Autonomous Oxygen Titration in Acutely Ill Adults: What the SAVE-O₂ AI Trial Actually Demonstrated

Hospital patient receiving supplemental oxygen via nasal cannula with pulse oximeter monitoring and automated oxygen titration device — illustrating the SAVE-O₂ AI randomized clinical trial on autonomous oxygen management in acutely ill adults.

SOLEIL RESEARCH HIGHLIGHT
Monday Edition | August 17, 2026 | Soleil Research Editorial Team | 10 min read

AT A GLANCE

Closed-loop autonomous oxygen titration using the O2matic PRO100 substantially increased the proportion of time acutely ill adults spent within a target normoxemia range compared with manual titration — but the trial was not powered to establish effects on mortality, respiratory failure, or hospital length of stay.

  • 300 participants across 4 tertiary hospitals
  • +21 percentage points in normoxemia time
  • Hyperoxemia: 9.2% vs 29.1%
  • No established hard clinical-outcome benefit

PLAIN-LANGUAGE SUMMARY

In hospitals, supplemental oxygen is commonly adjusted manually by clinical staff to keep patients within a target range. Too little oxygen (hypoxemia) and too much oxygen (hyperoxemia) can both be undesirable. This trial tested whether a device that automatically adjusts oxygen flow using real-time pulse oximetry readings could keep patients more consistently within a target range than standard manual care.

For that specific question, the answer was yes: the autonomous system performed substantially better at maintaining the target oxygen range. What the trial did not show — and was not designed or powered to show — is whether this improvement translates into better survival, fewer complications, or shorter hospitalization.

STUDY OVERVIEW

The SAVE-O₂ AI trial was a multicenter, individually randomized, parallel-group, open-label trial conducted across four tertiary hospitals in the United States. Three hundred adults hospitalized with acute respiratory illness, trauma, burns, or acute surgery who required 1–10 L/min of supplemental oxygen were assigned 1:1 to autonomous titration with the O2matic PRO100 or standard manual oxygen titration for up to 72 hours.

The trial was prospectively registered, used a prespecified protocol and statistical analysis plan, and operated under an FDA investigational device exemption. The primary endpoint was the proportion of time within the target normoxemia range (SpO₂ 90%–96%). The key secondary endpoint was time spent in hypoxemia (SpO₂ <88%).

Methodological note: A small proportion of time with SpO₂ above 96% while patients were already breathing room air (FiO₂ 21%) was considered nonmodifiable and incorporated with normoxemia in the analysis; this represented approximately 2% of observed time.

KEY FINDINGS

  • Autonomous titration achieved normoxemia 85% of the time vs 63% with manual titration — adjusted difference +21 percentage points (95% CI 18–25; P<.001).
  • Time in hypoxemia was 2.0% vs 3.6% — adjusted difference −1.3 percentage points (95% CI −2.0 to −0.5; P=.002).
  • Time in hyperoxemia was 9.2% vs 29.1%.
  • 28-day in-hospital mortality was 4 vs 5 patients (HR 1.03; 95% CI 0.22–4.84). This was exploratory, and the trial was not powered for mortality.
  • Exploratory clinical outcomes, including mortality, respiratory failure, hospital-free days, and time to room air, did not provide evidence of clinical benefit; the trial was not powered to establish effects on these outcomes.

STRENGTHS

  • Prospectively registered randomized clinical trial with prespecified protocol and SAP
  • Multicenter design across four tertiary hospitals
  • FDA investigational device exemption and regulatory oversight
  • Clear distinction between the primary physiological endpoint and exploratory clinical outcomes
  • Transparent funding and conflict-of-interest disclosures

LIMITATIONS

  • Open-label design: knowledge of treatment assignment could have influenced clinical behavior in the manual-titration group
  • Only one device, the O2matic PRO100, was evaluated; results cannot automatically be generalized to other closed-loop oxygen systems
  • The study was not powered to establish effects on mortality, respiratory failure, hospital-free days, or time to room air
  • Autonomous titration depends on the accuracy of the pulse oximeter signal; automation does not correct sensor bias, imprecision, or measurement error
  • Participants required 1–10 L/min of supplemental oxygen; applicability to higher-acuity respiratory support settings is not established

FUNDING

The study was sponsored by the Defense Health Agency Research and Engineering Directorate, Combat Casualty Care Portfolio, through the Medical Technology Enterprise Consortium. Additional support was provided by the National Center for Advancing Translational Sciences under award U24 TR004437. IDTS Medical served as sponsor of the FDA investigational device exemption, and PRO100 devices were rented from O2matic. The publication states that the funders and device sponsor did not participate in study design, data analysis, interpretation, manuscript preparation, or the decision to publish.

CONFLICTS OF INTEREST

The publication reports individual author disclosures that include external research grants and personal fees or honoraria. These disclosures are separate from the study funding and should be considered when interpreting the work. Full author-by-author disclosures are provided in the primary publication.

☀️ SOLEIL INSIGHT

The SAVE-O₂ AI trial demonstrates something important — and something the headline alone cannot convey.

Closed-loop automation substantially improved control of a clinically meaningful physiological variable. That is a real finding, and it matters: maintaining normoxemia while reducing both hypoxemia and hyperoxemia is a legitimate therapeutic target in acutely ill patients.

But automation performance is not the same as patient benefit. The trial was not designed or powered to determine whether better normoxemia control translates into reduced mortality, fewer complications, or shorter hospitalization. Those questions remain open.

There is a second layer worth examining. The PRO100 system titrates oxygen based on SpO₂ readings from a pulse oximeter. Automating the response to that signal does not make the signal itself more accurate. Pulse oximetry has known limitations, and the automation layer inherits rather than resolves measurement error or bias.

This is what responsible evaluation of AI and automated systems in medicine requires: distinguishing what the system controls from what the system improves, and what the measurement provides from what it cannot guarantee.

Automation can optimize a process. It cannot substitute for the validity of the measurement it depends on.

REFERENCE

Douin DJ, Rice JD, Xiao M, et al. Autonomous Oxygen Titration for Maintaining Normoxemia in Acutely Ill Adults: The SAVE-O₂ AI Randomized Clinical Trial. JAMA Internal Medicine. Published online August 3, 2026. doi: 10.1001/jamainternmed.2026.4023

Soleil Research™ — Advancing Science. Empowering Humanity. ☀️