Skip to main content

Foundational Trials

AKIKI

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Initiation Strategies for Renal-Replacement Therapy in the Intensive Care Unit
  • Acronym: AKIKI
  • Year: 2016
  • Journal published in: New England Journal of Medicine
  • Citation: Gaudry S, Hajage D, Schortgen F, Martin-Lefevre L, Pons B, Boulet E, et al. Initiation Strategies for Renal-Replacement Therapy in the Intensive Care Unit. N Engl J Med. 2016;375(2):122-133.

Context & Rationale

  • Background
    • Renal replacement therapy (RRT) is a high-frequency, high-stakes ICU intervention for severe acute kidney injury (AKI), but the optimal initiation timing in the absence of immediate life-threatening complications was uncertain.
    • Earlier initiation might prevent metabolic/volume complications and “get ahead” of multi-organ failure, but exposes patients to catheter-related harms, haemodynamic instability, electrolyte derangements, and resource use.
    • Prior evidence was dominated by heterogeneous observational studies (strong confounding by indication) and small RCTs, leaving equipoise for a large pragmatic trial with explicit “watchful waiting” triggers.
  • Research Question/Hypothesis
    • In critically ill adults with severe AKI (KDIGO stage 3) receiving mechanical ventilation and/or vasopressors, and without urgent indications for RRT, does a delayed, trigger-based strategy improve 60-day survival compared with immediate initiation?
    • Implicit secondary hypothesis: a delayed strategy would reduce exposure to RRT without excess harm.
  • Why This Matters
    • Defines whether “on-demand” RRT can be safely used in severe ICU AKI, potentially avoiding unnecessary dialysis and procedure-related complications.
    • Provides a reproducible trigger set for trialists and clinicians, supporting standardisation across future RRT-timing studies and guidelines.

Design & Methods

  • Research Question: Among ICU patients with severe AKI (KDIGO stage 3) without urgent indications, does delayed (trigger-based) initiation of RRT reduce 60-day mortality compared with early (immediate) initiation?
  • Study Type: Prospective, multicentre, parallel-group, open-label, investigator-initiated randomised trial in 31 ICUs (France); central web-based randomisation stratified by centre; group-sequential design with planned interim analyses.1
  • Population:
    • Setting: Medical and mixed ICUs; enrolment September 2013 to January 2016.
    • Key inclusion: Adults (≥18 years) with AKI compatible with acute tubular necrosis (ischaemic/toxic context) and KDIGO stage 3 (creatinine >354 μmol/L or ≥3× baseline; anuria >12 h; or oliguria <0.3 mL/kg/h or <500 mL/day for >24 h), plus invasive mechanical ventilation and/or continuous IV catecholamines.
    • Key exclusions (selected): Severe chronic kidney disease (creatinine clearance <30 mL/min), kidney transplant, obstructive/vascular AKI, tumour lysis, thrombotic microangiopathy, acute glomerulopathy, dialysable intoxication, Child–Pugh C cirrhosis, extracorporeal circulation, moribund state or a decision to withhold/withdraw active treatment.
    • Pre-randomisation urgent RRT indications (excluded at baseline): Oliguria/anuria >3 days; serum potassium >6 mmol/L (or >5.5 mmol/L despite medical therapy); serum urea >40 mmol/L; metabolic acidosis with pH <7.15; acute pulmonary oedema with severe hypoxaemia despite diuretics.
  • Intervention:
    • Early strategy: RRT initiated as soon as possible after randomisation (trial target: within 6 h of documentation of KDIGO stage 3).
    • Modality: Continuous or intermittent RRT chosen by the treating team according to local practice and haemodynamic status.
  • Comparison:
    • Delayed (trigger-based) strategy: No RRT at randomisation; initiate RRT only if at least one pre-specified criterion occurred: (1) serum potassium >6 mmol/L or >5.5 mmol/L despite medical treatment; (2) metabolic acidosis with pH <7.15; (3) acute pulmonary oedema due to fluid overload with severe hypoxaemia; (4) serum urea >40 mmol/L; or (5) oliguria/anuria persisted for >72 h after randomisation.
    • Co-interventions: Medical management of hyperkalaemia/acidosis and diuretics permitted; RRT started promptly once criteria were met.
  • Blinding: Unblinded (open-label); primary outcome (all-cause mortality) is objective, but co-interventions and decisions about RRT modality/dose could introduce performance bias.
  • Statistics: A total of 546 patients were required to detect a 15 percentage-point absolute reduction in 60-day mortality (from 55% to 40%) with early initiation (two-sided α=0.05; 90% power). Group-sequential design inflated the target to 560 with two interim analyses; with a planned 10% loss to follow-up, total planned enrolment was 620. Primary analysis was intention-to-treat using time-to-event methods.
  • Follow-Up Period: 60 days for the primary endpoint; secondary outcomes reported to day 28 and day 60.

Key Results

This trial was not stopped early. Two planned interim analyses were performed with a blinded DSMB recommendation to continue; recruitment completed to 620 participants with follow-up to day 60.

Outcome Early strategy Delayed strategy Effect p value / 95% CI Notes
Mortality at day 60 (primary) 150/309 (48.5%) 153/308 (49.7%) HR 1.03 95% CI 0.82 to 1.29; P=0.79 Denominators reflect available vital status (early 309/311; delayed 308/308 at day 60).
Mortality at day 60 (adjusted analysis) — — HR 1.02 95% CI 0.80 to 1.28; P=0.84 Adjusted for prespecified covariates (as reported).
Mortality at day 28 128/308 (41.6%) 133/306 (43.5%) Not reported Not reported Denominators reflect available vital status (early 308/311; delayed 306/308 at day 28).
Received any RRT 305/311 (98%) 157/308 (51%) Not reported P<0.001 Delayed strategy avoided RRT in 151/308 (49%).
RRT-free days to day 28 (median, IQR) 17 (2–26) 19 (5–29) Not reported P<0.001 Includes patients who never received RRT.
Renal recovery at day 28 in survivors (RRT discontinued ≥7 days) 82/181 (45.3%) 82/165 (49.7%) Not reported P=0.35 Assessed among survivors to day 28.
Dialysis dependence at day 60 in survivors 5/237 (2.1%) 11/231 (4.8%) Not reported P=0.12 Among survivors to day 60.
Catheter-related bloodstream infection 30/311 (10%) 15/308 (5%) Not reported P=0.03 Higher in early strategy; plausibly exposure-mediated.
Hypophosphataemia 68/311 (22%) 47/308 (15%) Not reported P=0.03 Consistent with greater RRT exposure in early strategy.
Haemorrhage requiring transfusion 28/311 (9%) 37/308 (12%) Not reported P=0.21 Bleeding events were frequent in both groups.
  • Despite a large reduction in RRT exposure (98% vs 51%), 60-day mortality was similar between strategies (HR 1.03; 95% CI 0.82 to 1.29).
  • The delayed strategy functioned as a true “watchful waiting” approach: 49% never required RRT, and median time from randomisation to first RRT was 57 h (vs 2 h early).
  • Early initiation increased exposure-related complications, including catheter-related bloodstream infection (10% vs 5%) and hypophosphataemia (22% vs 15%).

Internal Validity

  • Randomisation and allocation:
    • Central web-based randomisation with stratification by centre and variable block sizes (allocation concealed until assignment).
    • Eligibility confirmation and randomisation occurred shortly after meeting KDIGO stage 3 criteria (trial target within 5 h), limiting immortal time bias at entry.
  • Dropout or exclusions:
    • Randomised: 620 patients; 1 withdrew permission to use data.
    • Denominators for mortality indicate minimal missing vital status (day 28: 308/311 early and 306/308 delayed; day 60: 309/311 early and 308/308 delayed).
  • Performance/detection bias:
    • Open-label design makes performance bias plausible (e.g., diuretic and medical rescue therapies) but primary outcome was objective (mortality).
    • Trigger-based criteria reduced clinician discretion over “when” to start RRT in the delayed arm, improving protocol fidelity despite lack of blinding.
  • Protocol adherence and crossover:
    • Early strategy: 6/311 (1.9%) did not receive RRT (3 ineligible inclusions who survived; 2 died before RRT could be started; 1 assignment misread).
    • Delayed strategy: 5/308 (1.6%) received RRT without meeting criteria (misinterpretation of assignment/criteria).
    • These deviations were uncommon, supporting preserved randomised contrast.
  • Baseline characteristics:
    • Groups were well matched in severity (e.g., SOFA score at randomisation 11±3 vs 11±3; SAPS III 73±12 vs 74±12) and organ support (mechanical ventilation ~86–87%; vasopressors ~85–86%).
    • High prevalence of sepsis (~80%) and multi-organ failure supports biological plausibility for benefit if early RRT truly modified outcomes.
  • Heterogeneity:
    • RRT modality and prescription were pragmatic (centre/clinician determined), which may increase external validity but introduces within-group variability in delivered “dose”.
    • Stratification by centre mitigates, but does not eliminate, potential centre-level effect modification.
  • Timing and dose of the intervention:
    • Time to first RRT: median 2 h (IQR 1–3) early vs 57 h (IQR 25–83) delayed.
    • Trigger-to-RRT latency in delayed arm: median 4.7 h (IQR 3.0–7.8).
    • Delayed-arm triggers (first qualifying criterion leading to RRT): oliguria/anuria >72 h 36.3%; hyperkalaemia 26.8%; urea >40 mmol/L 18.5%; metabolic acidosis 15.3%; pulmonary oedema 2.6%.
    • RRT exposure: total sessions 1665 (early) vs 943 (delayed); sessions per treated patient 3 (IQR 2–6) vs 4 (IQR 2–8).
    • Biochemistry at initiation: creatinine 3.27 mg/dL (IQR 2.62–4.05) early vs 5.33 mg/dL (IQR 3.85–6.50) delayed; urea 52 mg/dL (IQR 40–65) vs 90 mg/dL (IQR 65–106); potassium 4.4 mmol/L (IQR 3.8–4.8) vs 5.1 mmol/L (IQR 4.4–5.7).
  • Adjunctive therapy use:
    • Consistent with “watchful waiting”, delayed strategy used more diuretics (1.3% vs 36.5%) and more medical treatments for hyperkalaemia (5.5% vs 22.9%), reflecting predictable protocol consequences.
  • Outcome assessment and statistical rigour:
    • Primary outcome (all-cause mortality at day 60) is objective, reducing ascertainment bias.
    • Sequential design with planned interim analyses and intention-to-treat primary analysis supports inferential integrity; the trial was powered for a large absolute mortality difference (15%), so smaller effects remain statistically uncertain.

Conclusion on Internal Validity: Overall, internal validity appears moderate to strong: randomisation and follow-up were robust with excellent separation in RRT exposure, but open-label conduct and pragmatic co-interventions introduce some performance-bias risk (mitigated by objective mortality and protocolised delayed triggers).

External Validity

  • Population representativeness:
    • Represents a common ICU phenotype: severe AKI with multi-organ failure (high rates of sepsis, mechanical ventilation, vasopressors).
    • However, applicability is limited to patients who have reached KDIGO stage 3 and are sick enough to require organ support; it does not address earlier AKI (KDIGO 1–2) or ward/non-ICU settings.
    • Exclusions (e.g., creatinine clearance <30 mL/min, kidney transplant, Child–Pugh C cirrhosis, dialysable intoxication) narrow generalisability to those subgroups.
  • Applicability across systems:
    • The delayed strategy requires close biochemical/clinical monitoring and the capability to initiate RRT rapidly when triggers occur; centres with constrained access to RRT may not replicate the safety profile.
    • Modality mix (intermittent vs continuous) and catheter-related complication rates may differ in other healthcare systems, influencing net benefit/harm.

Conclusion on External Validity: Generalisability is moderate—the findings apply well to adult, high-acuity ICU patients with KDIGO stage 3 AKI and ready access to RRT, but less clearly to earlier AKI, advanced chronic kidney disease, or resource-limited environments.

Strengths & Limitations

  • Strengths:
    • Large, multicentre pragmatic randomised trial with high follow-up completeness.
    • Clinically meaningful primary endpoint (60-day mortality) and clear protocolised triggers for delayed RRT.
    • Marked separation in RRT exposure (98% vs 51%) supports a valid test of the clinical strategy.
  • Limitations:
    • Open-label conduct with potential co-intervention and behavioural effects (e.g., diuretics and medical rescue therapies differed).
    • RRT modality and prescription were not standardised, increasing heterogeneity in delivered treatment.
    • Powered for a large mortality effect (15% absolute); smaller but clinically relevant differences cannot be excluded.
    • Restricted inclusion (KDIGO stage 3 plus organ support) limits inference for earlier AKI or less severely ill populations.

Interpretation & Why It Matters

  • Clinical message
    • For severe ICU AKI (KDIGO stage 3) without urgent indications, “watchful waiting” with explicit triggers can avoid RRT in ~49% of patients without increasing 60-day mortality.
    • Early routine initiation increases exposure-related harms (catheter infection, hypophosphataemia) and resource use, without demonstrated survival benefit.
    • Practically, AKIKI supports an “on-demand” model: start promptly when clear biochemical/clinical triggers arise, rather than initiating solely because AKI has reached stage 3.

Controversies & Subsequent Evidence

  • Defining “early” and “delayed” across trials:
    • The accompanying NEJM editorial emphasised that timing strategies are inseparable from patient selection, competing risks, and iatrogenic harms of RRT, motivating an “indications-based” approach rather than routine early initiation.2
    • NEJM correspondence highlighted that apparently discordant findings between AKIKI and the single-centre ELAIN trial likely reflect different enrolment thresholds (ELAIN enrolled KDIGO stage 2 and used a later trigger set broadly similar to AKIKI’s early arm), complicating naïve trial-to-trial comparisons.34
  • Subsequent RCTs:
    • IDEAL-ICU (septic shock with severe AKI) found no mortality benefit to early initiation and was stopped early for futility, reinforcing the AKIKI signal in a sepsis-enriched population.5
    • STARRT-AKI (large international RCT) similarly showed no reduction in 90-day mortality with accelerated initiation, while increasing RRT exposure and adverse events—convergent evidence supporting “standard/on-demand” initiation.6
    • AKIKI2 tested whether “more-delayed” initiation (beyond a trigger-based delayed strategy) was safe and suggested potential harm with excessive delay, supporting a balanced “wait, but not too long” paradigm.7
  • Evidence syntheses and guideline direction:
    • An individual patient data meta-analysis found no survival benefit to early initiation across RCTs, with increased RRT exposure in early strategies.8
    • A contemporary Cochrane review reached similar conclusions regarding mortality and underscored the trade-off between earlier biochemical control and higher exposure to RRT-related harms.9
    • More recent ICU-focused guidance trends towards starting RRT for conventional urgent indications and avoiding routine early initiation in their absence, aligning with AKIKI’s trigger-based approach.10

Summary

  • In severe ICU AKI (KDIGO stage 3) without urgent indications, a delayed, trigger-based RRT strategy produced similar 60-day mortality to immediate initiation (HR 1.03; 95% CI 0.82 to 1.29).
  • The delayed strategy avoided RRT entirely in 49% of patients (51% vs 98% received RRT), emphasising that “delayed” is operationally “conditional/on-demand”.
  • Delayed initiation led to more advanced biochemical derangement at the point of RRT start (e.g., creatinine 5.33 vs 3.27 mg/dL; potassium 5.1 vs 4.4 mmol/L), but without detectable harm in mortality.
  • Early initiation increased exposure-related complications, including catheter-related bloodstream infection (10% vs 5%) and hypophosphataemia (22% vs 15%).
  • Subsequent large RCTs and meta-analyses largely support an indications-based, non-accelerated initiation approach, with caution against “excessive delay”.

Overall Takeaway

AKIKI established that, in severe ICU AKI without immediate life-threatening complications, a protocolised “watchful waiting” strategy can avoid RRT in nearly half of patients without worsening 60-day survival. Its real contribution is shifting the default from “dialyse early because AKI is severe” to “dialyse when clear triggers emerge”, while recognising that excessive delay may be harmful in selected patients.

Overall Summary

  • No mortality benefit to routine immediate RRT in KDIGO stage 3 ICU AKI without urgent indications.
  • Trigger-based delayed initiation avoids unnecessary RRT (49%) and reduces exposure-related harms.
  • Subsequent RCTs and meta-analyses broadly corroborate an indications-based, non-accelerated initiation strategy.

Bibliography