Skip to main content

Foundational Trials

EPaNIC

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Early versus Late Parenteral Nutrition in Critically Ill Adults
  • Acronym: EPaNIC (Early Parenteral Nutrition Completing Enteral Nutrition in Adult Critically Ill Patients)
  • Year: 2011
  • Journal published in: The New England Journal of Medicine
  • Citation: Casaer MP, Mesotten D, Hermans G, Wouters PJ, Schetz M, Meyfroidt G, et al. Early versus late parenteral nutrition in critically ill adults. N Engl J Med. 2011;365(6):506-17.

Context & Rationale

  • Background
    Early enteral nutrition (EN) was advocated in critical illness, but real-world delivery often fell short of caloric targets, prompting common “top-up” parenteral nutrition (PN) practices.
    • Prior international guidance diverged on when to initiate PN (early supplementation versus deferral), reflecting equipoise and limited randomised data.
    • Biological plausibility existed for both strategies: preventing cumulative energy/protein deficits versus avoiding early overfeeding, hyperglycaemia, infectious risk, and metabolic/liver complications.
  • Research Question/Hypothesis
    In adult ICU patients at nutritional risk, does early caloric completion with PN (from ICU day 3) compared with withholding PN until ICU day 8 (if EN remains insufficient) improve recovery (time to ICU discharge alive) and reduce morbidity without increasing mortality?
  • Why This Matters
    Nutrition timing is a high-volume, high-cost ICU intervention with potential to influence infections, organ failure recovery, length of stay, and downstream rehabilitation; EPaNIC directly tested a widely used strategy (early supplemental PN) against a deliberately “permissive underfeeding” approach during the first ICU week.

Design & Methods

  • Research Question: Whether late initiation of PN (day 8 if EN insufficient) versus early initiation of PN (day 3 to meet calculated caloric goal) shortens time to ICU discharge alive and reduces ICU morbidity in nutritionally at-risk adult ICU patients.
  • Study Type: Randomised, multicentre (7 ICUs in Belgium), investigator-initiated, open-label, parallel-group trial with stratified randomisation; ICU setting.
  • Population:
    • Adults admitted to participating ICUs, nutritionally at risk (NRS 2002 score ≥3) on ICU admission; central venous access required (patients without a central catheter excluded).
    • Key exclusions included: age <18 years; DNR/moribund; BMI <17 kg/m2; transfer from another ICU with an established nutrition regimen; short bowel syndrome; home ventilation; pregnancy/lactation; ketoacidotic/hyperosmolar coma; not critically ill (no indication for central line / ready for oral nutrition on admission).
    • Randomisation stratified by 16 diagnostic categories; allocation by sequentially numbered sealed opaque envelopes, later replaced by an identical digital system at added sites.
  • Intervention:
    • Late initiation strategy (“late PN”): 5% glucose IV (hydration; volume matched to theoretical PN volume) during ICU days 1–7; EN attempted in both groups; if EN remained insufficient after 7 ICU days, PN commenced on ICU day 8 to reach calculated caloric goal.
    • Micronutrients (trace elements/minerals/vitamins) were provided early in ICU stay per protocol to mitigate refeeding-related depletion.
  • Comparison:
    • Early initiation strategy (“early PN”): 20% glucose IV (targets: 400 kcal on ICU day 1; 800 kcal on day 2); PN initiated on ICU day 3 and titrated daily to achieve 100% of calculated caloric goal via combined EN+PN (with reduction/cessation when EN/oral intake covered ≥80% of goal).
    • PN products were standardised commercial 3-in-1 mixtures (OliClinomel or Clinimix); maximum caloric goal capped at 2880 kcal/day.
  • Blinding: Open-label to treating clinicians/patients; outcome adjudicators were unaware of study-group assignment; prespecified risk-factor adjusted analyses were planned.
  • Statistics: Sample size 4640 (2320/arm) to detect a between-group change of 1 ICU day with ≥80% power (two-sided α=0.05) and to detect a 3% change in ICU mortality with ≥70% power; analyses were intention-to-treat; Cox models for time-to-event outcomes with prespecified adjusted and unadjusted analyses.1
  • Follow-Up Period: ICU and hospital outcomes to discharge; vital status assessed to 90 days; functional outcomes assessed at hospital discharge (where reported).

Key Results

This trial was not stopped early. One interim safety analysis was performed after the first 1500 patients had been discharged from ICU; the DSMB advised continuation.

Outcome Late initiation (withhold PN to day 8)
(N=2328)		 Early initiation (PN from day 3)
(N=2312)		 Effect p value / 95% CI Notes
Time to discharge alive from ICU (primary) Median ICU stay 3 (2–7) days Median ICU stay 4 (2–9) days HR 1.06 95% CI 1.00 to 1.13; P=0.04 Cox model for time to ICU discharge alive; deaths censored after last surviving discharge (per protocol).
Discharged alive from ICU within 8 days 1750/2328 (75.2%) 1658/2312 (71.7%) OR 1.271 95% CI 1.080 to 1.495; P=0.007 Prespecified dichotomised “≤8 days” endpoint; OR from online supplement.
ICU mortality 141/2328 (6.1%) 146/2312 (6.3%) Not reported P=0.76 No mortality difference despite shorter ICU stay metrics.
Hospital mortality 242/2328 (10.4%) 251/2312 (10.9%) Not reported P=0.63 Unadjusted comparison reported in primary manuscript.
90-day mortality 257/2289 (11.2%) 255/2268 (11.2%) Not reported P=1.00 Some non-Belgian survivors discharged before day 90 were censored at hospital discharge in survival analyses (per protocol).
Any new infection 531/2328 (22.8%) 605/2312 (26.2%) Not reported P=0.008 Composite of adjudicated new infections during ICU stay.
Airway or lung infection 381/2328 (16.4%) 447/2312 (19.3%) Not reported P=0.009 Most frequent infection category; adjudicated outcome.
Bloodstream infection 142/2328 (6.1%) 174/2312 (7.5%) Not reported P=0.05 Borderline statistical significance; site-specific attribution not shown here.
Wound infection 64/2328 (2.7%) 98/2312 (4.2%) Not reported P=0.006 Consistent with lower infection burden in late-PN arm.
Hypoglycaemia during intervention (glycaemia <40 mg/dL) 81/2328 (3.5%) 45/2312 (1.9%) Not reported P=0.001 Occurred despite less insulin requirement in late-PN arm; reflects intensive insulin protocol context.
Mechanical ventilation duration Median 2 (1–5) days Median 2 (1–5) days Not reported P=0.02 Distributional difference with same medians; weaning HR reported separately (not significant at P=0.07).
Duration of renal replacement therapy (RRT recipients) Median 7 (3–16) days Median 10 (4–19) days Not reported P=0.008 Any RRT use: 8.6% vs 8.9% (P=0.77); duration differs among recipients.
Hospital stay (overall) Median 14 (8–28) days Median 16 (9–30) days Not reported P=0.004 Time-to-discharge-alive-from-hospital also favoured late PN (HR 1.064; 95% CI 1.001 to 1.131; P=0.04).
Total incremental healthcare costs Mean €16,863 (IQR 11,184–29,887) Mean €17,973 (IQR 11,617–34,568) Not reported P=0.04 Cost accounting context is Belgium-specific; PN acquisition costs were not directly deducted under reimbursement rules (see supplement).
Post-hoc subgroup: surgical contraindication to EN (N=517) — time to ICU discharge alive ICU stay 6 (2–16) days ICU stay 7 (3–19) days HR 1.198 95% CI 0.999 to 1.437; Pinteraction=0.1197 Patients predictably received virtually no EN by day 7 (0 [0–163] kcal/day); safety endpoint “alive ICU discharge within 8 days” OR 1.749 (95% CI 1.141 to 2.683).
    • Late initiation was associated with faster discharge alive from ICU (median 3 [2–7] vs 4 [2–9] days; HR 1.06; 95% CI 1.00 to 1.13; P=0.04) and fewer new infections (22.8% vs 26.2%; P=0.008).
    • Mortality was similar (ICU 6.1% vs 6.3%; hospital 10.4% vs 10.9%; 90-day 11.2% vs 11.2%), supporting a morbidity/efficiency signal rather than survival benefit.
    • Late initiation had more hypoglycaemia (3.5% vs 1.9%; P=0.001) in the context of tight glycaemic control, underscoring context dependence of harms/benefits.

Internal Validity

    • Randomisation and allocation: Stratified randomisation across 16 admission diagnostic strata; sequential sealed opaque envelopes (later digital system) with permuted blocks of 10; clinicians/nurses were unaware of block size (reducing selection bias).
    • Dropout/exclusions: All randomised patients were included in the intention-to-treat analysis; protocol violation occurred in 15 late-initiation patients (inadvertent PN ≥1 L/day for ≥2 days during intervention window), but retained in analysis.
    • Performance/detection bias: Open-label feeding strategy creates potential for co-intervention and discharge decision bias; mitigated by blinded outcome adjudicators and protocolised ICU care elements (weaning guidance, EN escalation, insulin infusion target).
    • Protocol adherence and separation: Clear temporal separation (early PN from ICU day 3 vs no PN until ICU day 8 if EN insufficient), with matched-volume IV glucose in the late group; a PDMS calculated daily volumes to operationalise separation.
    • Separation of metabolic exposure (objective markers): Late-initiation required less insulin (median 31 IU/day [IQR 19–48] vs 58 IU/day [IQR 33–95]) and had lower mean blood glucose (102±14 vs 107±18 mg/dL; P<0.001), yet more hypoglycaemia episodes (3.5% vs 1.9%).
    • Baseline comparability: Groups were reported as well matched on baseline demographic and illness severity variables; large sample size reduces chance imbalance.
    • Heterogeneity: Broad mixed ICU population with high representation of post-cardiac surgery; subgroup interactions for prespecified higher-risk strata (NRS ≥5; BMI extremes; sepsis; cardiac surgery) were not statistically persuasive (interaction testing at P<0.10 threshold).
    • Timing and dose: Intervention timing was early by design (glucose days 1–2; PN from day 3 in early arm); caloric goals were equation-based with a maximum cap (2880 kcal/day) and incorporated protein energy, raising risk of early full feeding in some phenotypes.
    • Outcome assessment: Primary endpoint used time-to-discharge-alive metrics with defined censoring rules; however, discharge readiness criteria included nutritional intake thresholds (see Controversies), making the primary endpoint not fully independent of the intervention.
    • Statistical rigour: Prespecified ITT analyses, time-to-event modelling, and adjusted/unadjusted comparisons were described; multiple secondary endpoints were tested without multiplicity correction (interpretation should prioritise consistency and biological plausibility over isolated P values).

Conclusion on Internal Validity: Overall, internal validity is moderate-to-strong: randomisation and allocation concealment were robust and separation was substantial, but the open-label design and discharge-related endpoints introduce plausible performance and endpoint-dependence bias that particularly affects length-of-stay interpretations.

External Validity

    • Population representativeness: Predominantly high-income, well-resourced Belgian ICUs with a large cardiac surgery cohort; excluded severely malnourished patients (BMI <17) and those without central access or already on oral nutrition at admission.
    • Care context: Delivered within an intensive insulin therapy paradigm (target 80–110 mg/dL) and structured EN escalation; these may differ from contemporary glycaemic targets and feeding practices in many ICUs.
    • Intervention feasibility: Strategy of deliberately delaying PN until day 8 is operationally feasible in most ICUs; however, the balance of benefit versus harm may shift in populations with higher baseline malnutrition, prolonged EN intolerance, or different metabolic management.
    • Health system transferability: Cost findings are sensitive to national reimbursement and costing methods; clinical endpoints are more transferable than economic endpoints.

Conclusion on External Validity: Generalisability is moderate: results are most applicable to well-nourished or mildly at-risk mixed ICU cohorts in high-resource settings, and less certain for markedly malnourished patients, settings with different glycaemic strategies, or contexts where ICU length of stay is driven by non-clinical constraints.

Strengths & Limitations

  • Strengths:
    • Large sample size (N=4640) with stratified randomisation and high protocol operationalisation via PDMS-enabled daily dosing.
    • Clinically relevant endpoints spanning infections, organ support, functional outcomes, and costs; blinded outcome adjudication.
    • Pragmatic, implementable strategies reflecting common ICU feeding dilemmas (early “completion” PN versus delayed PN).
  • Limitations:
    • Open-label design with endpoints (ICU/hospital discharge readiness) potentially influenced by nutrition strategy and clinician behaviour.
    • High proportion of post-cardiac surgery and low-to-moderate nutritional risk patients may limit inference for malnourished or prolonged-stay phenotypes.
    • Tight glycaemic control and equation-based caloric targets represent a specific care context; macronutrient composition and protein dosing may not reflect current best practice in all settings.
    • Multiple secondary endpoints without multiplicity adjustment increase risk of overinterpreting isolated statistically significant findings.

Interpretation & Why It Matters

  • Practice-changing inference
    In a large, nutritionally at-risk but predominantly non-malnourished adult ICU cohort, a strategy of deferring PN until ICU day 8 (while pursuing EN and providing micronutrients) was associated with fewer infections and faster recovery markers without a mortality penalty.
    • EPaNIC shifted the “default” away from routine early supplemental PN in the first ICU week, reframing early caloric debt as potentially adaptive rather than uniformly harmful.
    • It sharpened the question from “EN vs PN” to “which patients, when, and how much macronutrient”, anticipating later phenotype-based and indirect calorimetry-guided approaches.
    • Its strongest contemporary relevance is in low nutritional risk patients where early full feeding may represent overfeeding; its applicability is less certain in profound malnutrition or prolonged EN intolerance.

Controversies & Subsequent Evidence

    • Endpoint dependence and discharge criteria: Editorial and correspondence highlighted that “readiness for ICU discharge” incorporated nutritional intake (ability to receive ≥2/3 caloric requirements orally), which may be influenced by early PN via appetite suppression, making the primary endpoint not fully orthogonal to the intervention.234
    • Care-context sensitivity (glycaemia and “full feeding”): Commentaries emphasised that intensive insulin therapy (80–110 mg/dL target) and early delivery of near-goal calories could potentiate harms via hypoglycaemia and overfeeding, raising uncertainty about translation to contemporary glycaemic targets and more conservative early caloric strategies.24
    • Population phenotype (nutritional risk versus nutritional reserve): Post-hoc analysis explored whether disease type or macronutrient dose modified effects; the overarching signal remained that early completion feeding was not broadly beneficial, supporting the interpretation that patient selection (and baseline reserve) matters more than universal timing rules.5
    • Mechanistic debate (autophagy and muscle outcomes): The trialists proposed suppression of autophagy as a mechanistic explanation; a prespecified neuromuscular subanalysis found late PN was associated with fewer patients developing ICU-acquired weakness at the first evaluation (absolute difference 8.0%; 95% CI 4.1 to 11.9; P<0.0001).6
    • Health-economic implications: A formal cost analysis reported higher mean healthcare costs per patient with early PN (Crit Care 2012;16:R96), supporting resource stewardship arguments when clinical benefit is absent.7
    • Contrasting RCT signals with different targeting strategies:
      • Heidegger et al used indirect calorimetry-guided targets and began supplemental PN on ICU day 4 in patients not meeting energy goals; nosocomial infection was reduced (Lancet 2013;381:385–393), illustrating that timing effects may interact with patient selection and energy targeting strategy.8
      • Doig et al studied patients with short-term relative contraindications to early EN and found early PN increased energy/protein delivery without clear harm (JAMA 2013;309:2130–2138), supporting context-specific use when EN is not feasible.9
      • CALORIES (NEJM 2014;371:1673–1684) found no significant difference in 30-day mortality between early PN and early EN routes, suggesting route alone is less decisive than dose/timing/phenotype.10
      • NUTRIREA-2 (Lancet; pages 133–143, January 2018) found similar 28-day mortality with early EN versus early PN in ventilated adults with shock, but more gastrointestinal complications with EN (e.g., bowel ischaemia 19 [2%] vs 5 [1%]; P=0.007; colonic pseudo-obstruction 11 [1%] vs 1 [<1%]; P=0.009), complicating “EN-first” dogma in shock states.11
      • PEPaNIC (NEJM 2016;374:1111–1122) extended the “withhold PN in week 1” concept to paediatric ICU, reporting clinical superiority of late PN in children, reinforcing a generalisable biological hypothesis while still leaving adult malnutrition phenotypes unresolved.12
    • Meta-analytic synthesis: Subsequent syntheses generally report no mortality difference between EN and PN routes but fewer bloodstream infections with EN; however, heterogeneity in dose, timing, and glycaemic strategies limits “one-size” conclusions.131415
    • Guideline evolution: Post-EPaNIC guidance increasingly separates (i) “PN when EN is contraindicated” from (ii) “supplemental PN to meet targets”, with timing influenced by nutritional risk (e.g., SCCM/ASPEN 2016 recommend withholding PN for the first 7 days in low-risk patients; ESPEN 2019 suggests starting PN within 3–7 days when EN is contraindicated).1617181920

Summary

    • In 4640 nutritionally at-risk adult ICU patients, delaying PN until ICU day 8 (if EN insufficient) was associated with faster discharge alive from ICU (median 3 [2–7] vs 4 [2–9] days; HR 1.06; 95% CI 1.00 to 1.13; P=0.04).
    • Late PN reduced new infections (22.8% vs 26.2%; P=0.008) without altering ICU, hospital, or 90-day mortality.
    • Late PN had more hypoglycaemia (3.5% vs 1.9%; P=0.001) despite lower insulin doses and lower mean glucose, reflecting the tight glycaemic-control environment.
    • Economic analysis found higher mean healthcare costs with early PN, reinforcing a stewardship argument when clinical benefits are absent.
    • Interpretation is tempered by open-label design and discharge-readiness criteria that included nutritional intake, potentially inflating length-of-stay effects.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • EPaNIC tested timing of macronutrient PN completion (with early provision of micronutrients in both groups) in a tightly protocolised metabolic care environment; translation to modern practice should explicitly consider nutritional risk, expected ICU duration, and contemporary glycaemic targets.

Overall Takeaway

EPaNIC is a landmark because it challenged the prevailing reflex to “close the calorie gap” early in critical illness, showing that deferring PN until the second ICU week can be associated with fewer infections and faster recovery markers without an evident mortality trade-off. Its enduring contribution is conceptual: early full feeding may be harmful in some ICU phenotypes, and nutrition strategy should be individualised by nutritional reserve, expected duration of EN failure, and metabolic context.

Overall Summary

  • In a large adult ICU cohort at nutritional risk, delaying PN until day 8 (if EN insufficient) improved discharge and infection outcomes without changing mortality.
  • Interpretation of length-of-stay endpoints requires caution due to open-label design and discharge-readiness criteria incorporating nutritional intake.
  • Subsequent trials and guidelines support phenotype- and context-dependent PN timing rather than universal early supplementation.

Bibliography