Ovid-MEDLINE, PubMed and CENTRAL databases were searched for articles without time or language restriction up to November 15, 2016 using high sensitive search strategies (S1 Table). References from pertinent studies and eminent full-reviews were screened for additional articles. The search was designed and performed by three Authors (DB, GB and RB).

We included any randomized controlled trial (RCT) and quasi-RCT (trials in which allocation to treatment was made by alternation, use of alternate medical records, date of birth or other expected methods) testing the effects of any antioxidant supplementation (including but not limited to vitamin A, vitamin C, vitamin E, selenium, zinc, methionine or ubiquinone, either alone or in combination to other treatments) on renal endpoints in patients with diabetic kidney disease (DKD).

Studies were considered regardless of dosage or duration of administration of antioxidants and type of comparator. For cross-over studies the first period was considered.

DKD was defined as evidence of renal damage (chronic kidney disease, CKD) related to diabetic disease. CKD, in turn, was defined according to the National Kidney Foundation-Kidney Disease Outcomes Quality Initiative (NKF KDOQI) guidelines by the presence of a reduced glomerular filtration rate (GFR) <90 mL/min/1.73 m² or the persistence of hyperfiltration and/or urinary abnormalities such as pathological albuminuria, proteinuria or hematuria in subjects with GFR≥90 mL/min/1.73 m² [9].

The primary endpoint of interest was progression to end-stage kidney disease (ESKD), defined as need for chronic renal replacement therapy, kidney transplantation or doubling of serum creatinine from baseline values. Secondary outcomes were change in albuminuria, proteinuria, serum creatinine and renal function (creatinine clearance/eGFR). Information on any adverse event was also collected, when reported.

Studies were excluded if: 1) dealing with diabetic patients with no evidence of kidney disease or dealing with CKD patients without diabetes; 2) focusing on DKD patients on chronic or acute renal replacement therapy (e.g. hemodialysis or peritoneal dialysis); 3) testing the effects of synthetic antioxidants (e.g. bardoxolone methyl) or antioxidant mixtures which exact composition was not defined; 4) not providing short or long-term data on the outcomes of interest.

Studies where at least part of the population satisfied the above criteria were included in the review.

Titles and abstracts were screened independently by two authors (VC, GG) who discarded studies that were not pertinent to the topic. Case reports, reviews, editorials and studies performed on children (age<18) were excluded from qualitative analyses but screened for potential additional references. Two Authors (VC, GG) independently assessed the retrieved abstracts and the full text of these studies to determine eligibility according to the inclusion/exclusion criteria.

A third reviewer (GD) solved possible discrepancies on study judgments. Data extraction and analysis were carried out by two reviewers (VC, GG) and independently verified by another (GD).

Pooled meta-analyses were performed for outcomes in which data were available from more than two studies. Data on outcomes reported by single studies or in a descriptive way were reported narratively. The effects of treatment on continuous variables were assessed as mean difference (MD) or standardized mean difference (SMD), as appropriate.

Data were pooled using the random-effects model. To ensure robustness of the model and susceptibility to outliers pooled data were also analyzed with the fixed-effects model. Data expressed as median and range were converted to mean and SD by applying the Hozo formula [10]. Heterogeneity was assessed by the Chi² test on N-1 degrees of freedom, with an alpha of 0.05 considered for statistical significance and the Cochrane-I² [11]. I² values of 25%, 50% and 75% were considered to correspond to low, medium and high levels of heterogeneity, respectively. Possible sources of heterogeneity were explored by sensitivity analysis.

Publication bias was evaluated by the Egger’s regression test and by visual inspection of funnel plot. Statistical analyses were performed by GD, VC and GG using Review Manager (RevMan; Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014) and Stata/IC (Version 13.1, Stata Corp LP, Texas, USA).

Likelihood of bias in the single RCTs was evaluated according to the checklist developed by the Cochrane Renal Group that considers the presence of potential selection bias (random sequence generation and allocation concealment), performance bias (blinding of investigators and participants), detection bias (blinding of outcome assessors), attrition bias (incomplete outcome data), reporting bias (selective reporting) and possible other sources of bias.

A “Summary of findings” table summarizing pooled evidence for the main outcomes was constructed according to the GRADE method [12]. The five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness and publication bias) were taken into account to assess the quality of a body of evidence for the main pre-specified outcomes. All decisions to downgrade or upgrade the quality of studies were justified using footnotes and comments were made, when appropriate, to help readers' understanding of the review.

Fig 1 shows the flow diagram of the study selection process. Thirteen thousand five hundred and nineteen potentially relevant citations were initially found. By screening titles and abstracts, a total of 13491 references were excluded for various reasons (search overlap, study population or intervention not pertinent, review articles or case reports). Amongst the 28 studies selected for full text examination, 13 studies were excluded because not RCTs (n = 6) or because the study population (n = 6) or the intervention (n = 1) did not fulfil the review criteria. A total of 15 articles referring to14 full studies (4345 participants) were reviewed in detail. Eight studies [13–20] (327 participants) providing suitable numerical data on the outcomes of interest contributed to pooled meta-analyses. Main characteristics of the included studies are described in Table 1.

Among the 14 RCTs reviewed [13–27], four had a crossover design [14, 16, 17, 21]. One study was multicenter [24, 27]. The study population ranged from ten [16] to 3654 [24] participants.

Ten studies [13–17, 19, 20, 22, 24, 25] were double-blind and three were open label [18, 23, 26]; in another study [21] blinding of participants and personnel was not specified.

Two RCTs [13, 16] enrolled subjects with insulin-dependent diabetes mellitus (IDDM), nine studies [14, 15, 17–21, 23, 25] were conducted on individuals with non-insulin-dependent diabetes mellitus (NIDDM) and two studies [22, 24] included both IDDM and NIDDM patients. In Jadhav et al [26] information on the type of diabetes was not provided. Study follow-up varied from 4 weeks [13, 14, 20] to 4.5 years [24]. The mean age of patients ranged from 18.9 [16] to 65.4 [24] years. Male gender spanned from 27% [20] to 100% [13]. Diabetes vintage varied from 4.7 [20] to 13.5 [19] years. Single antioxidant therapy was vitamin C [13, 22], vitamin E [16, 20, 23, 24], zinc [17, 18] and sylimarin [25]. In two studies [14, 26] patients received vitamin C in combination with vitamin E. One study [19] tested the effects of lipoic acid in combination with pyridoxine while in another one [21] vitamin C was combined to vitamin E and beta-carotene. Farvid et al. [15] tested the effects of a combined regimen of zinc and magnesium against vitamin C, vitamin E and the combination of both substances. Eleven studies [13–17, 19–22, 24, 25] compared antioxidants to placebo while in three studies [18, 23, 26] the control group did not take any treatment.

The daily dose of administered vitamin C ranged from 500 mg [22] to 6 g [13]; the dose of vitamin E varied from 400 [24] to 1200 mg [16]; zinc supplements ranged from 30 [15, 17] to 50 mg [18] lipoic acid was administered at a daily dose of 800 mg [19]. Jadhav et al. [26] did not provide information on the administered dose of antioxidants.

Risk of bias of RCTs is summarized in Table 2. Information of the random sequence generation was provided in five studies [15, 17, 19, 20, 24], as well as allocation concealment [13, 15–17, 22].

Ten RCTs were double blind [13–17, 19, 20, 22, 24, 25] and two studies were open label [23, 26]. In two studies, blinding was unclear [18, 21]. Attrition bias was low in all but one study [21] in which the overall drop-out rate was as high as 29%. Reporting bias was low in two studies [20, 24] and unclear in the remainder. Risk of funding bias was potentially high in one study [24]. No further sources of bias were identified.

Data on progression to end-stage kidney disease defined as chronic renal replacement therapy was available in only one RCT [24]. Conversely, information on the need for kidney transplantation or doubling of serum creatinine was not available in any of the included trials. Change in urinary albumin excretion was analyzed in all RCTs [13–26]. Two studies, respectively, provided data on change in proteinuria [14] and serum creatinine [15]. Three studies [13, 16, 17] reported information on renal function. Data on adverse events were available in eight RCTs [13–17, 19, 20, 24].

In Lonn [24] et al., the number of patients experiencing needing to start chronic dialysis due to DKD progression was not different between the intervention and the control group (p = 0.96).