#NephJC Chat
Tuesday, April 15, 2025, 9 pm Eastern on Bluesky
Am J Kidney Dis. 2025 Feb;85(2):196-204.e1. doi: 10.1053/j.ajkd.2024.08.012. Epub 2024 Nov 7.
Stopping Versus Continuing Metformin in Patients With Advanced CKD: A Nationwide Scottish Target Trial Emulation Study
Emilie J Lambourg, Edouard L Fu, Stuart McGurnaghan, Bryan R Conway, Neeraj Dhaun, Christopher H Grant, Ewan R Pearson, Patrick B Mark, John Petrie, Helen Colhoun, Samira Bell; Scottish Diabetes Research Network Epidemiology Group
PMID: 39521399
Introduction
Metformin has been a cornerstone in the treatment of type 2 diabetes mellitus (T2DM) for more than 60 years. This humble biguanide derived from Galega officinalis (or goat’s-rue herb, French lilac) became the first line oral glucose-lowering therapy thanks to its efficacy, low cost, and safety profile. Since its early approval in Europe in the 1950s—and much later in the U.S. in 1995—it has consistently shown benefits beyond glycemic control, including weight neutrality and potential cardioprotective effects (Bailey J in Diabetologia. 2017; Diabetes Obes Metab. 2024; UKPDS 34, Lancet 1998).
From Diabetes management in chronic kidney disease: a consensus report by the ADA and KDIGO. Boer IH, et al. Kidney International, 2022.
Today, metformin is the most prescribed oral antidiabetic worldwide. Apart from T2DM, ongoing research is exploring metformin’s potential in other conditions such as cancer, memory loss, bone disorders, immunological diseases, polycystic ovarian syndrome, and non-alcoholic fatty liver disease (NAFLD). Despite its widespread use, metformin’s application in patients with reduced kidney function has been limited due to concerns about lactic acidosis. Until 2016, the US FDA had an absolute contraindication for metformin use if creatinine was > 1.5 mg/dL (men) or > 1.4 mg/dL (women), which would correspond to ~ GFR 60, an atrociously poor threshold considering it was based on fears of MALA from the phenformin data, and in an era when metformin was the only agent with any end organ benefit in T2DM. This was relaxed in 2016 to allow metformin use to a GFR 30, but still not allowing initiation below 45, and insisting on stopping at GFR 30. We have discussed some of this rationale in 2018 (Lalau et al, Diab Care 2018 | NephJC summary).
Most guidelines still recommend discontinuing metformin when eGFR falls below 30 mL/min/1.73 m², despite limited evidence supporting this threshold (Yang et al eClin Med 2024). Yet, the actual risk of metformin-associated lactic acidosis (MALA) appears to be extremely rare (approximately 3 cases per 100,000 patient-years). Unfortunately, when MALA does occur, it has a high mortality rate (36%), however, prompt recognition and management, including hemodialysis, can result in good patient outcomes despite severe laboratory findings. Perhaps most importantly, robust clinical evidence supporting strict discontinuation thresholds in CKD are currently lacking. It is also not clear if the GFR threshold is all that matters since MALA usually occurs in the setting of heart failure, sepsis, shock, or AKI (which can occur at any GFR).
Dosage adjustment for eGFR< 45 ml/min/1.73m2 per package inserts. Boer IH, et al. Kidney International, 2022.
In real-world practice, especially in resource-limited settings, metformin is often continued even in advanced CKD—sometimes out of necessity, sometimes based on clinical judgment.. Perhaps most concerning: switching to other glucose-lowering medications may increase the risk of hypoglycemia without offering the same cardiometabolic benefits as metformin does. Importantly, many of these studies have also found no increase in lactic acidosis risk, the original reason for avoiding metformin in CKD (e.g. Yang A., et al, Hsu WH, et al, Leyco T, et al. Bradley JN, et al.).
Despite the lack of definitive evidence, Lambourg et al applied a target trial emulation design to a nationwide Scottish cohort of patients with T2DM who progressed to stage 4 CKD (eGFR <30 mL/min/1.73 m²) to evaluate whether continuing versus stopping metformin impacted clinical outcomes. Such a study had to be done in a jurisdiction that was more friendly to a rational use of metformin and not arbitrary dictats. So the question today: Is it truly safer to stop metformin when kidney function declines, or have we overestimated the risk?
The Study
Methods
Study design: The study emulates a target trial using observational data to address the question of whether to continue or stop metformin in patients with advanced CKD. As previously discussed (see discussion on NephJC, YouTube rounds from Edouard Fu), this approach is designed to reduce bias inherent in observational studies. It included all adults with type 2 diabetes and incident stage 4 CKD in Scotland who were treated with metformin between January 2010 and April 2019.
Data sources: The study uses national databases like Scottish Care Information Diabetes Collaboration (SCI-DC), Scottish Renal Registry (SRR), Scottish Morbidity Records (SMR01), and National Records of Scotland (NRS) to identify diabetes diagnoses, kidney disease status, hospitalizations, and mortality.
Outcome: MACE was defined as a composite of fatal and non-fatal cardiovascular events determined using ICD-10 codes.
Eligibility criteria and treatment strategies: The inclusion criteria involved patients with type 2 diabetes and incident CKD stages 4 and 5 while excluding those on kidney replacement therapy or missing baseline data.
Adapted after table S1: Emulated Trial Protocol, from Lambourg et al, Am J Kidney Dis, 2025
The treatment strategies compared were continuing versus stopping metformin within 6 months of reaching CKD stage 4.
Follow-up: started at the assignment of a strategy and ended at the earliest of outcomes of interest, death, loss of follow-up, administrative censoring at 3 years, whichever came first.
Addressing confounding with the clone-censor-weight method: To address time-dependent confounding, the clone-censor-weight method was used. This method creates “clones” of each patient, assigning them to hypothetical treatment strategies ( continue vs stop metformin), and censors them based on adherence. For example, clones assigned to continue metformin were censored at six months if they had not done so in practice. The censoring mimics the loss of follow-up that might occur in a real randomized clinical trial. Informative censoring, where the reason for censoring is related to the outcome, can introduce bias. The inverse probability of censoring weighting (IPCW) was used to mitigate this. IPCW upweights uncensored replicates, rebalancing the dataset to account for the missing data. Weighted pooled logistic regression and weighted Cox models were used to adjust for baseline and time-updated covariates.
Figure S3. Three steps of the clone-censor-weight method, from Lambourg et al, Am J Kidney Dis, 2025
Sensitivity analyses used marginal structural models (MSMs). While clone-censor-weight method addresses early deviations from assigned strategies, MSMs help account for what happens after 6 months, when patients continue or stop treatment at different times. MSMs allow adjustment for changes in patient health and treatment over time, which can influence both if continuing metformin and the risk of outcomes. By reweighting patients based on their changing likelihood of remaining on treatment, MSMs provide a more robust estimate of the effect of sustained metformin use throughout the full follow-up.
Covariate balance assessment: Standardized mean difference (SMD) is calculated to evaluate covariate balance between treatment arms before and after weighting, with a threshold of 10% indicating imbalance. Love plots are used to depict covariate balance visually.
Ethical considerations: The study was conducted with ethical approval from relevant committees, and all datasets were de-identified prior to analysis.
Funding: The study was funded by the Chief Scientist Office for Scotland. The funders did not have a role in study design, data collection, analysis, reporting, or the decision to submit for publication.
Results
Starting from over 370,000 patients with T2D, only 4.5% progressed to CKD stage 4 during follow-up. Among these, 9510 patients were metformin-exposed before CKD stage 4, of whom only 45% (N=4278) were both persistent and adherent in the year prior. This final group represents 25.8% of all CKD stage 4 cases constituting the study cohort. The medium follow-up was 2.5 years.
Figure 1. Study flowchart, from Lambourg et al, Am J Kidney Dis, 2025
The study included an elderly (>¾ having more than 70 years), high-risk T2D population: long-standing diabetes (median duration of 14 years), 44% ever smokers, almost 40% with ischemic heart disease, 10% with cerebrovascular disease, and 25% having other circulatory conditions, 17% had cancer. The medium eGFR was 27 ml/min/1.73m2 (24.2-28.7), with reasonable glycemic control. Over 80% of the cohort were treated with ACEi/ARBs and lipid-lowering drugs, while 70% were on anticoagulants. Regarding glucose-lowering therapy, insulin and sulfonylureas are common, while SGLT2i (0.4%) and GLP-1RA (4%) are underused, reflecting the pre-2019 timeline.
Table 1. Baseline characteristics, from Lambourg et al, Am J Kidney Dis, 2025
Table 2 shows that before weighting, patients who continued metformin differed from those who stopped it. Continuers had slightly higher eGFR, lower HbA1C, and were less likely to be on insulin or sulfonylureas. Continuers had fewer lab tests and hospitalizations. After weighting, the two groups were balanced across measured characteristics, with some residual imbalance: HbA1c (SMD=0.109).
During the 6-month grace period at the start of the follow-up, 40% (N=1713) of the patients stopped metformin, and of these, only 4% restarted it later during follow-up. Only 44% of those who continued metformin during 6 months, continued until the end of follow-up.
Figure S2. Sankey diagram, from Lambourg et al, Am J Kidney Dis, 2025
All cause mortality and MACE
During the follow-up, 1702 patients died. The main causes of death were cardiovascular (34%), cancer (17%), and respiratory diseases (10%).
At 3 years, survival was higher for patients who continued metformin for at least 6 months (70.3%, 95% CI, 67.9-72.8) versus those who stopped within 6 months of reaching CKD stage 4 (63.4%, 95% CI, 60.5-66.5).
Figure 2. Weighted cumulative incidence mortality by treatment strategy, from Lambourg et al, Am J Kidney Dis, 2025
After adjusting for confounding, stopping metformin was associated with higher mortality, with an HR of 1.23 (1.08-1.41). Covariate balance was achieved at the end of the grace period. (Fig S4)
Figure S4. Love plot depicting covariate balance between the two treatment arms before and after weighting, from Lambourg et al, Am J Kidney Dis, 2025
A “love plot” might sound like the story arc for a new Fifty Shades of Grey sequel, but that’s not how it is used in this instance. In statistics, a love plot is a graphical tool used to visualize the effect of adjustments on covariate balance. In other words, the goal is to assess how well the treatment and control groups are balanced. Ideally, a love plot will show that the standardized mean difference (SMD or x-axis above) is reduced after adjustment, indicating that the treatment and control groups are more similar. On the other hand, it might reveal which covariates are not well balanced, even after adjustment.
During follow-up, 915 MACE events occurred. No association was found between stopping metformin and MACE risk (HR 1.05, 95% CI 0.88-1.26).
Figure S5. Weighted cumulative incidence for MACE, from Lambourg et al, Am J Kidney Dis, 2025
Cause-specific mortality
Stopping metformin was associated with an increased risk of respiratory death (HR 1.51, 95% CI 1.06-2.12) but not cancer mortality (HR 1.07, 95% CI, 0.80-1.44).
Sensitivity analyses
Using marginal structural models, the associations remained consistent:
All-cause mortality: HR 1.34 (95% CI, 1.08-1.67)
MACE: HR 1.04 (95%CI, 0.81-1.33)
Discussion
This study offers compelling observational evidence supporting the continued use of metformin in patients with advanced CKD. Using a target trial emulation approach the authors show that continuing metformin after progression to CKD stage 4, was associated with improved survival without an increased risk of cardiovascular events or lactic acidosis. The results are particularly compelling (even in an observational cohort) given the large, elderly, high-risk population with long-standing diabetes and substantial cardiovascular burden—precisely the patients who are often excluded from randomized trials and in whom real-world guidance is needed.
The key finding of a 7% survival benefit in patients who continued metformin challenges the current dogma. While the study is observational in nature, the methodology strengthens its validity. Importantly, the mortality benefit aligns with results from previous studies (Yang A, et al, EClin Medicine, 2024) where survival consistently favored patients who remained on metformin. Others (eg Bradley JN, et al.Diab 2017 and Yang A, et al EClin Med 2024) have also reported worsening glycemic control following metformin discontinuation. Ironically, these findings are coming to light in an era when their relevance is diminishing, with the advent of drugs (flozins and GLP1RAs) which can be safely used below GFR 30, and which also have evidence of clinical outcome benefits.
Apart from the inherent renalism of capricious cutoffs for GFR, one should reflect on the scores of patients who had their metformin stopped at unnecessarily high GFR thresholds and have suffered from the clinical consequences for the last few decades. This is as clear an example as can be seen of regulatory overreach and pharmacological pusillanimity in the misdirected pursuit of safety.
Interestingly, the current study also found a significantly higher risk of respiratory mortality among patients who discontinued metformin. While this may reflect residual confounding, it raises the possibility that metformin’s anti-inflammatory and immunomodulatory effects could extend to the lungs, a hypothesis that warrants further investigation. In a similar vein, metformin is also being studied as a prophylactic for sepsis-associated AKI (LiMiT AKI trial) due to anti-inflammatory properties and regulation of mitochondrial injury (Saraiva IE, et al. BMJ Open, 2024). It is fascinating that metformin is being considered in patients in the ICU with multisystem organ dysfunction; patients that we currently withhold metformin due to fear of MALA.
The limited use of flozins and GLP-1 RAs in this cohort reflects the study period (pre-2019) and highlights that metformin was not competing with modern therapies. As practice evolves, the place of metformin in CKD care may shift, but this study reinforces that metformin still has a seat at the table—particularly in patients where access to newer agents is limited or contraindicated.
Strengths
Robust methodology using target trial emulation, reducing confounders.
Large, national cohort.
Real-world relevance, especially in elderly and high-risk populations.
Limitations
Observational design: residual confounding and selection bias cannot be entirely excluded.
Limited generalizability to modern practice: low use of SGLT2i and GLP-1 RA.
Biologically plausibility needed on the respiratory mortality association.
Conclusion
While this study does not replace the need for an RCT (which realistically will not happen in this space), it challenges the practice of automatic metformin discontinuation at eGFR <30 mL/min/1.73 m². In real-world practice, especially in the resource-limited settings, these findings may help support more nuanced, individualized decisions.
As newer therapies like flozins and GLP-1 RAs reshape the treatment landscape in diabetes and CKD, metformin remains relevant. In the end, the question is not simply whether metformin should be stopped, but rather: Which patient, at what stage, and under what conditions might patients still benefit from metformin continuation?