Skip to main content

Foundational Trials

AKIKI-2

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Comparison of two delayed strategies for renal replacement therapy initiation for severe acute kidney injury (AKIKI 2): a multicentre, open-label, randomised, controlled trial
  • Acronym: AKIKI 2
  • Year: 2021
  • Journal published in: The Lancet
  • Citation: Gaudry S, Hajage D, Martin-Lefevre L, et al. Comparison of two delayed strategies for renal replacement therapy initiation for severe acute kidney injury (AKIKI 2): a multicentre, open-label, randomised, controlled trial. Lancet. 2021;397:1293-1300.

Context & Rationale

  • Background
    • In critical illness, KDIGO stage 3 AKI frequently prompts consideration of renal replacement therapy (RRT), but the “right” moment remains uncertain because many patients recover without RRT and RRT itself carries risk (catheters, haemodynamic instability, electrolyte shifts).
    • Earlier RCTs comparing early/accelerated initiation versus delayed/standard initiation generally showed no mortality benefit from earlier RRT, while demonstrating that a substantial proportion of patients assigned to delayed/standard strategies never require RRT.
    • However, “delayed/standard” strategies still include upper limits (e.g., prolonged oliguria and/or urea/urea nitrogen thresholds), and the safety of waiting beyond these limits (a “more-delayed” strategy) was not established.
  • Research Question/Hypothesis
    • In critically ill patients with severe (KDIGO stage 3) AKI who have already met commonly used “delayed” triggers, does postponing RRT even further (until urgent complications or a higher urea nitrogen threshold) increase days alive and free of RRT at day 28, without worsening patient-centred outcomes?
  • Why This Matters
    • If a “more-delayed” strategy safely avoids RRT in more patients, it could reduce iatrogenesis, conserve resources, and better align treatment with patient trajectories.
    • If “more-delayed” initiation is harmful, AKIKI-2 would help define a practical upper boundary for watchful waiting in severe ICU AKI.
    • The trial also tests whether a urea-based threshold (alongside urgent indications) can operationalise a reproducible strategy when classic “timing” is clinically ambiguous.

Design & Methods

  • Research Question: Among critically ill adults with KDIGO stage 3 AKI who meet “delayed initiation” triggers, does a “more-delayed” RRT initiation strategy (urgent indications or higher urea nitrogen threshold) improve days alive and RRT-free at day 28 versus initiating RRT promptly after those triggers?
  • Study Type: Multicentre, open-label, randomised, controlled trial in 39 ICUs (France); investigator-initiated; protocol published prospectively1.
  • Population:
    • Screened ICU patients with AKI receiving (or having received) vasoactive agents or invasive mechanical ventilation (or both), who reached KDIGO stage 3 AKI.
    • Two-step enrolment: (1) monitoring stage after KDIGO stage 3 onset; (2) randomisation only if predefined “delayed strategy” criteria occurred without earlier urgent need for RRT.
    • Randomisation criteria (triggering allocation): blood urea nitrogen (BUN) >112 mg/dL, or oliguria/anuria persisting for >72 hours.
    • Key exclusions included: immediate indication for RRT at screening; severe chronic renal failure; RRT already given for the current AKI episode; moribund state or treatment limitation; post-cardiac arrest without awakening; specific AKI aetiologies (e.g., obstruction, tumour lysis, thrombotic microangiopathy, acute glomerulopathy); poisoning by a dialysable agent; kidney transplant; Child–Pugh class C cirrhosis; pregnancy; under curatorship.
  • Intervention:
    • More-delayed strategy (n=141): after randomisation, RRT deferred until either (a) an urgent indication occurred, or (b) BUN reached 140 mg/dL for 1 day.
    • Urgent indication thresholds: serum potassium >6 mmol/L, or >5.5 mmol/L despite medical treatment; arterial pH <7.15 (with protocol-specified ventilation constraints); acute pulmonary oedema from fluid overload despite diuretics with severe hypoxaemia (oxygen flow >5 L/min to keep SpO2 >95%, or FiO2 >50% on invasive or non-invasive ventilation).
  • Comparison:
    • Delayed strategy (n=137): RRT initiated within 12 hours after randomisation (i.e., promptly after meeting delayed-trigger criteria).
    • RRT modality (intermittent haemodialysis vs continuous RRT), anticoagulation, and dose were at clinician/centre discretion.
  • Blinding: Open-label (no blinding of clinicians or participants; outcomes largely objective, but initiation/discontinuation of RRT is clinician-driven).
  • Statistics: Power calculation: 270 patients required to detect a 4-day increase in mean RRT-free days at day 28 (from 17 to 21) with 80% power at the 5% two-sided significance level (allowing ~5% attrition); primary analysis by intention-to-treat using non-parametric comparison for RRT-free days; survival analyses used time-to-event methods.
  • Follow-Up Period: Primary endpoint assessed to day 28; key secondary outcomes (mortality, renal recovery, functional status) assessed to day 60.

Key Results

This trial was not stopped early. Recruitment ran from May 7, 2018 to Oct 11, 2019, with planned follow-up completed to day 60.

Outcome More-delayed Delayed Effect p value / 95% CI Notes
Randomised, n 141 137 Not reported Not applicable Intention-to-treat cohort
Primary: RRT-free days (days alive and free of RRT) to day 28, median (IQR) 10 (0–24) 12 (0–25) Not reported P=0.93 Composite influenced by both death and RRT exposure
Received any RRT, n (%) 111 (79%) 134 (98%) Not reported P<0.0001 30/141 (21%) avoided RRT entirely in the more-delayed strategy
Time from randomisation to RRT initiation, hours, median (IQR) 33 (24–60) 3 (2–5) Not reported P<0.0001 Demonstrates strong protocol separation for initiation timing
Mortality at day 60, n (%) 77 (55%) 60 (44%) Not reported P=0.071 Not the primary endpoint; signal towards harm
Mortality at day 60 (multivariable Cox model) Not reported Not reported HR 1.65 95% CI 1.09 to 2.50; P=0.018 Adjusted analysis identified higher hazard of death with more-delayed strategy
Metabolic acidosis (arterial pH <7.15), n (%) 42 (30%) 25 (18%) Not reported P=0.022 Complication potentially related to delayed solute control
Acute coronary syndrome (among patients receiving RRT), n (%) 4 (4%) 0 (0%) Not reported P=0.026 Small numbers; hypothesis-generating harm signal
Catheter-related bloodstream infection, n (%) 15 (11%) 18 (13%) Not reported P=0.52 Similar between strategies
  • Despite major separation in RRT initiation timing (median 33 vs 3 hours) and fewer patients receiving RRT (79% vs 98%), the primary endpoint (RRT-free days to day 28) was essentially identical (median 10 vs 12; P=0.93).
  • There was a clinically important signal towards worse outcomes with “more-delayed” initiation: higher metabolic acidosis (30% vs 18%; P=0.022) and higher day-60 mortality (55% vs 44%; P=0.071), with a statistically significant adjusted hazard for death (HR 1.65; 95% CI 1.09 to 2.50; P=0.018).
  • Most “more-delayed” RRT starts were triggered by the urea nitrogen threshold (BUN >140 mg/dL) rather than classic urgent indications, and 21% avoided RRT altogether.

Internal Validity

  • Randomisation and allocation: Centralised computer randomisation, stratified by centre; allocation concealment via a web-based system is likely to have minimised selection bias.
  • Attrition / exclusions: 278 patients randomised (141 more-delayed; 137 delayed) and analysed by intention-to-treat; day-60 mortality ascertainment was not reported as incomplete.
  • Performance/detection bias: Open-label design; although initiation triggers were strategy-defined, clinician discretion (including physician approval for starting RRT) could influence decisions at the margin.
  • Protocol adherence (initiation): Clear separation achieved: received any RRT 111/141 (79%) vs 134/137 (98%); time to RRT 33 (24–60) vs 3 (2–5) hours.
  • Protocol adherence (delivered RRT): Among those treated, intensity surrogates were similar: number of RRT sessions 5 (2–10) vs 5 (2–10); duration of RRT 5 (2–10) vs 5 (2–10); first-day modality distribution broadly similar (intermittent 58–60%; continuous 39–40%).
  • Baseline characteristics: Groups were broadly comparable for age and illness severity (SAPS III ~72–73; SOFA ~11–12); notable numeric differences included higher creatinine (5.9 vs 5.0 mg/dL) and higher BUN (107 vs 92 mg/dL) at randomisation in the more-delayed group, and lower vasopressor use (57% vs 69%).
  • Heterogeneity: Pragmatic multicentre delivery with centre-level practice variation (modality and dose left to clinicians); randomisation stratified by centre helps, but residual clinical heterogeneity (AKI aetiologies/trajectories) is intrinsic.
  • Timing: The trial tested “late versus later” initiation; randomisation occurred only after patients met delayed criteria (oliguria/anuria >72 h or BUN >112 mg/dL), limiting inference about earlier initiation windows.
  • Separation of the variable of interest: Strong separation for initiation timing (33 vs 3 hours) and exposure (79% vs 98% receiving RRT) did not translate into separation in RRT-free days (10 vs 12), implying that competing risk (death) and/or discontinuation practices strongly influenced the primary outcome.
  • Outcome assessment: Mortality and most complications were objective; the primary endpoint (RRT-free days) embeds clinician-driven discontinuation and the competing risk of death.
  • Statistical rigour: Sample size target achieved (n=278 vs planned n=270); intention-to-treat analysis used for the primary outcome; adjusted mortality analysis reported a significant hazard ratio, but mortality was not the primary endpoint.

Conclusion on Internal Validity: Overall, internal validity appears moderate to strong given robust randomisation and clear separation in initiation/exposure, but is limited by the open-label design and a primary outcome that is sensitive to clinician-driven discontinuation and the competing risk of death.

External Validity

  • Population representativeness: Critically ill patients with KDIGO stage 3 AKI plus advanced organ support (vasopressors and/or invasive ventilation), with a high proportion of septic shock; this matches many general ICUs but represents a severely ill subset rather than all AKI.
  • Important exclusions: Patients with immediate classic indications for RRT, severe chronic renal failure, and selected AKI aetiologies were excluded, limiting applicability to those groups.
  • Setting and resources: Conducted in 39 French ICUs with ready access to intermittent haemodialysis and continuous RRT; applicability may be reduced in resource-limited settings where thresholds and monitoring intensity differ.
  • Operationalisability: Strategy-based thresholds (BUN triggers plus urgent indications) are implementable, but unit conventions (BUN vs urea) and clinician comfort with very high urea burdens may vary internationally.

Conclusion on External Validity: Findings are moderately generalisable to high-resource ICUs managing severe KDIGO stage 3 AKI on organ support, but less applicable to milder AKI, advanced CKD, immediate-indication AKI, or settings without close biochemical/clinical monitoring.

Strengths & Limitations

  • Strengths:
    • Pragmatic multicentre randomised design directly addressing a clinically common decision point (“how long can we safely wait?”).
    • Clear protocol-driven separation in initiation timing and RRT exposure (33 vs 3 hours; 79% vs 98% receiving RRT).
    • Clinically meaningful secondary outcomes including complications and functional status (Barthel index at day 60).
    • Prospectively published protocol and pre-specified analytic framework1.
  • Limitations:
    • Open-label strategy trial; clinician behaviour can influence initiation and discontinuation, affecting RRT-free days.
    • Two-step design selects a minority of screened patients (those who both reached KDIGO stage 3 and then met delayed criteria), narrowing inference.
    • Primary endpoint is a composite sensitive to death (competing risk) and clinician-driven discontinuation practices.
    • Mortality difference did not reach conventional significance in the unadjusted comparison (P=0.071) and the trial was not powered primarily for mortality.

Interpretation & Why It Matters

  • Defines a practical “upper bound” for watchful waiting
    In patients already meeting delayed triggers (BUN >112 mg/dL or oliguria/anuria >72 h), delaying further to BUN 140 mg/dL (or urgent indications) did not improve RRT-free days and raised concern for harm (more metabolic acidosis; mortality signal).
  • Avoiding unnecessary RRT remains achievable, but not at any cost
    One in five patients (21%) in the more-delayed arm avoided RRT entirely, reinforcing heterogeneity of AKI trajectories; however, “more-delayed” thresholds may permit clinically consequential uraemic/acid–base derangements in others.
  • Implications for strategy-based care pathways
    AKIKI-2 supports structured strategy pathways (clear urgent-indication thresholds and biochemical ceilings) rather than purely time-based triggers, while cautioning against excessively high urea nitrogen thresholds as a default rule.

Controversies & Subsequent Evidence

  • Endpoint choice and interpretability: The primary endpoint (RRT-free days) is strongly influenced by both death and clinician-driven discontinuation of RRT; commentary highlighted that discontinuation practices were not tightly standardised, potentially diluting strategy separation at the level of the primary endpoint2.
  • Mortality signal (unadjusted vs adjusted): Day-60 mortality was higher in the more-delayed group (55% vs 44%; P=0.071), and an adjusted Cox model reported higher hazard of death (HR 1.65; 95% CI 1.09 to 2.50; P=0.018); interpretation remains debated because randomisation should balance measured and unmeasured confounders, and the trial was not primarily powered for mortality.
  • “Timing” versus “strategy” framing: Correspondence argued that AKI trials should be interpreted as testing initiation strategies anchored to physiology/trajectory rather than clock-time alone, and the authors’ reply emphasised the clinical framing of strategy thresholds and the hazards of pushing watchful waiting beyond commonly used ceilings34.
  • Position within the broader RCT landscape: Earlier large ICU RCTs generally found no mortality benefit from earlier/accelerated RRT initiation compared with delayed/standard strategies (AKIKI, IDEAL-ICU, STARRT-AKI)567; ELAIN (single-centre, predominantly surgical population) reported benefit with earlier initiation and continues to contribute to heterogeneity across trials8.
  • Meta-analytic synthesis: High-level syntheses incorporating modern ICU RCTs generally report no survival advantage with earlier initiation while showing increased exposure to RRT (and its attendant risks) when initiation is accelerated; Cochrane and large meta-analyses support a conservative/standard initiation approach absent urgent indications910.
  • Guideline alignment: Contemporary sepsis and kidney guidance emphasises starting RRT for life-threatening complications (e.g., refractory hyperkalaemia, acidosis, pulmonary oedema) and using biochemical/clinical trajectories rather than isolated creatinine thresholds to guide initiation decisions1112.

Summary

  • AKIKI-2 tested “late versus later” RRT initiation in severe ICU AKI: delayed initiation within 12 hours after meeting delayed triggers versus a more-delayed strategy deferring RRT until urgent indications or BUN 140 mg/dL for 1 day.
  • The more-delayed strategy reduced RRT use (79% vs 98%) and substantially delayed initiation (median 33 vs 3 hours) but did not improve RRT-free days at day 28 (median 10 vs 12; P=0.93).
  • There was a clinically concerning signal for harm with more-delayed initiation: higher metabolic acidosis (30% vs 18%; P=0.022) and higher day-60 mortality (55% vs 44%; P=0.071), with higher adjusted hazard of death (HR 1.65; 95% CI 1.09 to 2.50; P=0.018).
  • Approximately 21% of patients in the more-delayed group avoided RRT entirely, highlighting heterogeneity and the potential value of better prognostic enrichment.
  • Overall, AKIKI-2 supports structured watchful waiting but cautions against pushing urea nitrogen thresholds too high as a default strategy in severe ICU AKI.

Overall Takeaway

AKIKI-2 suggests that in severe ICU AKI patients who have already crossed “delayed strategy” thresholds, pushing RRT initiation even later (to very high urea nitrogen levels unless urgent indications occur) does not improve days alive and RRT-free and may introduce clinically important harm signals. The trial therefore helps define a pragmatic upper boundary for watchful waiting and reinforces strategy-based initiation using urgent indications and trajectory rather than simply “waiting longer”.

Overall Summary

  • More-delayed RRT (urgent indications or BUN 140 mg/dL) reduced RRT exposure but did not improve RRT-free days and raised concern for harm (metabolic acidosis; mortality signal).
  • In severe ICU AKI, “watchful waiting” should remain structured and bounded rather than progressively deferred without clear physiological justification.

Bibliography