Genetic Study Links Metformin to Kidney Disease Risk, but the Evidence Is Shaky
A Mendelian randomization study reports metformin raises CKD risk with an implausible OR of 145. Confounding by diabetes likely explains it, so don't change practice.
Summary
Researchers used Mendelian randomization on European genetic data to ask whether metformin causally raises chronic kidney disease (CKD) risk. Using 24 genetic variants linked to metformin use, the main analysis gave an odds ratio of about 145 (95% CI 6.5–3216), while four other MR methods were non-significant. A second analysis, summary-data-based MR (SMR), found eight metformin-target genes tied to CKD, including ANPEP, STK11, ACACB and RPS6KB1. The authors conclude clinicians should be cautious prescribing metformin to people with kidney problems. However, the exposure is self-reported medication use, which mostly tracks type 2 diabetes itself. The enormous effect size and wide confidence interval suggest these results reflect confounding, not a harmful drug effect.
Detailed Summary
Metformin is the most widely prescribed type 2 diabetes drug, yet its use in chronic kidney disease (CKD) is debated. Concerns centre on drug accumulation and lactic acidosis when kidney function falls. Guidelines now permit dose-adjusted use down to an eGFR of about 30, so a genetic approach to this question is appealing. Whether it can answer it is a separate matter.
The team ran a two-sample Mendelian randomization (MR) analysis. The exposure was the UK Biobank GWAS 'ukb-a-159' (metformin treatment, 8,392 cases and 328,767 controls). The outcome was a FinnGen CKD GWAS (3,902 cases). They selected 24 independent SNPs (p<5×10⁻⁸, r²<0.001) as instruments and ran IVW, MR-Egger, weighted median, simple mode and weighted mode methods. They also ran a summary-data-based MR (SMR) analysis. It combined blood cis-eQTLs from eQTLGen (31,684 people) with FinnGen R10 CKD data (10,039 cases, 396,706 controls). Significant genes were intersected with metformin targets from DrugBank, DGIdb, SwissTargetPrediction, TargetNet and the literature.
The IVW estimate was OR 144.7 (95% CI 6.5–3216; p=0.002). MR-Egger, weighted median, simple mode and weighted mode gave similarly large point estimates but were all non-significant, with extremely wide intervals. Tests for heterogeneity (Cochran's Q p=0.51), pleiotropy (Egger intercept p=0.93; MR-PRESSO global p=0.54) and leave-one-out analysis showed no obvious problems. SMR flagged eight metformin-related genes: MGLL, ANPEP, STK11, ACACB, PRKAB1, SIRT1, KCNJ11 and RPS6KB1. The discussion highlights four (ANPEP, STK11, ACACB, RPS6KB) as CKD risk genes, speculating about inflammation, AMPK signalling, lipid metabolism, oxidative stress and mTOR/S6K1 activity.
The authors conclude metformin may raise CKD risk and urge caution, especially in people with renal impairment. They also review standard guidance: dose reduction at eGFR 30–45, avoidance below 30, regular eGFR monitoring, and the low lactic acidosis rate (~0.03 per 1,000 patient-years).
These conclusions are not well supported. The exposure is self-reported metformin use, so the instruments probably capture genetic liability to type 2 diabetes and related metabolic traits, which independently cause CKD. This is confounding by indication, and it violates the MR exclusion-restriction assumption. The 'effect' is not the pharmacological effect of metformin. An OR near 145 is biologically implausible, and the confidence interval spans three orders of magnitude. Only IVW reached significance. The SMR results also look internally inconsistent: three of the four highlighted genes have negative SMR betas (higher expression linked to lower CKD risk), yet the paper presents them as risk genes with ORs above 1. Genes such as STK11 and PRKAB1 are AMPK pathway components, where metformin is expected to be protective. This paper should not change prescribing, and existing trial and observational evidence on metformin and kidney outcomes remains more informative.
Key Findings
- IVW MR gave OR 144.7 (95% CI 6.5–3216) for metformin and CKD; the other four MR methods were non-significant.
- The instruments proxy self-reported metformin use, which likely reflects type 2 diabetes genetics rather than drug effects.
- SMR with blood eQTLs identified eight metformin-target genes linked to CKD, including ANPEP, STK11, ACACB, PRKAB1, SIRT1 and RPS6KB1.
- Several highlighted genes had negative SMR betas, which conflicts with the paper's framing of them as CKD risk genes.
- Pleiotropy and heterogeneity tests were unremarkable, but they cannot rule out confounding by diabetes as the indication.
Methodology
Two-sample MR (TwoSampleMR) used 24 SNPs from UK Biobank metformin-use GWAS against a FinnGen CKD GWAS, with IVW as the primary method and Egger, MR-PRESSO, Cochran's Q and leave-one-out sensitivity checks. A separate SMR/HEIDI analysis integrated eQTLGen blood cis-eQTLs with FinnGen R10 CKD data and intersected hits with metformin drug-target databases.
Study Limitations
The exposure is self-reported metformin use, not metformin exposure or drug-target perturbation, so confounding by indication (type 2 diabetes) is almost certain. The OR of ~145 with a very wide CI, plus non-significant results from other MR methods, indicates implausible and unstable estimates. SMR used blood rather than kidney eQTLs, and reported gene directions and ORs are inconsistent between results and discussion. The study covers only European ancestry, and the paper reports inconsistent counts for the number of genes and the outcome sample size.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.
Enter your email to subscribe: