Skip to main content

Foundational Trials

BOX - Oxygenation

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Oxygen Targets in Comatose Survivors of Cardiac Arrest
  • Acronym: BOX (oxygenation domain)
  • Year: 2022
  • Journal published in: New England Journal of Medicine
  • Citation: Schmidt H, Kjaergaard J, Hassager C, et al. Oxygen Targets in Comatose Survivors of Cardiac Arrest. N Engl J Med. 2022;387(16):1467-1476.

Context & Rationale

  • Background
    • Comatose survivors of out-of-hospital cardiac arrest (OHCA) are routinely mechanically ventilated and often exposed to high inspired oxygen fractions immediately after return of spontaneous circulation (ROSC).
    • Experimental ischaemia–reperfusion models suggested hyperoxia can exacerbate oxidative injury and worsen neurological dysfunction, while hypoxaemia is unequivocally harmful.
    • Observational human studies reported associations between supranormal arterial oxygen tension (high PaO2) and worse outcomes, but causal inference was limited by confounding (severity of illness, post-resuscitation physiology, treatment limitations).
    • Guidelines before BOX generally recommended avoiding both hypoxaemia and hyperoxaemia (often targeting SpO2 94–98%), but the certainty of evidence for an optimal PaO2 target early after arrest was low.
  • Research Question/Hypothesis
    • Whether targeting low-normal PaO2 (restrictive) versus high-normal PaO2 (liberal) in the first ICU phase after OHCA reduces the risk of death or severe neurological disability/coma.
    • Whether a pragmatic oxygenation strategy within “common clinical grounds” can meaningfully change patient-centred neurological outcomes when embedded in contemporary post-cardiac arrest care.
  • Why This Matters
    • Oxygen is universally administered after cardiac arrest, immediately modifiable, and potentially neurotoxic at excess levels; even small effect sizes could have major population impact.
    • Defining safe, evidence-based targets could standardise care, reduce practice variation, and inform guideline thresholds.
    • Prior signals were dominated by observational data; a randomised trial was required to address confounding and treatment indication bias.

Design & Methods

  • Research Question: In comatose adult OHCA survivors, does a restrictive PaO2 target (9–10 kPa) compared with a liberal PaO2 target (13–14 kPa) reduce the composite of death within 90 days or discharge from hospital with severe disability/coma (CPC 3–4)?
  • Study Type: Investigator-initiated, randomised, pragmatic, two-centre (Denmark), 2×2 factorial platform trial (BOX); oxygenation intervention open-label; concurrent blood-pressure intervention double-blind; randomisation stratified by site.
  • Population:
    • Setting: specialised cardiac arrest ICUs in two Danish tertiary centres.
    • Key inclusion: adults (≥18 years) with OHCA of presumed cardiac cause; sustained ROSC; comatose at assessment; intubated/ventilated; screened within 240 minutes of ROSC.
    • Key exclusions (high-level): non-cardiac cause of arrest; regained consciousness during screening; unwitnessed asystole; suspected/confirmed acute intracranial bleeding or stroke; pre-arrest CPC 3–4; treatment limitations/expected survival <180 days; systolic blood pressure <80 mmHg despite support; admission temperature <30°C; pregnancy-related exclusions (female of childbearing potential without negative hCG); logistical constraints.
  • Intervention:
    • Restrictive oxygen target: PaO2 9–10 kPa.
    • Delivery: At ICU admission, initial FiO2 set to 0.30 (unless SpO2 <93%, in which case FiO2 increased); FiO2 subsequently titrated to achieve assigned PaO2 using arterial blood gases.
    • Monitoring: Arterial blood gases at baseline (pre-randomisation) and prespecified timepoints (0, 2, 4, 6, 8, 10, 12, 18, 24, 30, 36, 48 hours), with continued sampling per protocol while arterial catheter in situ (up to 120 hours).
  • Comparison:
    • Liberal oxygen target: PaO2 13–14 kPa.
    • Delivery: At ICU admission, initial FiO2 set to 0.60 (unless SpO2 <93%, in which case FiO2 increased); FiO2 subsequently titrated to achieve assigned PaO2 using arterial blood gases.
    • Co-interventions: Ventilation settings (including PEEP and recruitment manoeuvres), sedation, temperature management, and haemodynamic support were delivered according to local post-arrest care pathways, within the factorial BOX platform framework.
  • Blinding: Oxygenation assignment was not blinded; the co-enrolled blood-pressure target was blinded using calibrated BP modules; blinding of 90-day neurological/cognitive outcome assessors was not reported in the main paper.
  • Statistics: Planned sample size 800; 732 patients required to detect a 10 percentage-point absolute reduction in the primary outcome (assumed 28% to 18%) with 80% power at two-sided alpha 0.05 (final alpha 0.047 after interim monitoring); primary analysis was a modified intention-to-treat (excluding participants who denied/deferred consent) using time-to-event modelling (Cox regression with adjusted hazard ratios).
  • Follow-Up Period: 90 days for clinical/neurological outcomes; 48 hours for neuron-specific enolase; protocolised early physiological sampling to 48 hours (and longer while arterial line present, up to 120 hours).

Key Results

This trial was not stopped early. The oxygenation comparison was reported for 394 patients in the restrictive group and 395 in the liberal group (modified intention-to-treat).

Outcome Restrictive PaO2 (9–10 kPa) Liberal PaO2 (13–14 kPa) Effect p value / 95% CI Notes
Primary composite (death within 90 days OR discharge with CPC 3–4) 126/394 (32.0%) 134/395 (33.9%) aHR 0.95 95% CI 0.75 to 1.21; P=0.69 Time-to-event analysis; event time at death or hospital discharge with CPC 3–4 (whichever occurred first)
Death within 90 days 113/394 (28.7%) 123/395 (31.1%) aHR 0.93 95% CI 0.72 to 1.19; P=Not reported Secondary outcome
Acute kidney injury requiring renal-replacement therapy 35/394 (8.9%) 48/395 (12.2%) aHR 0.72 95% CI 0.49 to 1.05; P=Not reported Secondary outcome; also reported among adverse events
CPC at 90 days (median, IQR) 1 (1–4) 1 (1–4) Not reported Not reported Secondary neurological outcome (ordinal)
Modified Rankin Scale at 90 days (median, IQR) 1 (0–4) 1 (0–4) Not reported Not reported Secondary neurological outcome (ordinal)
Montreal Cognitive Assessment at 90 days (median, IQR; assessed survivors) 28 (24–30) 27 (24–29) Not reported Not reported Cognitive testing among available survivors; missingness non-trivial
Neuron-specific enolase at 48 hours (ng/mL; median, IQR; alive at 48 hours) 17 (11–36) 18 (11–34) Not reported Not reported Biomarker of neuronal injury; secondary
Any bleeding (adverse event) 64/394 (16.2%) 41/395 (10.4%) RR 1.56 95% CI 1.07 to 2.26; P=0.02 Safety signal; multiplicity considerations apply
Uncontrolled bleeding (adverse event) 23/394 (5.8%) 11/395 (2.8%) RR 2.10 95% CI 1.05 to 4.22; P=0.04 Defined by treating team; ascertainment potentially influenced by open-label care
Arrhythmia (adverse event) 182/394 (46.2%) 194/395 (49.1%) RR 0.94 95% CI 0.80 to 1.11; P=0.47 No signal of harm/benefit
Infection (adverse event) 133/394 (33.8%) 127/395 (32.2%) RR 1.05 95% CI 0.86 to 1.28; P=0.65 No signal of harm/benefit
  • The restrictive and liberal PaO2 targets produced similar rates of the primary composite outcome (32.0% vs 33.9%) with no evidence of benefit (aHR 0.95; 95% CI 0.75 to 1.21; P=0.69).
  • Across secondary clinical outcomes (mortality, neurological scales, cognition, NSE), no clinically important between-group differences were reported.
  • A higher frequency of bleeding events was observed in the restrictive group (any bleeding 16.2% vs 10.4%; RR 1.56; 95% CI 1.07 to 2.26; P=0.02), warranting cautious interpretation given multiple safety comparisons.

Internal Validity

  • Randomisation and allocation
    • Web-based allocation with variable block sizes (2, 4, 6) and stratification by site; allocation concealment before randomisation was appropriate for selection-bias control.
    • 2×2 factorial design; oxygen target open-label; blood-pressure target blinded (reducing co-intervention bias in that domain).
    Drop-out / exclusions and follow-up completeness
    • Randomised: 802 patients; one patient was erroneously randomised twice (flow correction).
    • Modified intention-to-treat for oxygenation: 394 restrictive and 395 liberal, after exclusions for consent denial/deferment (6 per group).
    • Loss to follow-up was minimal: two participants transferred outside Denmark; outcomes censored on days 12 and 13.
    Performance / detection bias
    • Open-label oxygenation introduces potential for differential co-interventions (ventilator weaning, transfusion thresholds, anticoagulation decisions, timing of extubation) and for clinician-influenced outcomes (e.g., “uncontrolled bleeding”).
    • The primary outcome combined death (objective) with CPC at discharge (clinically assessed; potentially sensitive to discharge practices), but within a single health system and limited sites, practice variation may be constrained.
    Protocol adherence and separation of the variable of interest
    • Assigned strategies were initiated at ICU admission using distinct starting FiO2 (0.30 restrictive vs 0.60 liberal), then titrated by arterial blood gases.
    • Physiological separation was demonstrated in arterial oxygen saturation: mean difference (restrictive minus liberal) at 24 hours −2.27185% (95% CI −2.76441 to −1.77929); at 36 hours −2.73074% (95% CI −3.18019 to −2.28128); at 48 hours −0.84024% (95% CI −1.55180 to −0.12867).
    • Ventilator setting differences were small: mean PEEP difference at 24 hours −0.5151 cmH2O (95% CI −0.8254 to −0.2049); at 36 hours −0.5220 cmH2O (95% CI −0.8351 to −0.2090); at 48 hours −0.0487 cmH2O (95% CI −0.4720 to 0.3746).
    Baseline characteristics and illness severity
    • Groups were well balanced on key prognostic variables: age 62±13 (restrictive) vs 63±14 (liberal); male sex 325/394 (82.5%) vs 312/395 (79.0%).
    • Cardiac arrest phenotype was typical of specialised OHCA ICUs: shockable rhythm 335/394 (85.0%) vs 334/395 (84.5%); STEMI 166/394 (42.1%) vs 158/395 (40.0%); time to ROSC 21±13 vs 21±14 minutes; bystander CPR 351/394 (89.2%) vs 339/395 (85.8%).
    Timing and “dose” of intervention
    • Randomisation occurred early: median time from arrest to randomisation 146 minutes (IQR 123–189) vs 146 (IQR 122–191).
    • Both targets were within low-to-high normal PaO2 ranges, meaning the tested “dose” was a comparison of two normoxaemic strategies rather than avoidance of extreme hyperoxia.
    Adjunctive therapy use
    • Use of vasoactive and sedative agents was broadly similar (e.g., norepinephrine 367/395 [93%] vs 359/394 [91%]; propofol 365/394 [92.6%] vs 365/395 [92.4%]).
    Statistical rigour
    • Primary effect reported as adjusted hazard ratio with confidence intervals; no evidence of interaction by pre-specified subgroups (including blood-pressure allocation) was evident in reported subgroup estimates.

Conclusion on Internal Validity: Overall, internal validity appears moderate-to-strong for estimating the effect of comparing two normoxaemic PaO2 targets in this setting, given robust randomisation, near-complete follow-up, and demonstrated physiological separation; limitations are primarily the open-label oxygenation intervention and the reliance on discharge-based CPC within the composite primary endpoint.

External Validity

  • Population representativeness
    • Participants were predominantly OHCA of presumed cardiac cause with high rates of witnessed events/bystander CPR and shockable rhythms, managed in high-resource cardiac arrest centres.
    • Results may not extrapolate to in-hospital cardiac arrest, arrests of non-cardiac aetiology (e.g., asphyxial), prolonged no-flow/low-flow physiology, or patients with severe acute respiratory failure driving oxygen requirements.
    Applicability
    • Intervention is feasible in ICUs with access to arterial blood gas monitoring and protocolised ventilation/oxygen titration.
    • Practice environments with different thresholds for withdrawal of life-sustaining therapy, discharge, or neurological rehabilitation pathways may observe different behaviour of the discharge-based CPC component.

Conclusion on External Validity: Generalisability is moderate to similar tertiary-centre, ventilated OHCA populations; it is more limited for non-cardiac arrest phenotypes and settings without routine ABG-guided oxygen titration.

Strengths & Limitations

  • Strengths:
    • Randomised, pragmatic evaluation of a ubiquitous therapy embedded within a factorial platform trial.
    • Early randomisation (median ~2.4 hours post-arrest), targeting a mechanistically plausible window for reperfusion injury modulation.
    • High follow-up completeness with clinically meaningful neurological and cognitive endpoints at 90 days.
    • Demonstrated separation in oxygenation-related physiological measures across the early ICU course.
  • Limitations:
    • Open-label oxygenation assignment increases susceptibility to differential co-interventions and ascertainment bias for some adverse events.
    • Only two centres in one country; practice patterns and patient mix may not reflect all systems of care.
    • Both strategies targeted normoxaemia; findings do not directly answer whether avoiding early extreme hyperoxia (very high PaO2) improves outcomes.
    • Composite primary endpoint includes discharge CPC, which can be influenced by local discharge and rehabilitation pathways.
    • Cognitive testing (MoCA) available only in a subset of survivors, raising potential for informative missingness.

Interpretation & Why It Matters

  • Clinical implications
    • Among ventilated, comatose OHCA survivors managed in specialised ICUs, targeting PaO2 9–10 kPa did not improve survival or neurological recovery compared with targeting 13–14 kPa.
    • The results support the view that, within normoxaemic ranges, fine-tuning PaO2 targets alone is unlikely to be a dominant determinant of outcome relative to other post-arrest care components (haemodynamics, temperature management, reperfusion strategy, prognostication practices).
    • The bleeding signal cautions against assuming “lower is always safer” and reinforces the need to assess unintended consequences when modifying ubiquitous supportive therapies.

Controversies & Subsequent Evidence

  • Magnitude of physiological contrast and the “tested question”
    • The trial intentionally compared two targets that sit within guideline-consistent normoxaemia; this enhances acceptability and safety but narrows the mechanistic “dose” of oxygen exposure being tested, which may partly explain neutral clinical results.1
    Power assumptions versus plausible effect sizes
    • Sample size planning targeted a relatively large absolute reduction (10 percentage points). If the true effect of small PaO2 shifts within normoxaemia is smaller, the study would be expected to produce wide confidence intervals spanning clinically meaningful benefit and harm.
    Safety signal: bleeding
    • Bleeding events were more frequent in the restrictive group (any bleeding RR 1.56; uncontrolled bleeding RR 2.10), raising hypotheses about chance findings, ascertainment in an open-label trial, or biologically mediated effects; the study was not designed to adjudicate bleeding mechanisms.
    Evidence synthesis after BOX
    • A recent systematic review/meta-analysis focused on restrictive oxygenation targets after OHCA, incorporating BOX among available randomised data, reported no clear improvement in patient-centred outcomes with restrictive targets, supporting a “avoid extremes” approach rather than endorsing a single narrow PaO2 target.2
    • Broader meta-analysis across acutely ill adults reported increased mortality with liberal oxygen therapy, reinforcing the general principle that unnecessary hyperoxia is not benign (though indirect to the post-arrest population).3
    Guideline positioning
    • Contemporary post-resuscitation guidance has historically recommended avoiding hypoxaemia and avoiding routine high oxygen exposure after ROSC (often targeting SpO2 94–98%), and BOX provides randomised evidence that two normoxaemic PaO2 targets yield similar outcomes in this context.4

Summary

  • BOX Oxygenation randomised comatose OHCA survivors to PaO2 9–10 kPa (restrictive) versus 13–14 kPa (liberal) during early ICU care.
  • No difference was observed in the primary composite of death within 90 days or discharge with CPC 3–4 (32.0% vs 33.9%; aHR 0.95; 95% CI 0.75 to 1.21; P=0.69).
  • Secondary outcomes (90-day mortality, neurological scales, MoCA, NSE) were similar between groups.
  • Physiological separation was achieved (e.g., SaO2 mean difference at 24 hours −2.27% in restrictive vs liberal), indicating the intervention produced measurable differences in oxygenation exposure.
  • Bleeding events were more frequent in the restrictive group (any bleeding 16.2% vs 10.4%; RR 1.56; 95% CI 1.07 to 2.26), a key safety observation requiring cautious interpretation.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • Where DOI links were not available from the trial documents, PubMed search links are provided to maintain verifiability without inventing identifiers.
  • BOX Oxygenation addresses a narrow but pragmatic comparison (two normoxaemic PaO2 targets); ongoing work (e.g., Mega-ROX) was highlighted as likely to better address broader oxygenation strategies in critically ill populations including hypoxic–ischaemic brain injury after arrest.1

Overall Takeaway

BOX Oxygenation is a landmark post-cardiac arrest trial because it randomised a ubiquitous ICU therapy (oxygen titration) to two explicit PaO2 targets early after OHCA and found no difference in survival or neurological recovery within 90 days. Its results imply that, within normoxaemic ranges commonly used in practice, the precise PaO2 target is unlikely to be a major driver of outcome, and future trials may need greater physiological separation and/or different populations to identify clinically meaningful oxygen effects.

Overall Summary

  • In ventilated, comatose OHCA survivors, PaO2 9–10 kPa versus 13–14 kPa produced similar 90-day patient-centred outcomes, with a notable bleeding safety signal in the restrictive group.

Bibliography