Comparators continued
Ekhlasi, 2022 [83]	Dulaglutide vs. liraglutide	No	HbA1c, hypoglycaemia	Hypoglycaemia	Hypoglycaemia, nephropathy, retinopathy, myocardial infarction, stroke	None stated	No switch	No switch/NA
Eliasson, 2022 [60]	Semaglutide-O vs. empagliflozin, sitagliptin	Yes	HbA1c, systolic blood pressure, total cholesterol, LDL, HDL, triglycerides, BMI, heart rate, white blood cell count, estimated glomerular filtration rate, cardiovascular death, myocardial infarction, stroke, heart failure	Non-severe hypoglycaemic event, severe hypoglycaemic event	Background retinopathy, proliferative retinopathy, macular edema, severe vision loss, neuropathy, peripheral vascular disease, amputation, microalbuminuria, macroalbuminuria, end-stage renal disease, ischaemic heart disease, myocardial infarction, stroke, heart failure, non-severe hypoglycaemic event, severe hypoglycaemic event	Base case: UKPDS 82, Eastman [99], Bagust [100]; SA: UKPDS 68	After HbA1c 8.0% exceeded = stop comparators and initiate basal, after HbA1c 8.0% exceeded again bolus INS was added	Comparators discontinued
Franch-Nadal, 2022 [63]	Semaglutide-O vs. empagliflozin, sitagliptin, liraglutide	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, BMI, severe hypoglycaemic event, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Severe hypoglycaemic event, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Myocardial infarction, angina, heart failure, stroke, peripheral vascular disease, hemodialysis, peritoneal dialysis, renal transplant, non-severe hypoglycaemic event, severe hypoglycaemic event, cataract, blindness, neuropathy, amputation, gangrene, ulcer	Base case: UKPDS 68; SA: UKPDS 82	After HbA1c 7.5% exceeded = replace with basal INS	Comparators discontinued
Hu, 2022 [79]	Semaglutide-I vs. dulaglutide	Yes	HbA1c, weight, BMI, systolic blood pressure, heart rate	None stated	Ischaemic heart disease, myocardial infarction, heart failure, stroke, amputation, blindness, renal failure, ulcer, end-stage renal disease	Base case: UKPDS OM2; SA: none stated	After 5 years = substitute with basal INS	Comparators discontinued
Malkin, 2022 [61]	Semaglutide-O vs. empagliflozin, dulaglutide	Yes	HbA1c, systolic blood pressure, total cholesterol, HDL, BMI, non-severe hypoglycaemic event, severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event, severe nocturnal hypoglycaemic event, non-severe nocturnal hypoglycaemic event	Myocardial infarction, angina, heart failure, stroke, peripheral vascular disease, microalbuminuria, gross proteinuria, hemodialysis, peritoneal dialysis, renal transplant, background retinopathy, proliferative retinopathy, macular edema, severe vision loss, cataract, neuropathy, ulcer, amputation, non-severe hypoglycaemic event, severe hypoglycaemic event, severe nocturnal hypoglycaemic event, non-severe nocturnal hypoglycaemic event	Base case: UKPDS 68; SA: UKPDS 82	After HbA1c 7.5% exceeded = replace with basal INS	Comparators discontinued
Ruan, 2022 [81]	Semaglutide-I vs. dulaglutide	Yes	HbA1c, systolic blood pressure, total cholesterol, LDL, HDL, triglycerides, BMI, heart rate, white blood cell count, estimated glomerular filtration rate, ischaemic heart disease, myocardial infarction, stroke, heart failure, cardiovascular death, non-severe hypoglycaemic event, severe hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event	Background retinopathy, proliferative retinopathy, macular edema, severe vision loss, neuropathy, peripheral vascular disease, amputation, microalbuminuria, macroalbuminuria, end-stage renal disease, ischaemic heart disease, myocardial infarction, stroke, heart failure, non-severe hypoglycaemic event, severe hypoglycaemic event	Base case: UKPDS 68 and 82 and SweNDR, Fremantle	After 1 years = replace with basal INS	Comparators discontinued
Stafford, 2022 [73]	Semaglutide-I vs. canagliflozin	Yes	HbA1c, systolic blood pressure, total cholesterol, LDL, HDL, triglycerides, BMI, estimated glomerular filtration rate, non-severe hypoglycaemic event, severe hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event	Background retinopathy, proliferative retinopathy, macular edema, blindness, neuropathy, ulcer, amputation, microalbuminuria, macroalbuminuria, chronic kidney disease, end-stage renal disease, myocardial infarction, ischaemic heart disease, heart failure, stroke, peripheral vascular disease, non-severe hypoglycaemic event, severe hypoglycaemic event	Base case: UKPDS 82; SA: UKPDS 68	After HbA1c 8% reached = replace with basal INS, and after HbA1c 8% exceeded again, basal-bolus	Comparators discontinued
Viljoen, 2022 [62]	Semaglutide-I vs. dulaglutide	Yes	HbA1c, BMI	Set to 0	Myocardial infarction, angina, heart failure, stroke, peripheral vascular disease, hemodialysis, peritoneal dialysis, renal transplant, non-severe hypoglycaemic event, severe hypoglycemic event, cataract, blindness, neuropathy, amputation, gangrene, ulcer, microalbuminuria, gross proteinuria, background retinopathy, proliferative retinopathy, macular edema	Base case: UKPDS 68; SA: UKPDS 82	After 3 years = replace with basal INS	Comparators discontinued
Chien, 2020 [75]	No main vs. classes: MET, SU, DPP4, SGLT2, GLP1, Unspecified INS	Yes	HbA1c, weight, systolic blood pressure, total cholesterol, HDL, non-severe hypoglycaemic event, severe hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event, urinary tract infection, gastrointestinal	Ischaemic heart disease, myocardial infarction, congestive heart failure, stroke, amputation, blindness, end-stage renal disease, severe hypoglycaemic event	Base case: UKPDS 68; SA: none stated	Allocation to dual therapy after HbA1c 8.01% reached. Unclear when what threshold led to allocation to triple therapy or INS+MET	Comparators continued, then discontinued
Hu, 2021 [76]	Dapagliflozin + saxagliptin combination therapy vs. dapagliflozin monotherapy vs. saxagliptin monotherapy	Yes	HbA1c, systolic blood pressure, fasting plasma glucose, HDL, weight	Severe hypoglycaemic event, non-severe hypoglycaemic event, discontinuation associated with AEs	Ischaemic heart disease, myocardial infarction, congestive heart failure, stroke, blindness, end-stage renal disease, amputation, ulcer, hypoglycaemia, genitourinary infection, urinary tract infection	Base case: UKPDS OM2; SA: none stated	Not stated	No switch/NA
Lin, 2021 [78]	No main vs. linagliptin 5 mg, saxagliptin 5 mg, alogliptin 25 mg, sitagliptin 100 mg, and vildagliptin 50 mg	No	HbA1c	None stated	Myocardial infarction, stroke, congestive heart failure, atherosclerotic cardiovascular disease, cardiovascular death, end-stage renal disease, blindness, neuropathy, amputation	None stated	No switch	No switch/NA
Zupa, 2021 [71]	Empagliflozin vs. semaglutide-I	No	Heart failure, myocardial infarction, stroke, nephropathy, end-stage renal disease, all-cause mortality	Genitourinary infection	Heart failure, stroke, myocardial infarction, nephropathy, end-stage renal disease	None stated	No switch	No switch/NA
Chakravarty, 2018 [65]	Dapagliflozin vs. liraglutide, SU, DPP4, pioglitazone	Yes	HbA1c, weight, systolic blood pressure, hypoglycaemia	Hypoglycaemia	None stated	None stated	NA	No switch/NA
Neslusan, 2018 [66]	Canagliflozin vs. dapagliflozin	Yes	HbA1c, systolic blood pressure, BMI, total cholesterol, LDL, HDL, triglycerides, non-severe hypoglycaemic event, severe hypoglycaemic event, male genital mycotic infection, female genital mycotic infection, lower urinary tract infection, upper urinary tract infection, volume depletion-related AEs, osmotic diuresis-related AEs, discontinuation associated with AEs	Non-severe hypoglycaemic event, severe hypoglycaemic event, male genital mycotic infection, female genital mycotic infection, lower urinary tract infection, upper urinary tract infection, volume depletion-related AEs, osmotic diuresis-related AEs, discontinuation associated with AEs	Myocardial infarction, ischaemic heart disease, congestive heart failure, stroke, peripheral vascular disease, CVD, proliferative retinopathy, background retinopathy, macular edema, blind one eye, blind two eyes, neuropathy, ulcer, amputation, macroalbuminuria, microalbuminuria, chronic kidney disease in stages3a–5, end-stage renal disease	Base case: UKPDS 82 and 68, CDC, WESDR, REP; SA: none stated	INTENSIFY after: >7% HbA1c/>140 SBP/>2.6 LDL (comparators discontinued)	Comparators discontinued
Hou, 2019 [74]	Canagliflozin vs. dapagliflozin	No	HbA1c, systolic blood pressure, total cholesterol, HDL	Urinary tract infection, gastrointestinal	Myocardial infarction, stroke, congestive heart failure, end-stage renal disease, blindness, neuropathy, amputation	Base case: COMT model	No switch	No switch/NA
Ramos, 2019 [40]	Empagliflozin vs. sitagliptin, saxagliptin	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, estimated glomerular filtration rate, non-severe hypoglycaemic event, severe hypoglycaemic event, genitourinary infection	Non-severe hypoglycaemic event, severe hypoglycaemic event, genitourinary infection	Myocardial infarction, angina, peripheral vascular disease, stroke, congestive heart failure, microalbuminuria, gross proteinuria, end-stage renal disease, background retinopathy, proliferative retinopathy, macular edema, severe vision loss, cataract, ulcer, amputation, neuropathy, non-severe hypoglycaemic event, severe hypoglycaemic event, hemodialysis, peritoneal dialysis, laser treatment, renal transplant, genitourinary infection, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event, gangrene	Base case: UKPDS 82; SA: none stated	All switch to basal bolus when 8.5% HbA1c reached	Comparators discontinued
Ramos, 2020 [45]	Empagliflozin vs. SoC and liraglutide	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, estimated glomerular filtration rate, non-severe hypoglycaemic event, severe hypoglycaemic event, severe hypoglycaemic event	Non-severe hypoglycaemic event, severe hypoglycaemic event, severe hypoglycaemic event, genitourinary infection	Myocardial infarction, angina, heart failure, stroke, peripheral vascular disease, microalbuminuria, gross proteinuria, hemodialysis, peritoneal dialysis, renal transplant, background retinopathy, proliferative retinopathy, macular edema, severe vision loss, cataract, neuropathy, ulcer, amputation, severe nocturnal hypoglycaemic event, severe hypoglyceemic event, severe nocturnal hypoglycaemic event, severe hypoglycaemic event, laser treatment, blindness, gangrene	Base case: UKPDS 82 and 68, ADVANCE, SweNDR, ARIC, Fremantle, PROCAM; SA: none stated	After HbA1c 8.5% exceeded = meal INS (57 IU) and INS glargine (94 IU) started, uncertain if liraglutide and empagliflozin and MET still on	Comparators discontinued
Ramos, 2020 [46]	Empagliflozin vs. semaglutide-O	Yes	(1) HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, BMI; (2) HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, BMI, heart failure	Severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event, severe hypoglycaemic event, severe hypoglycaemic event	Myocardial infarction, angina, heart failure, stroke, peripheral vascular disease, microalbuminuria, macroalbuminuria, hemodialysis, peritoneal dialysis, renal transplant, background retinopathy, proliferative retinopathy, macular edema, severe vision loss, cataract, neuropathy, ulcer, amputation, non-severe nocturnal hypoglycaemic event, severe nocturnal hypoglycaemic event, non-severe hypoglycaemic event, severe hypoglycaemic event, laser treatment, blindness	Base case: UKPDS 68, UKPDS 82; SA: none stated	After HbA1c 7.5% exceeded = basal INS started, semaglutide-I and empagliflozin and MET still on	Comparators continued
Reifsnider, 2020 [69]	Empagliflozin vs. sitagliptin	Yes	(1) Without CVD: HbA1c, weight, systolic blood pressure (2) With CVD: Risk equations from EMPA-REG OUTCOME	Urinary tract infection, genital mycotic infection, upper respiratory tract infection, nasopharyngitis, headache	Myocardial infarction, stroke, heart failure, ischaemic heart disease, blindness, amputation, renal failure, cardiovascular death, transient ischaemic attack, revascularisation, macroalbuminuria, renal injury, angina, ulcer	Base case: UKPDS OM2; SA: none stated	Addition of either sitagliptin or empagliflozin as third-line, and INS as third-line. Rates from Montvida [101]	Comparators continued
Van der Linden, 2020 [54]	Dapagliflozin vs. DPP4	Yes	HbA1c, systolic blood pressure, total cholesterol, weight, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, non-severe hypoglycaemic event, urinary tract infection, gastrointestinal, discontinuation in general	Non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, non-severe hypoglycaemic event, urinary tract infection, gastrointestinal, discontinuation in general	Ischaemic heart disease, myocardial infarction, congestive heart failure, stroke, amputation, blindness, end-stage renal disease, non-severe hypoglycaemic event, non-severe nocturnal hypoglycaemic event, severe hypoglycaemic event, urinary tract infection, gastrointestinal	Base case: UKPDS68; SA: none stated	After HbA1c 8% exceeded = replace comparators with basal INS; when that fails, add bolus minus SU	Comparators discontinued
Ehlers, 2021 [56]	Empagliflozin vs. liraglutide	Yes	HbA1c et al. not reported, cardiovascular death, all-cause mortality, composite endpoint, heart failure, stroke, myocardial infarction	Non-severe hypoglycaemic event, severe hypoglycaemic event, severe nocturnal hypoglycaemic event, non-severe nocturnal hypoglycaemic event, genitourinary infection	Myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, hemodialysis, peritoneal dialysis, renal transplant, non-severe hypoglycaemic event, severe hypoglycaemic event, laser treatment, cataract, blindness, neuropathy, amputation, gangrene, ulcer, microalbuminuria, end-stage renal disease, background retinopathy	Base case: UKPDS 82; SA: none stated	After 9 years (HbA1c = 8.5%), both arms switched to basal-bolus therapy	Comparators discontinued
Gourzoulidis, 2021 [55]	Empagliflozin vs. dapagliflozin	No	Cardiovascular death, myocardial infarction, stroke, composite renal outcome, genital mycotic infection, acute kidney injury, severe hypoglycaemic event	Severe hypoglycaemic event, genital mycotic infection, acute kidney injury	Myocardial infarction, stroke, heart failure, cardiovascular death, composite renal outcome, severe hypoglycaemic event	None stated	No switch	No switch/NA
Ramos, 2021 [77]	Empagliflozin vs. liraglutide, sitagliptin	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, LDL, triglycerides, BMI, estimated glomerular filtration rate, non-severe hypoglycaemic event, severe hypoglycaemic event, genitourinary infection	Non-severe hypoglycaemic event, severe hypoglycaemic event, genitourinary infection	Myocardial infarction, angina, congestive heart failure, stroke, peripheral vascular disease, hemodialysis, peritoneal dialysis, renal transplant, non-severe hypoglycaemic event, severe hypoglycaemic event, laser treatment, cataract, blindness, neuropathy, amputation, gangrene, ulcer	Base case: UKPDS 82; SA: none stated	After exceeding threshold: HbA1c 8.5% = replace therapy with basal-bolus INS	Comparators discontinued
Reifsnider, 2021 [67]	Empagliflozin vs. dapagliflozin, canagliflozin	No	Cardiovascular death, myocardial infarction, stroke, heart failure, progression of albuminuria, composite renal outcome, genital mycotic infection, acute kidney injury, amputation, bone fracture, severe hypoglycaemic event	Severe hypoglycaemic event, genital mycotic infection, acute kidney injury, amputation, bone fracture	Cardiovascular death, myocardial infarction, stroke, heart failure, angina, transient ischaemic attack, revascularisation, progression of albuminuria, composite renal outcome, genital mycotic infection, acute kidney injury, amputation, bone fracture, severe hypoglycaemic event	Inhouse, based on EMPA-REG OUTCOME	No switch	No switch/NA
Lasalvia, 2022 [82]	Dapagliflozin vs. DPP4	Yes	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, haematocrit,weight	Non-severe hypoglycaemic event, severe hypoglycemic event, genitourinary infection	Ischaemic heart disease, myocardial infarction, congestive heart failure, stroke, amputation, blindness, end-stage renal disease, ulcer, non-severe hypoglycaemic event, severe hypoglycaemic event, genitourinary infection	Base case: UKPDS 68; SA: UKPDS 82	After HbA1c 8.3% or 8.1% = ‘receive rescue therapy with basal INS’	Comparators discontinued
Peng, 2022 [80]	SGLT2 vs. DPP4	No	HbA1c, systolic blood pressure, diastolic blood pressure, total cholesterol, HDL, hematocrit, weight	Unclear	Myocardial infarction, heart failure, stroke, all-cause mortality	None stated	No switch	No switch/NA
Reifsnider, 2022 [72]	Empagliflozin vs. liraglutide	Yes	HbA1c, weight, systolic blood pressure, cardiovascular death, myocardial infarction, stroke, heart failure, macroalbuminuria, renal injury, renal failure	Genital mycotic infection, urinary tract infection, nausea, hypoglycaemia, injection site reaction	Myocardial infarction, stroke, heart failure, ischaemic heart disease, blindness, ulcer, amputation, renal failure, cardiovascular death, angina, revascularisation, macroalbuminuria, renal injury, renal failure, hypoglycaemia	Base case: UKPDS 68 and 82; SA: none stated	Addition of either liraglutide or empagliflozin as third-line, and INS as fourth-line. Rates from Montvida [101]	Comparators continued

AE adverse events, ARIC Atherosclerosis Risk in Communities, ADVANCE Action in Diabetes and Vascular Disease: preterax and diamicron-MR controlled evaluation, BMI body mass index, CDC Centers for Disease Control and Prevention, COMT Chinese Outcomes Model for T2DM, CVD cardiovascular disease, DPP4 dipeptidylpeptidase-4 inhibitors, Fremantle The Fremantle Diabetes Study, GLP1 glucagon-like peptide-1 receptor agonist, HDL high-density lipoprotein, IDegLira combination of liraglutide and INS degludec, INS insulin, LDL low-density lipoprotein, MACE major adverse cardiac events, MDI multiple daily injections, MET metformin, NA not available, PROCAM Prospective Cardiovascular Münster, QW every week, REP Rochester Epidemiology Project, SA sensitivity analysis, SoC standard of care, SU sulfonylurea, SGLT2 sodium-glucose cotransporter-2 inhibitors, SweNDR Swedish National Diabetes Register, UKPDS UK Prospective Diabetes Study, UKPDS OM UK Prospective Diabetes Study outcomes model, WESDR Wisconsin Epidemiologic Study of Diabetic Retinopathy

Source of Baseline Characteristics of the Modelled Cohorts

The baseline characteristics of the modelled cohorts were sourced from randomised controlled trials (RCTs) in most studies (42/50), some of which focused on cardiovascular outcomes. Real-world data from observational studies or data from other CEAs were only used in eight studies [39, 50, 64, 65, 69, 72, 80, 83]. Baseline tables describing the characteristics of the modelled cohort were unavailable in one study [38, 39, 49, 50, 67, 71, 74, 78, 80, 83]. The characteristics of the cohorts in the studies were most frequently sourced from the SUSTAIN 7 [92] (n = 10) and PIONEER 2 (n = 10) trials [93], followed by EMPA-REG OUTCOME [94] (n = 7), and PIONEER 3 (n = 5) [95]. These trials tested the efficacy of injectable semaglutide, oral semaglutide, and empagliflozin.

However, there was considerable variation in the patient eligibility criteria across these trials. For example, SUSTAIN 7 excluded patients with HF, CKD, and retinopathy, while PIONEER 2 excluded patients with renal impairment, retinopathy, and pancreatitis, and EMPA-REG OUTCOME only included patients with established CVD (see Table 3 for more details on which trials were used as the basis for the modelled cohorts).

Subgroup Analysis

Explicit subgroup analysis was performed in 11 studies. Four studies analysed a subgroup with body mass index (BMI) ≥30 [41–43, 53]; two of these studies performed their base-case analysis on a cohort with BMI ≥35 [41, 42], while the other two performed a separate subgroup analysis on individuals with BMI ≥30 [43, 53]. A subgroup analysis on individuals with and without a history of DRCs, primarily CVD, was performed in four studies [69, 71, 72, 80]. One study used four different subgroups from the EXTRA study, and the two remaining studies used a cohort with different characteristics than the base-case cohort to test their results [38, 75].

Model Inputs: Treatment Effect and Adverse Events

The treatment effect and AEs were sourced from multiple sources, with RCTs being the most common (n = 36), followed by network meta-analyses (n = 16), indirect treatment comparisons (n = 8), or other CEAs (n = 6). The studies frequently used treatment effects from the SUSTAIN 7 (n = 9), PIONEER 2 (n = 9), EMPA-REG OUTCOME (n = 8), and PIONEER 3 (n = 6) trials.

The most frequently used treatment effects were changes in HbA1c (n = 47), systolic blood pressure (n = 41), total cholesterol (n = 33), and high-density lipoprotein (n = 33). Similar treatment effects were used between the studies, although those with SGLT2 as the main comparator had a broader range of treatment effects than those with GLP1 as the main comparator. Five studies with empagliflozin as the main comparator [46, 55, 56, 67, 72] and four studies with oral semaglutide [60, 64, 70, 81] used treatment effects specific to the reduction of CVD as an additional separate part of the treatment effect parameters. One study comparing the DPP4 class of products also included CVD-related effects [71]. Studies with GLP1 as the main comparator focused more on hyperglycaemia than those with SGLT2.

Most studies included drug-related AEs, and the most frequently used were variations of hypoglycaemic events: severe (n = 35), non-severe (n = 33), severe nocturnal (n = 13), and non-severe nocturnal (n = 12). Only one study, with GLP1 as the main comparator, included diabetic ketoacidosis [52]. For more information on treatment and adverse effects, see Table 3; for the sources of model input, see Online Resource Appendix A2.

Key Model Assumptions

Two key model assumptions used in the included studies are reported in this section: (1) the risk equations used to predict the development of DRCs over time; and (2) the time until, and conditions leading to, the treatment switch. There were notable differences in how these assumptions were applied in the studies (see Table 3 for details).

The UKPDS 68 [96] and 82 [88] risk equations were the most frequently used. However, there were differences in what risk equations were used in the base-case analysis, which depended on the class of the main comparator. Among the 27 studies using UKPDS 68 in the base-case scenario, 20 studies had GLP1 as the main comparator [34, 37–39, 41, 42, 47, 48, 50–53, 57–59, 61–63, 68, 81] and six studies had SGLT2 as the main comparator [45, 46, 54, 66, 72, 82], while one study [75] had no main comparator. Of the 20 times UKPDS 82 was used in the base-case, eight studies had SGLT2 as the main comparator [40, 45, 46, 56, 66, 67, 69, 72, 77], 10 studies had GLP1 as the main comparator [35, 39, 58–60, 64, 68, 73, 79, 81], and one study [76] had no main comparator. In eight of the studies mentioned above, both risk equations were used for the base-case; half of these studies had GLP1 as the main comparator [58, 59, 68, 81] and the other half had SGLT2 as the main comparator [45, 46, 66, 72]. GLP1 was the main comparator in all of the 18 studies that used these risk equations in the SA, with three using UKPDS 68 [60, 68, 73], 13 using UKPDS 82 [37–39, 41, 47, 48, 51, 53, 57, 61–63, 82], and two using both [43, 49]. Only one study with SGLT2 as the main comparator used UKPDS 82 in the SA [82].

Explicit reasoning for the choice of risk equation was only present in a few cases. Three studies with SGLT2 as the main comparator cited model fit as the reason [40, 45, 56], while eight GLP1 studies cited advice from model proprietors as the reason [37, 38, 47, 49, 57, 61–63]. Different aspects of the risk equations were sometimes used simultaneously. For example, in the study by Neslusan et al. [66], the change in HbA1c, systolic blood pressure (SBP), and lipids was modelled using UKPDS 68, macrovascular complications were modelled using UKPDS 82, and microvascular complications were modelled using a combination of four different risk equations [97–100]. Six studies either did not use risk equations or had unclear reporting of their use [55, 65, 71, 78, 80, 83]. For details on the risk equations used, see Table 3.

Regarding treatment switch, it occurred either at a predefined period after treatment initiation of the main comparators [36–39, 41–43, 47, 49, 62, 68–70, 72, 79, 81], when the HbA1c levels of the cohort reached a certain threshold [34, 35, 40, 46, 48, 50–52, 54, 57–60, 62–64, 66, 73, 75, 82], or a combination of the two [44, 45, 53, 56, 77]. These variations in treatment switch occurred in 16, 20, and 5 studies, respectively. The time to treatment switch was shorter for studies with GLP1 as the main comparator (n = 20, mean = 3.95 years, standard deviation = 2.7 years) than for the SGLT2 studies (n = 3, mean = 10 years, standard deviation = 2.6 years). At the point of treatment switch, the comparators were either discontinued and replaced with a different medication, often INS [34–45, 47–51, 53, 54, 56, 57, 60–63, 66, 68, 70, 73, 76, 77, 81, 82], continued while different medications could be added [46, 58, 59, 64, 69, 72], or continued for one step and then discontinued at a later step [52, 75], as reported in 33, 6, and 2 studies, respectively. Studies with GLP1 as the main comparator had the comparators discontinued 84% of the time, while those with SGLT2 had the comparators discontinued 47% of the time. Nine did not report a treatment switch [55, 65, 67, 74, 76].

Model Outputs—Diabetes-Related Complications

In the conducted CEAs, the most common output related to DRC was a summary of incidence or onset, most frequently both. Only two studies did not include DRCs as the output [53, 65], while the majority had similar items within the category. Overall, the difference in DRC outputs between the studies was small when present. SGLT2 studies reported HF and cardiovascular death more often and had more varied outputs relating to nephropathy, while GLP1 studies had more varied outputs regarding hypoglycaemia and retinopathy (see Table 3 for all extracted data).

The following categories represented the DRCs: CVD, nephropathy, neuropathy, retinopathy, AEs, and others. The most-reported DRC category was CVD, represented by stroke, myocardial infarction, HF or congestive HF, peripheral vascular disease, angina, ischaemic heart disease, CV death, composite major adverse cardiac events, and transient ischaemic attack. The nephropathic complications were represented by microalbuminuria, end-stage renal disease, haemodialysis, renal transplant, peritoneal dialysis, gross proteinuria, macroalbuminuria, composite renal outcome, renal injury, renal failure, progression of albuminuria, nephropathy, and acute kidney injury. Similar to the treatment effects, AEs as the model output were almost exclusively hypoglycaemia, divided into subcategories of severe and non-severe, and sometimes further divided into non-nocturnal and nocturnal. AEs associated with certain classes were also present, such as genitourinary infection for the SGLT2 class and gastrointestinal issues for GLP1. The neuropathy group consisted of amputation, ulcer, and neuropathy. Macular oedema, blindness, cataract, vision loss, background and proliferative retinopathy, and non-proliferative retinopathy were included in the retinopathy group, whereas laser treatment, gangrene, and revascularisation were included in the ‘others’ category.

Quality of the Included Studies

The answers to the essential questions of the quality assessment checklist are provided in Table 4. Answers to questions 1, 3, 6, 7, and 8 were adequate in almost all studies; however, answers to questions 2, 4, 5, 9, and 10 had some inadequacies. The description of comparators (question 2) was frequently partially adequately described due to the omission of relevant comparators to the research question (e.g., GLP1 products were compared only with GLP1 products and no SGLT2 products). Additionally, most studies lacked a societal perspective in their analyses (question 4). The accuracy of cost and consequence measurement (question 5) was often unclear due to the use of data that may not represent current treatment practice (i.e., the UKPDS risk equations) and the frequent use of short times to treatment switch, which may not reflect modern treatment practice. The heterogeneity of patient groups was rarely explored when the studies addressed uncertainty (question 9). Finally, the discussion sections (question 10) had inadequacies, mainly regarding comparison with other studies and generalisability.

Discussion

Main Findings

This systematic review aimed to provide an overview of the cost-effectiveness outcomes and the underlying methodological choices of the most recent studies published. In the included studies, SGLT2 and GLP1 products were the comparators of choice among the nNIADs. Empagliflozin and semaglutide (oral and injectable) were the most frequently used main comparators and were found to be cost effective compared with other products within their respective classes. This aligns with the recent ADA reports recommending using products from the SGLT2 and GLP1 classes based on the patient’s risk level and treatment goals [15, 16]. However, only 38% of studies compared SGLT2 and GLP1 directly. The lack of direct comparisons could be due to a lack of direct comparison data, although several of the included studies have performed indirect treatment comparisons or network meta-analyses, or used existing comparisons from the literature.

The cost effectiveness findings can be seen from both a class and a product level. From the class-level perspective, one class was not favoured over the other, but GLP1 never dominated SGLT2. On the other hand, SGLT2 dominated GLP1 in six comparisons. The dominance of SGLT2 over GLP1 seems limited to the older GLP1 products, as only one of the dominant results was between empagliflozin and the newer line of GLP1 products, oral semaglutide [46]. From the product perspective, semaglutide (oral or injectable) was more frequently the main comparator against empagliflozin (n = 11) than vice versa (n = 2) and was cost effective in the majority of these comparisons.

Sponsorship bias may be inferred from the fact that all industry-funded studies found their product to be cost effective or dominating, except for two studies that compared their competitors’ product against their own (i.e. injectable and oral semaglutide were compared with empagliflozin), and found the competitor to be not cost effective [58, 59]. In the non-industry-sponsored studies where SGLT2 and GLP1 were compared, oral semaglutide was found to be cost effective against all comparators except empagliflozin [64]. In contrast, the other study found injectable semaglutide to be cost effective against empagliflozin [71].

There were considerable regional differences in the choice of main comparators, with Europe favouring GLP1, the Asian region favouring SGLT2 slightly more, and the North American region seemed to favour both equally. Most included studies were conducted in Europe, possibly explained by its increased focus on prioritising healthcare resources. However, Asia and North America may benefit more from increased prioritisation of healthcare resources, as the current and projected incidence of T2DM is much higher than in the European region [1].

It can be seen as a shortcoming that only 11 studies explicitly defined the subgroups they modelled. Presenting results for different subgroups might help differentiate SGLT2 and GLP1 products more. Emphasising the differences in their respective effects might be especially important since the Danish Medicines Counsel has adopted the simplified assumption that their effects are equivalent [102]. The modelled cohorts were almost exclusively in a broad category described as patients with T2DM uncontrolled on one or several first-line medications. Most studies with empagliflozin as the main comparator based their cohort on the EMPA-REG OUTCOME trial, which only included patients with established CVD and focused more on explicitly incorporating the reduction of cardiorenal DRCs as separate treatment-effect parameters. Including the effect as separate parameters could lead to more accurate models since GLP1 and SGLT2 have demonstrated a reduction of DRCs that occur independent of the antihyperglycaemic effect [15]. However, one of the included studies argued that including both the direct and indirect effects carries the risk of double-counting, which could overestimate the effect [61]. The majority of the included studies focused on translating the differences between the comparators in terms of how they changed common physiological markers (HbA1c, SBP, cholesterol, etc.) into differences in DRCs over time. Some have critiqued the approach of translating these frequently minor differences into large effects over a time horizon that might be too long [25].

Regarding differences in the inclusion of AEs, this review highlighted the omission of diabetic ketoacidosis as an AE when SGLT2 was a comparator. It is a rare but well-established and potentially deadly AE, especially for INS-dependent T2DM [15, 94, 103–107], that included only one study [52] with GLP1 as the main comparator included. These differences between the models, which in some cases seem to be dependent on the class of the main comparator, are problematic, and decision makers need to pay special attention to this, as the results of the economic analysis could be greatly affected by them.

Most studies (48%) of DAMs used in the health economic evaluations were based on the CDM from the IQVIA™ [84, 85]. Still, the high representation of this model does not necessarily reflect its superiority compared with other models available. Studies with GLP1 as the main comparator use this model 61% of the time, while those with SGLT2 use it 33% of the time. The Mount Hood meetings exemplify that there are several complex simulation models specific to diabetes, each with own its strengths and weaknesses [108–111].

Development of the physiological parameters over time in the context of the disease is primarily governed by risk equations that represent assumptions about the progression of the disease. The present review focused on which risk equations were used and whether it was used for the base-case analysis or the SA. The most commonly used risk equations were UKPDS 68 and 82, which are based on the UKPDS study [96] from 1977 to 1997, while UKPDS 82 uses 10 years of follow-up beyond that [88]. While these risk equations are currently widely used, contemporary studies may be able to produce more accurate risk equations that better reflect modern treatment practices. The choice of risk equation seemed connected to the class of the main comparator, and the reasoning behind the choice was rarely explicit in the studies. The GLP1 studies that were explicit in their reasoning cited recommendations from the model proprietors (i.e. of the CDM). The authors of this paper could not find any such recommendations in the model validation paper, where the UKPDS 82 has been incorporated since 2014 [88]. The three studies with SGLT2 cited model fit, and reference to model proprietor recommendations was absent, even though they also used the CDM. We recommend that the reasoning behind the choice of risk equations is always stated explicitly, as it may be difficult to explain why using UKPDS 68 was the better choice since the UKPDS 82 risk engine is an updated and improved version based on a larger dataset over an extended period, with more significant predictors and event types, and, in general, predicted fewer DRC events [88].

Another key assumption relates to the time from initiation of the comparators in the study until they are either switched out, additional medication is added, or are replaced by rescue therapy. Studies with GLP1 as the main comparator and predefined time to switch often cited a report that states that the mean treatment duration of GLP1 was 29.35 months [112]. However, treatment guidelines known to the authors of this paper recommend lifelong treatment with the nNIADs, and only stopping if their use was contraindicated. Models that use a short preset time-to-treatment switch where the comparator was replaced with INS might not be representative of real-world practice, and the assumption might lead to misinforming decision makers about the long-term consequences of utilising that comparator.

The quality of the included studies was, in large part, adequate. The inadequate answers were about the omission of relevant comparators, lack of societal perspective, and the accuracy of measuring costs and consequences. These quality issues may however be a limitation of available data and not a lack of consideration from the authors of the included studies. However, the methods used in other studies were rarely discussed, except for some recent studies [58, 59, 61–63, 72, 73], where the authors directly commented on the advantages and disadvantages of their own and others’ methods.

Connecting Main Findings to Other Studies

Other reviews in this area find that SGLT2 and GLP1 appear to be cost effective for patients with T2DM uncontrolled on MET and other background oNIADs. Ruan et al. [30] focused on DPP4 and found it to be cost effective as a second-line treatment compared with sulfonylurea (SU) and INS, but not SGLT2 and GLP1. Bagepally et al. [28] found GLP1 to be cost effective compared with SU, DPP4, and TZD in high-income countries, while Bagepally et al. [29] found SGLT2 to be cost effective against SU, but not DPP4, in high-income countries. Zozaya et al. [31], Rahman et al. [26], and Yoshida et al. [27] focused on SGLT2 and found the class to be cost effective against SU, TZD and DPP4. Additionally, Yoshida et al. found SGLT2 to be cost effective against α-glucosidase inhibitors, INS, and SoC. However, Yoshida et al. and Zozaya et al. cautioned that the heterogeneity of the studies included in their studies made it challenging to determine in which treatment scenarios nNIADs were most cost effective.

The lack of CEAs or systematic reviews focusing on DPP4 seemingly conflicts with a global report regarding NIAD market shares that found DPP4s to be the most commonly used nNIAD, followed by GLP1 and SGLT2 [113].

Limitations of the Study

The present review has several limitations. The literature search was restricted to a brief period, and it could be argued that not including non-DAM methodologies is a limitation as valuable information was lost, even if DAMs are considered the best tool to inform decision makers. It could also be argued that systematic reviews should always strive to provide a unified conclusion based on the literature instead of a descriptive overview and broad qualitative synthesis of the results. However, the diversity of the methodological choices, settings, and the high risk of sponsorship bias studies, makes attempting to provide a conclusion on which nNIADs are cost effective challenging to interpret for decision makers. Only including studies that compare nNIADs with other nNIADs was also a limitation. Previous reviews conclude that nNIADs were cost effective compared with oNIADs; however, given the high cost of nNIADs, this is probably limited to second-line treatment. One study [114] has estimated that SGLT2 and GLP1 need a price reduction of 70% to be cost effective compared with MET. The choice of quality assessment checklist in the present study could be considered a limitation as it was not model-specific; however, model specifics are reported as part of the main data extraction. An additional limitation is excluding non-English-language papers since this could likely have excluded many papers from the Asian region. Finally, it is a limitation of this review that we included all published papers as separate studies without discussing whether some papers should be interpreted as adaptations of the same modelling study to different settings. This distinction is not trivial however but underlines the same conclusion that simply counting the number of published papers in favour of a certain conclusion about cost effectiveness does not necessarily count as ‘more evidence’ of this finding.

Strengths of the Study

The present review has some key strengths. Extracting a broad range of information about the assumptions and input and output parameters of DAMs can provide valuable insight into the underlying methodological choices. In a field as complicated as modelling the cost effectiveness of T2DM treatment, it might be beneficial to go beyond only reporting a combination of the economic outcomes and using a checklist that mainly evaluates the included paper’s quality on an ordinal scale of high to low quality. More information for decision makers might provide a better foundation for making decisions.

Suggestions for Future Research

More studies should be conducted directly comparing the products of the SGLT2 and GLP1 classes. Additionally, the combination of SGLT2 and GLP1 as a single treatment intervention was not examined in any of the included studies, but combining these two highly effective drugs using different biological pathways [15] may be worth exploring. However, the most benefit could be gained from conducting more studies for middle- to low-income countries due to their higher prevalence and incidence of T2DM. If the Danish Medicines Counsel is right in their assumption that the SGLT2 and GLP1 can be seen as equivalent in effect [102], SGLT2 is likely to be a good choice for these countries.

However, concluding the cost effectiveness of nNIADs for treating T2DM must be cautiously approached due to the differences between the studies, lack of transparency, and the sponsorship bias outlined in the present and previous reviews. We advise future reviews aiming to conclude on this to take this into account and to specify in what context their conclusion applies. For systematic reviews to be better able to derive conclusions, we suggest establishing better practices and consensus for conducting CEAs evaluating the cost effectiveness of T2DM treatments with a DAM. The central clinically relevant model assumptions, parameters, and comparators must be included, and the explicit reasoning for choices must be provided and presented as clearly and transparently as possible. Implementing a base-case cohort in DAMs, based on a representative sample of the population large enough to examine the effect on different subgroups, could greatly benefit the ability of these studies to reflect the consequences of more usage of the nNIADs in a real-world context in contrast to the highly selected populations of the RCTs used in the included studies. Finally, more observational research is needed into the different aspects of the treatment switch assumption regarding the time until INS is added to, or replaces, the nNIADs and the proportion of addition compared with substitution. Involving clinicians could aid in addressing many of these issues, especially in clarifying and validating the underlying methodological choices in terms of how they reflect and can be transferred to a real-world context and understood by decision makers.

Conclusions

There are several challenges associated with drawing conclusions about which of the nNIADs is the cost-effective option for the second-line treatment of T2DM. First, the data used to model disease progression may not reflect modern treatment practice. Second, the lack of reasoning behind key methodological choices, some of which may conflict with treatment guidelines. Third, the assumptions regarding the time until changes to the comparators may not be representative of current practice. Finally, some methodological differences seem to depend on the class of the main comparator in the analysis, and sponsorship bias may also affect results.

Supplementary Information

Below is the link to the electronic supplementary material.

Declarations

Compliance with ethics guidelines

This article is based on previous studies and does not involve any new studies of human or animal subjects performed by the authors.

Funding

This study was supported by Boehringer Ingelheim, who had no role in the design, analysis, or interpretation of the results in this study. Boehringer Ingelheim was given the opportunity to review the manuscript for medical and scientific accuracy as it relates to Boehringer Ingelheim’s substances and intellectual property considerations.

Conflicts of Interest

Peter Vestergaard is the head of research at Steno Diabetes Center North Denmark, funded by the Novo Nordisk Foundation. Lars Holger Ehlers has received funding for research and compensation for lecture activities from Boehringer Ingelheim. Henrik Vitus Bering Laursen has received partial funding for his PhD from Steno Diabetes Center North Denmark, which in turn is funded by the Novo Nordisk Foundation, and additional funding from Boehringer Ingelheim. Emmelie Ploug Jørgensen declares that she has no conflicts of interest.

Author contributions

HVBL was responsible for the conception of the study question and the search strategy, along with collecting, storing, and analysing data, writing the manuscript, and screening the included studies. EPJ contributed to the screening of studies, collecting data, writing and commenting on the manuscript, and providing interpretations of the results. LHE and PV provided ideas for the conception of the study question, generated ideas regarding focus areas of the analysis, provided comments and corrections to the manuscript, and provided feedback throughout the writing period. All authors read and approved the final manuscript.

Data availability statements

The data collected and generated from the included studies can be sent upon request to the corresponding author.