Editorial 

Automated Modes to Improve Mechanical Ventilation Outcomes—The Ghost in the Machine

Michael C. Sklar, Niall D. Ferguson

JAMA Published Online: December 8, 2025

doi: 10.1001/jama.2025.24401

Invasive mechanical ventilation, one of the defining interventions in critical care medicine, is received by more than 20 million patients annually in intensive care units (ICUs) worldwide.¹ In any given ICU, usually half or more of patients require invasive ventilatory support.² Among those needing at least 2 days of mechanical ventilation, the ICU mortality rate is high (>30%), as is the proportion of patients with difficult or prolonged weaning (>35%).³

Although mechanical ventilation can be highly effective at ensuring adequate gas exchange, it can also cause injury to the lung, promote diaphragm dysfunction, and impede resolution of respiratory failure. These complications of mechanical ventilation increase mortality and can also lead to prolonged ventilation and critical illness, along with adverse long-term sequelae. These risks can be minimized through careful adjustments to ventilator settings to provide optimal timing, volume, and pressure of delivered breaths. However, such adjustments require considerable expertise and must be made regularly in response to a patient’s changing respiratory condition. Timely provision of such expertise to every ventilated patient is a challenge in many ICU settings.

Adaptive support ventilation (ASV), a closed-loop ventilation mode in which the ventilator automatically adjusts its settings in response to a patient’s changing lung mechanics, respiratory drive, and metabolic demand, represents an approach that might be less costly than—and potentially superior to—current approaches that rely on human expert supervision. Through continuous, automated adjustment of ventilator settings, the system aims to maintain lung protection while simultaneously preserving diaphragm activity and minimizing patient-ventilator asynchrony. In theory, adaptive control could reduce clinician workload and prevent underrecognized injurious settings. Despite physiological promise, translation into improved clinical outcomes has remained uncertain; until now, no large randomized clinical trial of early ASV had been performed.

In this issue of JAMA, Sinnige and colleagues⁴ present the results of one of the largest randomized clinical trials of mechanical ventilation in the ICU. The investigators randomly assigned 1514 patients to an automated closed-loop ventilation system using INTELLiVENT-ASV or a control group of protocolized mechanical ventilation. Both groups underwent mechanical ventilation according to a lung-protective strategy along with sedation and weaning protocols. The primary outcome of ventilator-free days at day 28 was not different at a median of 16.7 days and 16.3 days in the closed-loop and control groups, respectively. Rescue therapies were used more frequently in the control group (especially use of prone positioning: 14% vs 9%), but given the very low rate of acute respiratory distress syndrome (2% in both groups), it is unclear what the indications for this were. Quality of ventilation was also reported as significantly better in the closed-loop ventilation group, but it is important to recognize that this composite outcome was recorded only in a convenience sample of 152 patients (13%). Secondary outcomes of 28-day mortality and duration of ventilation among survivors were not different between groups, with point estimates slightly favoring the control group.

Although the closed-loop ventilation system did not improve ventilator-free days, this rigorously executed and clinically important trial highlights several salient issues deserving further discussion. First, is the question of study design. The investigators set out to test whether ASV was superior to usual care. One might argue that if the intervention were easier to use and potentially reduced human resource demands, then a demonstration of noninferiority would be sufficient. That would, however, require a different study design with careful consideration of the trial population of interest (as discussed further below) and the minimum clinically important difference between groups, often requiring a larger sample size.⁵

Second, examining the reasons for mechanical ventilation among randomized patients may also help in interpreting these neutral findings. The study included a broad population of mechanically ventilated adults. Only one-third of patients were ventilated for respiratory failure, while 25% were ventilated for neurological dysfunction and 25% following cardiac arrest. The trial case mix thus had varying respiratory physiologies, trajectories, and risks of prolonged ventilation. This is reflected in the physiological profile of the cohort with relatively preserved compliance (driving pressure of 13 cm H₂O) and oxygenation (Pao₂/Fio₂ just under 200 mm Hg) in both groups. How could this impact the study findings? Patients ventilated primarily for neurological impairment or following cardiac arrest are often quickly weaned to minimal ventilatory settings but remain ventilated due to depressed level of consciousness rather than respiratory failure.⁶ This creates a challenge for demonstrating improvement in ventilator-free days even if ASV were effective in other situations requiring more ventilator adjustments. Additional end points such as time to successful spontaneous breathing trial and documented reasons for nonextubation would help clarify this situation.

Third, regardless of indication, the majority of patients in this trial received mechanical ventilation for a short time; in both groups, about 20% died and 40% were extubated alive within 3 days of randomization. This is a much higher rate of both early death and early successful extubation compared with an international cohort with respiratory failure.² This raises the question of whether there was enough exposure to the ASV intervention for it to exert any beneficial effect; the benefits of an automated adaptive ventilation system being presumably greater when a longer duration of ventilation provides more impetus for and benefit from adaptation to changing clinical characteristics over time. Of note, these 2 factors—indication for ventilation and duration of ventilation—which in this case may be diluting a possible treatment effect, could be a major threat to trial validity if a noninferiority design were used, as they would systematically bias the results toward the (noninferiority) null.

Fourth, the trial was conducted in centers of excellence with experience in administering mechanical ventilation. For example, screening for readiness to wean and extubate 3 times daily, as implemented in this study, is not standard practice in many ICUs and exceeds usual care.⁷ As a result, the control group in this trial likely represents a form of optimized, expert-delivered conventional ventilation rather than the heterogeneous usual care encountered across most health systems. This again may have obfuscated any treatment effect of ASV.

In this trial, the investigators impressively screened, enrolled, and randomized patients within 1 hour of intubation. Rapid enrollment was accomplished using a deferred consent model that allowed the investigators to initiate the study-assigned ventilation protocol promptly and avoid significant contamination from “wild-type” ventilation for hours or days before enrollment. This process of enrollment aligns with consent processes during usual clinical care, in which technical issues that do not affect patient autonomy, such as ventilator mode, are usually not discussed with patients or their families.⁸ Unfortunately, the rapid enrollment was accompanied by withdrawal of nearly 20% of the randomized cohort when subsequent consent was not obtained. Postrandomization exclusions are a serious threat to study internal validity, particularly in open-label trials.⁹ This situation is akin to a trial with a 20% loss to follow-up for the primary outcome, but it is actually worse because the baseline characteristics of these randomized but subsequently excluded patients are not known. Even if the rates of withdrawal are similar across treatment groups, bias may still be present. We appreciate the ethical principles for requiring consent for use of data, but this requirement diminishes science and benefits neither these particular study patients nor society at large—both have the potential to benefit from a well-conducted study.^10-13

Is it time to simply “set it and forget it”—to let the ventilator do the thinking for us? Not just yet. While closed-loop modes hold significant promise, the field is not ready to put ventilator management on autopilot. The neutral findings of the trial should not be interpreted as evidence against closed-loop ventilation. Rather, one might hypothesize that the effectiveness of such physiologically driven systems may be greater in patients with more severe forms of respiratory failure. In this setting, a demonstration of noninferiority may be enough to drive adoption. Open questions remaining include (1) the most appropriate patient population—likely those with acute respiratory failure, higher elastance, elevated respiratory drive, and a greater risk of ventilator-induced lung injury or diaphragm dysfunction; (2) when to use—early in the course, but likely after some stability of clinical severity has been established; (3) how much human oversight of adaptive ventilation is required; (4) how artificial intelligence should be integrated into closed-loop systems; and (5) how clinician expertise should be maintained while implementing these technological advances.