Most critical care trials studying mortality have inconclusive results due to factors like ineffective interventions, small sample sizes, treatment heterogeneity, and unrealistic assumptions about effect sizes in power calculations. Determining sample size requires balancing statistical power with feasibility. Many ICU trials are underpowered, using mortality as the primary outcome despite often unrealistic expectations of treatment benefits. Alternatively, patient-centred outcomes should be considered.
ECMO patients are particularly hard to recruit for trials, leading a recent study to evaluate the DOSE (daily organ support for patients on ECMO) outcome as a new longitudinal ordinal measure, aiming to validate it against long-term patient-centred outcomes and determine the optimal duration for daily data collection.
This study used data from the Australian EXCEL registry on adult ICU patients receiving ECMO between February 15, 2019, and September 13, 2023. Survivors were contacted for 6- and 12-month telephone interviews for patient-reported outcomes.
The DOSE outcome was developed with input from ECMO survivors, researchers, statisticians, and ICU/ECMO specialists and reviewed at scientific meetings. It ranks patient status daily for 90 days after ECMO initiation on a 6-level hierarchy (from worst to best): 1) death, 2) on ECMO, 3) invasively ventilated off ECMO, 4) in ICU not ventilated, 5) in hospital ward, 6) discharged alive. If multiple statuses occurred in a day, the worst category was recorded, creating a longitudinal ordinal outcome for trials. The DOSE outcome was validated against death or new disability at 6 and 12 months (defined as death or a ≥10% increase in WHODAS score from baseline).
Of 1,375 patients in the registry, 1,372 (99.8%) were included across 28 centres. Patients had a median age of 52, 34% were women, and most (91%) were mechanically ventilated before ECMO. ECMO types were VA (45%), VV (35%), and eCPR (20%). Common comorbidities included diabetes (18%) and heart failure (11%). Median ECMO duration was 6 days. At 6 and 12 months, mortality was 46% and 49%, respectively, while death or new disability occurred in 73% at 6 months and 74% at 12 months. By Day 20, 4% (VA ECMO), 27% (VV ECMO), and 3% (eCPR) of patients remained on ECMO. By Day 90, mortality rates were 42% (VA ECMO), 29% (VV ECMO), and 66% (eCPR).
The DOSE outcome’s ability to predict death or new disability at six months improved over time, with AUROC >0.800 after Day 16 for all ECMO types and minimal gains after Day 28. Results were similar for 12-month outcomes. DOSE outperformed ECMO duration, ventilation duration, and ICU length of stay for prediction. It showed moderate correlation with death or new disability (confirming convergent validity) and negligible correlation with BMI (confirming divergent validity).
Compared to 28-day mortality, both static and longitudinal DOSE outcomes significantly increased power in trial simulations. For VV ECMO, achieving 80% power with an OR of 1.30 required ~150 patients/arm using DOSE versus 600/arm using 28-day mortality, saving 550–900 patients in total. Longitudinal DOSE increased power by up to 11% over static DOSE, further reducing required sample sizes. For example, with an OR of 1.20, ~300/arm (longitudinal DOSE) vs. 375/arm (static DOSE) were needed in VV ECMO. An OR of 1.15 in longitudinal DOSE translated to a 3–5% absolute mortality reduction, demonstrating efficiency and feasibility advantages for clinical trials.
This study introduces the DOSE longitudinal ordinal outcome for ECMO trials, capturing six daily categories from death to hospital discharge. DOSE showed strong predictive performance (AUROC >0.800) for death or new disability at 6 and 12 months across ECMO types, with daily collection up to Day 28 sufficient for prediction while reducing workload. Compared to static DOSE and 28-day mortality, longitudinal DOSE increased trial power. As ICU trials focused solely on mortality often yield inconclusive results due to small effect sizes, DOSE offers a patient-centred, efficient alternative for evaluating long-term outcomes in ECMO trials.
Clinical trials often collect detailed daily data but underuse it, particularly in critical care. While daily patient status has been used in COVID-19 and coronary disease trials, ICU trials typically use derived outcomes (e.g., ventilator-free days) that lose granular information. Leveraging full daily data with longitudinal models can increase statistical power, reduce required sample sizes, and better handle outcomes like death. Compared to methods like win-ratio or multistate models, the approach in this study effectively incorporates event timing, avoids ranking event importance, provides clear treatment effect estimates, and manages missing data.
The DOSE scale should employ the estimand framework to clearly define the clinical question and align analysis and power calculations accordingly. The composite outcome incorporates death and organ support trajectory, reflecting survival and physical recovery within 28 days. DOSE predicts physical impairments well but is less predictive of cognitive and social outcomes, which involve longer-term recovery processes. This highlights the need for separate predictive models and interventions addressing both physical and cognitive recovery after ICU. Future research could identify earlier markers to better predict cognitive and social outcomes alongside physical recovery.
Source: AJRCCM
Image Credit: iStock
