DINIRS: Digital Twin for Individualized Treatment Effects of Non-Invasive Respiratory Support Strategies

Abstract

Objective: Choosing between noninvasive respiratory support (NIRS) and invasive mechanical ventilation (IMV) for patients with acute respiratory failure is a complex, time-sensitive decision with heterogeneous treatment effects across patient subgroups. Although clinical trials and guidelines provide population-level guidance, it remains unclear which patients benefit more from NIRS than IMV. We developed and validated a censoring-aware Digital Twin framework for Individualized Treatment Effects of Non-Invasive Respiratory Support (DINIRS) to estimate individualized treatment effects (ITEs) that capture mortality and ventilation duration. Materials and Methods: We trained DINIRS on 23 baseline clinical variables from the first 24 ICU hours in 5,336 MIMIC-IV patients with acute respiratory failure. We used a transformer encoder with a survival attention gate to decompose 28-day ventilator-free days (VFD-28) into survival probability and conditional ventilation duration. A cross-fitted, doubly robust learner estimated ITEs. We externally validated DINIRS in 2,540 patients from the multi-site eICU-CRD dataset. Results: The DINIRS policy achieved a mean benefit of 2.07 ventilator-free days per patient (207 per 100 patients) compared with the observed practice. Predicted NIRS benefit was higher among patients with less organ dysfunction (88.4% versus 49.0%) and persisted across hypoxemia severity. External validation reproduced this pattern without retraining. Discussion: Our analysis revealed that the NIRS benefit stemmed from shorter ventilation among survivors rather than from reduced mortality, indicating that avoiding intubation-associated complications was the primary mechanism. Conclusion: This study demonstrated individualized estimation of NIRS benefit using ICU data, though prospective validation is needed before these estimates inform treatment decisions.