Benefit-risk assessment of incretin and other anti-diabetic agents in type 2 diabetes using a stochastic multicriteria acceptability analysis model.

Abstract

To the Editor: Incretin-based therapies for type 2 diabetes mellitus (T2DM) include incretin mimetics of glucagon-like peptide-1 receptor agonists (GLP-1 RAs) and incretin enhancers of dipeptidyl peptidase-4 (DPP-4) inhibitors.[1] With good hypoglycemic effects of incretin-based drugs that show no weight gain or hypoglycemia risk, these drugs are increasingly used in patients with T2DM. GLP-1 RAs are considered superior to DPP-4 inhibitors in controlling glycosylated hemoglobin, fasting blood glucose, and body weight.[2] However, the incidence of adverse events, such as adverse gastrointestinal reactions and dizziness caused by GLP-1 RAs, is higher than it is for DPP-4 inhibitors.[3] Unlike other anti-diabetic agents, safety concerns have been raised regarding the risk for gastrointestinal cancer associated with incretin-based treatments. Both a better understanding of their roles as second-line glucose-lowering treatments and a comprehensive assessment of their benefits and harms could inform the choice of treatment in clinical practice. Multicriteria decision analysis (MCDA), a general framework for constructing multicriteria decision models for benefit-risk assessment (BRA), has been widely used in health management and drug evaluation.[4,5] The stochastic multicriteria acceptability analysis (SMAA) model was developed from the traditional MCDA model to reduce the impact of the value preferences of decision-makers without requiring them to give subjective weight to decision-making indicators.[6,7] Previous studies have reviewed the multiple benefit and risk outcomes of GLP-1 RAs and DPP-4 inhibitors, but none have adopted the BRA evaluation model to synthesize multiple outcomes and obtain comprehensive comparison results. In this study, SMAA and network meta-analysis (NMA) were used to analyze the BRA of incretin-based treatments with other anti-diabetic agents to provide more comprehensive evidence for the clinical use of anti-diabetic agents in the treatment of T2DM. The risk/benefit outcomes were identified through reference to previous reviews, NMA, and drug information published by the Food and Drug Administration on incretin-based therapies. These included 26 outcomes, as shown in Figure 1 and Supplementary Table 1, https://links.lww.com/CM9/B446. Medline, Embase, ClinicalTrials.gov, and the Cochrane Library were searched from their inception to March 29, 2019. We used “Glucagon-Like Peptide-1 Receptors” and “Dipeptidyl-Peptidase IV Inhibitors” as keywords or MeSH terms, accompanied by EMTREE terms and relevant free words to search these databases.Figure 1: Value map of the BRA index of incretin-based therapies. Network plot presenting the trial data contributing evidence comparing incretin-based therapies for outcomes. (A) FPG, HbA1c, PPG, weight; (B) HDL, LDL, TC, TG; (C) DBP, heart rate, SBP; (D) Constipation, diarrhea, dyspepsia, gastroenteritis, nausea, vomiting; (E) All-cause death, hypertension, hypoglycemia, MACE, pancreatitis; (F) Arthralgia, cancers of digestive system, dizziness, headache. AGI: Alpha-glucosidase inhibitor; BRA: Benefit-risk assessment; DBP: Diastolic blood pressure; DPP-4: Dipeptidyl peptidase-4; FPG: Fasting plasma glucose; GLP-1 RAs: Glucagon-like peptide-1 receptor agonists; HbA1c: Glycosylated hemoglobin; HDL-C: High-density lipoprotein cholesterol; LDL-C: Low-density lipoprotein cholesterol; MACE: Major adverse cardiovascular events; Met: Metformin; PPG: Postprandial plasma glucose; SBP: Systolic blood pressure; SGLT-2: Sodium-glucose co-transporter 2; SU: Sulfonylureas; TC: Total cholesterol; TG: Triglyceride; TZD: Thiazolidinediones.The inclusion criteria were as follows: (1) reports written in English; (2) randomized clinical trials (RCTs); (3) the subjects were T2DM; (4) the intervention measures were GLP-1 RAs or DPP-4 inhibitors; (5) metformin (Met), sulfonylureas (SU), thiazolidinediones (TZD), alpha-glucosidase inhibitor (AGI), sodium-glucose co-transporter 2 (SGLT-2) inhibitor, insulin, or placebo as control measures; (6) GLP-1 RAs and DPP-4 inhibitors as control drugs; and (7) relevant indicators appearing in the study are included in the literature. Exclusion criteria were as follows: (1) reports not written in English; (2) non-RCTs; (3) the subjects were not T2DM; (4) animal research and other basic research; (5) no GLP-1 RAs and DPP-4 inhibitors in the intervention; and (6) ongoing or unfinished experimental studies. Data were extracted, including trial information (author, title, publication year, sample size, trial duration, types of intervention, and control), population characteristics (diabetes duration, age, baseline level of glycosylated hemoglobin A1c (HbA1c), background treatment), types of intervention and control measures, benefit and risk indicators, and relevant results. Two investigators (FS and SBC) extracted data independently in duplicate. The quality of studies was assessed using the Cochrane risk of bias tool (generation of random sequence, allocation hidden, blind method of the research object, results blind method, data integrity, selective reporting, and funded by company randomization). The Bayesian method was used to analyze each risk/benefit outcome. The measure of each risk/benefit outcome indicator was expressed as mean difference or odds ratio and its 95% confidence interval. The first main result of SMAA was acceptability, that is the probability of the scheme ranking in the evaluation system. The second main result was the confidence factor (CF). This refers to the probability that the alternatives rank first when the central weight vector is selected. Based on SMAA, the results of inconsistent comparison pairs in the NMA results were replaced with the direct comparison of the results of the meta-analysis, and indicators that could be related to each other in the outcome indicators were removed for sensitivity to test the stability of the model results. Of the 26 related BRA outcomes [Figure 1 and Supplementary Table 1, https://links.lww.com/CM9/B446], 11 were benefit indicators, including glycosylated HbA1c, fasting plasma glucose, postprandial plasma glucose, weight, blood lipids, blood pressure, and heart rate. In all, 15 were risk indicators, including all-cause death, pancreatitis, constipation, diarrhea, dyspepsia, gastroenteritis, nausea, vomiting, arthralgia, hypertension, cancers of the digestive system, dizziness, headache, hypoglycemia, and major adverse cardiovascular events. A total of 589 RCTs involving 295,908 patients with T2DM were included. In all, 34 RCTs (16,023 patients with T2DM) on GLP-1 RAs and DPP-4 inhibitors were head-to-head comparisons. In other comparisons, comparators included metformin, SU, TZD, AGI, SGLT-2 inhibitors, insulin, and placebo. The overall quality of the literature is high, except for the high risk associated with the research object, the use of blind methods of outcome evaluation, and corporate sponsorship. We compared the effects of DPP-4 inhibitors and GLP-1 RAs on benefit and risk indicators, respectively. The effects of incretin and insulin, Met, SGLT-2 inhibitors, SU, TZD, and AGI on the benefit and risk indicators were also compared. The acceptability of GLP-1 RAs was better than that of DPP-4 inhibitors in 84.5% of cases. The CF for GLP-1 RAs being better than DPP-4 inhibitors was 99.5%. The acceptability of DPP-4 inhibitors was better than that of insulin in 93.1% of cases, and that of GLP-1 RAs were better than insulin in 90.6% of cases. The CF for DPP-4 inhibitors being better than insulin was 99.5%, and for GLP-1 RAs being better than insulin was 99.9%. The acceptability of Met was better than that of DPP-4 inhibitors in 61.5% of cases, and that of GLP-1 RAs were better than Met in 70.4% of cases. The CF for Met being better than DPP-4 inhibitors was 88.7%, and for GLP-1 RAs being better than Met was 92.2%. The acceptability of SGLT-2 inhibitors was better than that of DPP-4 inhibitors in 93.5% of cases, and that of SGLT-2 inhibitors was better than that of GLP-1 RAs in 76.0% of cases. The CF for SGLT-2 inhibitors being better than DPP-4 inhibitors was 99.7%, and for SGLT-2 inhibitors being better than GLP-1 RAs was 92.9%. The acceptability of DPP-4 inhibitors was better than that of SU in 80.4% of cases, and that of GLP-1 RAs was better than that of SU in 94.2% of cases. The CF for DPP-4 inhibitors being better than SU was 96.4%, and for GLP-1 RAs being better than SU was 99.9%. The acceptability of TZD was better than that of DPP-4 inhibitors in 58.6% of cases, and that of GLP-1 RAs was better than that of TZD in 69.8% of cases. The CF for TZD being better than DPP-4 inhibitors was 90.9%, and that for GLP-1 RAs being better than TZD was 97.1%. The acceptability of DPP-4 inhibitors was better than that of AGI in 72.1% of cases, and that of GLP-1 RAs was better than that of AGI in 89.4% of cases. The CF for DPP-4 inhibitors being better than AGI was 84.6%, and that for GLP-1 RAs being better than AGI was 98.9%. Based on SMAA, GLP-1 RAs were more likely to be superior to DPP-4 inhibitors in terms of BRA. The acceptability of GLP-1 RAs was better than those of insulin, SU, TZD, and AGI but lower than that of SGLT-2 inhibitors. The acceptability of DPP-4 inhibitors was higher than those of insulin and SU but lower than that of Met and SGLT-2 inhibitors. SMAA is a derivative model of MCDA, which is widely used in various industries. As it is presently used in the medical field, MCDA has eight steps: clarifying the decision-making environment, determining the evaluation index, collecting the specific data from each index, normalizing the data collected by each index, giving each index weight, calculating BRA values, conducting sensitivity analyses, and interpreting the results. Unlike the traditional MCDA model, SMAA reduces the impact of the value preferences of evaluators on decision-making and does not need decision-makers to give subjective weight to decision-making indicators. This study had many advantages. First, a total of 589 RCT studies involving 295,908 patients with T2DM were systematically searched and included in it. The sample size was large, and the bias risk for each study was evaluated. Second, 26 risk/benefit outcome indicators related to the treatment of T2DM with incretin were included, and the indicators were comprehensive and representative. Third, the SMAA model used in this study can reduce the subjectivity of decision-makers as they weight the indicators, at least to a certain extent. Further analysis regarding patient-level characteristics and their values and preferences are warranted (Supplementary Link: https://links.lww.com/CM9/B369). (Table 1) Table 1 - Results of SMAA analysis. Drug name Rank acceptability index (%) CF (%) GLP-1 RAs-DPP-4 inhibitors 84.5 99.5 DPP-4 inhibitors-insulin 93.1 99.5 GLP-1 RAs-insulin 90.6 99.9 Met-DPP-4 inhibitors 61.5 88.7 GLP-1 RAs-Met 70.4 92.2 SGLT-2 inhibitors-DPP-4 inhibitors 93.5 99.7 SGLT-2 inhibitors-GLP-1 RAs 76.0 92.9 DPP-4 inhibitors-SU 80.4 96.4 GLP-1 RAs-SU 94.2 99.9 TZD-DPP-4 inhibitors 58.6 90.9 GLP-1 RAs-TZD 69.8 97.1 DPP-4 inhibitors-AGI 72.1 84.6 GLP-1 RAs-AGI 89.4 98.9 AGI: Alpha-glucosidase inhibitor; CF: Confidence factor; DPP-4: Dipeptidyl peptidase-4; GLP-1 Ras: Glucagon-like peptide-1 receptor agonists; Met: Metformin; SGLT-2: Sodium-glucose co-transporter 2; SMAA: Stochastic multi-criteria acceptability analysis; SU: Sulfonylureas; TZD: Thiazolidinediones. Funding This work was supported by a grant from the National Natural Science Foundation of China (No. 72074011). Conflicts of interest None.