Metformin Use Associated with Lower Acute Kidney Injury Risk in Patients With Type 2 Diabetes: Study

For evaluating the effect and predictive relevance of metformin, a total of 3,145 ICU patients with type 2 diabetes were analyzed in the MIMIC-IV database. Besides this, a validation cohort comprised of 100 ICU patients was independently taken from an external hospital setting. In the development cohort, 1,245 patients had fulfilled AKI criteria, while 1,059 patients had used metformin in the past before admission. Using propensity score matching, 760 well-matched cases of metformin users were created.

Conditional logistic regression analysis, LASSO feature selection, and clustering robust SE model was used for assessing AKI odds. For evaluating the incremental predictive relevance, base and augmented metformin machine learning models were developed, which included Clinical Logistic, LASSO Logistic, and XGBoost.

Key findings:

• The analysis of 3,145 MIMIC-IV subjects and external validation cohorts showed significant renal protection and improved machine learning performance.
• Among the 760 propensity-matched pairs, conditional logistic regression analysis showed that pre-admission use of metformin was significantly associated with lower odds of AKI (OR = 0.392, 95% CI: 0.306–0.502; P < 0.001).
• The protective effect of metformin use was consistent in matched-pair cluster-robust standard errors (OR = 0.408, 95% CI: 0.325–0.512; P < 0.001), after LASSO adjustment (OR = 0.343, 95% CI: 0.267–0.442; P < 0.001), and in sensitivity analysis excluding baseline ICU AKI (OR = 0.427, 95% CI: 0.338–0.538; P < 0.001).
• Metformin addition improved model performance.
• The XGBoost plus metformin model had the highest internal performance in MIMIC-IV, resulting in an ROC-AUC of 0.863 (95% CI: 0.849–0.877), a PR-AUC of 0.816, and a Brier score of 0.148.
• In the independent 100-patient external hospital cohort, the Clinical Logistic plus metformin model had the highest performance, resulting in an ROC-AUC of 0.955 (95% CI: 0.910–1.000), a PR-AUC of 0.932, and a Brier score of 0.138, with SHAP analysis indicating that metformin was a major negative predictor of AKI.

In summary, studies based on real-world registry and machine learning have proved that pre-admission metformin therapy is associated with a reduced probability of acute kidney injury among patients with type 2 diabetes admitted to the ICU. The conclusions drawn from this research highlight the significance of metformin in providing essential clinical information.

Reference:

Liu, Y., Wu, Y., Xie, M. et al. Metformin use and acute kidney failure risk in patients with type 2 diabetes: an explainable machine learning study with dual-cohort validation. BMC Nephrol (2026). https://doi.org/10.1186/s12882-026-05403-0