Early dialysis tied to higher death risk in sepsis kidney injury study

Starting continuous kidney replacement therapy within 12 hours of a sepsis-related acute kidney injury diagnosis was associated with a higher risk of death over the following 60 days than waiting longer, according to a retrospective cohort study published on 10 September 2026 in the open-access journal Critical Care: Sepsis and Severe Infection. The finding comes from an analysis of 629 adults with sepsis-associated acute kidney injury who all ultimately received the therapy, drawn from the MIMIC-IV intensive care database, version 3.1.

The research team, led by Yunfei Guo and Hongxia Yang, set out to address a question that has divided intensive care physicians for years: when should kidney replacement begin in patients whose kidneys are failing as a consequence of sepsis? Sepsis is the most common contributing factor to acute kidney injury in critically ill patients, accounting for roughly half of all such episodes in the intensive care unit. For hemodynamically unstable patients with severe disease, continuous renal replacement therapy, or CRRT, is the cornerstone of extracorporeal organ support, yet it carries substantial risks of its own, including drops in blood pressure during treatment, vascular access complications, and altered antimicrobial drug levels.

Large randomized controlled trials, including AKIKI, IDEAL-ICU, and STARRT-AKI, have consistently shown no mortality benefit from starting replacement therapy early. But those trials enrolled broad, heterogeneous ICU populations, and the authors of the new study argue that treatment effects specific to sepsis-associated kidney injury may be diluted within such mixed cohorts. Whether the neutral trial results extend to particular subpopulations of these patients, they write, warrants further investigation.

To probe that question, the researchers applied a target trial design framework to observational data. They screened adult intensive care patients and included those who met the Sepsis-3 criteria, had reached stage 2 or 3 acute kidney injury according to KDIGO creatinine criteria, and received CRRT during their ICU stay. Time zero was strictly defined as the exact moment of the stage 2 or 3 kidney injury diagnosis, and all baseline characteristics were measured up to that point. Patients were then classified as receiving early therapy, within 12 hours of time zero, or delayed therapy, more than 12 hours after it. Of the 629 eligible patients, 115 began CRRT early and 514 began it later.

Because physicians do not decide when to start dialysis at random, the two groups differed at baseline in ways that could independently affect survival. To address this, the team used propensity score overlap weighting, a statistical technique that emphasizes patients with the greatest clinical equipoise, meaning those who could realistically have been assigned to either strategy, and avoids the extreme weights that can destabilize traditional matching or inverse probability weighting. The primary outcome was 60-day all-cause mortality, analyzed with weighted Cox proportional hazards models, with the proportional hazards assumption verified using scaled Schoenfeld residuals.

After overlap weighting, early CRRT initiation was significantly associated with increased 60-day mortality, with a hazard ratio of 1.62, a 95 percent confidence interval of 1.17 to 2.25, and a p value of 0.004. In the unadjusted Kaplan-Meier analysis, survival probability was also significantly lower in the early group, with a log-rank p value of 0.002. A restricted mean survival time analysis with 60-day truncation estimated that patients in the early group lived an average of 11.2 fewer days within that window than those in the delayed group, a difference of 16.1 versus 27.2 days, with a bootstrap 95 percent confidence interval of 14.6 to 7.7 days and a permutation p value below 0.001.

The association held up across a battery of sensitivity analyses. A landmark analysis that excluded four patients who died within 12 hours of the kidney injury diagnosis, a step designed to guard against immortal time bias, yielded a hazard ratio of 1.48 with a 95 percent confidence interval of 1.06 to 2.06 and a p value of 0.023. Alternative timing thresholds ranging from 6 to 72 hours all produced directionally consistent estimates, with hazard ratios between 1.65 and 2.17 and p values below 0.05; using a 24-hour cutoff, the hazard ratio was 2.08. In analyses stratified by kidney injury stage, the point estimate was 1.07 for the 52 stage 2 patients and 1.45 for the 577 stage 3 patients, the latter falling just short of statistical significance.

Yet the picture is far from straightforward, and the authors themselves stress that the headline result depends heavily on the statistical method chosen. When the team ran a conventional multivariable Cox regression instead of the overlap-weighted model, the estimate was attenuated and no longer statistically significant, with a hazard ratio of 1.27, a 95 percent confidence interval of 0.94 to 1.72, and a p value of 0.124. The researchers flag this method-dependence explicitly in their limitations section, underscoring the need for cautious interpretation. Residual confounding by indication remains a major concern: even after weighting, four variables retained standardized mean differences of 0.1 or greater, namely stage 3 kidney injury, lactate level, vasopressor count, and baseline creatinine. These are strong markers of illness severity that could influence both the timing decision and mortality risk, and the authors acknowledge that the possibility that early CRRT merely identifies a sicker subgroup, rather than causing harm, cannot be excluded.

In exploratory, post-hoc subgroup analyses that were not adjusted for multiple comparisons, the association between early initiation and mortality was more pronounced in certain patient strata. It was strongest among patients with thrombocytopenia, defined as a platelet count below 100 multiplied by 10 to the ninth per liter, where the hazard ratio reached 3.06 with a 95 percent confidence interval of 1.68 to 5.57, and it was also elevated among patients with lower lactate levels, below 2 millimoles per liter, with a hazard ratio of 2.13. Vasopressor use did not significantly modify the effect. The authors suggest that maintaining circuit patency in thrombocytopenic patients is challenging, and that the obligatory use of anticoagulation combined with frequent circuit clotting may exacerbate underlying sepsis-induced coagulopathy, but they emphasize that this observation is hypothesis-generating and requires independent validation.

Why might earlier dialysis be linked to worse survival? The authors propose several hypothesized mechanisms, all of which they note require dedicated investigation. CRRT can induce intradialytic hypotension, potentially worsening ischemia in the renal medulla. Exposure of blood to the artificial membrane may trigger bioincompatibility reactions that amplify systemic inflammation and oxidative stress. Early initiation also alters the pharmacokinetics of antimicrobial agents, potentially leading to sub-therapeutic drug levels, while a delayed approach allows time for hemodynamic optimization and spontaneous renal recovery, potentially sparing patients from vascular access and anticoagulation complications.

The study has further constraints that the authors lay out in detail. It was restricted to patients who ultimately received CRRT, so it cannot answer the clinically central question of whether an early strategy should be preferred over a watchful-waiting approach in which many patients never need dialysis at all. The analysis also stops short of being a fully implemented target trial emulation, since it did not employ clone-censor-weight methodology or time-dependent treatment modeling. As a single-center database analysis, its findings may not generalize to other settings, and dynamic clinical variables were not captured at the granularity required for comprehensive adjustment. An E-value analysis was performed to quantify how strong an unmeasured confounder would need to be to explain away the association, and the authors note that a confounder linked to both treatment timing and mortality would need to be substantial to overturn the result.

Despite the consistent direction of the findings across sensitivity analyses, the authors are careful not to position their work as a challenge to the large randomized trials. They point out that their cohort was limited to patients who received the therapy, whereas the trials compared early initiation against watchful waiting, and they note that a recent independent analysis of sepsis-associated kidney injury patients in MIMIC-IV and the eICU-CRD similarly reported improved survival with later CRRT initiation, providing convergent evidence. Their conclusions are measured: the observations do not support clinical recommendations regarding CRRT timing, but they may inform the design of future prospective studies investigating phenotype-specific initiation strategies, particularly for subgroups such as patients with thrombocytopenia.

Subject of Research: Timing of continuous renal replacement therapy initiation in sepsis-associated acute kidney injury

Article Title: Overlap-weighted retrospective cohort study of early versus delayed renal replacement therapy in sepsis-associated AKI

Article References: Guo, Y., & Yang, H. (2026). Overlap-weighted retrospective cohort study of early versus delayed renal replacement therapy in sepsis-associated AKI. Critical Care: Sepsis and Severe Infection, 1(1), Article 13. https://doi.org/10.1186/s44541-026-00018-2

Image Credits: AI Generated

DOI: 10.1186/s44541-026-00018-2

Keywords: sepsis-associated acute kidney injury, continuous renal replacement therapy, CRRT timing, MIMIC-IV, overlap weighting, 60-day mortality, retrospective cohort study, propensity score, thrombocytopenia, critical care, renal replacement therapy, hypothesis-generating