Effects of sodium-glucose co-transporter 2 (SGLT2) inhibition on renal function and albuminuria in patients with type 2 diabetes: a systematic review and meta-analysis

Search strategy

We conducted a systematic search of PubMed, Embase and Cochrane Central Register of Controlled Trials (CENTRAL) databases through June 19th 2016. The search strategy is provided in Item S3; we used medical subject headings, as well as free-text search terms, including SGLT2 inhibitors, canagliflozin, dapagliflozin, empagliflozin, atigliflozin, ‘bi 44847’, ertugliflozin, ipragliflozin, luseogliflozin, remogliflozin, sergliflozin, sotagliflozin and tofogliflozin. We also conducted a manual search of references of existing reviews in this field to identify additional relevant studies.

Study selection

Two reviewers (LX and YL) independently screened the search results and retrieved relevant studies for further evaluation. The retrieved full-text articles were examined by two reviewers (LX and YL) in parallel for inclusion according to predetermined criteria. We included RCTs conducted on adult type 2 diabetic patients that compared SGLT2 inhibitors with either placebo or other antidiabetic drugs and reported changes in eGFR and/or urine albumin/creatinine ratio (ACR). Only manuscripts published in English were included. For studies reporting renal outcomes in forms other than eGFR and urine ACR (i.e., serum creatinine, urine protein excretion, etc.), an e-mail was sent to the corresponding author requesting further data. For multiple publications from the same study, only the first publication reporting renal outcomes was included. Disagreement was resolved by discussion and/or consultation with a third reviewer (PX).

Data extraction and validity assessment

Two reviewers (LX and JL) independently used a standard data extraction tool to record the following properties of each study: the study characteristics (author, year, study design, method of randomization, duration of follow-up, and number of dropouts), participant characteristics (sample size, age, sex, duration of diabetes, baseline HbA1C level, baseline blood pressure, eGFR and urine ACR), therapeutic intervention (type of SGLT2 inhibitor, dose, frequency and duration of treatment), concomitant therapies (concomitant antidiabetic therapy and RAAS inhibitors), comparison groups (placebo-controlled or active-controlled), outcomes of interest (means and standard deviations (SDs) of changes in eGFR and urine ACR in treatment and control groups), whether outcomes of chronic kidney disease (CKD) subjects (defined as eGFR < 60 ml/min per 1.73 m² or microalbuminuria or macroalbuminuria) were reported, and the funding source. The program g3data (www.frantz.fi/software/g3data.php) was used to extract relevant data that were reported in figures but not in the text.

Study quality was evaluated by two authors (LX and YL) independently using the ‘Risk of bias’ assessment tool from the Cochrane Handbook for Systematic Reviews of Interventions, version 5.1 (2011). The domains of assessment included random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting and other bias.

Statistical analysis

Outcome measures for each trial were mean differences of changes (calculated as (end of trial value for treatment group—baseline value for treatment group)—(end of trial value for control group—baseline value for control group)) in eGFR and urine ACR. For studies in which SD was not directly reported, SD was calculated from SE (standard error) or 95% confidence intervals (CI), or imputed as recommended in the Cochrane Handbook for Systematic Reviews of Interventions, version 5.1 using a correlation coefficient of 0.8 (with the formula provided in Item S4) (Higgins & Green, 2011). For studies with more than one SGLT2 treatment arm, these groups were combined to create a single treatment arm (with the formulae provided in Item S5) (Higgins & Green, 2011). The inverse variance method was used to estimate the pooled weighted mean differences (WMDs) in eGFR and urine ACR. As clinical and statistical heterogeneity were anticipated, we decided a priori to use the random-effects model in our data synthesis.

Statistical heterogeneity was quantified using the Cochrane Q test and I² statistic. Subgroup analysis was planned for the type of SGLT2 inhibitor, placebo or active control, concomitant use of RAAS inhibitors, trial duration, mean baseline age, mean duration of diabetes, mean baseline HbA1C level, whether the study population were CKD patients (eGFR < 60 ml/min per 1.73 m² or microalbuminuria or macroalbuminuria), and whether the study population had hyperfiltration (eGFR ≥ 125 ml/min per 1.73 m²) (Dahlquist, Stattin & Rudberg, 2001). Test for subgroup differences were carried out using RevMan 5.3 (The Nordic Cochrane Centre, The Cochrane Collaboration).

The possibility of publication bias was assessed using funnel plots and Egger’s test. Sensitivity analysis excluding trials with relatively high risk (defined as ≥1 item with high risk or ≥2 items with unclear risk in the ‘Risk of Bias’ assessment tool) was performed.

Statistical analyses were performed using the Stata 12.0 software package (StataCorp, LP, College Station, TX) and RevMan 5.3 (The Nordic Cochrane Centre, The Cochrane Collaboration). Statistical significance was set at p < 0.05 for all analyses.

Study selection and characteristics

The results of the literature search and study selection are shown in Fig. 1. Details of the study selection process can be found in Item S6. This search process led to inclusion of 47 studies (Bailey et al., 2015; Barnett et al., 2014; Bode et al., 2013; Bolinder et al., 2012; Cefalu et al., 2013; DeFronzo et al., 2015; Fonseca et al., 2013; Forst et al., 2014; Haring et al., 2014; Haring et al., 2013; Inagaki et al., 2014; Ji et al., 2015; Ji et al., 2014; Kadowaki et al., 2014; Kaku et al., 2014; Kashiwagi et al., 2015a; Kashiwagi et al., 2015b; Kashiwagi et al., 2015c; Kashiwagi et al., 2015d; Kohan et al., 2014; Kovacs et al., 2014; Lambers Heerspink et al., 2013; Lavalle-Gonzalez et al., 2013; Lewin et al., 2015; Lu et al., 2016; Nauck et al., 2011; Nishimura et al., 2015; Qiu, Capuano & Meininger, 2014; Ridderstrale et al., 2014; Rodbard et al., 2016; Roden et al., 2013; Rosenstock et al., 2016; Rosenstock et al., 2014; Rosenstock et al., 2015; Ross et al., 2015; Schernthaner et al., 2013; Schumm-Draeger et al., 2015; Sha et al., 2014; Strojek et al., 2011; Tikkanen et al., 2015; Wanner et al., 2016; Weber et al., 2016; Wilding et al., 2013a; Wilding et al., 2013b; Wilding et al., 2009; Wilding et al., 2012; Yale et al., 2013) with 22,843 participants in our meta-analysis.

The characteristics of the included studies are shown in Table 1. Of the 47 studies, 46 studies with 22,603 participants reported changes in eGFR, and 17 studies with 7,285 participants reported changes in urine ACR. Five SGLT2 inhibitors, including dapagliflozin, canagliflozin, empagliflozin, ipragliflozin and tofogliflozin, were assessed. A total of 38 studies were placebo-controlled, and 9 were controlled by other antidiabetic medications, including metformin, glimepiride, glipizide, linagliptin, and sitagliptin. The trial durations ranged from 4 weeks to 156 weeks. Six studies reported outcomes of CKD subjects. In other studies, the mean baseline eGFR ranged from 76.7 to 149.2 ml/min per 1.73 m², and the mean baseline urine ACR ranged from 6.7 to 143.7 mg/g. None of the studies reported outcomes in a group of patients with hyperfiltration. Collection of data on concomitant use of RAAS inhibitors was planned a priori; however, this was impeded as only six studies reported the number of patients using RAAS inhibitors (Barnett et al., 2014; Bode et al., 2013; Lambers Heerspink et al., 2013; Sha et al., 2014; Wanner et al., 2016; Weber et al., 2016), among which only one study reported outcomes with stratification according to RAAS inhibitor usage (Weber et al., 2016).

Quality assessment (Figs. S1 and S2) revealed that 32 studies described random sequence generation, and that 34 described allocation concealment. Twenty-seven studies described blinding of participants and personnel. All 47 studies had a low risk of detection or reporting bias, and 31 studies had a low risk of attrition bias.

Assessment of publication bias

Visual inspection revealed some asymmetry in the funnel plots for both eGFR (Fig. S3) and urine ACR (Fig. S4). Egger’s regression confirmed statistical significance of publication bias for urine ACR (p = 0.002), but not eGFR (p = 0.057).

Effects of SGLT2 inhibitors on eGFR

In pooled analysis of 46 studies reporting eGFR changes, no significant difference was observed between the SGLT2 treatment group and control group (calculated as ((end of trial eGFR for SGLT2 inhibition group—baseline eGFR for SGLT2 inhibition group)—(end of trial eGFR for control group—baseline eGFR for control group))) (WMD −0.33 ml/min per 1.73 m², (95% CI [−0.90 to 0.23]); Fig. 2). Substantial heterogeneity was detected across studies (I² = 63.1%, p < 0.001).

Pre-specified subgroup analyses were conducted to identify possible sources of heterogeneity (Fig. 3). Analysis of trial duration revealed that in the trials with a duration of 4–26 weeks, SGLT2 inhibition was associated with a larger eGFR reduction than control (WMD -0.98 ml/min per 1.73 m², (95% CI [−1.42 to −0.54]), I² = 0.9%, 29 studies with 10,946 patients); while in trials that lasted longer than 52 weeks, SGLT2 inhibition was associated with slower eGFR decline than control (WMD 2.01 ml/min per 1.73 m², (95% CI [0.86 to 3.16]), I² = 46.0%, 5 studies with 5,334 patients). SGLT2 inhibitors were observed with a larger eGFR reduction than placebo and a smaller eGFR reduction than active control. No significant eGFR difference was observed between the SGLT2 inhibitor group and control group in patients with CKD (WMD −0.78 ml/min per 1.73 m², (95% CI [−2.52 to 0.97]), I² = 65.0%, 5 studies with 1,574 patients). No significant subgroup differences were observed in subgroup analysis for the type of SGLT2 inhibitor, the mean duration of diabetes, or the mean baseline HbA1C level. Subgroup analyses for patients with hyperfiltration and RAAS inhibitor use had been planned but were hampered by lack of such stratification in the published data.

Effects of SGLT2 inhibitors on urine ACR

In pooled analysis of 17 studies evaluating the urine ACR, SGLT2 inhibition was not associated with statistically significant albuminuria reduction (WMD −7.24 mg/g, (95% CI [−15.54 to 1.06]), Fig. 4). Substantial heterogeneity was observed across studies (I² = 39.2%, p = 0.03).

Subgroup analyses (Fig. 5) suggested that SGLT2 inhibition was associated with a significant urine ACR reduction in the participants with CKD (WMD −107.35 mg/g, (95% CI [−192.53 to −22.18]), I² = 35.7%, 5 studies with 1,063 participants). Stratification according to the duration of diabetes mellitus revealed a trend of enhanced albuminuria reduction in the patients with a longer duration of diabetes mellitus (for the patients with a history of ≤5 years, WMD 23.21 mg/g, (95% CI [−0.23 to 46.65]), for the patients with a history of 5–10 years, WMD −7.06 mg/g, (95% CI [−13.93 to −0.20]), and for the patients with a history of >10 years, WMD −20.37 mg/g, (95% CI [−42.77 to 2.04]), p = 0.02 for subgroup difference). Subgroup analyses for the different SGLT2 inhibitors, use of placebo vs. active control, trial duration, mean baseline age and mean baseline HbA1C level did not reveal any significant subgroup difference.

Sensitivity analysis

Similar results were observed when analyses were limited to trials with relatively low risk (defined as no item with high risk and no more than 1 item with unclear risk) for eGFR (WMD −0.06 ml/min per 1.73 m², (95% CI [−0.90 to 0.78]), I² = 73.0%, 24 trials with 14,535 participants) and urine ACR (WMD −7.25 mg/g, (95% CI [−17.25 to 2.76]), I² = 48.0%, 12 trials with 5,308 participants).