A large retrospective cohort study evaluating patients newly initiated on oral anticoagulation has revealed that direct oral anticoagulants are associated with a statistically significant higher risk of acute kidney injury within thirty days of therapy initiation compared to traditional warfarin therapy. The clinical investigation, conducted by researchers at the Department of Pharmacy within The Third Xiangya Hospital of Central South University in Changsha, China, analyzed real-world electronic medical records to address critical knowledge gaps regarding early renal safety profiles across diverse clinical indications. While anticoagulant-related nephropathy was historically identified and characterized primarily in connection with vitamin K antagonists like warfarin, growing pharmacovigilance reports have documented kidney injury following the administration of direct-acting agents. Most historical observational inquiries concentrated on extended treatment periods or restricted their cohorts exclusively to non-valvular atrial fibrillation, potentially missing the vulnerable early period when patients remain highly susceptible to rapid hemodynamic fluctuations and adverse nephrotoxic events. By rigorously capturing renal function alterations over an initial thirty-day window, the investigators demonstrated that the initiation of direct oral anticoagulants correlated with an overall thirty-four point six percent relative elevation in acute kidney injury odds compared to warfarin, an excess hazard primarily driven by more severe manifestations of renal impairment.

To establish a balanced and reliable comparison between the treatment cohorts, the research team assembled a study population comprising 6,876 consecutive adult patients who initiated oral anticoagulant therapy between January 2014 and May 2025. Among these participants, 5,048 individuals were newly prescribed direct oral anticoagulants—consisting of rivaroxaban, edoxaban, dabigatran, or apixaban—while 1,828 individuals commenced therapy with warfarin. To minimize confounding by indication and ensure comparability across baseline health metrics, the investigators implemented a strict new-user design requiring that all eligible participants had no documented dispensing of any oral anticoagulant during the 365-day period preceding their index prescription date. In addition, inclusion mandated valid baseline serum creatinine measurements recorded during the preceding year alongside follow-up serum creatinine tests obtained between the index date and the earliest of hospital discharge or thirty days post-initiation. The patient cohort captured a broad cross-section of clinical practice, including individuals receiving anticoagulation for stroke prevention in non-valvular atrial fibrillation, acute deep vein thrombosis and pulmonary embolism treatment, and venous thromboembolism prophylaxis across varied medical and surgical hospital settings.

Addressing the baseline imbalances inherent to retrospective observational datasets required advanced statistical balancing techniques to approximate the exchangeability achieved in prospective trials. Before weighting was applied, patients in the direct oral anticoagulant group were notably older, exhibiting a median age of 64.00 years versus 58.00 years among warfarin recipients, had a greater representation of female participants at 51.3 percent compared to 42.8 percent, and presented with better baseline renal function reflected by a median estimated glomerular filtration rate of 94.50 mL/min/1.73 m2 versus 82.87 mL/min/1.73 m2 in the warfarin group. To eliminate these confounding differences across 81 baseline demographic, clinical, pharmacological, and procedural characteristics, the authors implemented overlap propensity score weighting derived from a generalized boosted model utilizing 10,000 regression trees. Generalized boosted modeling provided substantial algorithmic flexibility compared to conventional logistic regression, allowing the researchers to achieve excellent covariate balance across both exposure groups, reducing standardized mean differences for all 81 baseline variables to well below the established threshold of 0.200, while multiple imputation using predictive mean matching appropriately handled missing clinical values across the dataset.

Following propensity score overlap weighting, weighted logistic regression demonstrated a notable difference in the primary outcome of thirty-day acute kidney injury, defined according to the international Kidney Disease: Improving Global Outcomes serum creatinine criteria. Across the balanced sample, direct oral anticoagulant recipients exhibited an acute kidney injury incidence of 13.16 percent compared to 10.12 percent among patients treated with warfarin, corresponding to a statistically significant odds ratio of 1.346 with a 95 percent confidence interval spanning from 1.057 to 1.715 and a p-value of 0.016. Because minor residual imbalances persisted after overlap weighting for three procedural characteristics—namely history of fracture, cardiac surgery, and orthopedic surgery, each displaying standardized mean differences between 0.100 and 0.130—the researchers performed an additional multivariable adjustment within the weighted logistic regression model. This doubly robust model fully confirmed the primary association, yielding an adjusted odds ratio of 1.363 with a 95 percent confidence interval between 1.068 and 1.740 and a p-value of 0.013, solidifying evidence that direct oral anticoagulant initiation is tied to heightened short-term nephrotoxic risk in real-world clinical care.

When examining secondary clinical endpoints stratified by acute kidney injury severity stages, the investigators revealed that the elevated risk associated with direct oral anticoagulants was predominantly driven by high-grade renal impairment. For stage 1 acute kidney injury, the weighted incidence was numerically higher in the direct oral anticoagulant group at 10.27 percent compared to 8.39 percent among warfarin users, but this difference failed to achieve statistical significance, yielding an odds ratio of 1.249, a 95 percent confidence interval of 0.957 to 1.631, and a p-value of 0.102. Conversely, for severe acute kidney injury, defined collectively as stage 2 or stage 3 impairment under KDIGO criteria, the weighted event rate reached 2.89 percent in the direct oral anticoagulant cohort compared to only 1.73 percent in the warfarin cohort. This severe injury comparison yielded a statistically significant odds ratio of 1.694 with a 95 percent confidence interval ranging from 1.013 to 2.832 and a p-value of 0.044, demonstrating that patients exposed to direct oral agents faced a substantial excess risk of developing pronounced functional renal deterioration during early therapy.

The study further evaluated clinical indications to determine whether this excess risk manifested uniformly across different therapeutic contexts, identifying marked heterogeneity based on treatment purpose. Among the 1,884 patients prescribed anticoagulation for stroke prevention in atrial fibrillation, comprising 1,184 direct anticoagulant users and 700 warfarin recipients, the weighted incidence of acute kidney injury was 13.14 percent for direct agents versus 8.81 percent for warfarin, producing a statistically significant odds ratio of 1.566 with a 95 percent confidence interval of 1.034 to 2.369 and a p-value of 0.034. In contrast, among 1,566 patients treated for active venous thromboembolism, weighted incidence stood at 13.57 percent versus 9.89 percent, which did not attain statistical significance with an odds ratio of 1.431 and a confidence interval between 0.872 and 2.349. Similarly, among 3,426 individuals receiving venous thromboembolism prophylaxis, weighted event rates were comparable at 12.90 percent versus 11.73 percent, reflecting an odds ratio of 1.114 and a confidence interval of 0.768 to 1.616, indicating that the heightened risk profile was especially pronounced within the atrial fibrillation population.

To parse compound-specific contributions, the investigators conducted drug-stratified analyses comparing each specific direct oral anticoagulant against warfarin, which uncovered substantial variation among individual medications. Rivaroxaban represented the vast majority of the direct anticoagulant cohort with 4,178 treated patients and was the only individual agent to demonstrate a statistically significant increase in acute kidney injury risk compared to warfarin, producing an odds ratio of 1.392 with a 95 percent confidence interval of 1.075 to 1.804 and a p-value of 0.012. In contrast, comparative analyses for edoxaban across 551 patients yielded an odds ratio of 1.256 with a confidence interval of 0.809 to 1.951, dabigatran across 276 patients showed an odds ratio of 1.301 with a confidence interval of 0.745 to 2.272, and apixaban across 43 patients exhibited an odds ratio of 1.114 with a confidence interval of 0.484 to 2.565. The authors emphasized that because point estimates for edoxaban, dabigatran, and apixaban remained above unity, their non-significant p-values likely reflected reduced statistical power stemming from smaller sample sizes rather than definitive evidence of renal neutrality or absence of risk.

An exploratory dose-stratified evaluation of rivaroxaban revealed unexpected associations between prescribed daily doses and renal complications, highlighting complex prescribing patterns in hospital environments. Patients initiated on a lower-than-recommended daily dose of rivaroxaban relative to indication-specific reference dosing exhibited a markedly heightened risk of acute kidney injury compared to those receiving recommended standard doses, presenting an odds ratio of 2.792 with a 95 percent confidence interval of 1.524 to 5.114 and a p-value of 0.001. Conversely, higher-than-recommended dosing yielded a non-significant odds ratio of 1.539 with a confidence interval of 0.758 to 3.123, while off-label prescribing showed no association with an odds ratio of 1.048 and a confidence interval of 0.747 to 1.470. A direct propensity-score-weighted comparison between reduced-dose and standard-dose rivaroxaban users corroborated this finding, showing weighted incidence rates of 10.93 percent versus 4.25 percent and an adjusted odds ratio of 2.743 with a confidence interval of 1.495 to 5.031. The investigators cautioned that this pattern likely reflects channeling bias, where older patients with baseline renal impairment or vascular comorbidities were systematically directed toward reduced doses, although insufficient initial anticoagulation leading to microthrombotic ischemic renal events represents a potential biological contributor.

Comprehensive sensitivity analyses reinforced the overall stability of the research team’s primary conclusions across diverse methodological and modeling frameworks. Re-estimating treatment effects through inverse probability of treatment weighting based on generalized boosted model propensity scores produced an unadjusted odds ratio of 0.828 with a 95 percent confidence interval of 0.695 to 0.987; however, after incorporating residual unbalanced baseline covariates into the outcome model to achieve doubly robust estimation, the association reversed to demonstrate a significantly increased acute kidney injury risk with an odds ratio of 1.418 and a confidence interval of 1.100 to 1.827. In addition, combining logistic-regression-derived propensity scores with overlap weighting yielded a consistent, statistically significant elevation in risk, showing an odds ratio of 1.200 with a 95 percent confidence interval of 1.043 to 1.381 and a p-value of 0.011. In prespecified clinical subgroups, point estimates remained above 1 across most patient strata without significant treatment-by-subgroup interaction effects, though a statistically significant risk increase emerged specifically among patients presenting with baseline chronic kidney disease stage G3b, displaying an odds ratio of 2.833 and a confidence interval of 1.246 to 6.441.

Despite its rigorous analytical framework and extensive control of 81 baseline covariates, the study authors highlighted several methodological limitations that necessitate careful interpretation of these findings as hypothesis-generating evidence. Because the investigation was conducted retrospectively at a single tertiary medical center using electronic health records, generalizability to broader international populations and alternative healthcare settings remains unestablished. Furthermore, acute kidney injury was identified exclusively via changes in serum creatinine levels because urine output metrics were not reliably documented, which may have led to an underestimation of true injury rates. The retrospective dataset also lacked international normalized ratio values and time-in-therapeutic-range measurements for warfarin users, preventing assessment of anticoagulation quality, while absent renal biopsy, urine sediment, and hematuria data precluded definitive pathological confirmation of anticoagulant-related nephropathy. Given that direct oral anticoagulants undergo varying degrees of renal clearance—reaching approximately 66 percent for rivaroxaban and 80 percent for dabigatran—the authors concluded that vigilant baseline renal monitoring and routine early follow-up testing are clinically prudent when initiating these agents, especially among elderly patients and individuals with pre-existing renal disease.

Subject of Research: Comparative 30-day acute kidney injury risk between direct oral anticoagulants and warfarin

Article Title: Oral anticoagulants and the risk of acute kidney injury: a retrospective real-world cohort study using overlap propensity score weighting

Article References: Li, P., Tian, F.-Y., Liu, H.-B., Rao, M., Yin, W.-J., Zuo, X.-C., & Zhou, L.-Y. (2026). Oral anticoagulants and the risk of acute kidney injury: a retrospective real-world cohort study using overlap propensity score weighting. Frontiers in Pharmacology, 17, Article 1935787. https://doi.org/10.3389/fphar.2026.1935787

Image Credits: AI Generated

DOI: 10.3389/fphar.2026.1935787

Keywords: acute kidney injury, direct oral anticoagulants, warfarin, rivaroxaban, anticoagulant-related nephropathy, atrial fibrillation, pharmacoepidemiology, propensity score weighting, Oral, anticoagulants, risk, acute