Publication

• Title: Oxygen-saturation targets for critically ill adults receiving mechanical ventilation
• Acronym: PILOT
• Year: 2022
• Journal published in: New England Journal of Medicine
• Citation: Semler MW, Casey JD, Lloyd BD, Hastings PG, Hays MA, Stollings JL, et al. Oxygen-saturation targets for critically ill adults receiving mechanical ventilation. N Engl J Med. 2022;387:1759-1769.

Context & Rationale

• Background

  • Supplemental oxygen is near-universal in ICU care, but the “dose” of oxygen (FiO₂/PaO₂/SpO₂ exposure) is rarely treated as a titratable drug with defined therapeutic windows.
  • Hypoxaemia is an obvious proximate threat, yet hyperoxaemia is biologically plausible as harmful through oxidative injury, vasoconstriction, and absorption atelectasis; observational ICU data have repeatedly shown associations between higher PaO₂ exposure and mortality, with important residual confounding risk. ³
  • Prior randomised evidence was heterogeneous in setting and signal: single-centre conservative strategies suggested benefit in some contexts (e.g., Oxygen-ICU), while large multicentre trials showed neutral average effects (ICU-ROX; HOT-ICU); very conservative PaO₂ targets in ARDS raised safety concerns (LOCO2), and deliberate hyperoxia in septic shock suggested harm (HYPERS2S). ⁴⁵⁶⁷⁸
  • There remained a specific pragmatic gap: most bedside oxygen titration in ventilated patients is performed using pulse oximetry, yet robust comparative-effectiveness evidence for routine SpO₂ target ranges in an undifferentiated mechanically ventilated ICU population was limited.
• Research Question/Hypothesis

  • In invasively ventilated critically ill adults, do different SpO₂ target strategies (lower 88–92%; intermediate 92–96%; higher 96–100%) change ventilator-free days (VFDs) at day 28 and other patient-centred outcomes?
  • Hypothesis (implicit): a more conservative oxygenation strategy might improve outcomes by reducing hyperoxia-related harm, balanced against potential harm from increased hypoxaemic exposure.
• Why This Matters

  • Oxygen targeting is ubiquitous, low-cost, and system-level; even modest true effects could translate into large absolute population benefit or harm.
  • SpO₂ targets are readily implementable via alarms and workflows, but require evidence on safety and effectiveness in “real-world” mechanically ventilated patients.
  • A pragmatic design testing oxygen targets as delivered in routine care could clarify whether clinicians should deliberately tolerate lower saturations or avoid liberal oxygenation in ventilated ICU populations. ¹

Design & Methods

• Research Question: Among adults receiving invasive mechanical ventilation in the ED/medical ICU, what is the effect of three SpO₂ target strategies (88–92%, 92–96%, 96–100%) on ventilator-free days at day 28?
• Study Type: Pragmatic, single-centre, cluster-crossover randomised trial in an academic ED and medical ICU; 18 two-month periods with randomised assignment to one of three oxygen-saturation target strategies; prespecified analytic washout at the end of each period; open-label delivery of the intervention. ¹
• Population:
  • Setting: Emergency Department and medical ICU at Vanderbilt University Medical Center (academic centre).
  • Inclusion: Adults ≥18 years; first receipt of invasive mechanical ventilation in the ED or medical ICU (or in the ED with planned admission to the medical ICU).
  • Exclusion: Pregnancy; incarceration.
  • Enrolment approach: Pragmatic “all-eligible” enrolment aligned to unit-period assignment (cluster-crossover), minimising bedside selection.
• Intervention:
  • Lower target strategy: SpO₂ goal range 88–92% (target 90%); oxygen titration (primarily FiO₂) to maintain SpO₂ within range; pulse oximeter alarms set to match range.
  • Fallback when pulse oximetry unavailable/inaccurate: PaO₂ target 60 mm Hg.
  • Individual-level flexibility: Target could be modified at clinician discretion or patient/family request and documented.
• Comparison:
  • Intermediate target strategy: SpO₂ goal range 92–96% (target 94%); PaO₂ target 70 mm Hg if needed.
  • Higher target strategy: SpO₂ goal range 96–100% (target 98%); PaO₂ target 110 mm Hg if needed.
  • Co-interventions: Otherwise usual care for ventilation and ICU management; oxygen titration was the variable of interest.
• Blinding: Open-label (clinicians and bedside teams necessarily aware of the assigned SpO₂ target); key outcomes were largely objective and derived from the electronic health record; manual adjudication of prespecified safety events was performed in a subset with adjudicators blinded to assignment. ²
• Statistics: Planned sample size 2250 to provide 92% power (two-sided α=0.05; β=0.08) to detect an absolute difference of 2 ventilator-free days between any two groups; primary outcome analysed with a proportional-odds model including covariates of group assignment and time (study day) modelled with restricted cubic splines; pairwise comparisons reported with 95% CIs; prespecified interim analyses with Haybittle–Peto boundary (P<0.001) for efficacy. ¹
• Follow-Up Period: Outcomes assessed through day 28 (or hospital discharge) including ventilator-free days and in-hospital mortality before day 28.

Key Results

This trial was not stopped early. Two interim analyses were planned; no prespecified stopping boundary was met, and enrolment continued to trial completion.

• Despite clear stepwise differences in oxygen exposure (FiO₂ and frequency of SpO₂ 99–100%), there was no signal of benefit for lower or higher SpO₂ targets on ventilator-free days (overall P=0.81) or short-term mortality.
• Event rates for kidney outcomes were similar; a numerical trend toward fewer patients receiving RRT in the higher-target group (11.6%) vs lower/intermediate (14.8%/14.5%) did not exclude no effect (95% CIs crossed 1.0).
• Prespecified subgroup analyses did not demonstrate convincing heterogeneity of effect across major diagnostic strata (e.g., cardiac arrest, sepsis/septic shock, ARDS), though interaction testing is intrinsically underpowered in a three-arm pragmatic trial.

Internal Validity

• Randomisation and allocation: Cluster-crossover randomisation of 18 two-month periods to one of three SpO₂ target strategies; allocation concealment to clinicians was not feasible, but enrolment of all eligible mechanically ventilated adults in the ED/medical ICU reduced bedside selection bias.
• Drop out / exclusions: 2987 patients were enrolled; the prespecified primary analysis excluded 347 enrolled during washout (110 lower, 113 intermediate, 124 higher) and 99 patients with COVID-19 (33 lower, 36 intermediate, 30 higher), leaving 2541 analysed; sensitivity analyses including these patients were reported as consistent with the primary findings. ²
• Performance / detection bias: The intervention could not be blinded; clinician behaviour (e.g., oxygen titration intensity, extubation practices) might theoretically shift with perceived risk, but the primary outcome (VFDs) and mortality are relatively objective and were extracted from the EHR.
• Protocol adherence: Individual-level modification of the assigned SpO₂ target was uncommon: 30/808 (3.7%) in the lower-target periods, 21/859 (2.4%) in intermediate, and 60/874 (6.9%) in higher; modification occurred early (median day 1.5–1.6). ²
• Baseline characteristics: Groups were broadly comparable (e.g., median age 57–59 years; non-respiratory SOFA median 5 across groups; similar proportions with sepsis/septic shock, cardiac arrest, acute myocardial infarction, and baseline stage ≥II AKI), supporting balance at enrolment in this period-randomised design.
• Timing: Enrolment occurred at first receipt of invasive mechanical ventilation; median time from initiation of ventilation to enrolment was 0.0 hours (IQR approximately 0.0–5.5 hours across groups), maximising exposure to the assigned strategy during early critical illness.
• Heterogeneity: The trial intentionally enrolled a broad ventilated population (including cardiac arrest, sepsis, and respiratory failure), improving pragmatic relevance but increasing the risk that true benefits/harms in narrower phenotypes are diluted in the overall average treatment effect.
• Separation of the variable of interest:
  • SpO₂ exposure (all values): median 94 (IQR 92–97) vs 95 (93–97) vs 97 (95–99) in lower vs intermediate vs higher periods (differences: lower–intermediate −1.0; 95% CI −1.2 to −0.8; intermediate–higher −2.0; 95% CI −2.2 to −1.8; lower–higher −3.0; 95% CI −3.2 to −2.8). ²
  • Hyperoxaemic saturations: SpO₂ 99–100% occurred in 12.3% (lower) vs 14.7% (intermediate) vs 32.7% (higher) of measurements (difference lower–higher −20.4 percentage points; 95% CI −20.5 to −20.4). ²
  • FiO₂ exposure: median FiO₂ 0.30 (IQR 0.21–0.45) vs 0.36 (0.25–0.50) vs 0.45 (0.40–0.60) in lower vs intermediate vs higher periods (differences: lower–intermediate −0.06; 95% CI −0.07 to −0.05; intermediate–higher −0.09; 95% CI −0.10 to −0.09; lower–higher −0.15; 95% CI −0.16 to −0.14). ²
• Outcome assessment: Primary and key secondary outcomes were clinically meaningful and largely objective (VFDs, mortality, organ-support-free days); confidence intervals were not adjusted for multiplicity across multiple pairwise comparisons and exploratory outcomes, appropriately supporting cautious inference for secondary signals.
• Statistical rigor: Prespecified proportional-odds modelling with time adjustment; sample size target exceeded (2541 analysed vs 2250 planned); interim monitoring boundary prespecified; intraclass correlation estimate was low (0.007), supporting efficiency of the cluster-crossover design.

Conclusion on Internal Validity: Overall, internal validity is moderate-to-strong: pragmatic all-eligible enrolment and objective outcomes strengthen inference, while the open-label cluster-crossover design introduces potential performance bias and time-trend confounding that cannot be fully excluded despite prespecified modelling and washout.

External Validity

• Population representativeness: Broad mechanically ventilated adult population in an ED/medical ICU (including sepsis, cardiac arrest, acute myocardial infarction), aligning with many real-world mixed medical ICUs; pregnancy and incarceration excluded.
• Setting generalisability: Single academic centre; practice patterns (ventilation, weaning, sedation, renal-replacement thresholds) may differ across regions and healthcare systems, potentially modifying absolute event rates and treatment effects.
• Applicability of the intervention: SpO₂-target strategies are easily implementable via bedside alarms and oxygen titration; the fallback PaO₂ targets may be less feasible where arterial blood gas sampling is limited.
• Temporal applicability: Patients with COVID-19 were excluded from the primary analysis; direct extrapolation to COVID-era hypoxaemic respiratory failure should be cautious.

Conclusion on External Validity: Generalisability is moderate: the pragmatic approach supports translation to similar ED/medical ICU ventilated populations, but single-centre conduct and exclusion of COVID-19 constrain certainty across diverse systems and phenotypes.

Strengths & Limitations

• Strengths:
  • Large pragmatic trial (2541 analysed) with early enrolment at initiation of invasive ventilation and minimal bedside selection.
  • Cluster-crossover design tested system-level oxygen-target strategies as actually delivered by ICU workflows, enhancing implementation relevance.
  • Meaningful exposure separation was achieved, particularly in FiO₂ burden and time spent at SpO₂ 99–100%.
  • Patient-centred primary outcome (VFDs) and objective mortality outcomes reduce measurement subjectivity.
• Limitations:
  • Open-label design; ventilator discontinuation and other care processes could be influenced by clinicians’ perceptions of assigned oxygen targets.
  • Single-centre conduct increases vulnerability to site-specific practice effects and limits transportability.
  • SpO₂ targets exist on the haemoglobin–oxygen dissociation plateau for many patients, producing modest median saturation separation (94 vs 95 vs 97%) and raising the possibility of “insufficient biological contrast” for some mechanistic hypotheses.
  • Primary analysis excluded washout and COVID-19 periods (prespecified), which improves interpretability of stable-period comparisons but departs from a pure “all-enrolled” analysis.
  • Prespecified safety outcomes were manually adjudicated only in a subset, with later reliance on pragmatic EHR-derived outcomes. ²

Interpretation & Why It Matters

• Clinical meaning

  Across a broad ventilated ED/medical ICU population, targeting SpO₂ 88–92%, 92–96%, or 96–100% did not materially change ventilator-free days or short-term mortality, suggesting that within these ranges, average clinical outcomes are relatively insensitive to the chosen SpO₂ target strategy.
• What was actually “dosed”

  The clearest between-group differences were in oxygen exposure (FiO₂ distribution and time spent at SpO₂ 99–100%); PILOT therefore informs practice debates about liberal oxygen delivery and hyperoxaemic exposure, not merely alarm thresholds.
• Research implications

  Neutral average effects in a pragmatic design strengthen equipoise for personalised oxygen targeting (phenotype- and context-dependent), and underscore that future trials likely require either (i) substantially different PaO₂ exposure separation, (ii) enriched high-risk subgroups, or (iii) mechanistic endpoints to detect clinically meaningful oxygen-related toxicity or benefit.

Controversies & Subsequent Evidence

• Was the trial “negative” because saturation separation was too small? Median SpO₂ differed by only 1–3 percentage points across groups (94 vs 95 vs 97), consistent with physiologic plateau effects; this limits the contrast in oxygen content for many patients and could attenuate detectable outcome differences even when FiO₂ exposure differs substantially.
• Counterpoint—oxygen exposure separation was clinically meaningful: The higher-target strategy produced markedly more time at SpO₂ 99–100% (32.7% of measurements) than lower/intermediate (12.3%/14.7%), and substantially higher FiO₂ distributions, indicating that a “liberal oxygen” phenotype was achieved without clear harm in the primary endpoints.
• How PILOT fits with other RCTs: Large trials of conservative vs usual oxygen strategies (ICU-ROX; HOT-ICU) were broadly neutral, whereas trials exploring more extreme physiological targets in selected syndromes suggested potential harm signals at extremes (very conservative targets in ARDS in LOCO2; deliberate hyperoxia in septic shock in HYPERS2S). ⁵⁶⁷⁸
• Guideline alignment: Major oxygen-use guidelines generally advocate avoiding unnecessary hyperoxaemia while preventing clinically important hypoxaemia (e.g., BTS recommends 94–98% for most acutely ill patients and 88–92% for those at risk of hypercapnic respiratory failure; similar positioning is reflected in TSANZ guidance and the BMJ Rapid Recommendation against liberal oxygen in acute illness). PILOT provides pragmatic ICU-specific evidence that these “middle-ground” targets likely do not markedly change short-term outcomes in a broad ventilated population. ⁹¹⁰¹¹
• Interpretation debate—average effects vs personalisation: Contemporary commentary emphasised persistent equipoise and the possibility that individualised oxygen targets (rather than uniform saturation goals) may better match heterogeneous mechanisms of harm/benefit across ICU phenotypes. ¹²¹³

Summary

• PILOT was a pragmatic single-centre cluster-crossover trial randomising 18 two-month ICU/ED periods to SpO₂ targets of 88–92%, 92–96%, or 96–100% in invasively ventilated adults.
• Protocol adherence was high (SpO₂ target modified in 2.4–6.9% of patients) and oxygen exposure separation was clear (median FiO₂ 0.30 vs 0.36 vs 0.45; SpO₂ 99–100% in 12.3% vs 14.7% vs 32.7% of measurements).
• Primary outcome VFDs to day 28 did not differ among groups (overall P=0.81); pairwise proportional-odds estimates were close to null.
• In-hospital death before day 28 and major organ support outcomes (vasopressors, RRT, AKI) were similar across groups, with wide CIs excluding only large effects.
• Overall, PILOT supports a pragmatic “avoid extremes” approach and underlines the likelihood that any clinically important benefit of oxygen targeting is modest, phenotype-dependent, or requires larger exposure contrasts than routine SpO₂-based titration achieves.

Overall Takeaway

PILOT is a landmark pragmatic ICU oxygen-target trial because it tested three clinically recognisable SpO₂ strategies at scale, embedded in routine care, and achieved clear oxygen-exposure separation without detecting meaningful differences in ventilator-free days or short-term mortality. The results shift the field toward an “avoid extremes and personalise when needed” paradigm, while reinforcing that large benefits from modest routine SpO₂-range adjustments are unlikely in unselected ventilated populations.

Bibliography

• Semler MW, Casey JD, Lloyd BD, et al. Protocol and statistical analysis plan for the Pragmatic Investigation of Optimal Oxygen Targets (PILOT) trial: a cluster-crossover trial in critically ill adults with mechanical ventilation. BMJ Open. 2021;11:e052013.
• Semler MW, Casey JD, Lloyd BD, et al. Supplementary appendix. In: Oxygen-saturation targets for critically ill adults receiving mechanical ventilation. N Engl J Med. 2022;387:1759-1769.
• de Jonge E, Peelen L, Keijzers PJ, et al. Association between administered oxygen, arterial partial oxygen pressure and mortality in mechanically ventilated intensive care unit patients. Crit Care. 2008;12:R156.
• Girardis M, Busani S, Damiani E, et al. Effect of conservative vs conventional oxygen therapy on mortality among patients in an intensive care unit: the Oxygen-ICU randomized clinical trial. JAMA. 2016;316:1583-1589.
• ICU-ROX Investigators and the Australian and New Zealand Intensive Care Society Clinical Trials Group. Conservative oxygen therapy during mechanical ventilation in the ICU. N Engl J Med. 2020;382:989-998.
• Schjørring OL, Klitgaard TL, Perner A, et al. Lower or higher oxygenation targets for acute hypoxemic respiratory failure. N Engl J Med. 2021;384:1301-1311.
• Barrot L, Asfar P, Mauny F, et al. Liberal or conservative oxygen therapy for acute respiratory distress syndrome. N Engl J Med. 2020;382:999-1008.
• Asfar P, Schortgen F, Boisramé-Helms J, et al. Hyperoxia and hypertonic saline in patients with septic shock (HYPERS2S): a two-by-two factorial randomised clinical trial. Lancet Respir Med. 2017;5:180-190.
• O'Driscoll BR, Howard LS, Earis J, Mak V. BTS guideline for oxygen use in adults in healthcare and emergency settings. Thorax. 2017;72(Suppl 1):ii1-ii90.
• Beasley R, Chien J, Douglas J, et al. Thoracic Society of Australia and New Zealand oxygen guidelines for acute oxygen use in adults: “Swimming between the flags”. Respirology. 2015;20:1182-1191.
• Siemieniuk RAC, Chu DK, Kim LH, et al. Oxygen therapy for acutely ill medical patients: a clinical practice guideline. BMJ. 2018;363:k4169.
• Self WH, Semler MW, Rice TW. Oxygen targets for patients who are critically ill: emerging data and state of equipoise. Chest. 2020;157:487-488.
• Casey JD, Buell KG, Semler MW, et al. Individualized oxygen targets in ICU care. JAMA. 2024;332:198-199.