The research team, led by Takahiro Masuda of Institute of Science Tokyo Hospital together with biostatistician Akihiro Hirakawa, took a deliberately different analytical route from most earlier work. Previous studies typically reduced the question to a single binary endpoint: did the patient survive or die? That approach, the authors argue, conflates two fundamentally different events. A patient on ECMO can leave the circuit in one of two mutually exclusive ways: first liberation, meaning the first recorded decannulation, or death before liberation. Treating these as competing risks allows researchers to ask whether prolonged pre-cannulation ventilation mainly delays recovery, mainly increases early death, or both.

The technical machinery behind the analysis is substantial. The team estimated cumulative incidence functions with the Aalen-Johansen estimator, fitted cause-specific Cox proportional hazards models stratified by epidemic period, and adjusted for age, sex, and body mass index. Because patients treated in the same prefecture may share referral pathways and practice patterns, the models used robust standard errors clustered by prefecture. Expected time on ECMO was calculated as the restricted mean time in the on-ECMO state over 90 days, obtained by integrating the estimated probability of remaining on support. For the key seven-day comparison, the researchers also performed an inverse-probability-weighted descriptive analysis, with stabilized weights built from age, sex, body mass index, and epidemic period, truncated at the 99th percentile and re-estimated across 500 bootstrap replicates.

The cohort itself reflects the realities of pandemic-era ECMO. Median age was 58 years, nearly 79 percent were men, and the median pre-cannulation ventilation duration was just one day, with more than half of patients cannulated within 24 hours of intubation. Only 147 patients, about 10 percent, had been ventilated for more than seven days before ECMO began. Notably, prolonged ventilation before cannulation became steadily rarer as the pandemic progressed, falling from 12 percent of cannulations in the first epidemic period to just 2 percent in the most recent, a shift the authors attribute to changing cannulation thresholds and evolving standards of care.

The headline finding concerns the pace of liberation. Compared with patients cannulated within one day of intubation, the adjusted hazard ratio for first liberation was 0.87 for those ventilated two to three days, 0.65 for four to seven days, and 0.41 for eight to fourteen days, meaning the rate of leaving ECMO was less than half that of the earliest-cannulated group. The association was overwhelmingly concentrated in the first two weeks after cannulation: during days zero to fourteen, the hazard ratio for the eight-to-fourteen-day category dropped to 0.26. With the seven-day dichotomy, the hazard ratio for liberation was 0.36 in the first fortnight, weakened to 0.63 between days fifteen and thirty, and reached 1.16 after day thirty, indicating that the disadvantage was essentially an early-phase phenomenon.

Expected time on support told a parallel story. Patients cannulated within a day spent an average of 17.4 days on ECMO within the 90-day window, while those ventilated for eight to fourteen days beforehand spent 29.1 days, the longest of any category. Using the seven-day cutpoint that mirrors existing guidance, patients ventilated for more than a week before cannulation spent 26.9 expected days on ECMO versus 18.4 days for those cannulated sooner, a difference of 8.5 days. One third of the prolonged-ventilation group was still on the circuit at day thirty, compared with fewer than 15 percent of the early-cannulation group. For intensive care units, that difference translates directly into additional circuits, oxygenators, beds, and specialized nursing hours.

Death before liberation showed a strikingly different pattern. Although the exposure categories were jointly associated with the death rate, there was no monotonic gradient below fifteen days: hazard ratios hovered between 1.11 and 1.31 across the two-to-three, four-to-seven, and eight-to-fourteen-day groups. Only the small group ventilated for at least fifteen days, just 32 patients with 15 first liberations, showed a clearly elevated hazard of 2.51. The authors are emphatic that this endpoint is not overall mortality: median observation after liberation was only seven days, and just 5.3 percent of liberated patients were followed through day ninety, so deaths after decannulation could not be reliably compared.

The researchers are equally careful to defuse a potentially dangerous misreading. This cohort is conditioned on having received ECMO, so patients considered for support after prolonged ventilation but never cannulated are invisible to the analysis. A slower liberation rate is not futility: 47 percent of patients in the eight-to-fourteen-day category were still liberated within ninety days, and death before liberation did not rise steadily below fifteen days. The relevant Extracorporeal Life Support Organization criterion also combines ventilation duration with high ventilator settings, and plateau pressure was not recorded in the registry, so that criterion could not be evaluated directly. The study, in other words, identifies no threshold beyond which ECMO should be withheld.

Confounding remains the central interpretive challenge. Pre-cannulation ventilation duration is an indirect marker that may reflect the pace of lung injury, delayed referral, center-level practice, and the selection of patients judged likely to remain eligible despite waiting. The registry did not capture respiratory mechanics, ventilator intensity, vasopressor requirements, comorbidities, or vaccination status. Still, two observations argue against severity alone explaining the results: measured hypoxemia at cannulation did not worsen monotonically with ventilation duration, with the lowest median PaO2/FiO2 ratio actually in the earliest-cannulated group, and adjusting for gas exchange and positive end-expiratory pressure did not materially change the estimates. Prolonged injurious ventilation is biologically plausible as a driver of slower recovery, but the registry cannot establish that mechanism.

What the study offers, the authors conclude, is a forecasting tool rather than a gatekeeping rule. Knowing that a patient ventilated for more than a week will likely need roughly eight and a half additional days of circuit support helps clinicians plan staffing, anticipate resource occupancy, and give families honest expectations about the pace of separation from the machine. During pandemic surges, when ECMO capacity was rationed and contested, such numbers carried life-and-death weight for system planning; in routine practice, they remain inherent to running any ECMO program. The message is neither that late cannulation is harmless nor that it is hopeless, but that time on the ventilator before cannulation leaves a measurable imprint on the entire trajectory that follows.

Subject of Research: Pre-cannulation ventilation duration and outcomes of venovenous ECMO in COVID-19-related acute respiratory distress syndrome

Article Title: Pre-cannulation ventilation duration and the trajectory of venovenous extracorporeal membrane oxygenation in COVID-19-related acute respiratory distress syndrome: a nationwide competing-risks cohort study

Article References: Masuda, T., Hirakawa, A., & on behalf of Japan ECMOnet for COVID-19 (2026). Pre-cannulation ventilation duration and the trajectory of venovenous extracorporeal membrane oxygenation in COVID-19-related acute respiratory distress syndrome: a nationwide competing-risks cohort study. Journal of Artificial Organs, 29(4), Article 67. https://doi.org/10.1007/s10047-026-01600-8

Image Credits: AI Generated

DOI: 10.1007/s10047-026-01600-8

Keywords: ECMO, COVID-19, acute respiratory distress syndrome, mechanical ventilation, competing risks, liberation, decannulation, critical care, Japan registry, prognosis, extracorporeal life support, intensive care