Predictive Score for the Success of Discontinuation of Renal Replacement Therapy (RRT) in Intensive Care
Predictive Score for the Success of Discontinuation of Renal Replacement Therapy (RRT) in Intensive Care
The goal of this observational study is to evaluate when renal replacement therapy (RRT) can be safely discontinued in intensive care unit (ICU) patients with acute kidney injury (AKI). The main question it aims to answer is:
Can the UNDERSCORE predictive score accurately identify ICU patients with AKI who can successfully stop RRT without needing restart or experiencing early death?
Using routinely collected clinical data from the AP-HP Health Data Warehouse, the study will follow ICU patients with AKI treated with RRT to assess whether the UNDERSCORE score predicts successful RRT discontinuation within 7 days.
Background and Rationale
Acute kidney injury (AKI) is a frequent and severe complication among patients admitted to intensive care units (ICUs), and it is associated with high morbidity, mortality, prolonged length of stay, and long-term renal impairment. In cases of severe AKI, renal replacement therapy (RRT), also referred to as extracorporeal renal replacement therapy (EER), remains the only effective supportive treatment capable of correcting life-threatening metabolic disturbances such as hyperkalemia, metabolic acidosis, fluid overload, and uremia.
Over the last decade, substantial progress has been made in defining optimal strategies for initiating RRT. Large randomized controlled trials, including AKIKI, IDEAL-ICU, and STARRT-AKI, have demonstrated that in the absence of immediate life-threatening complications, a conservative or "watchful waiting" strategy is safe and avoids unnecessary exposure to RRT in a significant proportion of patients. As a result, delayed initiation strategies have become standard of care in many ICUs.
In contrast, the question of when and how to discontinue RRT once renal function begins to recover remains insufficiently addressed. Current clinical practice regarding RRT discontinuation is highly heterogeneous and largely based on individual clinician judgment rather than objective, validated criteria. Prolonging RRT beyond what is necessary may expose patients to avoidable complications, including catheter-related infections, bleeding related to anticoagulation, hemodynamic instability, thrombosis, and nutritional losses, while also increasing healthcare costs and resource utilization.
International guidelines, including those from KDIGO, explicitly acknowledge the lack of evidence-based criteria for stopping RRT and highlight the need for reliable markers to predict successful weaning. Previous studies that attempted to identify predictors of successful RRT discontinuation were conducted before conservative initiation strategies became standard practice and often included patients who may not have had a validated indication for RRT. Moreover, these studies varied widely in terms of timing of assessment (during RRT, immediately after cessation, or days later), further limiting their applicability.
An ancillary, yet unpublished, analysis of the AKIKI 2 randomized controlled trial led to the development of a pragmatic predictive score-UNDERSCORE-designed to estimate the probability of successful RRT discontinuation. Successful weaning was defined as the absence of RRT resumption and survival within seven days following a weaning attempt. This score relies on simple, routinely available clinical and biological variables, making it potentially suitable for bedside use. However, before such a tool can be recommended for clinical practice, robust external validation on large, real-world datasets is essential.
The AP-HP Health Data Warehouse (Entrepôt de Données de Santé, EDS) provides a unique opportunity to perform this validation. Covering nearly 11 million patients across 39 hospitals, including all AP-HP ICUs, the EDS integrates administrative, clinical, biological, procedural, and outcome data with precise time stamps. Leveraging this resource allows for a large-scale, multicenter external validation of the UNDERSCORE score under real-life conditions and across a wide spectrum of ICU practices.
Study Objectives Primary Objective
The primary objective of the SERENA study is to externally validate the UNDERSCORE predictive score for successful weaning from renal replacement therapy in ICU patients with severe acute kidney injury.
Successful weaning is defined as the absence of RRT resumption and survival within seven days following a weaning attempt.
Secondary Objectives
Secondary objectives include:
Characterizing trajectories of renal recovery in ICU patients with severe AKI treated with RRT.
Evaluating the performance of the UNDERSCORE score across clinically relevant subgroups, including age categories, severity of illness, presence of septic shock, mechanical ventilation status, and modality of RRT.
Exploring additional clinical or biological predictors of successful RRT discontinuation available in the EDS.
Developing and internally validating an enriched predictive model incorporating these additional variables, if relevant.
Study Design
This study is a retrospective, observational cohort study designed as a pre-specified external validation of an existing predictive score. No intervention is performed, and no change in patient management is induced by the study.
The study period spans from January 2017 to December 2025.
Study Population
The study population consists of adult patients admitted to an ICU within AP-HP hospitals who experienced severe acute kidney injury and received renal replacement therapy during their ICU stay.
Inclusion Criteria
Admission to an AP-HP intensive care unit.
Diagnosis of severe acute kidney injury, defined as KDIGO stage 2 or 3.
Receipt of renal replacement therapy (intermittent or continuous) during the ICU stay.
Exclusion Criteria
None beyond the absence of required data to compute the UNDERSCORE score or ascertain outcomes.
All ICUs within AP-HP are eligible, ensuring a broad and representative population.
Data Sources
All data will be extracted from the AP-HP Health Data Warehouse (EDS), which integrates:
Medico-administrative data from the PMSI (ICD-10 diagnoses, CCAM procedures),
Structured electronic health record data (Orbis),
Laboratory results with precise time stamps,
Medication prescriptions and administrations,
Procedural metadata related to RRT sessions,
ICU clinical documentation and outcome data.
Data extraction will be performed within the secure EDS infrastructure using Cohort360 and Jupyter environments. No individual-level data will be exported outside the EDS.
Variables Collected Demographic and Baseline Characteristics
Age
Sex
Pre-existing conditions (chronic kidney disease, hypertension, diabetes, congestive heart failure, coronary artery disease)
Physiological and Biological Parameters
Baseline serum creatinine prior to ICU admission
Serum creatinine at ICU admission and at diagnosis of severe AKI
Blood urea nitrogen, potassium, bicarbonate levels at AKI diagnosis
Severity and Clinical Status
SAPS III score
SOFA score
Presence of septic shock at admission
Invasive mechanical ventilation
Vasopressor use (epinephrine or norepinephrine)
RRT Characteristics and ICU Management
RRT modality (continuous or intermittent)
Duration of RRT exposure
Number of RRT sessions
Indications for RRT initiation
Diuretic use
Fluid balance and daily urine output
Nutritional and Inflammatory Status
Modified NUTRIC score
Albumin
Transthyretin
C-reactive protein
Outcomes
Successful or failed RRT weaning
Mortality at 28 days, 60 days, ICU discharge, and hospital discharge
Renal recovery
Catheter-related infections
RRT-related complications
Outcome Definition
The primary outcome is the performance of the UNDERSCORE score in predicting successful RRT discontinuation, defined as:
No resumption of RRT and
Survival within seven days after the weaning attempt.
Statistical Analysis Plan
The UNDERSCORE score will be reconstructed using EDS data according to its original definition. Predictive performance will be assessed as follows:
Discrimination: Area under the receiver operating characteristic curve (AUROC) with 95% confidence intervals obtained by bootstrap resampling.
Calibration: Calibration-in-the-large, calibration slope, and graphical calibration curves using loess smoothing or grouped risk bins.
Clinical Utility: Decision Curve Analysis to evaluate net benefit across relevant probability thresholds.
Subgroup Analyses: Stratified analyses by RRT modality, septic shock status, mechanical ventilation, age groups, and hospital sites.
Missing Data: Extent and patterns of missingness will be described. Multiple imputation will be used if appropriate, with sensitivity analyses performed.
Sensitivity Analyses: Alternative outcome windows (e.g., 28 days), alternative variable definitions, and exclusion of early discharges.
Analyses will be conducted in R and/or Python within the secure EDS environment, with full version control and reproducibility.
Biases and Limitations
Potential sources of bias include misclassification of RRT exposure, inter-site variability in practice, missing data, and imperfect measurement of predictors. These limitations will be mitigated through validated coding algorithms, standardized variable definitions, site-level analyses, robust statistical methods, and sensitivity analyses.
As this is a predictive-not causal-study, confounding by severity will be acknowledged but not interpreted causally.
Ethical and Regulatory Considerations
The study is conducted in accordance with MR-004 regulations and within the secure AP-HP EDS framework. No additional risk or burden is imposed on patients. No partners external to AP-HP are involved.
Public Health Impact
By providing a robust external validation of a pragmatic RRT weaning score, this study has the potential to standardize clinical decision-making, reduce unnecessary exposure to invasive therapies, improve patient safety, shorten ICU stays, and optimize resource utilization. Ultimately, it may contribute to improved outcomes and faster recovery for critically ill patients with acute kidney injury.
Inclusion Criteria:
Exclusion Criteria: