A systematic review comparing the evidence for kidney function outcomes between oral antidiabetic drugs for type 2 diabetes

Background: The development of kidney disease is a serious complication among people with type 2 diabetes mellitus, associated with substantially increased morbidity and mortality. We aimed to summarise the current evidence for the relationship between treatments for type 2 diabetes and long-term kidney outcomes, by conducting a systematic search and review of relevant studies. Methods: We searched Medline, Embase and Web of Science, between 1st January 1980 and 15th May 2018 for published clinical trials and observational studies comparing two or more classes of oral therapy for type 2 diabetes. We included people receiving oral antidiabetic drugs. Studies were eligible that; (i) compared two or more classes of oral therapy for type 2 diabetes; (ii) reported kidney outcomes as primary or secondary outcomes; (iii) included more than 100 participants; and (iv) followed up participants for 48 weeks or more. Kidney-related outcome measures included were Incidence of chronic kidney disease, reduced eGFR, increased creatinine, ‘micro’ and ‘macro’ albuminuria. Results: We identified 15 eligible studies, seven of which were randomised controlled trials and eight were observational studies. Reporting of specific renal outcomes varied widely. Due to variability of comparisons and outcomes meta-analysis was not possible. The majority of comparisons between treatment with metformin or sulfonylurea indicated that metformin was associated with better renal outcomes. Little evidence was available for recently introduced treatments or commonly prescribed combination therapies. Conclusions: Comparative evidence for the effect of treatments for type 2 diabetes on renal outcomes, either as monotherapy or in combination is sparse.

Type 2 diabetes mellitus (DM) increases an individual's risk for health problems including cardiovascular disease, blindness, chronic kidney disease (CKD), and nerve damage [1][2][3][4] . The development of kidney disease is associated with other complications of type 2 diabetes and with poorer outcomes 1,3,5 . Therefore, slowing the development of, or preventing kidney disease is one aim of therapy 2 . Type 2 diabetes drugs are thought to play a major role in protecting the kidneys by controlling blood sugar levels and may confer additional protective effects according to specific drug profiles 3 . However, as kidney function declines, type 2 diabetes drug options become limited due to prescribing restrictions 2,3,5-7 . This presents a challenge for treating type 2 diabetes in patients with non-diabetic related kidney disease, as well as those with renal diabetic complications.
Treatment choice reflects a complex balancing of expected risks and benefits. A recent systematic review focused on vascular outcomes, glyclated hemoglobin (HbA1c), body weight, hypoglycaemia and common adverse events 8 . Here we focus on kidney-related outcomes as another important aspect of clinical care that clinicians must consider when prescribing drugs for type 2 DM. Our aim was to provide a summary of the current evidence of long term kidney outcomes, from comparative, long terms studies of oral antidiabetic drugs. We included the following outcomes: change in kidney function (estimated glomerular filtration rate), progression or development of proteinuria, development of end-stage renal disease (ESRD) and composite outcomes compared between different oral drugs for the treatment of type 2 DM.

Methods
The protocol for this systematic review was submitted, reviewed and approved by PROSPERO (International prospective register of systematic reviews, ref. 2016: CRD42016036646). The study was conducted and is reported in accordance with the PRISMA protocol (Supplementary File 1) 9 .

Data sources and searches
We searched the databases; Medline, Embase and Web of Science for articles published between 1 st January 1980 and 15 th May 2018. The search comprised keywords and MESH terms relating to three broad themes: kidney function, type 2 diabetes drugs and clinical studies. We limited the search to English-language studies, and studies in humans. The search strategies are in Supplementary Table 1 and Supplementary  Table 2 (Supplementary File 2). The reference lists of relevant reviews identified through the search were also screened.

Study selection
One reviewer (SW) screened all citations identified in the searches. Titles and abstracts for all studies were compared to the selection criteria. Then the full-text of selected studies were reviewed against the inclusion and exclusion criteria. Reviewer two (MI) was blinded to the articles selected by reviewer one and screened a 20% sample of the articles selected by reviewer one after the title screen. The studies chosen by the two reviewers were compared.
We defined the search and screening strategies before completing the searches. Studies were eligible for inclusion if they were clinical studies that (i) compared two or more classes of oral therapy for type 2 DM; (ii) reported kidney outcomes as primary or secondary outcomes; (iii) included more than 100 participants, and (iv) followed participants for 48 weeks or more. We restricted the review to oral antidiabetic drugs recommended at the initiation and first intensification of treatment 6 .
We did not include studies that reported only placebocontrolled comparisons as we were interested in the difference in effects between active therapy regimes to reflect therapy choices made in routine clinical care; placebo-controlled studies would not estimate this difference. Our definition of a kidney outcome was broad to identify as many studies as possible. We accepted any kidney-related outcome, including the incidence of chronic kidney disease, reduced estimated Glomerular Filtration Rate (eGFR), increased creatinine, 'micro' and 'macro' albuminuria, proteinuria, end stage renal disease (ESRD) and composite kidney outcomes. We did not include composite microvascular outcomes that combined kidney outcomes with other microvascular outcomes such as retinopathy or neuropathy.
Data extraction and quality assessment After study selection, using a predefined data collection tool, we extracted data for the following items: number of participants, study design, calendar years covered by the study, length of follow-up, drug comparison, mean age of study population, exclusion criteria for study, kidney measurements taken at baseline, mean duration of diabetes, mean HbA1c at baseline, primary outcome for the study, kidney outcomes reported and results for kidney outcomes reported. Reviewer one (SW) assessed each study for quality, using the GRACE 2014 10 items for observational comparative effectiveness research and the Cochrane Collaboration tool for assessing risk of bias in randomised trials 11 for RCTs. Figure 1 details the study selection process through which we found 9,086 potentially eligible studies. The first reviewer (SW) completed the initial title screen and selected 1,896 articles. The second reviewer (MI) was blinded and reviewed a 20% random sample of these articles. The agreement between reviewers was good, reviewer two selected an additional paper that was rejected after discussion. After subsequent discussions (SW, MI and LT), we selected 15 studies.

Results
We identified 15 eligible studies, seven of which were randomised controlled trials (RCTs) 12-18 and eight were observational studies [19][20][21][22][23][24][25][26] . Across the 15 studies, three RCTs 16-18 and one observational study 22 , reported changes in eGFR as an outcome. All seven RCTs 12-18 and two observational studies 22,25 investigated albumin-creatinine ratio (ACR) as an outcome. Six observational studies reported kidney endpoints, including kidney failure, nephropathy, acute dialysis and composite endpoints with eGFR 19-21,23,24,26 . Comparisons made, and outcomes studied are summarised graphically in Figure 2. Given the range of the kidney function outcomes reported and the drug class comparisons made we did not complete a meta-analysis of the results, instead we provide a narrative summary of studies. Selected studies and their findings are summarised in Table 1 and Table 2.
In total, we identified 32 direct comparisons between oral drugs for the treatment of type 2 DM: 22 comparisons between monotherapies, three comparisons between dual therapy combinations, and seven comparisons between dual therapies and monotherapies, outlined in Table 3. One study compared many combination therapy options to metformin; we did not include the triple therapy combinations from this study in our results, details of the comparisons are in Supplementary One study found the risk of eGFR falling to below 60 mL/min/ 1.73m 2 was greater in the sulfonylurea group compared to the metformin group 22 . Three found higher risks of kidney failure outcomes (various composites of codes for nephropathy, dialysis, renal transplant, ESRD, and reductions in eGFR) for sulfonylurea compared to metformin 20,21,23 . One study, using proteinuria as an outcome, found no difference between drug classes 22 . One further study reported higher rates of acute dialysis for people initiating metformin compared to sulfonylureas 26 .

Sulfonylurea monotherapy vs. thiazolidinedione monotherapy.
Findings from two RCTs showed differences in ACR that were not statistically significant 12,16 . However, one of these studies also showed an increase in mean eGFR among patients treated with a TZD, but a fall in the SU group 16 .

Combination therapy comparisons
Only three studies compared combination therapies.

Metformin plus sulfonylurea vs. metformin plus thiazolidinedione.
One RCT compared metformin plus sulfonylurea to metformin plus a thiazolidinedione 15 . They reported that ACR decreased in the metformin plus thiazolidinedione group and increased in the metformin plus sulfonylurea group 15 .

Sulfonylurea plus metformin vs. sulfonylurea plus thiazolidinedione.
One RCT compared sulfonylurea plus metformin to sulfonylurea plus thiazolidinedione 13 . The study found that the ACR increased in the sulfonylurea plus metformin group, and decreased in the sulfonylurea plus thiazolidinedione group 13 .

Metformin plus sulfonylurea vs. metformin plus gliptin (DPP4i).
One observational study compared metformin plus sulfonylurea combination therapy to metformin plus sitagliptin 25 . The results showed weak evidence that metformin plus sitagliptin improved the likelihood of reductions in ACR, with an odds ratio of 1.20 (95% CI: 0.99-1.47, P = 0.063) 25 .

Dual therapy vs. monotherapy
Three observational studies made seven comparisons between monotherapy options and combination therapy 20,21,23 . One study indicated that people taking metformin were at a lower risk of renal failure compared to people taking metformin plus sulfonylurea 21 . Another study found the opposite, people taking metformin plus sulfonylurea were at lower risk of kidney failure compared to metformin 23 . The same study found no differences in the risk of kidney failure compared to metformin in people prescribed; i) metformin plus thiazolidinedione, and ii) metformin plus gliptin. They also reported that people prescribed sulfonylurea plus thiazolidinedione, and a sulfonylurea plus DPP4i were at higher risk for kidney failure compared to metformin 23 .
Another observational study found no difference in eGFR outcomes between sulfonylurea monotherapy and metformin plus sulfonylurea combination therapy 20 .

Study quality
We assessed each study for quality, using the GRACE 2014 10 items for observational comparative effectiveness research and the Cochrane Collaboration risk of bias tool for RCTs 11 Supplementary   most common drug for initiating treatment, and the addition of other drugs to metformin is likely to be associated with progression or poor control of type 2 DM, comparing metformin to drug prescribed at the first stage of intensification is problematic, particularly for renal outcomes. Those people receiving treatment intensification will tend to be sicker, and distinguishing between the effects of treatment and the effects of the underlying disease may not always be possible.

Key findings
Overall, we have found a lack of consistent evidence of longterm differences in kidney outcomes between T2DM drugs. In comparisons of treatments for type 2 DM, for thiazolidinediones vs metformin, there is some evidence of reduced proteinuria -of four comparisons with ACR as an outcome (in combination or monotherapy), three favoured TZD and one showed no difference. Most evidence from observational research also suggested that metformin is associated with better kidney outcomes than sulfonylureas.
Despite frequent use of combination therapies for the treatment of diabetes, we found few studies that compared commonly used dual therapies that investigated renal outcomes.

Previous work
The finding that thiazolidinediones may reduce proteinuria compared with metformin is aligned with observations of other authors and supported by animal studies 27,28 . Though previous evidence is limited, other work suggests that TZDs could exert reno-protective effects via a number of pathways, including reducing blood pressure 28 . TZDs may also act directly in the kidneys via proliferator-activated receptor gamma (PPARg), found in the kidney (and in other tissue) 27,28 . However, changes in estimated GFR may reflect changes in fluid status rather than true changes in renal function, which was not measured directly in any study 29 .

Strengths
To our knowledge, this is the first systematic review of the comparative research literature that investigated the effects of type 2 diabetes drug regimens on renal function. We have conducted an extensive and detailed search, with broad definitions of renal function.

Limitations
We have focused on renal outcomes only but recognize this is just one of many safety and effectiveness factors to be considered when deciding treatment options. Despite the importance of careful monitoring and maintenance of kidney function for people with diabetes, we identified just 15 long-term studies reporting renal outcomes. Renal complications of type 2 diabetes take many years to develop after the onset of diabetes and studies may not be adequately powered or have sufficient length of followup to detect differences. Therefore, many studies have used the surrogate marker of changes in proteinuria as a marker of clinical renal outcomes. Further, initial changes in kidney function may be misleading. One included study indicates benefits of canagliflozin over glimipiride for kidney function decline at 104 weeks: however these benefits were not apparent until 52 weeks 17, 30 . This and the EMPA-REG study 31 have indicated initial acute falls in eGFR with better outcomes compared to placebo only observed over the longer term so this would not be apparent in short-term studies.
Our review included both randomised and non-interventional studies. Whilst the unique inferential advantages of randomization are clear, our review highlights a large overall difference in population size depending on study type: randomised trials generally included hundreds of patients, whilst non-interventional studies often had tens of thousands of participants. Rarer outcomes such as ESRD are therefore more likely to be detected in non-interventional settings. This highlights their important role, but the evidence generated from them needs to be evaluated cautiously due to the potential for bias and confounding.
The available evidence does not reflect drugs currently prescribed in routine care. In our review, 69% (22/32) of the comparisons, contrasted different monotherapies, with just three comparisons between dual therapy combinations. In clinical practice, metformin is the most common first-line therapy, and GPs now rarely prescribe thiazolidinediones (EU marketing authorization for Rosiglitazone was suspended in 2010 32 , following concern regarding increased heart failure risk) 33 .
In the UK, NICE guidance recommends the addition of sulfonylureas, Dipeptidyl peptidase-4 inhibitors (DPP4is) Sodium-glucose Cotransporter 2 Inhibitors (SGLT2is), or TZDs to metformin, yet, just one study compared these combinations (MTF+SU vs MTF+DPP4i) 25, [33][34][35] . Recent studies that have shown potentially exciting improvements in renal outcomes for patients treated with SGLT2is were conducted against placebo and so were not eligible for this study 36,37 .
We found that definitions of kidney outcomes were not consistent across studies. Definitions of renal decline in the observational studies relied upon either codes for kidney disease (e.g. diabetic nephropathy, acute renal failure), surrogate markers (e.g. eGFR or proteinuria) or a combination of codes and tests, summarised in Supplementary Table 4 (Supplementary File 2). For the albuminuria data, which has a skewed distribution, most studies used logarithmic transformation to approximate normal, yet not all studies applied this method 18 . Such differences between outcomes will limit future opportunities for pooling effect estimates in meta-analyses. Different approaches to study design may also limit the validity of findings. We found two observational studies that made the same comparisons yet found different effects. In the randomised controlled studies, we found that eligibility criteria were strict. Many studies excluded people most at risk of kidney outcomes e.g. those with reduced kidney function or cardiovascular disease 12,13,15-18 . These restrictions limit the generalisability of study findings to routine clinical settings where people presenting with diabetes have complex comorbidities 38 . Further, as most individuals with type 2 diabetes will receive treatment for other comorbid conditions, prescribers need to know how diabetic therapies interact with concomitant drugs, yet this is not addressed by the studies identified in this review.

Clinical relevance
In clinical practice, kidney function is one of many considerations for treatment choice in type 2 DM. Some of the differences we found for albuminuria and eGFR between people taking different oral therapies for type 2 diabetes were statistically significant, but the clinical importance of these findings may be limited. Some surrogate outcomes such as a doubling of creatinine or 30% decline in eGFR are closely associated with risk of future ESRD 39,40 while ACR is not 39,41,42 . Outcomes that are clinically relevant need to be assessed in future studies. Ideally, these should include hard outcomes such as hospital admission with acute kidney injury or the development of ESRD. Therefore, large, well-designed studies with long follow up, including individuals that represent the typical type 2 diabetes population, will be required. However, the incidence of kidney outcomes is likely to be low in most randomised trials and therefore high-quality observational studies will also be needed.
Our review highlights a lack of rigorous studies comparing the effects of oral type 2 diabetes drugs on kidney outcomes, in particular, for the newer drug intensification options where prescribing is rapidly increasing.

Data availability
All data underlying the results are available as part of the article and supplementary material no additional source data are required.
Competing interests SW is funded by a GSK PhD scholarship. HS is an employee of and holds shares in GSK. LAT reports no competeing interests. IJD is funded by, holds stock in and has consulted for GSK. LS is funded by a fellowship from the Wellcome Trust and consults for GSK and AstraZeneca, has received grants from the European Union and is a Trustee of the British Heart Foundation.      . with great interest. The paper reports a systematic literature et al review of studies examining the kidney prognosis in patients treated with different combinations of antidiabetic drugs in Type II diabetes. The study found a lack of literature to draw firm conclusions. The topic is important, and the paper is well written and follows the PRISMA guidelines. The paper describes the elements of the search strategy and the authors reviewed an extensive amount of papers to end up with a small sample of relevant papers. Due to substantial variety in kidney function outcomes and drug class comparisons, the authors did not conduct a meta-analysis. We have only a few comments to the article: Potential uncontrolled confounding by indication (and contraindication) are probably the most important limitation when interpreting the findings of the included observational studies. In particular, because metformin is the recommended first-line treatment in patients without renal impairment. It could be more clear whether the estimates included in Table 1 "kidney outcomes " are adjusted for relevant confounders and what confounders that were included in recorded HR each study. Figure 2 is very illustrative and a good way to summarize data in this review. Unfortunately, it is not possible to see the strength of the associations in such a figure. Would it be possible to use different line thickness to illustrate the strength of the associations? The introduction states that the study focuses on "following outcomes: change in kidney function (estimated glomerular filtration rate), progression or development of proteinuria, development of (page 3). However in the result section end-stage renal disease (ESRD) and composite outcomes" following outcomes are mentioned "changes in eGFR […] albumin-creatinine ratio (ACR) […] kidney endpoints, including kidney failure, nephropathy, acute dialysis and composite endpoints (page 3) Finally, in Table 3 the studies are divided in the three groups " with eGFR" . ACR, eGFR, based on the study endpoints (Table 3). We suggest that the terms and Kidney outcomes" describing other kidney outcomes than ACR and eGRF are clearly defined and used consequently throughout the paper. It is not clear, whether the final search strings differed substantially from the first searches, which are described in supplementary Table 1 and 2.

Are the rationale for, and objectives of, the Systematic Review clearly stated? Yes
Are sufficient details of the methods and analysis provided to allow replication by others?