Submitted by A.S. Quenault on Tue, 12/05/2026 - 10:12
A major randomized, multicentre, adaptive trial evaluating high-flow nasal therapy (HFNT) after cardiac surgery — the Nasal Oxygen Therapy After Cardiac Surgery (NOTACS) trial —has recently published its final clinical results in JAMA Network Open. While the trial concluded that there was no statistically significant clinical difference between HFNT and standard oxygen therapy (SOT), the true legacy of NOTACS lies in its immense contribution to clinical trial methodology.
The international trial, led by the Royal Papworth Hospital NHS Foundation Trust, with methodological input from researchers at the MRC Biostatistics Unit (BSU), found that standard oxygen therapy is as effective as high-flow nasal oxygen therapy in the recovery of patients who had cardiac surgery. The trial was NIHR-funded in the UK and supported by generous local grants in Australia and New Zealand from the Medical Research Future Fund Australia and Heart Lung Foundation New Zealand (previously the Green Lane Research and Education Fund), respectively.
It is the largest study of its kind and will now guide best practice worldwide for people having major heart operations. It aimed to see if high-flow nasal oxygen therapy (HFNOT) - which sees oxygen gently pumped through the nose immediately post-surgery – led to shorter lengths of stay in hospital for patients and if it reduced readmissions to hospital by preventing chest infections and other respiratory complications, compared to standard oxygen therapy (SOT).
A total of 1,280 adults participated across 17 hospitals in the UK (755 participants), Australia (370) and New Zealand (155). All participants had heart surgery and were randomly assigned to receive either HFNOT or SOT for at least 16 hours immediately after surgery.
The results show that HFNOT did not improve recovery for patients, compared with SOT:
- Both groups had the same number of days at home without extra help. In fact, most people in both groups needed some help after their surgery. This means the type of oxygen they received didn’t make a difference.
- When looking at time spent at home, both groups did equally well. On average, people in both groups spent about 82 out of 90 days at home in the 90 days after surgery.
- Rates of complications, hospital readmissions, and quality of life were similar between groups.
For the researchers from the BSU – Sofía S. Villar, Dominique-Laurent Couturier, Mia S. Tackney, Sarah Dawson and Letao Yuan, the NOTACS trial served as a vital testing ground for analysing complex, patient-centric trial endpoints. The trial's primary outcome was "Days alive and at home" at 90 days (DAH90), a composite metric that combines initial hospital stay, readmissions, and mortality into a single score. Because DAH90 data is incredibly complex—often zero-inflated, left-skewed, and bimodal—it presents massive challenges for trial design, sample size calculations, and data analysis.
To solve these challenges, the researchers are announcing two upcoming methodological papers that will fundamentally change how future clinical trials model outcomes and handle missing data.
1. Redefining Trial Modelling and Sample Size Calculations In an upcoming paper titled "Beyond the Composite: Enhancing Trial Analysis through a Divide and Conquer Approach to 'Days Alive and at Home'", researchers Letao Yuan, Sofía S. Villar, and Dominique-Laurent Couturier introduce a novel framework for modelling complex endpoints.
Rather than treating DAH as a single, condensed metric, the BSU team developed a "Divide and Conquer" strategy that breaks the endpoint down into separate components (e.g., minimum hospital stay, extended stay, readmission, and death) which are modelled separately. The researchers demonstrate that this approach significantly improves model fit compared to existing alternatives. Crucially, this breakthrough allows for highly accurate DAH data generation, which future trialists can use to perform robust, simulation-based sample size calculations and properly evaluate the operating characteristics of their statistical tests.
2. Solving the Missing Data Problem to Prevent Type I Error Inflation In a second forthcoming paper titled "Component over Composite: Mitigating Type I Error Inflation when Imputing 'Days Alive and at Home'", researchers Mia S. Tackney, Sarah Dawson, Letao Yuan, Dominique-Laurent Couturier, and Sofía S. Villar tackle the critical issue of missing data. Because DAH combines hospital data (which is usually complete) with patient location diaries (which are prone to missingness), handling missing values is exceptionally difficult. The BSU team’s simulation study revealed a stark warning for the scientific community: using a naive approach—such as defining missingness at the composite level and directly imputing it with Predictive Mean Matching—can lead to severe Type I error inflation (false positives). To safeguard the integrity of future trials, the authors demonstrate that using the “Divide-and-Conquer” strategy by performing Multiple Imputation (MI) at the component level, successfully controls Type I error rate while maintaining statistical power.
Looking ahead, while the clinical intervention tested in NOTACS did not yield a significant patient benefit, the trial data has catalyzed a leap forward in biostatistics. The novel "Divide and Conquer" modelling and the insights into component-level imputation, pioneered by the MRC BSU, will equip future trialists wishing to use a recent patient centred outcome with the tools they need to design more efficient, accurate, and robust clinical trials worldwide.
Professor Sofía S. Villar, Group Leader at the MRC Biostatistics Unit and one of the researchers leading the methodological work, said:
"The NOTACS trial presented us with unique challenges, specifically the choice of a patient-centred primary outcome: 'days alive and at home.' This is a measure that truly mattered to patients, but it required a sophisticated design to be statistically robust.
We took these challenges on board and developed an adaptive trial design that accounted for the inherent uncertainty of scaling up from a pilot study to an international multi-centre trial. While we have learned a tremendous amount along the way, we hope that by sharing our experiences and our methodological framework for design and analysis of such outcome measures we can make it easier for future trials to adopt them in a way that is both interpretable and scientifically rigorous."