Multicenter validation of longitudinal SOFA-based prognostic models for ICU triage during scarcity in mechanically ventilated COVID-19 patients

Abstract The COVID-19 pandemic has highlighted the need for evidence-based allocation of scarce intensive care unit (ICU) resources during periods of extreme capacity strain. Although age-based criteria have been proposed as a last resort non-medical allocation strategy in crises, incorporating longitudinal changes in a patient's clinical condition may improve medical triage. We therefore evaluated whether longitudinal sequential organ failure assessment (SOFA) trajectories could improve prognostic assessment and theoretical ICU resource utilization compared with an age-based approach. We developed and externally validated three logistic regression models using three Dutch ICU cohorts of mechanically ventilated COVID-19 patients admitted during the first waves of the pandemic: (1) SOFA score on day 1, (2) SOFA trajectory up to day 3, and (3) SOFA trajectory up to day 6 after initiation of mechanical ventilation. The models were trained on the development cohort (Maastricht University Medical Center, n = 143) and validated in two hospitals of comparable ICU size, Rijnstate Hospital, Arnhem (n = 216) and St. Antonius Hospital, Nieuwegein (n = 221). Model performance was compared with age-based triage on the date of intubation and assessed using the area under the receiver operating characteristic curve (AUROC), calibration plots, and decision matrices that evaluated the trade-offs between percentage of bed days gained and false positives (patients predicted to die, but who survived). Incorporating longitudinal SOFA data improved discrimination, with AUROC increasing from 0.71 (day 1 model) to 0.78 (day 3) and 0.89 (day 6) in the development cohort, with similar results in the external validation cohorts (AUROC up to 0.82 for the day 6 model). Age-based triage yielded the greatest theoretical capacity gains, but was associated with substantially more false positives. In contrast, longitudinal SOFA assessment up to day 6 resulted in fewer false positives, while preserving clinically relevant theoretical ICU capacity gains. The longitudinal SOFA-based trajectory model improved prognostic discrimination compared with the age-based triage model in critically ill COVID-19 patients. These findings suggest that longitudinal SOFA-assessment up to six days after intubation may improve prognostic support during prolonged ICU scarcity and warrant further evaluation of longitudinal SOFA trajectories as a potential component of ICU triage protocols during future healthcare crises.