We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines to conduct this systematic review and meta-analysis[14].
Search Strategy and Selection Criteria
The PubMed, Embase, Cochrane Central Registry of Controlled Trials and Web of Science databases were searched by two independent authors (Y. L and Y.Y) to identify and select relevant articles published prior to 9th April 2020. The following keywords were used as search terms: (“sodium-glucose cotransporter-2 inhibitor” OR “SGLT-2 inhibitor” OR canagliflozin OR dapagliflozin OR empagliflozin OR ertugliflozin OR ipragliflozin OR luseogliflozin OR tofogliflozin) AND (heart failure) AND (“clinical trial” OR trial OR RCT). Additional studies were selected by manually screening the references of articles identified by the search. No restrictions were placed on time or language. Abstracts and presentations from all major conference proceedings between 2018 and 2020, including the European Society of Cardiology, American Heart Association, American College of Cardiology and American Diabetes Association, were also reviewed.
The titles and abstracts of retrieved articles were reviewed to exclude clearly irrelevant studies, and then, the full texts of the remaining articles were examined. The following predetermined inclusion criteria were used: (1) randomized controlled trials evaluating SGLT-2 inhibitors in patients with T2DM and HF; (2) the key outcomes of interest are the composite of CVD/HHF, individual CVD, individual HHF, and all-cause mortality (ACM); serious adverse events served as safety outcomes; (3) data included hazard ratios (HR, SGLT-2i vs. placebo) estimated by Cox regression analysis in terms of each outcome, or the event frequency, incident rate and sample size for different outcomes in both SGLT-2i and placebo groups were reported; (4) abstracts and conference proceedings from the previous 3 years were included if all relevant data were provided. We excluded studies if they (1) were observational or nonrandomized studies, (2) were reviews or case reports, or (3) used duplicate data.
Data extraction and outcome
Two independent investigators (Y. L and Y. Y) extracted data according to a predesigned data collection form, which included first author's name, year of publication, drugs, study design, trial identifier and name, age and sex of study participants, intervention and number of patients per arm, median follow-up, median observational time, data regarding cardiovascular events and serious adverse events. A third reviewer (J. X) was consulted to resolve disagreements.
The primary cardiovascular outcome was a composite of CVD (cardiovascular death) or HHF (hospitalization for heart failure), and the secondary outcomes were individual CVD, individual HHF or ACM. The safety outcomes were severe AEs reported by each trial.
Evaluation of Study Quality and Publication Bias
The quality of the included RCTs was assessed in accordance with the Cochrane quality assessment tool based on seven domains, including (1) randomized sequence generation, (2) allocation concealment, (3) blinding of participants and personnel, (4) blinding of outcome assessment, (5) incomplete outcome data, (6) selective outcome reporting, and (7) other sources of bias. Trials were rated as low risk, high risk, or unclear risk. We also tried to contact the authors of included RCTs to obtain more information for a more comprehensive analysis. To examine publication bias in the results of the meta-analysis, a funnel plot was used.
Statistical Analysis
Hazard ratios were used as the effect size measure for clinical cardiovascular outcomes, and they were further pooled using the generic inverse variance method[15]. The generic inverse variance method was used to combine log hazard ratios (log HR) and standard errors of the log HR (SElogHR). The log HRs were adjusted for a common set of confounders across studies, such as age, gender, and most importantly, time to events. Heterogeneity between studies was assessed using the Cochrane Q test and I2 statistic. As a guide, I2 <50% indicated low heterogeneity, 50–75% indicated moderate heterogeneity and >75% indicated high heterogeneity[16]. Studies with an I2 statistic of > 50% were considered to have substantial heterogeneity, and a random effects model was used to analyse these studies. Otherwise, a fixed effects model was initially used. We evaluated publication bias using funnel plots. A sensitivity analysis was performed by excluding one study that was only presented at a conference (CREDENCE trial). Alternatively, 3 studies reported the number of clinical outcomes (n/N) that occurred in both the SGLT-2i and placebo arms. For another sensitivity analysis, we calculated the incidence rate of outcome events and risk ratios (RRs) with data extracted from each RCT. Pooled RRs with 95% confidence intervals (95% CIs) were used as the effect size measure for clinical and safety outcomes by using the Mantel-Haenszel fixed effects model. A two‐tailed p value of less than 0.05 was considered statistically significant. Statistical analyses were performed using RevMan 5.3 (The Cochrane Collaboration, Copenhagen, Denmark).