Computational tools for personalizing treatment of acute respiratory failure, from machine learning to digital twins: a narrative review

Patient-specific computational tools hold great promise for the development of more personalized treatment strategies for acute respiratory failure. Such tools span a continuum from data-driven predictors, to patient-specific mechanistic models, and ultimately to fully realized digital twins with continuous bidirectional model-patient interactions. Data-driven prediction models apply machine learning to large-scale patient datasets to develop tools that can help clinicians identify patients who are likely, or unlikely, to benefit from a particular course of treatment. By incorporating detailed computational representations of disease pathophysiology, patient-specific mechanistic models can provide insights into the effects of existing or novel treatment strategies, support patient stratification and treatment personalization, and enable the design of in silico clinical trials of new interventions. Finally, fully realized dynamic digital twins of patients could provide real-time decision support and 'simulate-before-treat' capabilities at the bedside, helping clinicians optimize treatment as the patient's disease state evolves. This narrative review provides an overview of recent research applying these approaches in the context of acute respiratory failure, encompassing both respiratory and ventilatory support across neonatal, paediatric and adult populations, and pre-hospital, ward and intensive care environments.