Cardiovascular and Renal Benefits of Novel Diabetes Drugs by Baseline Cardiovascular Risk: A Systematic Review, Meta-analysis, and Meta-regression Free

Globally, diabetes is an increasingly prevalent disorder across all regions. In 2021, it was estimated that 537 million adults aged 20–79 years (10.5% of this age-group) live with diabetes, and by 2030, 643 million are projected to have diabetes. Type 2 diabetes accounts for the vast majority (over 90%) of these diabetes cases. Type 2 diabetes significantly increases the risk for atherosclerotic cardiovascular disease (ASCVD), heart failure, and end-stage renal disease (ESRD). In 2021, ∼6.7 million adults were estimated to have died due to diabetes or its complications (1). In the U.S., type 2 diabetes has been estimated to reduce life expectancy by ∼4 years, with ASCVD and heart failure accounting for ∼31% of all diabetes attributable death (2). Type 2 diabetes is also a driver of the growth in treated ESRD incidence across the world. Particularly in Asia, generally more than one-half of all ESRD cases were attributed to diabetes. In the U.S., ESRD has high morbidity and mortality, with dialysis associated with a >25-year shorter life span compared with the general population’s life expectancy (3). Prevention strategies targeting ASCVD such as major adverse cardiovascular events (MACE), heart failure, and adverse renal outcomes in patients with type 2 diabetes are thus critically important to reduce disease burden at the population level.

Two recent meta-analyses (4,5) showed significant relative risk reductions in these clinical outcomes with two recently introduced glucose-lowering drug classes: glucagon-like peptide 1 receptor agonists (GLP-1RA) and sodium–glucose cotransporter-2 inhibitors (SGLT2i). More recently published trials primarily focusing on improving heart failure and renal outcomes were not included in these meta-analyses (4,5), in part because these trials allowed for inclusion of participants without diabetes. Yet, the relative efficacy of both new drug classes for reducing cardiovascular and renal outcome rates has generally been shown not to depend on comorbid diabetes status and glycemic control at baseline (6). For the latter reason, eligibility for both drugs has now been expanded to patients with diabetes and high cardiovascular risk, heart failure, or chronic kidney disease and HbA_1c levels below recommended targets. Eligible patients also include those without prior first-line metformin use (7).

However, to what extent expected treatment benefits of GLP-1RA and SGLT2i differ by cardiovascular risk levels is not clear. Prices of both drug classes are much higher than traditional diabetes medications and cardiovascular preventive drugs such as statins. For example, in the U.S., it has been estimated that prices were up to 360 times higher than traditional diabetes medications (8). In other countries, including lower- and middle-income countries, these price ratios are more favorable, although prices of both drug classes remain in general higher (9–11). Wider use of these expensive novel glucose-lowering medications may thus yield uncertain net benefits and cost-effectiveness at the population level.

We performed an updated systematic review to include more recently published trials, followed by meta-analyses and meta-regression analyses. We investigated whether patient populations with varying risks differed in cardiovascular and renal benefits received from GLP-1RA and SGLT2i.

We updated the search strategies of the two aforementioned prior meta-analyses (4,5) that covered trial reports published through February 2020 using PubMed with publication dates through 7 November 2022. Screening of titles/abstracts, followed by full-text reviewing for eligibility, was performed by two independent reviewers (J.M.R.-V. and B.S.F.). Discrepancies for excluding articles including reasons for exclusion at full-text review were resolved by consensus. Eligible articles included reports of confirmatory randomized controlled trials of GLP-1RA and SGLT2i with ASCVD, heart failure, and/or adverse renal outcomes as primary end points to determine safety or efficacy. Our search strategy is detailed in Supplementary Table 1.

Study selection criteria included trials with an adult patient population with or without type 2 diabetes but not exclusively enrolling patients with type 1 diabetes or gestational diabetes mellitus. We required trial designs randomizing patients to an intervention arm in which GLP-1RA or SGLT2i were administered as a single drug added to existing therapy/standard of care. We required that the trials’ control arm be defined as existing therapy/standard of care with or without placebo. Furthermore, we required reporting of efficacy data for at least one of our primary or secondary outcomes. Primary outcomes were defined as cardiovascular mortality, MACE, hospitalization for heart failure, and/or a composite end point of adverse renal outcomes. The latter was based on first occurrence of worsening kidney function, ESRD, or death from renal causes. All-cause mortality, fatal or nonfatal myocardial infarction (MI), and fatal or nonfatal any stroke were defined as secondary outcomes.

Two reviewers (J.M.R.-V. and M.T.) independently extracted data from selected full-text reports. Disagreements were resolved through discussion or arbitration with a third reviewer (B.S.F.). We extracted hazard ratio (HR) and crude event rate data including 95% CIs for each outcome of interest. When HRs or event rates were reported not for the total group but, rather, by subgroup, we combined these data using fixed-effects meta-analysis. For stroke, we prioritized data extraction for events of any subtype of stroke to avoid heterogeneity due to combining efficacy data on different event types such as any stroke with ischemic stroke. In Dapagliflozin Effect on Cardiovascular Events trial (DECLARE-TIMI 58) investigators used ischemic stroke (12), and we used the HR for any stroke (13), as also used by McGuire et al. (5). For our heart failure outcome, we ignored the reported recurrent event HR data from the Effect of Sotagliflozin on Cardiovascular Events in Patients with Type 2 Diabetes Post Worsening Heart Failure (SOLOIST-WHF) trial in the meta-analysis of HRs. For the Effect of Sotagliflozin on Cardiovascular and Renal Events in Patients with Type 2 Diabetes and Moderate Renal Impairment Who Are at Cardiovascular Risk (SCORED) trial, however, we used reported binomial count data on first heart failure readmissions to obtain an HR estimate for time to first event (14). For the composite renal outcome, we prioritized extraction of efficacy data using the recommended definition of sustained ≥40% reduction in estimated glomerular filtration rate (eGFR) (15), need for renal-replacement therapy, or renal death (Supplementary Table 2). When study reports used a different end point definition, we made attempts to identify secondary analyses by additional trial-specific PubMed searches using a combination of search terms for trial and drug names. Otherwise, we used reported data on sustained ≥50% reduction in eGFR from baseline (16–23) or doubling of serum creatinine from baseline (24–30). A doubling of serum creatinine from baseline is assumed equivalent to a sustained 57% reduction in eGFR from baseline (15).

In addition, we extracted data on study characteristics, intervention characteristics, and patient characteristics, as well as glucose-lowering medications used at baseline. We assessed risk of bias using version 2 of the Cochrane risk-of-bias tool for randomized trials (31).

Summary log HRs and their 95% CIs for time-to-event outcomes were calculated for GLP-1RA and SGLT2i trials separately with use of meta-analysis with a random-effects model, in which the reported effect size of every study was weighted by the inverse of its variance.

For estimation of absolute treatment benefits, we computed 5-year absolute risk differences (ARDs). A 5-year time horizon was chosen for each trial because absolute risks and therefore differences in risk across study arms depend on follow-up duration. For each outcome, we first extracted the event rate per 100 person-years of the control arm and its SE. Subsequently, we computed the rate ratio comparing the intervention arm with the control arm and its SE. Finally, we used an exponential distribution for calculating cumulative 5-year risks in each arm. For estimation of summary ARDs for each drug class, we conducted random resampling for variance estimation of risk differences (32,33). Then we performed meta-analyses of risk differences using random-effects models weighted by the inverse of the variance. SCORED and SOLOIST-WHF, with use of recurrent events for the heart failure end point, were ignored for estimation of ARDs of hospitalization for heart failure.

We assessed interstudy heterogeneity with the Cochran Q and I² statistics. The Cochran Q test is based on a χ² distribution with a null hypothesis that is used to evaluate a common effect size shared by all trials. With the I² statistic one estimates the proportion of the total observed variance that reflects real differences in effect size. Commonly used I² thresholds for degree of heterogeneity are ≤25% (low), 26–50% (moderate), and >50% (high).

To assess heterogeneity of relative and absolute treatment benefits by baseline cardiovascular risk, we performed meta-regression using mixed-effects linear modeling. We used the reported cardiovascular mortality rate in the control group as a covariate and the log HR and 5-year ARD estimates, respectively, for each trial as outcomes. We chose the control arm’s cardiovascular mortality rate as covariate because it can be considered a good proxy for global cardiovascular risk, includes heart failure, and was available for all trials.

To assess the potential impact of differences in trial design and setting, we conducted sensitivity analyses for meta-regressions of 5-year ARDs by excluding studies with SGLT2i that also included inhibition of sodium–glucose cotransporter 1 (23,34). We also excluded those done in an acute care setting (25,26,34) and those that exclusively enrolled patients with established heart failure at baseline (18,19,22,34–37). MACE was generally defined as death from cardiovascular causes, nonfatal MI, or any nonfatal stroke. However, in DECLARE-TIMI 58, ischemic instead of any stroke was used. For this reason, we performed an additional analysis excluding this one trial. For similar reasons, we repeated the meta-regression for the composite renal outcome after removing trials with a divergent end point definition (16–19,22–26,34). Finally, we conducted analyses excluding studies with evaluation of drugs currently not available on the market (23,34,38–40).

For the analyses, we used the metafor and meta packages in R, version 4.1.1 (R Foundation for Statistical Computing [https://www.r-project.org]) and P < 0.05 for statistical significance. The study protocol was performed according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and registered on International prospective register of systematic reviews (PROSPERO) (CRD42022308907).

Statistical codes and data sets used for our analyses can be obtained from https://github.com/Modeling-NovelDiabetesMeds.

We included 15 reports on 13 trials that were already selected in the two previous systematic reviews. In our updated PubMed search we identified 1,179 citations, and 339 citations were additionally identified through trial-specific searches. After removing 134 duplicates, from the total of 1,384 articles, we ultimately included 34 full-text reports (Supplementary Fig. 1) on 22 trials (n = 154,649): 9 on GLP-1RAs (n = 64,326) vs. 13 on SGLT2i (n = 90,413). Although the Canagliflozin Cardiovascular Assessment Study (CANVAS) Program comprised an integrated analysis of two sister trials (CANVAS and CANVAS-Renal), we used its efficacy data as coming from a single trial. The nine trials that were not considered in the two previous meta-analyses (4,5) included one GLP-1RA trial (39) and eight SGLT2i trials (16,19,22,23,34–37,41).

Mean age of the trial participants ranged between 62 and 72 years, 36% of the trial participants were women, and mean BMI varied from 28.2 to 32.8 kg/m² across trials. Four trials only enrolled patients with diabetes, and in these trials mean duration of diabetes ranged from 9.2 to 15.8 years. Thirteen trials also enrolled patients without a history of cardiovascular disease. The lowest proportion of trial participants with cardiovascular disease at baseline was 37%, in Dapagliflozin And Prevention of Adverse outcomes in Chronic Kidney Disease (DAPA-CKD). Baseline eGFR varied between 43 and 85 mL/min/1.73 m², and the proportion of participants with micro- or macroalbuminuria ranged from 25 to 100% in Canagliflozin and Renal Events in Diabetes with Established Nephropathy Clinical Evaluation (CREDENCE) (Table 1). All included trials were deemed high quality and with a low risk of bias (Supplementary Tables 3 and 4).

Summary HRs were significant for GLP-1RA regarding cardiovascular mortality (HR 0.87, 95% CI 0.80–0.96), MACE (HR 0.87, 95% CI 0.79–0.97), hospitalization for heart failure (HR 0.89, 95% CI 0.81–0.99), and the composite renal outcome (HR 0.84, 95% CI 0.73–0.97). For SGLT2i, the summary HR for cardiovascular mortality was 0.86 (95% CI 0.81–0.92), and for MACE it was 0.88 (95% CI 0.82–0.95). The SGLT2i trials’ summary HRs for hospitalization for heart failure and the composite renal outcome were 0.70 (95% CI 0.67–0.74) and 0.65 (95% CI 0.58–0.74), respectively (Fig. 1). For the secondary outcomes, among GLP-1RA trials, the summary HR was 0.89 (95% CI 0.82–0.96) for all-cause mortality, 0.91 (95% CI 0.82–1.02) for MI, and 0.84 (95% CI 0.76–0.92) for any stroke. For SGLT2i, these HRs were 0.88 (95% CI 0.82–0.94), 0.89 (95% CI 0.78–1.00), and 0.92 (95% CI 0.75–1.13) (Supplementary Figs. 2–4). I² statistics generally remained <50% and Q statistics reached P < 0.05 only for HRs for MACE among GLP-1RA trials and for the composite renal outcome among SGLT2i trials.

Summary 5-year ARD estimates for cardiovascular mortality, MACE, hospitalization for heart failure, and the composite renal outcome were statistically significant and varied from −0.80% (95% CI −1.37 to 0.23) to −4.25% (95% CI −6.39 to 2.10) (Fig. 2). The largest risk reduction for hospitalization for heart failure was estimated among SGLT2i trials. I² statistics were >50% for SGLT2i trials for hospitalization for heart failure and the composite renal outcome. Also, Q statistics reached P < 0.01 for these outcomes among SGLT2i trials.

Five-year absolute risk reductions for all-cause mortality (−1.53%, 95% CI −2.45 to 0.60, for GLP-1RA and −1.84%, 95% CI −2.90 to 0.77, for SGLT2i) with both drug classes were larger than for cardiovascular mortality (−1.16%, 95% CI −1.92 to 0.40, for GLP-1RA and −1.33%, 95% CI −2.16 to 0.50, for SGLT2i). For MI and any stroke, summary 5-year ARDs were small and only statistically significant for any stroke with GLP-1RA (−0.83%, 95% CI −1.31 to 0.36) (Supplementary Figs. 5–7). For these secondary outcomes, all I² statistics were ∼0 and none of the Q statistics reached P < 0.05.

The cardiovascular mortality rate observed in the control arm ranged from 0.8 to 2.4 per 100 person-years among GLP-1RA trials and from 0.7 to 12.5 per 100 person-years among SGLT2i trials. We did not find any association between the control arm cardiovascular mortality rate and log-HRs for both drug classes. Five-year ARDs generally improved with the control arm cardiovascular mortality rate and most prominently for hospitalization for heart failure with SGLT2i (slope −1.44, P < 0.001), from −0.9% (95% CI −1.7 to 0.1) to a predicted −11.6% (95% CI −14.2 to 9.0) in high-risk populations (Fig. 3, Supplementary Tables 8 and 9, and Supplementary Fig. 8). These risk reductions correspond to number needed to treat (NNT), defined as the inverse of absolute risk reduction, of 109 (95% CI 59–159) and 9 (95% CI 7–11), respectively. For GLP1-RAs, none of the associations for ARDs were significant (Supplementary Table 9).

After removal of trials with investigation of SGLT2i and sodium–glucose cotransporter inhibitors (23,34) or acute setting trials (25,26,34), slopes for the association between 5-year ARDs and baseline cardiovascular mortality rates remained similar (Supplementary Figs. 9 and 10). In analyses omitting SGLT2i trials that exclusively enrolled patients with established heart failure at baseline, the slope for 5-year ARD of heart failure changed from −1.44 (P < 0.001) to −1.67 (P < 0.001). For other main outcomes, slopes generally steepened but did not achieve statistical significance (Supplementary Fig. 11).

In sensitivity analyses excluding trials with divergent definitions of MACE (12) or the composite renal outcome (16–19,22–26,34), as well as excluding trials with evaluation of drugs currently not available on the market (23,34,38–40), slopes remained similar (Supplementary Table 9 and Supplementary Figs. 12–14).

We summarized data from 22 trials with 154,649 participants including data from nine newer trials (16–19,22,23,25,34–37,39,41). The findings from our updated meta-analyses of HRs augment the existing evidence on relative cardiovascular and renal benefits of GLP-1RA and SGLT2i. We found statistically significant summary HRs for both drug classes for cardiovascular mortality, MACE, hospitalization for heart failure, and the composite renal outcome. For our secondary end points, all-cause mortality and MI, hazard rates were statistically significantly reduced by GLP1-RAs and SGLT2i as well. For stroke, the hazard rate reduction was statistically significant for GLP1-RAs. We generally did not find strong indications for heterogeneity in these relative effect estimates for both drug classes. Moreover, the meta-regression analyses of log HRs did not reveal important differences in relative treatment efficacy by baseline cardiovascular mortality rates. Finally, we estimated absolute benefits achieved with these agents and generally found more substantial 5-year absolute risk reductions for outcomes among high-risk participants. Statistically significant increasing trends for reduction of the 5-year risk of hospitalization for heart failure were estimated among SGLT2i trials.

When relative effects remain constant across the participants’ baseline risk levels, absolute risk reductions with efficacious treatments are expected to increase when the control arm’s risk is high. Thus, the NNT will improve (42). Indeed, it became apparent that the SGLT2i trials’ cardiovascular mortality rates in the control arm were indicative of rates of heart failure. This resulted in predictions of significantly larger 5-year heart failure risk reductions (>10%) in trials with higher cardiovascular mortality rates, >12 events per 100 person-years. Such absolute risk reductions will translate to lower NNT estimates, reaching levels <10. In comparison, NNT estimates for heart failure were >100 for SGLT2i trials where low-risk patients were enrolled.

Our estimates of absolute treatment benefits with GLP-1RA appear more favorable than those reported by Sattar et al. (4) for cardiovascular and renal outcomes. First, NNTs in this analysis were calculated over a weighted average median follow-up duration of 3.0 years. Second, we additionally included the FREEDOM Cardiovascular Outcomes (CVO) trial (39), with evaluation of continuous subcutaneous infusion of an exendin-4–based agonist (exenatide) but failure to demonstrate significant relative reduction of cardiovascular event rates. We believe that leaving out this trial would potentially lead to biased estimates of both summary HRs and ARDs. Moreover, concerns that exendin-4–based agonists are less effective than human glucagon-like peptide 1–based molecules have been refuted (4).

In a recent study where investigators explored absolute benefits for SGLT2i trials, investigators analyzed heart failure incidence rate differences among 10 trials and found more prominent rate reductions among trials with higher baseline risk (43). In another recent study investigators estimated anticipated absolute effects using time-specific risk ratios from nine SGLT2i trials and baseline risks for four theoretical populations of patients with heart failure. The baseline risks were based on external cohort study data to facilitate the application of their meta-analysis (44). They predicted that the number of hospitalizations for heart failure and cardiovascular deaths per 1,000 patients over a 2-year period would increase among patients diagnosed with heart failure at highest risk. Highest risk was defined according to new diagnosis with heart failure in the hospital.

In agreement with these studies, we believe that distinguishing between relative and absolute treatment benefits provides additional and clinically useful insights. Absolute treatment benefits can be estimated with an indirect estimation method combining relative efficacy estimates from trials and baseline rates derived from external data sources or trial subgroups. This method would be valid when such baseline rates are generalizable to clinical practice and when the trials’ relative effects are transportable. The latter assumption is difficult to ascertain. Therefore, we decided for our analyses to estimate ARDs directly from the reported trial data (42,45,46) while standardizing for differences in follow-up duration to a 5-year time horizon. However, investigators can still opt to use the summary HRs in combination with user-defined control event rates, for example, as incorporated in a state-transition model, to estimate ARDs.

Some limitations of our analyses need to be acknowledged. First, the range in cardiovascular mortality rates among the control arms of the GLP-1RA trials was more limited than among SGLT2i trials for which these rates could be much higher. The major reason was that trials with enrollment of patients with heart failure included only evaluation of SGLT2i, and these trials had the highest cardiovascular mortality rate. For generalizability of our results, the variation in baseline risk should reflect the risks of target patient populations as recommended in the clinical guidelines. Because SGLT2i are generally exclusively recommended for use in patients with existing heart failure, the 5-year ARD estimates observed in the high-risk trial populations could remain generalizable to these patients. On the other hand, we demonstrated that heart failure hospitalization rates can be decreased with GLP-1RA as well, and efficacy seems to be maintained in those with established heart failure (18,19,22,34–37). More research is needed to assess whether absolute treatment benefits with GLP-1RA for this outcome could increase to an extent similar to that observed for SGLT2i. Second, for our estimates of 5-year ARDs, we could not incorporate effects of competing death rates, which may have varied across study arms and therefore attenuate ARDs. Unfortunately, efficacy data for death rates other than ASCVD and hospitalization for heart failure are generally not reported. We recommend that trialists provide such data to improve estimation of absolute treatment benefits in meta-analyses. Third, we could not always estimate summary HRs and 5-year ARDs while including data from all trials. This was the case for outcomes other than all-cause and cardiovascular mortality. However, generally we obtained a sample size of at least five trials or more, which is considered reasonable for performing meta-analysis and meta-regression analysis (47). Fourth, we used combined end point definitions for many of the outcomes (e.g., mortality, MACE, stroke, renal outcomes). Heterogeneity within the distribution of single end points may have affected our summary effect estimates. For example, we used a combined stroke end point of fatal and nonfatal stroke subtypes. Yet, for SGLT2i, it has been established that relative effects are generally only significant for hemorrhagic strokes (48). As such, it can be expected that in SGLT2i trials for which the proportion of hemorrhagic strokes was high, HRs for any stroke were more protective. In addition, it should be noted that the 5-year ARD estimates of combined end points are sensitive to the baseline risk distribution of outcome subtypes. For example, for MACE, the overall ARD is mainly determined by the risk reduction of cardiovascular mortality, although this contribution may vary per trial. It was not feasible to evaluate all potential contributors of interstudy heterogeneity in our analyses, and we chose to focus our analyses on the impact of the cardiovascular mortality rate in the control arm. We assumed that some contributors would correlate with this parameter and that remaining sources of interstudy heterogeneity could be accounted for by using random effects. Fifth, we used aggregated data and therefore no causal inferences could be made about associations from our meta-regression analyses. As such, it is not possible to make conclusions about lack of effect modification across varying cardiovascular risk factor levels. Finally, due to limiting our search strategy to PubMed as defined by prior meta-analyses (4,5), there is a possibility of incomplete retrieval of data from secondary analyses of included clinical outcome trials (5).

Following the publication of earlier cardiovascular outcome trials (25–30,49–52), the American Diabetes Association initially recommended addition of GLP-1RA and SGLT2i in those with established ASCVD, heart failure, or CKD. However, these recommendations only apply to patients who have not reached their hemoglobin A_1c (HbA_1c) goal on metformin therapy (7). More recently these recommendations were expanded to those at elevated ASCVD risk due to end organ damage or having multiple cardiovascular risk factors without established disease. In addition, recommendations for initiation were made independent of baseline and individualized target HbA_1c levels, as well as metformin use (6). Also, in the recently published Joint American Diabetes Association/European Association for the Study of Diabetes guidelines for the management of hyperglycemia in type 2 diabetes, similar recommendations were included. These guidelines’ recommendations apply for those with no established disease but multiple cardiovascular risk factors (such as age ≥55 years, obesity, hypertension, smoking, dyslipidemia, or albuminuria). In addition, where such high-risk patients need additional cardiorenal risk reduction, combination treatment with both a GLP-1RA and an SGLT2i is suggested, regardless of background metformin use and HbA_1c levels (53).

The joint impact of multiple risk factors on expected treatment benefits is preferably accounted for by estimation of absolute risks for a scenario with and without initiation of treatment. When absolute risk or cumulative incidence curves are expressed over a longer-term horizon, differences in (restricted) event-free survival time can also be communicated in a meaningful way. For example, models have been developed to generate individualized benefits with lipid-lowering, blood pressure–lowering, and antithrombotic agents (54). Our findings of varying absolute treatment benefits by cardiovascular risk call for development of such risk tools to guide trial design and reporting of results using risk strata. Finally, these tools can also support more patient-centered decision-making on these agents.

In conclusion, relative effects of novel diabetes drugs are preserved across baseline cardiovascular risk, whereas absolute benefits increase in those at high cardiovascular risk, particularly regarding hospitalization for heart failure. Our findings suggest a need for baseline risk assessment tools to identify variation in absolute treatment benefits with GLP-1RA and SGLT2i to improve the decision-making and cost-effectiveness for these agents (55–57).

Acknowledgments. The authors thank Carrie Levinson, Reference & Instruction Librarian, Mount Sinai Health System, who provided expertise for developing the search strategy.

Funding. Research reported in this publication was supported by the National Heart, Lung, and Blood Institute of the National Institutes of Health under award no. R01HL153456. M.G.M.H. has received research funding from the Netherlands Organization for Health Research and Development, the German Innovation Fund, and Netherlands Educational Grant (“Studie Voorschot Middelen”), and additional research funding from the Gordon and Betty Moore Foundation. K.E.F. is the recipient of an Innovative Clinical or Translational Science award from the American Diabetes Association (1-18-ICTS-041).

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Duality of Interest. M.G.M.H. receives royalties from Cambridge University Press for a textbook on medical decision-making, reimbursement of expenses from the European Society of Radiology (ESR) for work on the ESR guidelines for imaging referrals, and reimbursement of expenses from the European Institute for Biomedical Imaging Research for membership on the Scientific Advisory Board. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.M.R.-V. and B.S.F. made the analysis plan, performed analyses, and were involved in writing of the manuscript. J.M.R.-V. and B.S.F. performed the systematic search and study selection. J.M.R.-V. and M.T. conducted data extraction. All other authors contributed to the discussion and reviewed and edited the manuscript. All authors had full access to all the data and had the final responsibility for the decision to submit for publication.

Prior Presentation. Parts of this study were presented in abstract form at the 82nd Scientific Sessions of the American Diabetes Association, New Orleans, LA, 3–7 June 2022.