Skip to main content

Foundational Trials

ACTiVE

Image: Shutterstock / ChaNaWiT

Publication

Context & Rationale

  • Background
    • Invasive ventilation requires repeated adjustment as respiratory mechanics, gas exchange and spontaneous effort change. Intermittent clinician adjustments can leave patients outside intended physiological targets or delay recognition that less support is needed.
    • INTELLiVENT-ASV combines automated control of ventilation and oxygenation with optional automated weaning. Its proposed benefits include more consistent protective ventilation, earlier transition to spontaneous breathing and reduced demands on staff.
    • Earlier evidence largely concerned feasibility and physiological outcomes. In a 220-patient cardiac surgery trial, automation increased the mean proportion of time receiving optimal ventilation by 29.7 percentage points (95% CI 22.1 to 37.4; P<0.001), but this selected postoperative population did not establish improved outcomes in a broad ICU population.1
    • A 2025 Cochrane review synthesised multiple automated systems and suggested shorter ventilation. Its evidence base preceded the ACTiVE results, and the different devices and comparators limit assumptions about a uniform effect of automation.2
  • Research Question/Hypothesis
    • Would initiating INTELLiVENT-ASV very early in invasive ventilation increase days alive and free from invasive ventilation by day 28 compared with protocolised conventional ventilation?
    • The trial tested the incremental benefit of automation when both groups received protective ventilation and structured sedation and liberation practices.
  • Why This Matters
    • ACTiVE moved evaluation of a widely available ventilator platform towards a large randomised comparison using a patient-centred endpoint.
    • The clinical and operational questions are distinct: better physiological target attainment might improve outcomes, reduce workload, do both, or do neither. Each claim requires appropriate measurement.

Design & Methods

  • Research Question:
    • In critically ill adults expected to require invasive ventilation for at least 24 hours, does early automated closed-loop ventilation increase ventilator-free days at day 28 compared with protocolised conventional ventilation?
  • Study Type:
    • Investigator-initiated, international, multicentre, parallel-group, 1:1 randomised superiority trial in seven ICUs in the Netherlands and Switzerland; recruitment ran from October 2020 to June 2025.
    • A password-protected web system allocated treatment using concealed permuted blocks of 4, 6 or 8, stratified by centre. The trial was registered as NCT04593810; its protocol was published in 2022.3
    • Amsterdam University Medical Centers sponsored and coordinated the trial; the Netherlands Organisation for Health Research and Development (ZonMw) funded it. The funding organisation had no role in trial conduct, analysis, interpretation or publication.
    • An independent data and safety monitoring board reviewed conduct and safety. The protocol provided for an initial review after 150 patients and subsequent six-monthly reviews; it did not specify a group-sequential efficacy stopping boundary.4
    • Two authors disclosed lecture fees from Hamilton Medical; Schultz disclosed part-time employment with Hamilton Medical from January 2022 to January 2023. These relationships warrant transparent consideration without implying that they invalidated the results.
  • Population:
    • Adults aged ≥18 years, receiving invasive ventilation in a participating ICU, with less than one hour elapsed since its initiation in the ICU and an anticipated requirement of ≥24 hours.
    • Exclusions included an unavailable compatible ventilator, pregnancy, >1 hour of ventilation in the ICU or >6 hours before ICU admission, another interventional trial, recent pneumonectomy or lobectomy, BMI >40 kg/m², pre-existing restrictive lung disease, ECMO, unreliable pulse oximetry, neuromuscular disorders expected to prolong ventilation, previous randomisation, or inability to obtain consent.
    • Acute brain injury and cardiac arrest were included; the neuromuscular exclusion should not be interpreted as excluding all neurological patients.
    • Of 3,641 patients assessed, 1,514 were randomised: 752 to automation and 762 to conventional ventilation. Removal of 150 versus 163 patients without deferred consent left 602 versus 599 in the modified intention-to-treat population.
    • Median age was 63 years in both groups; women comprised 36.4% versus 35.7%. Median SAPS II was 54 versus 52 and SOFA was 8 versus 8.
    • Respiratory failure prompted intubation in 30.4% versus 33.4%, cardiac arrest in 27.7% versus 25.0%, and neurological disease in 25.9% versus 24.9%. Medical admissions accounted for 86.0% versus 86.3%.
  • Intervention:
    • INTELLiVENT-ASV, a Hamilton Medical rule-based closed-loop platform, was started as soon as possible after allocation. The platform remained unchanged during recruitment; this was not a trial of a learning neural-network system.
    • Clinicians entered sex and measured height to determine predicted body weight (PBW). The controller adjusted tidal volume, respiratory rate and support to achieve minute ventilation using respiratory mechanics and end-tidal CO₂ feedback; SpO₂ feedback guided FiO₂ and PEEP.5
    • Default targets were end-tidal CO₂ 35–45 mmHg and SpO₂ 92–96%, with patient-specific modification permitted. Clinicians could select ARDS, chronic hypercapnia or brain injury presets and set safety limits.
    • The brain injury preset required tighter CO₂ control and manual PEEP selection. QuickWean was unavailable with that preset, an important qualification to the extent of automation in neurological patients.5
    • QuickWean was recommended; automated spontaneous breathing trials (SBTs) were optional. Clinicians retained responsibility for reviewing SBT results and deciding on extubation.
    • Clinicians could override the controller or change mode whenever necessary. Sensor failure disabled the relevant automatic controller and generated alarms; automation therefore required continuing bedside oversight.
    • The allocated strategy continued until successful extubation, death or day 28, whichever occurred first.
  • Comparison:
    • Conventional volume-controlled, pressure-controlled or pressure-support ventilation, adjusted by clinicians. Other automated modes, including ASV, SmartCare/PS, NAVA and proportional assist ventilation, were prohibited.
    • Protective ventilation recommendations included tidal volume 6–8 mL/kg PBW, plateau pressure ≤30 cm H₂O, appropriate PEEP, SpO₂ 92–96% and the lowest FiO₂ providing adequate oxygenation.5
    • Patients receiving controlled ventilation were assessed at least three times daily for transition to pressure support. Both groups underwent extubation-readiness assessments at least three times daily.
    • Conventional SBTs used a T-piece or pressure support <10 cm H₂O. Success required ≥30 minutes with respiratory rate <35/min, SpO₂ >90%, increases in heart rate and blood pressure <20%, and no anxiety or diaphoresis.4
    • Shared extubation criteria included an awake, responsive patient, an adequate cough, temperature 36–38°C, PaO₂/FiO₂ >150 mmHg on FiO₂ ≤0.40, haemodynamic stability with no or low-dose vasopressors, and respiratory rate <35/min. Successful testing prompted clinical review rather than automatic tube removal.
    • Both groups received local sedation practices favouring analgesia, avoidance of unnecessary deep sedation and regular sedation assessment. Daily sedation interruption was encouraged.
    • Tracheostomy was generally deferred until at least day 10 unless indicated otherwise. Thromboprophylaxis, fluid and nutrition management followed shared recommendations and local practice; selective digestive decontamination depended on national practice.
  • Blinding:
    • Treating clinicians and investigators were unblinded; the trial statistician was masked to group identity.
    • Unblinded decisions could influence extubation, rescue treatment and diagnostic investigation, even though death and ventilation dates are comparatively objective.
  • Statistics:
    • Power calculation: A total of 1,200 patients, allowing 5% dropout, was required to detect 1.5 additional ventilator-free days from an assumed mean of 20 days and common SD of 9, with 80% power (β=0.20) and a two-sided α=0.05.
    • Ventilator-free days combined survival and liberation: deaths within 28 days and ventilation beyond day 28 received zero; after repeated intubations, only the final successful liberation counted. Liberation had to last at least 24 hours; non-invasive ventilation did not count as invasive support.
    • The primary analysis used a mixed-effects cumulative logistic model with centre as a random effect, reporting a common odds ratio for a more favourable ventilator-free-day distribution. OR >1 favoured automation; the OR was not a ratio of ventilation durations.
    • The analysis was modified intention-to-treat, excluding patients without deferred consent. Available cases were analysed without imputation; the final SAP specified additional missing-data analyses if primary-outcome missingness exceeded 5% within the analysis population.6
    • The per-protocol analysis required the assigned strategy for ≥80% of ventilation time during the first five days. A covariate-adjusted analysis included age, sex and illness severity.
    • Ventilation quality was the percentage of classified time in optimal, acceptable or critical zones during the first six hours. Centre-adjusted linear models compared zones, with P<0.017 required for the three zone comparisons. A further cumulative logistic analysis weighted observations by each patient’s classified breath count.
    • Binary secondary outcomes used centre-adjusted models reporting absolute differences; mortality used Cox models with centre frailty. Ventilation duration among survivors and lengths of stay used median regression.
    • The SAP specified Holm–Bonferroni correction across secondary outcomes; the supplement applied it to 19 comparisons. Reported confidence intervals were not multiplicity-adjusted.65
    • Six prespecified treatment-interaction analyses examined admission type, neurological indication, cardiac arrest, hypoxaemic respiratory failure, BMI and illness severity.
  • Follow-Up Period:
    • The primary endpoint was assessed at day 28; mortality follow-up extended to day 90.
    • Detailed ventilation and treatment measurements were concentrated in the first five days. High-resolution ventilation-quality recording was available only in a subset from three centres.

Key Results

This trial was not stopped early. Recruitment reached the planned analysable cohort: 1,201 patients remained after deferred-consent exclusions, with primary-outcome data available for 601 versus 595 patients.

Values are automation versus conventional ventilation. Absolute differences are published model estimates, in percentage points for binary outcomes. Primary P values are unadjusted; secondary rows give nominal and Holm-adjusted P values where available. All confidence intervals are unadjusted for multiplicity.5

Outcome INTELLiVENT-ASV Conventional ventilation Effect p value / 95% CI Notes
Primary: ventilator-free days, day 28; median (IQR) 16.7 (0–26.1); n=601 16.3 (0–26.5); n=595 Common OR 0.91 95% CI 0.77 to 1.06; P=0.23 Higher is better; no demonstrated superiority
Ventilator-free days; mean (SD) 13.1 (12.2) 13.4 (12.3) Mean difference −0.3 days 95% CI −1.7 to 1.0 Descriptive presentation of the primary endpoint; not a separate confirmatory test
28-day mortality 224/602 (37.2%) 218/597 (36.5%) HR 1.04 95% CI 0.86 to 1.25; nominal P=0.69; adjusted P=1.00 Absolute difference 0.7 points; 95% CI −4.8 to 6.1
Ventilation duration in 28-day survivors; median (IQR), days 3.3 (1.0–9.4); n=377 2.6 (1.0–8.7); n=377 Median difference 0.7 days 95% CI −0.0 to 1.4; nominal P=0.054; adjusted P=0.810 Survivor-only comparison; not evidence of shorter ventilation
Time in optimal ventilation zone, first 6 h; mean (SD) 48.4% (40.2); n=78 36.7% (38.4); n=64 Mean difference 12.9 points 95% CI −0.2 to 25.9; P=0.055 Did not meet the zone-specific P<0.017 threshold
Time in acceptable ventilation zone, first 6 h; mean (SD) 40.1% (36.6); n=78 27.5% (31.6); n=64 Mean difference 12.5 points 95% CI 1.2 to 23.9; P=0.032 Did not meet the zone-specific P<0.017 threshold
Time in critical ventilation zone, first 6 h; mean (SD) 11.6% (26.9); n=78 35.8% (45.0); n=64 Mean difference −24.6 points 95% CI −36.6 to −12.6; P<0.001 Physiological subset endpoint; met the zone-specific threshold
Severe hypoxaemia: PaO₂ <55 mmHg 96/599 (16.0%) 126/596 (21.1%) Absolute difference −5.1 points 95% CI −9.4 to −0.7; nominal P=0.022; adjusted P=0.352 Not statistically significant after multiplicity correction
Severe hypercapnia: PaCO₂ >55 mmHg and pH <7.35 119/599 (19.9%) 144/596 (24.2%) Absolute difference −4.1 points 95% CI −8.7 to 0.5; nominal P=0.080; adjusted P=1.00 No established reduction
Any rescue strategy 86/599 (14.4%) 121/596 (20.3%) Absolute difference −6.0 points 95% CI −10.2 to −1.7; nominal P=0.006; adjusted P=0.108 Recruitment manoeuvres, prone positioning or bronchoscopy
Prone positioning 55/599 (9.2%) 83/596 (13.9%) Absolute difference −4.8 points 95% CI −8.4 to −1.2; nominal P=0.009; adjusted P=0.153 Clinician-dependent treatment decision
New ARDS after 48 h 15/599 (2.5%) 12/596 (2.0%) Absolute difference 0.5 points 95% CI −1.2 to 2.2; nominal P=0.566; adjusted P=1.00 Not the prevalence of ARDS at enrolment
Ventilator-associated pneumonia 13/599 (2.2%) 21/596 (3.5%) Absolute difference −1.4 points 95% CI −3.3 to 0.4; nominal P=0.131; adjusted P=1.00 New infection after 48 h; no established reduction
Pneumothorax requiring drainage 6/599 (1.0%) 2/596 (0.3%) Absolute difference 0.7 points 95% CI −0.3 to 1.6; nominal P=0.157; adjusted P=1.00 Sparse events; uncertainty about uncommon harm remains
Extubation failure within 24 h 32/413 (7.7%) 30/412 (7.3%) Absolute difference 0.5 points 95% CI −3.0 to 4.1; nominal P=0.766; adjusted P=1.00 Denominator restricted to patients assessed after extubation
ICU stay; median (IQR), days 4.5 (1.8–11.9); n=595 4.4 (1.8–12.4); n=594 Median difference 0.1 days 95% CI −0.9 to 1.1; nominal P=0.850; adjusted P=1.00 No established shortening
Hospital stay; median (IQR), days 12.2 (3.5–27.5); n=564 12.0 (4.0–25.3); n=581 Median difference 0.3 days 95% CI −0.8 to 1.4; nominal P=0.598; adjusted P=1.00 More incomplete data than for the primary endpoint
90-day mortality 246/600 (41.0%) 237/592 (40.0%) HR 1.04 95% CI 0.87 to 1.24; nominal P=0.656; adjusted P=1.00 Absolute difference 0.9 points; 95% CI −4.6 to 6.5
  • The slightly higher median ventilator-free-day value did not represent a favourable overall distribution: the common OR was 0.91 and the mean difference was −0.3 days. Per-protocol results were also inconclusive: OR 1.01 (95% CI 0.83 to 1.23; P=0.93), with 423 versus 567 patients; the adjusted analysis gave OR 0.93 (95% CI 0.76 to 1.14; P=0.49).5
  • The overall ventilation-quality ordinal comparison favoured automation, OR 1.50 (95% CI 1.43 to 1.57; P<0.001), and remained significant after Holm correction. This does not establish reductions in hypoxaemia, rescue treatment or mortality; those clinical secondary outcomes did not survive multiplicity correction.5
  • No prespecified subgroup interaction was detected. For neurological indications, OR was 0.74 (95% CI 0.49 to 1.10; interaction P=0.14); for hypoxaemic respiratory failure, OR was 1.00 (95% CI 0.57 to 1.78; interaction P=0.57). These imprecise subgroup results neither identify a benefiting phenotype nor exclude clinically relevant heterogeneity.

Internal Validity

  • Randomisation and Allocation:
    • Concealed web allocation, variable block sizes and centre stratification provided sound protection against selection before randomisation.
    • These safeguards apply to the original randomised groups. Their protection is less secure after substantial exclusions determined after allocation.
  • Dropout and Post-randomisation Exclusions:
    • Deferred-consent exclusions removed 150/752 patients assigned to automation and 163/762 assigned to conventional ventilation. Baseline characteristics and outcomes of excluded patients were unavailable.
    • Subsequent primary-outcome loss was small: one patient in the retained automated group and four in the retained conventional group lacked day-28 ventilator-free-day data.
    • The principal attrition concern is therefore the consent-related removal of approximately one-fifth of randomised patients, rather than the five missing primary outcomes among retained participants.7
    • Different outcomes used different available-case denominators; for example, 90-day mortality was available for 600 versus 592 patients. A single statement of near-complete follow-up should not be applied indiscriminately to every endpoint.
  • Performance and Detection Bias:
    • Clinicians knew allocation and decided when to change mode, perform rescue procedures, investigate complications and extubate.
    • Shared protocols and a masked statistician reduce some bias, but do not remove the influence of treatment knowledge on clinical decisions. Blinded adjudication of pulmonary complications was not reported.
  • Protocol Adherence and Crossover:
    • Median daily time receiving the assigned strategy was 95.8% (IQR 70.8–100) versus 100% (100–100). The per-protocol populations were 423 versus 567, showing appreciably more departures from assigned treatment in the automated arm.5
    • At one hour, INTELLiVENT-ASV was recorded in 427/591 patients (72.3%) assigned to automation and 23/594 (3.9%) assigned to conventional ventilation. At the day-one assessment, corresponding values were 432/512 (84.4%) and 17/501 (3.4%).5
    • These observations demonstrate meaningful separation with incomplete early delivery and some control exposure. A complete patient-level count and breakdown of reasons for crossover were not reported; daily snapshots should not be mistaken for unique crossover totals.
  • Baseline Characteristics and Heterogeneity:
    • Groups were broadly comparable in age, sex, severity and baseline respiratory measurements. Median PaO₂/FiO₂ was 190 versus 184.1 mmHg and median driving pressure was 13 cm H₂O in both groups.
    • Sepsis was recorded in 83/571 (14.5%) versus 105/564 (18.6%). Covariate adjustment for age, sex and severity did not materially change the primary inference.
    • Patients were substantially ill, as reflected by approximately 37% day-28 mortality, but many required ventilation for reasons other than sustained pulmonary failure. Overall severity and the opportunity to benefit from ventilator adjustment are different characteristics.
  • Timing and Dose:
    • Median ventilation time before randomisation was 0.6 hours in both groups. This early allocation minimised prolonged exposure to an uncontrolled ventilation strategy.
    • The one-hour mode data show that early randomisation did not ensure immediate receipt of automation in every intervention patient. Median assigned-strategy exposure also concealed a lower-adherence tail.
    • The biological dose depended on both time under automated control and the functions actually enabled. In particular, use of the mode did not guarantee delivery of automated weaning or SBTs.
  • Separation of the Variable of Interest:
    • At one hour, median minute ventilation was 7.4 versus 8.3 L/min, with a published difference of −0.9 L/min (95% CI −1.1 to −0.6); end-tidal CO₂ was 37.5 versus 34.5 mmHg, difference 3.0 mmHg (95% CI 1.5 to 4.5).5
    • At the same assessment, tidal volume was 6.6 versus 6.5 mL/kg PBW and driving pressure 13 versus 13 cm H₂O. Mode separation was therefore greater than separation in these particular protective-ventilation variables.
    • On day one, median tidal volume was 7.1 versus 6.7 mL/kg PBW, respiratory rate 17 versus 18/min, PEEP 7 versus 6 cm H₂O, and FiO₂ 0.30 versus 0.35. Automation did not consistently mean lower tidal volume.5
  • Key Delivery Aspects:
    • A formal weaning trial was recorded in 413/602 patients (68.6%) versus 416/599 (69.4%); median time to the first formal trial was two days in both groups.5
    • QuickWean plus automated SBT was documented in 28/602 (4.7%) versus 3/599 (0.5%), QuickWean alone in 28/602 (4.7%) versus 11/599 (1.8%), and automated SBT alone in 7/602 (1.2%) versus 2/599 (0.3%). Categories could overlap.
    • Weaning strategy was unknown, undocumented or not performed in 273 patients in each group. These data prevent a confident estimate of how consistently the automated liberation functions were used.5
  • Adjunctive Therapies:
    • Propofol exposure was broadly similar: median cumulative dose 6,141 versus 6,205 mg and median two treatment days in each group. Reported daily drug and fluid measurements do not demonstrate a large, consistent co-intervention imbalance.5
    • Conventional ventilation involved more rescue therapy, particularly prone positioning. As these were post-randomisation responses, they are part of the total effect of the assigned strategies rather than baseline confounders to be adjusted away.
    • The unblinded nature of rescue decisions and their loss of statistical significance after correction limit causal claims about why they differed.
  • Outcome Assessment and Statistical Rigour:
    • Ventilator-free days were explicitly defined, and both their distribution and components were presented. The cumulative logistic primary model accounted for centre and matched the final SAP.
    • Per-protocol and covariate-adjusted analyses supported the primary conclusion. The per-protocol comparison nevertheless selected patients according to events after randomisation and could not restore the original randomised contrast.
    • Multiplicity correction was available in the supplement and must govern interpretation of secondary clinical outcomes. The ventilation-quality subset and changes between protocol versions require separate scrutiny.

Conclusion on Internal Validity: Moderate. Allocation and the primary analysis were well organised, but substantial consent-related exclusions, unblinded clinical decisions and incomplete documentation of automated weaning reduce confidence in a fully unbiased estimate for all randomised patients. The physiological subset provides more limited evidence than the main clinical comparison.

External Validity

  • Population Representativeness:
    • The findings apply most directly to early invasive ventilation in adult, predominantly medical ICU patients in centres familiar with both INTELLiVENT-ASV and protocolised conventional ventilation.
    • Acute neurological disease and cardiac arrest were prominent; elective postoperative ventilation was uncommon. The study should not be treated as an ARDS-specific trial or a trial restricted to patients with difficult weaning.
    • ECMO, BMI >40 kg/m², restrictive lung disease and neuromuscular causes of prolonged ventilation were excluded. Children, pregnancy and patients already ventilated for longer periods were not studied.
    • Device availability excluded 696 screened patients, and ventilation timing excluded another 545. These practical selection mechanisms restrict representativeness beyond the formal clinical criteria.
  • Applicability Across Healthcare Systems:
    • Seven centres in two high-income European countries provide more breadth than a single-centre study, but limited evidence about ICUs with different staffing, training or access to compatible ventilators and reliable sensors.
    • At least three daily readiness assessments and reinforced protective-ventilation practice created an active comparator. The incremental effect may differ where these processes are less consistent, but ACTiVE does not establish either greater benefit or adequate safety in those settings.
    • The tested strategy remained dependent on clinicians selecting targets, interpreting arterial gases, managing sedation and airway protection, and responding to alarms. Its results do not justify reducing staffing or extending autonomous operation beyond the conditions studied.
    • Generalisability is specific to this platform and its implementation. Other automated modes, future software versions and systems that control different parts of the ventilation pathway require their own evidence.

Conclusion on External Validity: Moderate for comparable adult ICUs using structured ventilation and liberation protocols; limited for advanced respiratory failure, excluded populations, later weaning, and settings with substantially different resources. A benefit under less protocolised care remains an untested hypothesis.

Strengths & Limitations

  • Strengths:
    • Large, international, investigator-initiated randomised comparison with concealed allocation, independent safety oversight and a patient-centred primary endpoint.
    • Very early randomisation, a common protective-ventilation framework and meaningful separation in ventilator mode.
    • Accessible protocol versions, a dated final SAP, detailed process data, sensitivity analyses and explicit secondary-outcome multiplicity correction.
    • Clinical follow-up to 90 days and complementary physiological measurements helped distinguish target attainment from patient benefit.
  • Limitations:
    • Substantial exclusions after randomisation and no available clinical description of excluded patients.
    • Unblinded decisions about liberation, rescue treatment and complication ascertainment.
    • Incomplete delivery and documentation of automated weaning, with many patients having short ventilation exposure or non-respiratory barriers to extubation.
    • A small, selected high-resolution subset, differing quality-analysis denominators and an endpoint based on physiological thresholds rather than validated clinical surrogacy.
    • No comprehensive measurement of staff time, cognitive workload, costs or longer-term functional recovery; some originally planned outcomes were omitted.

Interpretation & Why It Matters

  • Clinical efficacy
    • Routine early use of INTELLiVENT-ASV did not improve ventilator-free days over the conventional strategy tested. There was no established reduction in mortality, length of stay or extubation failure.
    • The primary result does not support adopting this platform with an expectation of faster liberation in a similar broad ICU population.
  • Physiological performance
    • Automation reduced exposure to the predefined critical ventilation zone in the measured subset. This is evidence that the controller changed physiological management.
    • A surrogate improvement does not itself establish prevention of lung injury, preservation of respiratory muscle function or better recovery. The observed clinical results did not demonstrate those downstream benefits.
  • What remains unresolved
    • ACTiVE was a superiority trial. Failure to demonstrate superiority cannot establish clinical equivalence, non-inferiority or interchangeable safety.
    • Operational value remains plausible but incompletely measured. Fewer manual adjustments, fewer alarms, less staff time and lower total cost are separate outcomes; none should be inferred solely from a ventilation-quality score.

Controversies & Other Evidence

  • Deferred Consent: Ethical Justification Does Not Ensure Statistical Independence:
    • Emergency enrolment allowed intervention within an hour, a genuine design advantage. The problem is the subsequent exclusion mechanism: legal necessity does not demonstrate that consent availability was unrelated to prognosis or treatment experience.
    • The investigators stated that exclusions occurred independently of allocation and clinical outcomes. Similar exclusion counts and balanced measured characteristics among retained patients cannot verify that assertion; excluded-patient characteristics and outcomes were unavailable.7
    • The version-5 protocol sought consent within 72 hours and specifically contemplated retaining data when patients died before consent could be obtained. The final report does not describe implementation of this provision by country or centre. It would therefore be incorrect to assume that all early deaths were excluded.4
    • Replacing patients without consent helped achieve the planned retained sample size, but replacement cannot repair selection introduced after randomisation. Nor can per-protocol analysis recover the effect in patients whose data were removed.
  • Case Mix and the Causal Pathway to Liberation:
    • The accompanying editorial emphasised that many patients were ventilated after cardiac arrest or for acute neurological disease. Their extubation may depend predominantly on consciousness, airway protection and neurological prognosis rather than ventilator adjustments.7
    • The editorial estimated that approximately 20% had died and 40% had been extubated within three days. Short exposure narrows the opportunity for a preventive ventilation strategy to alter subsequent recovery; the high mortality should not be confused with a uniformly high burden of modifiable pulmonary injury.
    • Time to a successful SBT, subsequent delay to extubation and the reasons for that delay would help distinguish physiological readiness from non-respiratory barriers. ACTiVE reported time to the first formal weaning trial, but not these more discriminating measures.
    • Enriching future studies for sustained pulmonary dysfunction is a defensible research strategy. ACTiVE’s interaction tests did not establish that ARDS, hypoxaemia or another subgroup benefits, and such enrichment should not be presented as a proven treatment indication.
  • How Much Automated Liberation Was Actually Tested?:
    • A ventilation mode can automate gas exchange while leaving important parts of liberation dependent on clinicians. QuickWean was recommended rather than compulsory, automated SBTs were discretionary, and the brain injury preset disabled QuickWean.5
    • The low documented use of particular automated weaning functions and extensive unknown or unperformed entries make the delivered intervention less completely characterised than the description “fully automated weaning” implies.
    • Accordingly, ACTiVE estimates the effect of offering and using this platform under the observed implementation conditions. It does not isolate the efficacy of consistently enabled automated SBTs in patients who are otherwise ready for extubation.
  • Protocol Evolution and Reporting Transparency:
    • The change summary from protocol version 1 to version 5 documents removal of ventilation quality from the primary objective and hypothesis, placing it under secondary objectives. The 2022 published protocol already designated ventilator-free days as the primary endpoint; the quality result should not be promoted retrospectively to a second primary success.43
    • The original protocol proposed a zero-inflated modelling approach; the final SAP used a cumulative logistic model. A justified change before analysis is acceptable, and the published primary analysis matched the dated final plan.46
    • The SAP records an August 2024 revision and a final version dated 12 January 2025, before recruitment ended. It shortened the principal ventilation-quality window from 24 to 6 hours and removed maximal inspiratory pressure because reliable collection was not feasible.6
    • Day-28 quality of life was included in the protocol but was not reported in the main results or supplement, and its omission was not explained in the supplied change summary. The loss of respiratory muscle testing also leaves a proposed biological mechanism untested.
    • The breath-weighted ordinal quality comparison was not described in the final SAP’s quality-analysis section, which specified patient-level zone comparisons. It is useful supporting information, but should be distinguished from that prespecified analysis.
  • Ventilation Quality: Selection, Definitions and Precision:
    • High-resolution data were exported from 305 patients, but recording required a researcher to connect a memory box during office hours. Only 78 versus 64 patients contributed to the combined six-hour quality analysis; later windows included different, larger subsets.5
    • The main table’s footnote describes 152 patients while listing 78 versus 64. The supplement clarifies that individual-parameter time analyses included 80 versus 72, whereas the combined six-hour endpoint used 78 versus 64. These denominators should be stated explicitly rather than treated as interchangeable.
    • A breath was critical if any available parameter crossed a threshold: tidal volume ≥12 mL/kg PBW, maximum airway pressure ≥36 cm H₂O, end-tidal CO₂ <25 or ≥51 mmHg, or SpO₂ <85%. These thresholds describe a composite physiological state; they do not diagnose ventilator-induced lung injury.
    • Incomplete observations could still be classified as critical when an available parameter was critical, whereas otherwise incomplete classifications were missing. Because percentages excluded unclassified data, signal completeness could influence the composite; arm-specific completeness is important to its interpretation.5
    • Higher overall quality did not mean improvement in every component. Time in the optimal tidal-volume zone was 75.2% versus 86.8%; the published difference was −11.6 points (95% CI −21.0 to −2.1; nominal P=0.018). Optimal end-tidal CO₂ and SpO₂ time favoured automation. These exploratory component findings argue against reducing “quality” to tidal-volume protection alone.5
    • Fixed CO₂ and oxygen thresholds may also differ from deliberately individualised targets, including permissive hypercapnia or brain-protective ventilation. The original protocol anticipated patient-category-specific zones; the main results used the common published zones.
    • The very precise breath-weighted OR should be interpreted alongside the patient-level mean differences. Repeated breaths do not create additional independent randomised patients, and the handling of within-patient dependence in that ordinal analysis was not explicitly described.
  • Multiplicity, Power and Interpretation of the Primary Endpoint:
    • The reductions in rescue treatment, prone positioning and severe hypoxaemia had nominal P values of 0.006, 0.009 and 0.022, but Holm-adjusted values of 0.108, 0.153 and 0.352. Their nominal confidence intervals cannot override the planned family-wise testing procedure.5
    • For the three prespecified quality-zone comparisons, only critical-zone time met P<0.017. The optimal and acceptable zone results should not both be described as statistically established improvements.
    • Observed mean ventilator-free days were 13.1 and 13.4, with SDs of 12.2 and 12.3, rather than the assumed mean of 20 and SD of 9. Thus, achieving the planned sample did not guarantee the original operating characteristics.
    • The published mean-difference interval, −1.7 to 1.0 days, does not include the planned 1.5-day mean benefit. This weighs against that anticipated average gain, but does not establish a prespecified acceptable margin of harm; the primary inferential model also estimated a common OR rather than a mean difference.
    • Ventilator-free days appropriately penalise death but combine distinct clinical processes. Mortality and survivor ventilation duration must therefore accompany the composite; survivor-only duration remains conditioned on a post-randomisation event.8
    • The editorial proposed that future non-inferiority designs could address whether acceptable clinical outcomes accompany operational gains. Such trials would need a justified margin, sufficient adherence and evidence that the population and endpoint can discriminate clinically meaningful differences; the same short exposure and case mix could otherwise make non-inferiority too easy to demonstrate.7
  • Rescue Treatments and the ARDS Denominator:
    • Prone positioning was used in 9.2% versus 13.9%, whereas new ARDS was recorded in 2.5% versus 2.0%. These figures prompted questions about rescue-treatment indications.7
    • The ARDS endpoint counted only disease developing after 48 hours. It is not a measure of all ARDS present during ventilation, so the apparent discrepancy does not by itself show inappropriate proning.
    • Less rescue treatment could reflect better oxygenation, different clinician thresholds or other post-randomisation changes. With unblinded decisions and multiplicity-adjusted uncertainty, the trial cannot resolve these mechanisms.
  • Randomised Evidence Beyond ACTiVE:
    • INTELLiPOWER randomised 96 patients in a crossover comparison of three-hour ventilation periods. Overall mechanical power was 15.8 versus 16.1 J/min, with a mean difference of −0.44 J/min (95% CI −1.17 to 0.29; P=0.24). This supports caution about assuming that algorithmic optimisation necessarily reduces total ventilatory energy.9
    • PROMIZING studied a different intervention—PAV+ versus pressure support—in 573 randomised patients already able to receive partial support. Median time to successful liberation was 7.3 versus 6.8 days (P=0.58). It provides parallel evidence that a sophisticated mode may not add clinical benefit over structured conventional care, but does not replicate ACTiVE’s device or treatment phase.10
  • Systematic Review Evidence:
    • The 2025 Cochrane update included 62 trials and 5,052 participants across 11 automated systems. Across 51 trials, the published relative reduction in ventilation duration was 24% (95% CI 18% to 30%), with substantial heterogeneity (I²=87%); mortality RR was 0.94 (95% CI 0.82 to 1.07).2
    • That synthesis did not incorporate the ACTiVE outcome results. Differences in devices, patient populations, timing and comparator protocols preclude treating its pooled estimate as the expected effect of INTELLiVENT-ASV in ACTiVE’s setting.
    • The discordance supports updated, device-specific and comparator-sensitive synthesis. It does not justify either discarding ACTiVE or assuming that every automated system is ineffective.
  • Observational Evidence and Care Processes:
    • WEAN SAFE followed 5,869 patients across 481 ICUs in 50 countries. Among patients with separation attempts, 22.4% had a delay of at least five days before the first attempt; deeper sedation and delayed initiation were associated with weaning failure.11
    • These observational findings establish variation and potentially modifiable barriers in usual practice. They make comparator quality relevant, but do not prove that automation overcomes those barriers.
  • Subsequent Debate and Operational Outcomes:
    • A 2026 commentary argued that future automated-ventilation trials should measure caregiver time, cognitive burden, training and system efficiency alongside patient outcomes. It explicitly recognised that automation could reduce, increase or redistribute workload.12
    • That is an appropriate research agenda, but the commentary’s description of clinical outcomes as equivalent exceeds what ACTiVE’s superiority design established. Operational benefit and acceptable clinical safety require direct, separate demonstration.
    • A 2026 exploratory ACTiVE analysis included 13 automated and 24 conventionally ventilated neurocritical care patients. Workload-relevant alarms were 3.28 versus 3.73 per hour (P=0.81), while alarm-management interventions were 0.14 versus 0.21 per hour (P=0.01).13
    • This small post hoc subset suggests a possible reduction in one category of bedside intervention. It did not establish fewer alarms, lower total nursing time, safer staffing reductions or improved clinical outcomes.
  • Guideline Context:
    • The AARC adult liberation guideline recommends a standardised daily readiness assessment and, when appropriate, an SBT before noon; SBTs may use pressure support or unassisted breathing, and FiO₂ should not be increased during testing.14
    • This 2024 guideline predates ACTiVE and cannot be presented as incorporating its findings. Its continuing relevance is the shared foundation of systematic assessment and timely liberation; ACTiVE does not replace those practices with a requirement for automated ventilation.

Summary

  • ACTiVE randomised 1,514 adults in seven ICUs; 602 versus 599 remained after consent-related exclusions. It compared early INTELLiVENT-ASV with protocolised conventional ventilation.
  • Ventilator-free days were not improved: median 16.7 versus 16.3 days; common OR 0.91 (95% CI 0.77 to 1.06; P=0.23). Mortality and lengths of stay were not demonstrably reduced.
  • Critical-zone ventilation time was lower in the measured subset, but clinical secondary signals for hypoxaemia and rescue treatment did not remain significant after multiplicity correction.
  • The main interpretative constraints are substantial post-randomisation exclusions, prominent neurological and cardiac-arrest case mix, incomplete automated-weaning documentation and selected physiological sampling.
  • The trial informs use of this platform in comparable protocolised ICUs. It establishes neither equivalence nor staffing or economic benefit; those questions require targeted prospective evaluation.

Overall Takeaway

ACTiVE is an important large trial showing that early INTELLiVENT-ASV did not improve ventilator-free days over protocolised conventional ventilation in the population studied, despite better performance on a physiological subset endpoint. Its lasting contribution is to separate physiological automation from proven patient benefit and to define the clinical, implementation and operational questions that remain unresolved.

Overall Summary

  • Early automated closed-loop ventilation improved selected physiological measures but did not establish better patient outcomes, clinical equivalence or reduced staffing requirements compared with protocolised conventional ventilation.

Bibliography

Last updated September 10th, 2026