Browse the corpus

Age and Sex Differences in Efficacy of Treatments for Type 2 Diabetes: A Network Meta-Analysis. IMPORTANCE: Sodium-glucose cotransporter 2 (SGLT2) inhibitors, glucagon-like peptide-1 (GLP-1) receptor agonists, and dipeptidyl peptidase 4 (DPP4) inhibitors improve hyperglycemia, and SGLT2 inhibitors and GLP-1 receptor agonists reduce the risk of major adverse cardiovascular events (MACEs) among individuals with type 2 diabetes. It is not clear whether efficacy varies by age or sex. OBJECTIVE: To assess whether age or sex are associated with differences in the efficacy of SGLT2 inhibitors, GLP-1 receptor agonists, and DPP4 inhibitors. DATA SOURCES AND STUDY SELECTION: The MEDLINE and Embase databases and US and Chinese clinical trial registries were searched for articles published from inception to November 2022; in August 2024, the search was updated to capture the trial results. Two reviewers screened for randomized clinical trials of SGLT2 inhibitors, GLP-1 receptor agonists, or DPP4 inhibitors vs a placebo or active comparator in adults with type 2 diabetes. DATA EXTRACTION AND SYNTHESIS: Individual participant data and aggregate data were used to estimate age × treatment interactions and sex × treatment interactions in multilevel network meta-regression models. MAIN OUTCOME AND MEASURES: Hemoglobin A1c (HbA1c) and MACEs. RESULTS: Of the 601 eligible trials identified (592 trials with 309 503 participants reported HbA1c; mean age, 58.9 [SD, 10.8] years; 42.3% were female and 23 trials with 168 489 participants reported MACEs; mean age, 64.0 [SD, 8.6] years; 35.3% were female), individual participant data were obtained for 103 trials (103 reported HbA1c and 6 reported MACEs). The use of SGLT2 inhibitors (vs placebo) was associated with less HbA1c lowering with increasing age for monotherapy (absolute reduction [AR], 0.24% [95% credible interval {CrI}, 0.10% to 0.38%] per 30-year increment in age), for dual therapy (AR, 0.17% [95% CrI, 0.10% to 0.24%]), and for triple therapy (AR, 0.25% [95% CrI, 0.20% to 0.30%]). The use of GLP-1 receptor agonists was associated with greater HbA1c lowering with increasing age for monotherapy (AR, -0.18% [95% CrI, -0.31% to -0.05%] per 30-year increment in age) and for dual therapy (AR, -0.24% [95% CrI, -0.40% to -0.07%]), but not for triple therapy (AR, 0.04% [95% CrI, -0.02% to 0.11%]). The use of DPP4 inhibitors was associated with slightly better HbA1c lowering in older people for dual therapy (AR, -0.09% [95% CrI, -0.15% to -0.03%] per 30-year increment in age), but not for monotherapy (AR, -0.08% [95% CrI, -0.18% to 0.01%]) or triple therapy (AR, -0.01% [95% CrI, -0.06% to 0.05%]). The relative reduction in MACEs with use of SGLT2 inhibitors was greater in older vs younger participants per 30-year increment in age (hazard ratio, 0.76 [95% CrI, 0.62 to 0.93]), and the relative reduction in MACEs with use of GLP-1 receptor agonists was less in older vs younger participants (hazard ratio, 1.47 [95% CrI, 1.07 to 2.02]). There was no consistent evidence for sex × treatment interactions with use of SGLT2 inhibitors and GLP-1 receptor agonists. CONCLUSIONS AND RELEVANCE: The SGLT2 inhibitors and GLP-1 receptor agonists were associated with lower risk of MACEs. Analysis of age × treatment interactions suggested that SGLT2 inhibitors were more cardioprotective in older than in younger people despite smaller reductions in HbA1c; GLP-1 receptor agonists were more cardioprotective in younger people.

Over the past 2 decades, new glucose-lowering agents have altered the management of type 2 diabetes. The efficacy of sodium-glucose cotransporter 2 (SGLT2) inhibitors and glucagon-like peptide-1 (GLP-1) receptor agonists in improving cardiovascular and kidney outcomes has been established,1,2 with widespread use in clinical practice and inclusion in clinical guidelines.3 However, the possibility that treatment effects may differ depending on participant characteristics has led to questions about applying trial findings to individuals less represented in trials, such as older people and women.4,5,6 Global estimates indicate that 1 in 5 people older than 65 years have diabetes,7 and almost half of those with type 2 diabetes are older than 65 years.8,9 Moreover, age-related functional limitations and conditions (such as frailty) typically manifest earlier in people with type 2 diabetes.10 The risk of complications from diabetes increases with age, potentially increasing the absolute benefits of treatment. Conversely, older adults also may be more susceptible to hypoglycemia with intensive glycemic targets.11,12 Among females, absolute risk of type 2 diabetes and cardiovascular disease is lower than in males, but diabetes is associated with a greater relative increase in cardiovascular risk in females than in males.13,14 Female patients also have different patterns of cardiovascular complications and less intensive management of cardiovascular risk factors than male patients.15 It is, therefore, important to determine whether treatment effects differ by age and sex.8,16,17

increase in cardiovascular risk in females than in males.13,14 Female patients also have different patterns of cardiovascular complications and less intensive management of cardiovascular risk factors than male patients.15 It is, therefore, important to determine whether treatment effects differ by age and sex.8,16,17 Clinical guidelines do not currently recommend different diabetes therapies for male and female patients, nor across different age groups. The guidelines have highlighted the uncertainty that comes from the underrepresentation of female participants and older people within trials.3,18 We aimed to perform a systematic review and meta-analysis of both aggregate and individual participant trial data to estimate whether the efficacy of SGLT2 inhibitors, GLP-1 receptor agonists, and dipeptidyl peptidase 4 (DPP4) inhibitors for type 2 diabetes differs by age and sex.

This systematic review and network meta-analysis followed a prespecified protocol (CRD42020184174).19 The protocol covers a wider project for calibration of the network meta-analysis to a community sample, seeking to provide estimates of efficacy reflecting representative samples. The current report presents findings from the assessment of age × treatment interactions and sex × treatment interactions prior to calibration to a community sample. Findings are reported according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline.20 Eligible studies were randomized clinical trials that enrolled adults aged 18 years or older diagnosed with type 2 diabetes and assessed the efficacy of SGLT2 inhibitors, GLP-1 receptor agonists, or DPP4 inhibitors on either glycated hemoglobin A1c or major adverse cardiovascular events (defined as cardiovascular death, nonfatal myocardial infarction, or nonfatal stroke) compared with either a placebo or an active comparator of any other drug class. We excluded within-class comparisons and unregistered trials. We included trials regardless of whether they assessed superiority or noninferiority. For trials with crossover designs, we included only data before the crossover.

fatal stroke) compared with either a placebo or an active comparator of any other drug class. We excluded within-class comparisons and unregistered trials. We included trials regardless of whether they assessed superiority or noninferiority. For trials with crossover designs, we included only data before the crossover. We searched 2 electronic databases (MEDLINE and Embase) using both keywords and Medical Subject Headings (full search terms appear in the eMethods in Supplement 1) as well as US and Chinese clinical trial registries from inception to November 2022 (Figure 1). All titles and abstracts were screened, retaining all potentially eligible studies for full-text review. All stages of screening were completed by 2 reviewers (E.B. and P.H. or L.W.) working independently, with conflicts resolved by consensus and involving a third reviewer (P.H. or L.W.) when required. In August 2024, we updated the search to include results of identified, eligible registered trials published after the initial search date. Any trials not published in English or Chinese were excluded due to a lack of available translation. For all eligible trials, we assessed whether individual participant data were available for analysis by third-party researchers through the Vivli clinical data repository and applied to the appropriate independent steering committees for access.

Any trials not published in English or Chinese were excluded due to a lack of available translation. For all eligible trials, we assessed whether individual participant data were available for analysis by third-party researchers through the Vivli clinical data repository and applied to the appropriate independent steering committees for access. Drug names, doses, and regimens were extracted from text strings obtained from ClinicalTrials.gov and published documents (articles and clinical study reports). Age and sex at baseline were obtained from published documents for trials with aggregate data. For trials with individual participant data available, we summarized age and sex directly from the trial data. The hemoglobin A1c results were extracted from ClinicalTrials.gov or published documents. For trials with individual participant data, the hemoglobin A1c values at baseline and at the time of the primary outcome were extracted. When the primary outcome values were missing, the last available observation was carried forward. In the sensitivity analysis, the baseline observation was carried forward. For major adverse cardiovascular events, the results were obtained via manual extraction from published documents (including subgroups for age and sex). For trials with individual participant data, the definition for major adverse cardiovascular events was harmonized across trials using adjudicated events. For the trials with aggregate data only, the findings for major adverse cardiovascular events that were consistent with the definition used in the individual participant data were extracted to allow consistent comparison across the trials. Individual-level trial data were cleaned and harmonized in the Vivli repository.

For the trials with aggregate data only, the findings for major adverse cardiovascular events that were consistent with the definition used in the individual participant data were extracted to allow consistent comparison across the trials. Individual-level trial data were cleaned and harmonized in the Vivli repository. Data on adverse events were also extracted from the trials with individual participant data, focusing on serious adverse events and the events with established associations with each drug class. For each trial, incident serious adverse events were identified as well as gastrointestinal adverse events, urinary tract infections, hypoglycemic episodes, amputations, and ketoacidosis. Adverse events were not assessed in the trials with aggregate data only due to a lack of harmonized definitions. Risk of bias was assessed in each study using the Cochrane Risk of Bias tool.21

ied as well as gastrointestinal adverse events, urinary tract infections, hypoglycemic episodes, amputations, and ketoacidosis. Adverse events were not assessed in the trials with aggregate data only due to a lack of harmonized definitions. Risk of bias was assessed in each study using the Cochrane Risk of Bias tool.21 A detailed description of the statistical analysis appears in the eMethods in Supplement 1. The age and sex distributions were summarized for each trial using individual participant data (when available) or from published summary statistics. Multilevel network meta-regression models were fitted for hemoglobin A1c and major adverse cardiovascular events using the multinma package in R,22 as previously described.19 This modeling approach was chosen because it does not disrupt randomization; makes less stringent assumptions than a standard network meta-analysis; and can (without causing aggregation bias) accommodate individual participant data, aggregate-level trial data, and subgroup-level trial data in models estimating treatment × covariate interactions.

chosen because it does not disrupt randomization; makes less stringent assumptions than a standard network meta-analysis; and can (without causing aggregation bias) accommodate individual participant data, aggregate-level trial data, and subgroup-level trial data in models estimating treatment × covariate interactions. For hemoglobin A1c, the network meta-analyses were separately fit for trials of monotherapy, dual therapy, and triple therapy, reflecting different indications for the drugs in question. All trials including major adverse cardiovascular events were analyzed together because their participants were selected based on cardiovascular risk. Treatment groups evaluating the combined effect of 2 or more treatments were excluded. For SGLT2 inhibitors, GLP-1 receptor agonists, DPP4 inhibitors, and metformin, the treatment groups were categorized by drug and dose. Insulin was modeled as a single category. For the remaining drug classes, groups within the same trial with different doses of the same drug were combined into a single group. For all models, placebo was the reference treatment. Trial-level regression models of each outcome by age, sex, and treatment were fitted for trials with individual participant data, and any age × treatment interactions and sex × treatment interactions were assessed. Linear regression models were fitted for hemoglobin A1c, and included hemoglobin A1c at baseline as a covariate. The last recorded value was carried forward in participants who did not complete the trial.

h individual participant data, and any age × treatment interactions and sex × treatment interactions were assessed. Linear regression models were fitted for hemoglobin A1c, and included hemoglobin A1c at baseline as a covariate. The last recorded value was carried forward in participants who did not complete the trial. Cox regression models were fitted for the major adverse cardiovascular event outcomes. Noncardiovascular death was treated as a competing event in the analyses of the major adverse cardiovascular event outcomes. Cause-specific hazard ratios (HRs) are presented. The cause-specific HRs for the competing event were also estimated for noncardiovascular mortality (defined when death occurred prior to the first major adverse cardiovascular event). The Cox proportional hazards regression model assumptions were checked by plotting scaled Schofield residuals. Residual plots and restricted cubic splines of age were inspected for nonlinearity for the outcomes of hemoglobin A1c and major adverse cardiovascular events (additional details appear in the eResults in Supplement 1).

roportional hazards regression model assumptions were checked by plotting scaled Schofield residuals. Residual plots and restricted cubic splines of age were inspected for nonlinearity for the outcomes of hemoglobin A1c and major adverse cardiovascular events (additional details appear in the eResults in Supplement 1). Estimates of the individual participant data and the aggregate trial–level data were included in the meta-analyses along with the age and sex distributions of each trial. For the trials of major adverse cardiovascular events, we also included subgroup-level data. For adverse event data, quasi-Poisson regression models and negative binomial regression models were fitted for incident events within the individual participant data. Summary statistics from each of these models were then used in the meta-analysis. Models were summarized using the posterior mean and 95% credible interval (CrI) for the main effect and for the age × treatment interactions and the sex × treatment interactions. The 95% CrIs indicate a plausible range of values; hence, when the 95% CrI includes the null (0 for the comparisons of hemoglobin A1c and 1 for the comparisons of major adverse cardiovascular events), no effect or no interaction is among the plausible interpretations. To allow comparisons across the outcomes, we repeated the main analyses, restricting the data to the 14 trials with individual participant data or aggregate data for both hemoglobin A1c and major adverse cardiovascular events. None of the analyses used formal adjustment for multiple testing.

plausible interpretations. To allow comparisons across the outcomes, we repeated the main analyses, restricting the data to the 14 trials with individual participant data or aggregate data for both hemoglobin A1c and major adverse cardiovascular events. None of the analyses used formal adjustment for multiple testing. All analyses were conducted using R version 4.4 (R Foundation for Statistical Computing).

Eligible studies were randomized clinical trials that enrolled adults aged 18 years or older diagnosed with type 2 diabetes and assessed the efficacy of SGLT2 inhibitors, GLP-1 receptor agonists, or DPP4 inhibitors on either glycated hemoglobin A1c or major adverse cardiovascular events (defined as cardiovascular death, nonfatal myocardial infarction, or nonfatal stroke) compared with either a placebo or an active comparator of any other drug class. We excluded within-class comparisons and unregistered trials. We included trials regardless of whether they assessed superiority or noninferiority. For trials with crossover designs, we included only data before the crossover. We searched 2 electronic databases (MEDLINE and Embase) using both keywords and Medical Subject Headings (full search terms appear in the eMethods in Supplement 1) as well as US and Chinese clinical trial registries from inception to November 2022 (Figure 1). All titles and abstracts were screened, retaining all potentially eligible studies for full-text review. All stages of screening were completed by 2 reviewers (E.B. and P.H. or L.W.) working independently, with conflicts resolved by consensus and involving a third reviewer (P.H. or L.W.) when required. In August 2024, we updated the search to include results of identified, eligible registered trials published after the initial search date. Any trials not published in English or Chinese were excluded due to a lack of available translation.

We searched 2 electronic databases (MEDLINE and Embase) using both keywords and Medical Subject Headings (full search terms appear in the eMethods in Supplement 1) as well as US and Chinese clinical trial registries from inception to November 2022 (Figure 1). All titles and abstracts were screened, retaining all potentially eligible studies for full-text review. All stages of screening were completed by 2 reviewers (E.B. and P.H. or L.W.) working independently, with conflicts resolved by consensus and involving a third reviewer (P.H. or L.W.) when required. In August 2024, we updated the search to include results of identified, eligible registered trials published after the initial search date. Any trials not published in English or Chinese were excluded due to a lack of available translation. For all eligible trials, we assessed whether individual participant data were available for analysis by third-party researchers through the Vivli clinical data repository and applied to the appropriate independent steering committees for access.

Drug names, doses, and regimens were extracted from text strings obtained from ClinicalTrials.gov and published documents (articles and clinical study reports). Age and sex at baseline were obtained from published documents for trials with aggregate data. For trials with individual participant data available, we summarized age and sex directly from the trial data. The hemoglobin A1c results were extracted from ClinicalTrials.gov or published documents. For trials with individual participant data, the hemoglobin A1c values at baseline and at the time of the primary outcome were extracted. When the primary outcome values were missing, the last available observation was carried forward. In the sensitivity analysis, the baseline observation was carried forward. For major adverse cardiovascular events, the results were obtained via manual extraction from published documents (including subgroups for age and sex). For trials with individual participant data, the definition for major adverse cardiovascular events was harmonized across trials using adjudicated events. For the trials with aggregate data only, the findings for major adverse cardiovascular events that were consistent with the definition used in the individual participant data were extracted to allow consistent comparison across the trials. Individual-level trial data were cleaned and harmonized in the Vivli repository.

A detailed description of the statistical analysis appears in the eMethods in Supplement 1. The age and sex distributions were summarized for each trial using individual participant data (when available) or from published summary statistics. Multilevel network meta-regression models were fitted for hemoglobin A1c and major adverse cardiovascular events using the multinma package in R,22 as previously described.19 This modeling approach was chosen because it does not disrupt randomization; makes less stringent assumptions than a standard network meta-analysis; and can (without causing aggregation bias) accommodate individual participant data, aggregate-level trial data, and subgroup-level trial data in models estimating treatment × covariate interactions. For hemoglobin A1c, the network meta-analyses were separately fit for trials of monotherapy, dual therapy, and triple therapy, reflecting different indications for the drugs in question. All trials including major adverse cardiovascular events were analyzed together because their participants were selected based on cardiovascular risk. Treatment groups evaluating the combined effect of 2 or more treatments were excluded. For SGLT2 inhibitors, GLP-1 receptor agonists, DPP4 inhibitors, and metformin, the treatment groups were categorized by drug and dose. Insulin was modeled as a single category. For the remaining drug classes, groups within the same trial with different doses of the same drug were combined into a single group. For all models, placebo was the reference treatment.

ts, DPP4 inhibitors, and metformin, the treatment groups were categorized by drug and dose. Insulin was modeled as a single category. For the remaining drug classes, groups within the same trial with different doses of the same drug were combined into a single group. For all models, placebo was the reference treatment. Trial-level regression models of each outcome by age, sex, and treatment were fitted for trials with individual participant data, and any age × treatment interactions and sex × treatment interactions were assessed. Linear regression models were fitted for hemoglobin A1c, and included hemoglobin A1c at baseline as a covariate. The last recorded value was carried forward in participants who did not complete the trial. Cox regression models were fitted for the major adverse cardiovascular event outcomes. Noncardiovascular death was treated as a competing event in the analyses of the major adverse cardiovascular event outcomes. Cause-specific hazard ratios (HRs) are presented. The cause-specific HRs for the competing event were also estimated for noncardiovascular mortality (defined when death occurred prior to the first major adverse cardiovascular event). The Cox proportional hazards regression model assumptions were checked by plotting scaled Schofield residuals. Residual plots and restricted cubic splines of age were inspected for nonlinearity for the outcomes of hemoglobin A1c and major adverse cardiovascular events (additional details appear in the eResults in Supplement 1).

We identified 672 eligible trials and included 601 in the network meta-analysis (Figure 1). Of the 601 trials included in the network meta-analysis, 498 only had aggregate data available (303 311 participants) and 103 had individual participant data available (92 182 participants; all 103 reported hemoglobin A1c and 6 reported major adverse cardiovascular events). There were 592 trials (309 503 participants) that reported hemoglobin A1c (mean age, 58.9 [SD, 10.8] years; 42.3% were female) and 23 trials (168 489 participants) that reported major adverse cardiovascular events (mean age, 64.0 [SD, 8.6] years; 35.3% were female); 14 trials reported both outcomes (hemoglobin A1c and major adverse cardiovascular events). Additional details (including all model outputs and analysis code) and risk of bias appear in the online project repository. The trials that reported hemoglobin A1c (by drug class and aggregate baseline characteristics) appear in Table 1. The baseline characteristics were similar for the trials with individual participant data available and for those with only aggregate data available. Trials that reported major adverse cardiovascular events appear in Table 2. There were more male participants than female participants and the age range of almost all trial participants was 40 to 80 years (including in the trials that only recruited people >70 years of age, in which participants were generally between 70 and 80 years of age; eFigure 1 and eTable 1 in Supplement 1). After collapsing groups that had different doses for the same agents.

The trials that reported hemoglobin A1c (by drug class and aggregate baseline characteristics) appear in Table 1. The baseline characteristics were similar for the trials with individual participant data available and for those with only aggregate data available. Trials that reported major adverse cardiovascular events appear in Table 2. There were more male participants than female participants and the age range of almost all trial participants was 40 to 80 years (including in the trials that only recruited people >70 years of age, in which participants were generally between 70 and 80 years of age; eFigure 1 and eTable 1 in Supplement 1). After collapsing groups that had different doses for the same agents. The data do not sum to the No. of trials because some trials included more than 1 treatment and drug class comparison. Category excludes insulins. Category excludes insulins (eg, repaglinide). This trial did not have a placebo group. This trial did not have an acronym. The main treatment effects for hemoglobin A1c comparing each treatment vs placebo in a standard network meta-analysis without covariates appear in eFigure 2 in Supplement 1. The treatments included in the trials reduced hemoglobin A1c by −0.5% to −1.5% (absolute reductions). The main treatment effects for major adverse cardiovascular events show a reduced risk of major adverse cardiovascular events for SGLT2 inhibitors and GLP-1 receptor agonists compared with placebo, with null findings for DPP4 inhibitors (eFigure 3 in Supplement 1).

emoglobin A1c by −0.5% to −1.5% (absolute reductions). The main treatment effects for major adverse cardiovascular events show a reduced risk of major adverse cardiovascular events for SGLT2 inhibitors and GLP-1 receptor agonists compared with placebo, with null findings for DPP4 inhibitors (eFigure 3 in Supplement 1). The age × treatment and sex × treatment interactions for hemoglobin A1c and major adverse cardiovascular events appear in Figure 2. The use of SGLT2 inhibitors (vs placebo) was associated with less hemoglobin A1c lowering with increasing age for monotherapy (absolute reduction, 0.24% [95% CrI, 0.10% to 0.38%] per 30-year increment in age), for dual therapy (absolute reduction, 0.17% [95% CrI, 0.10% to 0.24%]), and for triple therapy (absolute reduction, 0.25% [95% CrI, 0.20% to 0.30%]). There was no evidence for nonlinearity in the age × treatment interaction (eFigure 4 in Supplement 1). The results were also similar confining the analysis to trials with greater than or equal to 6 months of follow-up (eFigure 5 in Supplement 1).

for triple therapy (absolute reduction, 0.25% [95% CrI, 0.20% to 0.30%]). There was no evidence for nonlinearity in the age × treatment interaction (eFigure 4 in Supplement 1). The results were also similar confining the analysis to trials with greater than or equal to 6 months of follow-up (eFigure 5 in Supplement 1). The horizontal lines show the 95% credible interval. Age was modeled as a continuous variable and divided by 30 (so that the coefficient reflects the difference in efficacy over a 30-year age difference). Estimates to the left of the line of no effect (dotted vertical line) indicate that the treatment is more efficacious in those with an older age or male sex. Estimates to the right of this line indicate the inverse. The size of the data markers reflects the proportion of participants in the analysis who had been allocated to a drug in that class. For dual therapy and triple therapy, participants were required or permitted to also take additional antidiabetic medications. The fixed and random effects refer to the main treatment effects (eg, 300 mg of canagliflozin).

ts the proportion of participants in the analysis who had been allocated to a drug in that class. For dual therapy and triple therapy, participants were required or permitted to also take additional antidiabetic medications. The fixed and random effects refer to the main treatment effects (eg, 300 mg of canagliflozin). The use of GLP-1 receptor agonists was associated with greater hemoglobin A1c lowering with increasing age for monotherapy (absolute reduction, −0.18% [95% CrI, −0.31% to −0.05%] per 30-yer increment in age) and for dual therapy (absolute reduction, −0.24% [95% CrI, −0.40% to −0.07%]), but not for triple therapy (absolute reduction, 0.04% [95% CrI, −0.02% to 0.11%]). The use of DPP4 inhibitors was associated with slightly better hemoglobin A1c lowering among older people for dual therapy (absolute reduction, −0.09% [95% CrI, −0.15% to −0.03%] per 30-year increment in age), but no evidence of variation in efficacy for monotherapy (absolute reduction, −0.08% [95% CrI, −0.18% to 0.01%]) or for triple therapy (absolute reduction, −0.01% [95% CrI, −0.06% to 0.05%]). There was a small difference in efficacy of SGLT2 inhibitors favoring males for triple therapy only (absolute reduction, −0.06% [95% CrI, −0.10 to −0.02]).

n in efficacy for monotherapy (absolute reduction, −0.08% [95% CrI, −0.18% to 0.01%]) or for triple therapy (absolute reduction, −0.01% [95% CrI, −0.06% to 0.05%]). There was a small difference in efficacy of SGLT2 inhibitors favoring males for triple therapy only (absolute reduction, −0.06% [95% CrI, −0.10 to −0.02]). In trials of SGLT2 inhibitors, older age was associated with a greater relative reduction in major adverse cardiovascular events (HR, 0.76 [95% CrI, 0.62-0.93] per 30-year increment in age). In trials of GLP-1 receptor agonists, older age was associated with a lower relative reduction in major adverse cardiovascular events (HR, 1.47 [95% CrI, 1.07-2.02] per 30-year increment in age). The estimate for DPP4 inhibitors included the null (HR, 0.73 [95% CrI, 0.52-1.00]). For major adverse cardiovascular events, the use of DPP4 inhibitors was less efficacious in male participants (HR, 1.65 [95% CrI, 1.25-2.21] for males vs females), although this association was attenuated after including sex subgroup data in the analysis (HR, 1.22 [95% CrI, 1.04-1.42]) and, after excluding the only trial of DPP4 inhibitors with individual participant data, the 95% CrI included the null (eFigure 6 in Supplement 1). There was no evidence of a sex × treatment interaction for GLP-1 receptor agonists (HR, 1.17 [95% CrI, 0.87-1.58] for males vs females) or for SGLT2 inhibitors (HR, 0.95 [95% CrI, 0.86-1.06] for males vs females). Additional models did not show nonlinearity of the age × treatment interaction within the range of ages included in the trials (eFigure 7 in Supplement 1).

ction for GLP-1 receptor agonists (HR, 1.17 [95% CrI, 0.87-1.58] for males vs females) or for SGLT2 inhibitors (HR, 0.95 [95% CrI, 0.86-1.06] for males vs females). Additional models did not show nonlinearity of the age × treatment interaction within the range of ages included in the trials (eFigure 7 in Supplement 1). The sensitivity analyses that included or excluded age and sex subgroup data in the model had an effect on the hemoglobin A1c findings in an analysis that excluded 1 of the 4 trials of SGLT2 inhibitors with individual participant data (eFigure 6 in Supplement 1). The greater relative reduction in major adverse cardiovascular events risk at older ages was preserved or greater in all sensitivity analyses. Similar results were obtained in the analyses restricting the data to the 14 trials with individual participant data for both hemoglobin A1c and major adverse cardiovascular events (eFigure 8 in Supplement 1). The results from the analyses of major adverse cardiovascular events differed depending on the inclusion or exclusion of single trials of GLP-1 receptor agonists and DPP4 inhibitors with individual participant data and the inclusion or exclusion of subgroup data (eFigure 6 in Supplement 1).

(eFigure 8 in Supplement 1). The results from the analyses of major adverse cardiovascular events differed depending on the inclusion or exclusion of single trials of GLP-1 receptor agonists and DPP4 inhibitors with individual participant data and the inclusion or exclusion of subgroup data (eFigure 6 in Supplement 1). There was no age × treatment interaction or sex × treatment interaction between any class of medication and gastrointestinal adverse events, hypoglycemia, or urinary tract infections (eFigure 9 in Supplement 1). There was no age × treatment interaction or sex × treatment interaction with serious adverse events for SGLT2 inhibitors, GLP-1 receptor agonists, or DPP4 inhibitors (eFigure 9 in Supplement 1). Death was uncommon across trials with individual participant data (eFigure 10 in Supplement 1), and there was no evidence for any age × treatment interaction or sex × treatment interaction for noncardiovascular death (eFigure 11 in Supplement 1). There were too few events within the individual participant trial data to fit models for amputation or ketoacidosis.

idual participant data (eFigure 10 in Supplement 1), and there was no evidence for any age × treatment interaction or sex × treatment interaction for noncardiovascular death (eFigure 11 in Supplement 1). There were too few events within the individual participant trial data to fit models for amputation or ketoacidosis. The associations between age × treatment and sex × treatment interactions and the overall age- and sex-specific relative efficacy for each drug class vs placebo appear in Figure 3. The SGLT2 inhibitors were associated with reduced major adverse cardiovascular events in older people regardless of sex (HR, 0.84 [95% CrI, 0.76-0.93] for 75-year-old females; HR, 0.81 [95% CrI, 0.73-0.89] for 75-year-old males; HR, 0.91 [95% CrI, 0.85-0.97] for 65-year-old females; and HR, 0.88 [95% CrI, 0.80-0.96] for 65-year-old males). These graphs are based on a model that included all available trials (some of the trials only had aggregate data available and others had individual participant data available) and sex subgroup data. The hazard ratios (data points) and 95% credible intervals show the age- and sex-specific estimates for the effect of each treatment compared with placebo on the risk for major adverse cardiovascular events. The data points and whiskers are offset from their time point to avoid superimposition. The density plots indicate the proportion of trial participants by sex and age.

als show the age- and sex-specific estimates for the effect of each treatment compared with placebo on the risk for major adverse cardiovascular events. The data points and whiskers are offset from their time point to avoid superimposition. The density plots indicate the proportion of trial participants by sex and age. There was no association in GLP-1 receptor agonists for a significant reduction in major adverse cardiovascular events among males (HR, 0.99 [95% CrI, 0.89-1.11] in 65-year-old males) or older people (HR, 0.91 [95% CrI, 0.79-1.05] for 75-year-old females; HR, 1.03 [95% CrI, 0.87-1.20] for 75-year-old males). There was a decreased risk of major adverse cardiovascular events associated with use of GLP-1 receptor agonists among younger females (HR, 0.85 [95% CrI, 0.81-0.91] for 55-year-old females; HR, 0.88 [95% CrI, 0.82-0.95] for 65-year-old females). These findings should be interpreted with caution because the age × treatment interaction for GLP-1 receptor agonists was sensitive to the exclusion of the trial for which we had individual participant data (eFigure 6 in Supplement 1). Also, although the GLP-1 receptor agonists class showed an overall benefit for major adverse cardiovascular events (eFigure 3 in Supplement 1), the effect on major adverse cardiovascular events for some of the drugs within this class was null (eFigure 12 in Supplement 1).

l participant data (eFigure 6 in Supplement 1). Also, although the GLP-1 receptor agonists class showed an overall benefit for major adverse cardiovascular events (eFigure 3 in Supplement 1), the effect on major adverse cardiovascular events for some of the drugs within this class was null (eFigure 12 in Supplement 1). Similarly, even though there were some differences in efficacy across age and sex for DPP4 inhibitors, these findings should be interpreted with caution because these agents showed a null overall effect on major adverse cardiovascular events. All interaction estimates were sensitive to the inclusion of specific trials. The heterogeneity estimates for all of the random-effects models appear in eTable 2 in Supplement 1. The adverse events by trial appear in eTable 3 in Supplement 1.

The trials that reported hemoglobin A1c (by drug class and aggregate baseline characteristics) appear in Table 1. The baseline characteristics were similar for the trials with individual participant data available and for those with only aggregate data available. Trials that reported major adverse cardiovascular events appear in Table 2. There were more male participants than female participants and the age range of almost all trial participants was 40 to 80 years (including in the trials that only recruited people >70 years of age, in which participants were generally between 70 and 80 years of age; eFigure 1 and eTable 1 in Supplement 1). After collapsing groups that had different doses for the same agents. The data do not sum to the No. of trials because some trials included more than 1 treatment and drug class comparison. Category excludes insulins. Category excludes insulins (eg, repaglinide). This trial did not have a placebo group. This trial did not have an acronym.

The main treatment effects for hemoglobin A1c comparing each treatment vs placebo in a standard network meta-analysis without covariates appear in eFigure 2 in Supplement 1. The treatments included in the trials reduced hemoglobin A1c by −0.5% to −1.5% (absolute reductions). The main treatment effects for major adverse cardiovascular events show a reduced risk of major adverse cardiovascular events for SGLT2 inhibitors and GLP-1 receptor agonists compared with placebo, with null findings for DPP4 inhibitors (eFigure 3 in Supplement 1).

The age × treatment and sex × treatment interactions for hemoglobin A1c and major adverse cardiovascular events appear in Figure 2. The use of SGLT2 inhibitors (vs placebo) was associated with less hemoglobin A1c lowering with increasing age for monotherapy (absolute reduction, 0.24% [95% CrI, 0.10% to 0.38%] per 30-year increment in age), for dual therapy (absolute reduction, 0.17% [95% CrI, 0.10% to 0.24%]), and for triple therapy (absolute reduction, 0.25% [95% CrI, 0.20% to 0.30%]). There was no evidence for nonlinearity in the age × treatment interaction (eFigure 4 in Supplement 1). The results were also similar confining the analysis to trials with greater than or equal to 6 months of follow-up (eFigure 5 in Supplement 1). The horizontal lines show the 95% credible interval. Age was modeled as a continuous variable and divided by 30 (so that the coefficient reflects the difference in efficacy over a 30-year age difference). Estimates to the left of the line of no effect (dotted vertical line) indicate that the treatment is more efficacious in those with an older age or male sex. Estimates to the right of this line indicate the inverse. The size of the data markers reflects the proportion of participants in the analysis who had been allocated to a drug in that class. For dual therapy and triple therapy, participants were required or permitted to also take additional antidiabetic medications. The fixed and random effects refer to the main treatment effects (eg, 300 mg of canagliflozin).

The associations between age × treatment and sex × treatment interactions and the overall age- and sex-specific relative efficacy for each drug class vs placebo appear in Figure 3. The SGLT2 inhibitors were associated with reduced major adverse cardiovascular events in older people regardless of sex (HR, 0.84 [95% CrI, 0.76-0.93] for 75-year-old females; HR, 0.81 [95% CrI, 0.73-0.89] for 75-year-old males; HR, 0.91 [95% CrI, 0.85-0.97] for 65-year-old females; and HR, 0.88 [95% CrI, 0.80-0.96] for 65-year-old males). These graphs are based on a model that included all available trials (some of the trials only had aggregate data available and others had individual participant data available) and sex subgroup data. The hazard ratios (data points) and 95% credible intervals show the age- and sex-specific estimates for the effect of each treatment compared with placebo on the risk for major adverse cardiovascular events. The data points and whiskers are offset from their time point to avoid superimposition. The density plots indicate the proportion of trial participants by sex and age.

This network meta-analysis of 601 trials, including individual participant data from 103 trials, assessed whether the efficacy of 3 newer drug classes (SGLT2 inhibitors, GLP-1 receptor agonists, and DPP4 inhibitors) varied by age or sex in people with type 2 diabetes. For hemoglobin A1c, use of SGLT2 inhibitors showed modestly reduced efficacy with increasing age, with attenuation of the treatment effect compared with placebo by approximately 0.25% at 75 years of age compared with 45 years of age. In contrast, the reduction in major adverse cardiovascular events with SGLT2 inhibitors was greater in older people compared with younger people. There was some evidence that hemoglobin A1c lowering was greater with use of GLP-1 receptor agonists among older trial participants, whereas efficacy for cardiovascular outcomes was greater among younger female participants.

vascular events with SGLT2 inhibitors was greater in older people compared with younger people. There was some evidence that hemoglobin A1c lowering was greater with use of GLP-1 receptor agonists among older trial participants, whereas efficacy for cardiovascular outcomes was greater among younger female participants. Previous studies assessing heterogeneity in efficacy of treatments for type 2 diabetes (ie, treatment × age or treatment × sex interaction) have generally used aggregate or subgroup data from randomized clinical trials or relied on observational (nonrandomized) data. A prior meta-analysis16 of differences between male and female participants in the efficacy of SGLT2 inhibitors and GLP-1 receptor agonists found no statistically significant difference in efficacy for cardiovascular outcomes, but the investigators speculated on possible reduced cardiovascular efficacy among female patients due to the greater statistical uncertainty in the estimates for this group. The current analysis, which included a larger and more comprehensive group of studies and incorporated individual participant data, provides greater precision, and more clearly demonstrated that sex is not associated with any important difference in the efficacy of these medication classes for the treatment of type 2 diabetes.

ent analysis, which included a larger and more comprehensive group of studies and incorporated individual participant data, provides greater precision, and more clearly demonstrated that sex is not associated with any important difference in the efficacy of these medication classes for the treatment of type 2 diabetes. A recent systematic review and network meta-analysis2 assessed the efficacy of type 2 diabetes treatment across a range of clinical outcomes, including heart failure, end-stage kidney disease, and medication-related harms, which were not included in the current analysis. The prior network meta-analysis2 showed that, in addition to major adverse cardiovascular events, use of SGLT2 inhibitors and GLP-1 receptor agonists reduced the risk of admission to the hospital for heart failure and reduced the risk of end-stage kidney disease, with superior efficacy with use of SGLT2 inhibitors in reducing end-stage kidney disease. Harms with treatment were generally specific to the medication class and included genital infections with SGLT2 inhibitors and gastrointestinal complications with GLP-1 receptor agonists. However, the prior network meta-analysis2 did not assess heterogeneity by age and sex and did not include an analysis of individual participant data.

ent were generally specific to the medication class and included genital infections with SGLT2 inhibitors and gastrointestinal complications with GLP-1 receptor agonists. However, the prior network meta-analysis2 did not assess heterogeneity by age and sex and did not include an analysis of individual participant data. One likely explanation for the reduction in glycemic efficacy with use of SGLT2 inhibitors, and in older age, is age-related decline in kidney function. For example, a recent double-blind 3-way crossover study45 comparing use of DPP4 inhibitors vs SGLT2 inhibitors demonstrated that participants with estimated glomerular filtration rates of 60 to 90 mL/min/1.73 m2 (compared with those with estimated glomerular filtration rates >90 mL/min/1.73 m2) had lower hemoglobin A1c while taking DPP4 inhibitors than while taking SGLT2 inhibitors. In this context, it is notable that the reductions in major adverse cardiovascular events with SGLT2 inhibitors were greater in older people, despite lower glycemic efficacy. This highlights the limitation of surrogate outcomes (such as hemoglobin A1c) in determining the risks of major adverse cardiovascular events, for which hyperglycemia is a less important risk factor than hypertension or dyslipidemia.46

with SGLT2 inhibitors were greater in older people, despite lower glycemic efficacy. This highlights the limitation of surrogate outcomes (such as hemoglobin A1c) in determining the risks of major adverse cardiovascular events, for which hyperglycemia is a less important risk factor than hypertension or dyslipidemia.46 The results from the current network meta-analysis are also consistent with the established efficacy of SGLT2 inhibitors for improving cardiovascular outcomes in conditions other than diabetes (such as heart failure or chronic kidney disease), which are not characterized by hyperglycemia. Current clinical guidelines3,47 recommend less stringent glycemic targets in older people living with multiple long-term conditions or frailty due to greater risks of adverse events. The current findings highlight the need to consider the cardioprotective effects of therapies when treating older people in addition to safety, tolerability, and the priorities of patients.

gent glycemic targets in older people living with multiple long-term conditions or frailty due to greater risks of adverse events. The current findings highlight the need to consider the cardioprotective effects of therapies when treating older people in addition to safety, tolerability, and the priorities of patients. Although the current findings demonstrate similar or better cardiovascular efficacy among older people within the included trials, generally clinical trials only rarely enroll people older than 80 years of age. There are also likely to be unmeasured differences between trial participants and the patients considered for treatment in routine care. For example, age-associated states (such as frailty48), which increase the risk of both cardiovascular events and complications,11,49 were not quantified in the included trials. The current analysis does not, therefore, assess whether efficacy is similar in people at older ages (ie, >80 years) or who are living with frailty. Patients with older age are a group in which the balance of risks and benefits is the most uncertain.

s and complications,11,49 were not quantified in the included trials. The current analysis does not, therefore, assess whether efficacy is similar in people at older ages (ie, >80 years) or who are living with frailty. Patients with older age are a group in which the balance of risks and benefits is the most uncertain. Moreover, it is likely that the effect of age on treatment efficacy is moderated through other measurable age-related characteristics (such as kidney function or the presence and extent of comorbidities). Accounting for such characteristics in future work may allow more nuanced understanding of the likely benefits of treatments according to more specific characteristics, determining not only the overall treatment efficacy in older people, but in older people with different physiological and clinical characteristics. There is a need for trials that (1) recruit and retain older people and those living with frailty and (2) explicitly measure and report functional status.

characteristics, determining not only the overall treatment efficacy in older people, but in older people with different physiological and clinical characteristics. There is a need for trials that (1) recruit and retain older people and those living with frailty and (2) explicitly measure and report functional status. This network meta-analysis has limitations. First, although the primary strength was the use of individual participant data to estimate age × treatment interactions and sex × treatment interactions, these data were not available for all included trials. Individual participant data improves statistical power and allows integration of the individual participant data and the aggregate data within a network meta-analysis to preserve randomization and avoid aggregation bias. We also followed rigorous systematic review methods to identify eligible studies and have made all model outputs and analysis code publicly available to facilitate replication of the findings. However, despite the inclusion of a large volume of individual participant data, these data were only available for 17% of the trials (103/601). Furthermore, the trials for which we did have individual participant data were not a random sample of the included trials because the data availability depended on the sponsor’s data sharing arrangements. We did not attempt to obtain additional individual participant data through direct contact with study authors.

01). Furthermore, the trials for which we did have individual participant data were not a random sample of the included trials because the data availability depended on the sponsor’s data sharing arrangements. We did not attempt to obtain additional individual participant data through direct contact with study authors. Second, the use of multilevel network meta-regression also meant that aggregate data were used for all treatment comparisons (within class, between class, and vs placebo) to estimate the interactions, regardless of whether individual participant data were available. Treatment effects within the drug classes were estimated independently; drugs within a class were not assumed to have the same or similar efficacy. However, to estimate the interactions from the available data, the current approach assumed that interactions were common across drugs in the same class, and in practice, it also requires at least some trials with individual participant data for each drug class. Third, even though we included a large number of trials, a relatively small proportion of these assessed cardiovascular outcomes. Fourth, trial groups were dropped that had multiple drug classes because of software limitations (did not allow for explicit modeling of components within groups), and because the focus was on drug class–level interactions. Fifth, although we assessed glycemic and cardiovascular efficacy, which are clinically relevant outcomes, the analysis did not include other clinical outcomes (such as kidney events).

mitations (did not allow for explicit modeling of components within groups), and because the focus was on drug class–level interactions. Fifth, although we assessed glycemic and cardiovascular efficacy, which are clinically relevant outcomes, the analysis did not include other clinical outcomes (such as kidney events). Sixth, even though we assessed whether the association between these medications and established risks varied by age and sex, these analyses were limited by the small number of events within the trial data. Furthermore, we did not attempt to identify novel associations between these agents and specific adverse events. Such analyses would ideally draw on both trial data and routine health care data, in which identification of rarer events is more feasible. Seventh, we did not present major adverse cardiovascular events in terms of absolute risks. In most settings, it is likely that major adverse cardiovascular events occur more frequently with older age, which would tend to increase the absolute benefits of treatment. However, competing risks (eg, noncardiovascular mortality) are also likely to occur more frequently with older age. Consequently, the absolute benefit of treatment in older people will depend not only on the relative treatment effects, but also on the rates of major adverse cardiovascular events, adverse events, and competing events in the target population.

This network meta-analysis has limitations. First, although the primary strength was the use of individual participant data to estimate age × treatment interactions and sex × treatment interactions, these data were not available for all included trials. Individual participant data improves statistical power and allows integration of the individual participant data and the aggregate data within a network meta-analysis to preserve randomization and avoid aggregation bias. We also followed rigorous systematic review methods to identify eligible studies and have made all model outputs and analysis code publicly available to facilitate replication of the findings. However, despite the inclusion of a large volume of individual participant data, these data were only available for 17% of the trials (103/601). Furthermore, the trials for which we did have individual participant data were not a random sample of the included trials because the data availability depended on the sponsor’s data sharing arrangements. We did not attempt to obtain additional individual participant data through direct contact with study authors.

The SGLT2 inhibitors and GLP-1 receptor agonists were associated with lower risk of major adverse cardiovascular events. Analysis of age × treatment interactions suggested that SGLT2 inhibitors were more cardioprotective in older than in younger people despite smaller reductions in hemoglobin A1c; GLP-1 receptor agonists were more cardioprotective in younger people.