Editorial
Patient, Treatment, Outcome, Large Simple Trials of Common Therapies
Kevin P. Seitz, Jonathan D. Casey, Matthew W. Semler
JAMA Published Online: June 12, 2025
doi: 10.1001/jama.2025.9657
Each year, as many as 20 million critically ill adults receive mechanical ventilation worldwide, and approximately 1 in 3 die before hospital discharge.1-3 For every patient receiving mechanical ventilation, the fraction of inspired oxygen (Fio2) must be titrated to maintain arterial oxygen saturation. Targeting higher oxygen saturations (96% to 100%) provides a margin of safety against hypoxemia but may increase exposure to excess Fio2, hyperoxemia, and tissue hyperoxia, which can cause oxidative damage, inflammation, and organ injury. Targeting lower oxygen saturations (88% to 92%) may minimize these risks but may increase exposure to hypoxemia and tissue hypoxia. Yet for decades, no randomized trial examined which approach to oxygen therapy produces the best patient outcomes.1 Why?
Historically, the design, conduct, and oversight of randomized trials have centered on the development of new drugs and devices. Such trials aim to protect participants from the unknown risks of experimental therapies that would not be received outside of research. They often cost tens or hundreds of millions of dollars to conduct—a sensible investment for a pharmaceutical company aiming to profit from a new drug, but a potentially insurmountable barrier to the examination of common and inexpensive therapies that patients already receive in clinical care.4 As a result, up to 90% of treatments that are recommended by guidelines and received by patients in clinical care have never been evaluated in a randomized trial; some of these treatments may be ineffective or even harmful.5 Addressing this profound moral problem requires new methods for highly efficient trials comparing the effectiveness of existing treatments.
In this issue of JAMA, Martin and colleagues6 demonstrate one such method: the large simple trial embedded within clinical care.7 The Intensive Care Unit Randomized Trial Comparing Two Approaches to Oxygen Therapy (UK-ROX) compared the use of a lower oxygen saturation target of 90% (range, 88% to 92%) to usual care among 16 500 adults receiving invasive mechanical ventilation in 97 intensive care units in the UK. Patients were enrolled early in their course of mechanical ventilation (median, 5 hours), were severely ill (median predicted risk of death, 35%), had a range of critical illnesses (eg, >5000 patients with sepsis and >1500 patients with hypoxic-ischemic encephalopathy), and were experiencing significant hypoxemia (eg, >11 000 patients with a Pao2:Fio2ratio consistent with acute respiratory distress syndrome). The primary outcome of all-cause, all-location mortality by 90 days did not differ between the lower oxygen saturation target group (35.4%) and the usual care group (34.9%; adjusted absolute risk difference, 0.7 percentage points; 95% CI, −0.7 to 2.0).
To fully understand what the UK-ROX trial means both for oxygen therapy in critical illness and for future research evaluating common treatments, it is helpful to consider the 3 core aspects of a large simple trial (enrollment, separation between groups, and outcome collection) and a future opportunity for such trials (personalization).
Because tens of millions of patients receive oxygen therapy during mechanical ventilation each year, even a small difference in outcomes between the available approaches could save thousands of lives. To be able to detect small but important between-group differences in outcomes, the UK-ROX trial enrolled 16 500 patients, making it one of the largest patient-level randomized trials in the history of critical care. The trial achieved this level of enrollment by embedding screening, eligibility assessment, and randomization within clinical care. Because research personnel could not feasibly obtain written informed consent prior to enrollment, a waiver of the requirement for informed consent was granted. This allowed treating clinicians to identify, enroll, and randomize their eligible patients. Research personnel subsequently provided patients or their representatives with an opportunity to discontinue participation and requested consent to access outcome data from a preexisting registry. Only 66 of the 16 500 participants (0.4%) requested the removal of their data. This ethical framework was well-matched to the low- or minimal-risk nature of randomizing patients to one of two standard-of-care strategies during emergency care. Embedding enrollment within clinical care resulted in both a highly representative trial population and an almost unprecedented level of efficiency in trial conduct. Having enrolled 16 500 patients over 44 months (375 patients per month) at a total cost of approximately $2.3 million US ($140 US dollars per patient) makes UK-ROX approximately 300 times as cost-efficient as the average clinical trial in the US (for which the median total cost is $40 000 per patient).4,8
To meaningfully compare outcomes between groups in a trial, the treatments that patients receive in each group must be sufficiently different. This requires (1) that participants experience sufficient exposure to the treatments (eg, sufficient duration of oxygen therapy during mechanical ventilation), (2) that the treatments assigned in each trial group be sufficiently distinct (eg, the oxygen saturation targets differ by a clinically meaningful amount), and (3) that patients and clinicians adhere to the assigned treatment (eg, oxygen is consistently titrated to a target of 88% to 92% in the lower oxygen saturation target group). In the lower oxygen saturation target group of UK-ROX, patients’ oxygen saturation values were adherent to the assigned target 89.4% of the time (at least among the 15.1% of participants for whom data about oxygen therapy were collected; collecting data on oxygen therapy for all participants would have been preferable but was determined to be logistically infeasible). This high level of adherence resulted in a mean oxygen saturation of 93.3%, which is comparable with mean oxygen saturations achieved in the lower target groups of prior trials.3,9,10
Despite this strong adherence to the lower oxygen saturation target, the differences in oxygen saturation (1.9%) and Fio2 (0.04) between the trial groups in UK-ROX were about half the magnitude of some prior trials.2,9 This is because, unlike prior trials that compared discrete, nonoverlapping lower and higher oxygen saturation targets (eg, 88% to 92% vs 96% to 100%),2,3,9-11 the UK-ROX trial compared the use of a lower oxygen saturation target vs usual care. Usual care often is not a single treatment strategy but a mix of practice patterns that vary by clinician, unit, hospital, and region.12 The mean oxygen saturation of 95.1% in the usual care group of the UK-ROX trial was lower than in the usual care groups of prior trials13 and nearly identical to the mean oxygen saturation of 95.0% among patients treated with an intermediate target of 92% to 96% in a recent trial.3 This may diminish the trial’s ability to test its foundational hypothesis that “too much oxygen risks damaging the lungs and other vital organs” (trial protocol)6 and may limit the generalizability of the neutral findings to settings in which usual care includes exposure to higher oxygen saturations.
Large simple trials may be facilitated by the collection of outcome data from preexisting, rigorously curated clinical, administrative, or governmental registries or databases. For all patients, the UK-ROX trial collected data on the primary outcome from the Civil Registrations of Death, a federal database of all deaths in England and Wales. This ensured complete data on the primary outcome for 16 394 patients (99.8%) without the costs of traditional after discharge follow-up. For the 15 058 patients (91.6%) who consented to data linkage, ICU and hospital outcomes were extracted from the Case Mix Programme dataset, a national clinical audit of patient outcomes from adult critical care in England, Northern Ireland, and Wales. Especially in settings without integrated electronic health records, obtaining patient outcomes from existing longitudinal registries may be a crucial efficiency for large, comparative effectiveness trials.
Like previous smaller trials, UK-ROX found that use of a lower oxygen saturation target was safe, but did not improve outcomes, overall2,3,9,11,14 or in any prespecified subgroup.2,3,9,11 The optimal oxygenation target, however, may differ for different patients based on their individual characteristics. In a recent secondary analysis, a machine learning model that calculates the predicted effect of treatment with a lower vs higher oxygen saturation target on mortality for individual patients based on 24 baseline characteristics was derived in one randomized trial and validated in a second randomized trial.15 Even though neither trial demonstrated a significant average treatment effect (like UK-ROX), the treatment effects for individual patients ranged from a decrease in mortality of more than 20% with use of a higher target to a decrease in mortality of more than 20% with use of a lower target. Future research should evaluate whether this model, or others, can identify patients in the UK-ROX trial who experienced benefit or harm from the lower oxygen saturation target and evaluate whether the use of personalized oxygenation targets derived from randomized trial data can improve outcomes compared with clinical care. By enrolling large trial populations that represent the full diversity of patients in clinical care, collecting detailed data on baseline characteristics that may modify the effect of treatment on outcomes and analyzing the effect of treatment on outcomes for individual patients (rather than simply on average), future large simple trials may help deliver medicine that is both evidence-based and personalized.
For decades patients and clinicians lacked evidence to inform the basic clinical question of which approach to oxygen therapy results in the best outcomes for critically ill patients receiving mechanical ventilation. Together with the results from recent smaller trials,2,3,9-11,13,14 the findings of UK-ROX make clear that, for critically ill adults receiving mechanical ventilation, the use of an oxygen saturation target as low as 90% is safe but does not improve outcomes overall. The UK-ROX trial also demonstrates that by enrolling large and representative trial populations using fit-for-purposes regulatory approaches, collaborating with clinicians to achieve separation between groups, and leveraging existing sources of outcome data, large simple trials embedded within clinical care can help us finally fulfill our ethical obligation to understand the effects of common treatments on patient outcomes.
