Skip to main content

Foundational Trials

ELAIN

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Effect of Early vs Delayed Initiation of Renal Replacement Therapy on Mortality in Critically Ill Patients With Acute Kidney Injury: The ELAIN Randomized Clinical Trial
  • Acronym: ELAIN
  • Year: 2016
  • Journal published in: JAMA
  • Citation: Zarbock A, Kellum JA, Schmidt C, Van Aken H, Wempe C, Pavenstädt H, Boanta A, Gerß J, Meersch M. Effect of Early vs Delayed Initiation of Renal Replacement Therapy on Mortality in Critically Ill Patients With Acute Kidney Injury: The ELAIN Randomized Clinical Trial. JAMA. 2016;315(20):2190-2199.

Context & Rationale

  • Background
    • Optimal timing of renal replacement therapy (RRT) for ICU-associated acute kidney injury (AKI) was uncertain, with wide practice variation between early/pre-emptive initiation and “watchful waiting”.
    • Earlier RRT could plausibly improve physiological control (solute, acid–base, electrolytes, fluid balance) and mitigate downstream organ dysfunction.
    • Earlier RRT could also expose patients who might recover without RRT to avoidable harms (vascular access complications, haemodynamic instability, anticoagulation-related bleeding, electrolyte disturbances) and increased resource use.
    • Prior evidence consisted largely of heterogeneous observational studies and small randomised trials, with inconsistent findings and variable definitions of “early” vs “late”.
    • Biomarker-based enrichment (e.g. neutrophil gelatinase–associated lipocalin [NGAL]) was proposed to identify high-risk patients more likely to progress and therefore more likely to benefit from earlier RRT.
  • Research Question/Hypothesis
    • In critically ill adults with KDIGO stage 2 AKI and elevated NGAL plus additional severity criteria, does initiating continuous RRT within 8 hours reduce 90-day all-cause mortality compared with delaying RRT until KDIGO stage 3 or conventional absolute indications?
  • Why This Matters
    • Timing decisions drive both exposure to RRT (including potentially avoidable initiation) and key patient-centred outcomes (survival, renal recovery, dialysis dependence, ICU/hospital trajectory).
    • If early RRT were beneficial, it would justify initiating RRT at moderate AKI severity (KDIGO 2) in selected high-risk patients rather than waiting for complications or further progression.
    • If not beneficial, it would support deferring RRT to reduce unnecessary treatment and iatrogenic complications.

Design & Methods

  • Research Question: In critically ill adults with enriched-risk KDIGO stage 2 AKI, does early initiation of continuous RRT (within 8 hours) reduce 90-day all-cause mortality versus a delayed strategy (initiate within 12 hours of KDIGO stage 3 or absolute indications)?
  • Study Type: Single-centre, randomised, investigator-initiated, parallel-group controlled trial; stratified block randomisation; open-label; surgical ICU at a tertiary university hospital (Münster, Germany); enrolment August 2013 to June 2015; registry DRKS00004367.
  • Population:
    • Inclusion: age 18–90 years; KDIGO stage 2 AKI (creatinine 2.0–2.9× baseline and/or urine output <0.5 mL/kg/h for ≥12 h); plasma NGAL >150 ng/mL; plus ≥1 of: severe sepsis; vasopressors >0.1 μg/kg/min; refractory fluid overload with PaO2/FiO2 <300 and/or cumulative fluid balance >10% body weight; development/progression of non-renal organ dysfunction (non-renal SOFA ≥2).
    • Exclusion: eGFR <30 mL/min/1.73m2; prior RRT or kidney transplant; pregnancy; AIDS with CD4 <0.05×109/L; haematologic malignancy with neutrophils <0.05×109/L; hepatorenal syndrome; AKI due to renal artery occlusion, glomerulonephritis, interstitial nephritis, vasculitis, post-renal obstruction, HUS/TTP; concurrent participation in another interventional trial.
    • Stratification at randomisation: cardiovascular SOFA score (0–2 vs 3–4) and presence/absence of oliguria.
  • Intervention:
    • Strategy: initiate RRT within 8 hours of meeting eligibility (KDIGO stage 2 with enrichment criteria).
    • Modality: continuous venovenous haemodiafiltration (CVVHDF); effluent target 30 mL/kg/h; pre-dilution replacement; dialysate-to-replacement ratio 1:1; blood flow >110 mL/min; citrate anticoagulation.
    • Stopping criteria: urine output >400 mL/24 h without diuretics (or >2100 mL/24 h with diuretics) plus creatinine clearance >20 mL/min; switch to intermittent modalities permitted if RRT continued beyond 7 days.
  • Comparison:
    • Strategy: initiate RRT within 12 hours after KDIGO stage 3 AKI, or within 12 hours if an absolute indication occurred.
    • KDIGO stage 3 triggers: urine output <0.3 mL/kg/h for ≥24 h and/or ≥3-fold creatinine increase or creatinine ≥4.0 mg/dL with acute rise ≥0.5 mg/dL within 48 h.
    • Absolute indications: blood urea nitrogen >100 mg/dL; potassium >6 mEq/L and/or ECG changes; magnesium >8 mEq/L; urine output <200 mL/12 h or anuria; organ oedema in diuretic-resistant AKI.
    • Implication: delayed-group participants could avoid RRT if they never met these triggers (11/119 did not receive RRT).
  • Blinding: Unblinded (open-label); mortality outcome is objective, but several secondary outcomes (e.g. discharge/length of stay) are susceptible to performance bias.
  • Statistics: A total of 230 patients were required to detect an 18% absolute reduction in 90-day mortality (from 55% to 37%) with 80% power at the 5% significance level; group sequential design with one interim analysis planned at 50% of expected deaths; primary analysis in the full analysis set (intention-to-treat); additional per-protocol analysis restricted to those receiving RRT.
  • Follow-Up Period: Primary endpoint at 90 days; secondary outcomes at 28 and 60 days; lengths of stay analysed with censoring at 90 days.

Key Results

This trial was not stopped early. The planned sample size (n=231) was enrolled; a group-sequential interim analysis was pre-specified but the trial completed accrual.

Outcome Early RRT (n=112) Delayed RRT (n=119) Effect p value / 95% CI Notes
90-day all-cause mortality 44 (39.3%) 65 (54.7%) HR 0.66; abs diff −15.4% HR 0.45 to 0.97; abs diff −28.1 to −2.6; P=0.03 Primary endpoint
28-day mortality 34 (30.4%) 48 (40.3%) OR 0.64; abs diff −10.0% OR 0.37 to 1.11; abs diff −22.2 to 2.3; P=0.11 Secondary endpoint
60-day mortality 43 (38.4%) 60 (50.4%) OR 0.61; abs diff −12.0% OR 0.36 to 1.03; abs diff −24.8 to 0.7; P=0.07 Secondary endpoint
Duration of RRT (days), median (Q1–Q3) 9 (4–44) 25 (7–>90) HR 0.69; abs diff −18 days HR 0.48 to 1.00; abs diff −41 to 4; P=0.04 Among those receiving RRT (112 vs 108)
Renal recovery at day 90 60 (53.6%) 46 (38.7%) OR 0.55; abs diff 14.9% OR 0.32 to 0.93; abs diff 2.2 to 27.6; P=0.02 Deaths counted as “no recovery”
Renal recovery at day 90 (excluding deaths) 60/68 (88.2%) 46/54 (85.2%) OR 0.77; abs diff 3.1% OR 0.27 to 2.17; abs diff −9.1 to 15.2; P=0.62 Conditioning on survival attenuated between-group differences
Requirement of RRT at day 90 9/67 (13.4%) 8/53 (15.1%) OR 0.87; abs diff −1.7% OR 0.31 to 2.44; abs diff −14.3 to 11.0; P=0.80 Among survivors with available data
Duration of mechanical ventilation (hours), median (Q1–Q3) 125.5 (41–203) 181.0 (65–413) Abs diff −60.0 h Abs diff −110.0 to −22.0; P=0.002 Secondary endpoint; vulnerable to practice-pattern effects
Hospital length of stay (censored at day 90), median (Q1–Q3) 51 (31–74) 82 (67–>90) HR 0.34; abs diff −37 days HR 0.22 to 0.52; abs diff −∞ to −19.5; P<0.001 Time-to-discharge analysis with death censoring
ICU length of stay (censored at day 90), median (Q1–Q3) 19 (9–29) 22 (12–36) HR 0.85; abs diff −3.0 days HR 0.61 to 1.19; abs diff −12.0 to 4.5; P=0.33 No significant difference
Serious adverse events (excluding death) among RRT recipients 1/112 (0.9%) 0/108 (0%) Not reported Not reported Serious event: new-onset arrhythmia
  • Early initiation was associated with lower 90-day mortality: 39.3% vs 54.7% (HR 0.66; 95% CI 0.45 to 0.97; P=0.03).
  • Renal recovery at day 90 appeared higher when deaths were included (53.6% vs 38.7%; OR 0.55; 95% CI 0.32 to 0.93; P=0.02), but was similar among survivors (88.2% vs 85.2%; P=0.62).
  • Non-mortality outcomes favoured early RRT (mechanical ventilation 125.5 vs 181.0 hours; P=0.002; hospital stay censored 51 vs 82 days; HR 0.34; 95% CI 0.22 to 0.52; P<0.001), while ICU stay did not differ.

Internal Validity

  • Randomisation and allocation
    • Computer-generated 1:1 randomisation using stratified block randomisation (block size 10).
    • Strata were cardiovascular SOFA (0–2 vs 3–4) and presence/absence of oliguria, addressing prognostically important haemodynamic and renal phenotype features.
    • Allocation concealment appears plausible via the computerised assignment process, but the single-centre nature limits assessment of implementation fidelity beyond what is reported.
  • Dropout/exclusions after randomisation
    • No post-randomisation exclusions were reported for the full analysis set (n=231); follow-up for the primary endpoint was effectively complete.
    • Per-protocol analyses were pre-specified to include only those who received RRT (112 vs 108), but the primary inference was from the full analysis set.
  • Performance/detection bias
    • Open-label design introduces risk of performance bias for co-interventions and care processes (e.g. ventilation weaning, discharge decisions), particularly for length-of-stay endpoints.
    • Mortality is objective and less vulnerable to observer bias, but downstream processes influencing deaths (e.g. goals-of-care decisions) can still be indirectly influenced by knowledge of group assignment.
  • Protocol adherence and “separation” of the intervention
    • All early-strategy patients received RRT (112/112) versus 108/119 in the delayed strategy (11/119 never required RRT), indicating a clinically meaningful “avoidance” component in the delayed arm.
    • Time from KDIGO stage 2 to RRT was separated: median 6.0 (Q1–Q3 4.0–7.0) hours vs 25.5 (18.8–40.3) hours; absolute difference −21.0 hours (95% CI −24.0 to −18.0); P<0.001.
    • Physiological status at RRT initiation differed: creatinine 1.9 (0.6) vs 2.4 (1.0) mg/dL; difference −0.5 (95% CI −0.7 to −0.3); P<0.001.
    • Blood urea nitrogen at initiation: 38.5 (15.5) vs 47.5 (21.6) mg/dL; difference −9.0 (95% CI −14.1 to −3.9); P=0.001.
    • Urine output at initiation: 445.0 (175.0–807.5) vs 270.0 (112.5–670.0) mL; difference 115.0 (95% CI 25.0 to 220.0); P=0.01.
  • Baseline characteristics and “room to benefit”
    • Groups were broadly similar in severity: SOFA 15.6 (2.3) vs 16.0 (2.3); APACHE II 30.6 (7.5) vs 32.7 (8.8); mechanically ventilated 87.5% vs 88.2%; vasopressors 85.7% vs 90.8%.
    • The cohort was enriched for severe illness and surgical presentations (e.g. 50.0% vs 43.7% cardiac admissions; 30.4% vs 37.0% abdominal admissions), supporting biological plausibility for high event rates and potential treatment effect detection.
    • Some baseline imbalances existed (e.g. men 69.6% vs 57.1%; hypertension 86.6% vs 77.3%; arrhythmia 33.0% vs 44.5%), which may matter in a modest-sized single-centre RCT.
  • Heterogeneity
    • Single-centre conduct reduces between-centre heterogeneity but increases susceptibility to centre-specific practice effects and limits exploration of treatment-effect heterogeneity across settings.
    • The trial used an enriched entry phenotype (KDIGO 2 + NGAL + additional severity criteria), which reduces clinical heterogeneity but narrows inference.
  • Timing and dose of the intervention
    • Early initiation was operationalised relative to KDIGO stage 2 (≤8 hours), with delayed initiation tied to stage 3 or absolute indications (≤12 hours after trigger).
    • Delivered CRRT intensity in those receiving RRT was similar by group: effluent volume 26.6 (4.7) vs 26.6 (5.8) mL/kg/h; session duration 22.6 (1.7) vs 22.4 (2.1) hours/day.
  • Crossover and contamination
    • No conventional crossover from delayed to early strategy was reported; delayed initiation occurred per protocol triggers.
    • The “avoidance” of RRT in 11/119 delayed-strategy patients is a feature of the intervention contrast rather than contamination.
  • Outcome assessment
    • Primary endpoint (90-day all-cause mortality) is objective and clinically meaningful.
    • Renal recovery definitions and RRT dependence outcomes are clinically relevant, but interpretation is sensitive to competing risk of death (illustrated by attenuation in survivor-only analyses).
    • Length-of-stay and ventilation duration outcomes are clinically important but vulnerable to unblinded care and discharge practices.
  • Statistical rigour
    • Sample size target was met (n=231) and group sequential monitoring was pre-specified.
    • The primary effect estimate was expressed as a hazard ratio with 95% CI and corresponding absolute risk difference; P value was nominally significant (P=0.03), consistent with a potentially fragile result in a modest-sized trial.

Conclusion on Internal Validity: Internal validity is moderate: randomisation was structured and follow-up was complete with a hard primary endpoint, but open-label single-centre conduct, modest sample size with some baseline imbalances, and susceptibility of several secondary outcomes to performance bias limit confidence in causal interpretation beyond mortality.

External Validity

  • Population representativeness
    • Predominantly surgical ICU cohort (cardiac and abdominal admissions), with very high illness severity and high vasopressor use.
    • Eligibility required NGAL measurement and additional severity enrichment; this phenotype is not routinely defined or operationalised in many ICUs.
    • Exclusion of advanced chronic kidney disease (eGFR <30 mL/min/1.73m2) limits applicability to a common ICU subgroup with AKI-on-CKD.
  • Applicability across systems and settings
    • Intervention used CVVHDF with citrate anticoagulation and specified dose targets; centres with different modalities (e.g. intermittent haemodialysis) or anticoagulation protocols may not replicate exposure and effects.
    • The early strategy implies initiating RRT at KDIGO stage 2 in an enriched high-risk phenotype; this may not be feasible (biomarker availability) or desirable (resource constraints) in many healthcare systems.

Conclusion on External Validity: External validity is limited to moderate: findings apply most directly to severely ill, predominantly surgical ICU patients with KDIGO stage 2 AKI enriched by NGAL and clinical severity criteria, managed with continuous RRT in a high-resource setting.

Strengths & Limitations

  • Strengths:
    • Randomised evaluation of a clinically important, hard primary outcome (90-day mortality) in a high-risk ICU cohort.
    • Clear protocolised contrast in initiation strategy with strong separation in timing (median 6.0 vs 25.5 hours from KDIGO stage 2).
    • Enrichment strategy using NGAL and clinical severity criteria aimed to reduce heterogeneity and increase event rates.
    • Complete inclusion of randomised patients in the full analysis set and minimal missingness for the primary outcome.
  • Limitations:
    • Single-centre, open-label design limits generalisability and increases vulnerability to centre-specific practice effects and performance bias.
    • Modest sample size with a large observed effect size increases risk of overestimation and fragility.
    • Biomarker-driven selection (NGAL threshold) and additional severity criteria constrain applicability and raise questions about transportability to unselected ICU AKI populations.
    • Several secondary outcomes (ventilation duration, hospital discharge) are highly process-dependent and particularly susceptible to unblinded care.

Interpretation & Why It Matters

  • Clinical signal
    • Within this enriched, severely ill cohort, an early RRT strategy at KDIGO stage 2 was associated with lower 90-day mortality and shorter ventilation and hospital stays.
    • Renal recovery differences at day 90 were largely driven by survival differences rather than improved recovery among survivors.
  • Mechanistic inference
    • ELAIN operationalised a hypothesis that earlier solute/fluid control (and potentially immunomodulation) might improve multi-organ outcomes in high-risk AKI.
    • However, causal attribution is complicated by the unblinded single-centre context and subsequent discordant multicentre evidence.
  • Trial design lesson
    • Enrichment using biomarkers and severity criteria can improve event rates and feasibility, but may narrow inference and exacerbate between-trial discrepancies if enrichment identifies a phenotype not shared across studies.

Controversies & Subsequent Evidence

  • Large effect size despite a relatively modest median separation in initiation timing (6.0 vs 25.5 hours from KDIGO stage 2) raised plausibility concerns, particularly for process-dependent secondary outcomes such as hospital length of stay.1
  • Biomarker-driven enrolment (NGAL threshold) and surgical ICU predominance were emphasised as key reasons that findings might not transport to broader ICU AKI populations, especially where NGAL testing is not routine.12
  • Correspondence identified ambiguity in the reporting of adjusted time-to-event modelling and covariate handling; the authors clarified that the primary hazard ratio presented was unadjusted and that the Methods text was corrected post-publication.34
  • Contemporaneous and later multicentre RCTs using different “early/late” definitions often did not reproduce a mortality benefit for early initiation; these trials highlighted that many “delayed/standard” patients may recover without ever requiring RRT.5
  • In septic shock with severe AKI, IDEAL-ICU did not support early RRT and was stopped early for futility, reinforcing uncertainty about benefit outside ELAIN’s enriched phenotype.6
  • STARRT-AKI reported no survival benefit with an accelerated strategy and increased RRT exposure, shaping contemporary scepticism about routine early initiation without conventional indications.7
  • AKIKI-2 explored even more delayed initiation strategies, informing the boundary conditions of “how late is too late” and emphasising that “late” strategies vary materially in trigger thresholds.8
  • Systematic reviews that incorporate ELAIN alongside larger multicentre trials generally report no consistent mortality benefit from early/accelerated initiation, while early strategies reliably increase RRT use; heterogeneity is driven by differing definitions, enrichment, and trigger criteria.91011
  • Guidelines published after ELAIN generally recommend initiating RRT for life-threatening complications and in the context of overall clinical trajectory rather than routinely at earlier KDIGO stages in the absence of indications.1213

Summary

  • ELAIN randomised 231 critically ill adults with enriched-risk KDIGO stage 2 AKI (NGAL >150 ng/mL plus severity criteria) to early (≤8 h) vs delayed (stage 3/absolute indication) initiation of continuous RRT.
  • Early strategy achieved clear timing separation (median 6.0 vs 25.5 hours from KDIGO stage 2) and initiated at lower creatinine and blood urea nitrogen levels.
  • 90-day mortality was lower with early initiation: 39.3% vs 54.7% (HR 0.66; 95% CI 0.45 to 0.97; P=0.03).
  • Renal recovery at day 90 differed when deaths were included (53.6% vs 38.7%; P=0.02) but not among survivors (88.2% vs 85.2%; P=0.62).
  • Subsequent multicentre trials and syntheses have not consistently reproduced a survival benefit from early/accelerated initiation, narrowing ELAIN’s influence to hypothesis generation and phenotype-specific discussions.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • The “delayed” strategy in ELAIN allowed avoidance of RRT (11/119 never initiated RRT), a design feature that complicates comparisons with trials in which nearly all participants ultimately receive RRT.
  • Renal recovery analyses illustrate strong competing-risk structure: between-group differences diminish when restricting to survivors.

Overall Takeaway

ELAIN is a landmark because it tested an unusually early, biomarker- and severity-enriched strategy for initiating continuous RRT at KDIGO stage 2 and reported a mortality signal alongside improved process outcomes. Its single-centre open-label design and later discordant multicentre trial evidence mean its most durable contribution is the demonstration of phenotype enrichment and the framing of “timing” as a bundle of triggers, avoidance, and competing risks rather than a simple early–late dichotomy.

Overall Summary

  • In enriched high-risk KDIGO stage 2 AKI, early RRT was associated with lower 90-day mortality, but subsequent multicentre evidence supports a more conservative, indication-driven approach in most ICU populations.

Bibliography