NutriSighT: Interpretable Transformer Model for Dynamic Prediction of Underfeeding Enteral Nutrition in Mechanically Ventilated Patients

AbstractAchieving adequate enteral nutrition among mechanically ventilated patients is challenging, yet critical. We develop NutriSighT, a transformer model using learnable positional encodings to predict which patients would be underfed (receive less than 70% daily caloric requirements) between days 3-7 of mechanical ventilation and compared its performance against XGBoost. Using retrospective data from two ICU databases (3284 patients from AmsterdamUMCdb for development and 6456 from MIMIC-IV for external validation), we included adults mechanically ventilated for at least 72 h. NutriSighT achieved AUROC of 0.81 (95% CI: 0.81 – 0.82) and AUPRC of 0.70 (95% CI: 0.70 – 0.72) internally. External validation yielded AUROC of 0.76 (95% CI: 0.75 – 0.76) and an AUPRC of 0.70 (95% CI: 0.69 – 0.70). In comparison, XGBoost achieved AUROC of 0.58 (95% CI: 0.58 – 0.59) and AUPRC of 0.48 (95% CI: 0.46 – 0.50). This approach may help clinicians personalize nutritional therapy in critical care. IntroductionOptimal enteral nutrition (EN) is vital for critically ill patients requiring mechanical ventilation to meet their metabolic needs while mitigating complications1,2. Critical care guidelines recommend initiating early enteral nutrition in critically ill patients, but there is heterogeneity in the recommended caloric targets for the first week of intensive care unit (ICU) stay. For example, the European Society of Parenteral and Enteral Nutrition (ESPEN) advocates for underfeeding, as defined by receipt of nutrition with less than 70% daily caloric requirements during the first week of ICU stay, but the American Society of Parenteral and Enteral Nutrition (ASPEN) recommends a broader caloric intake range of 12 to 25 kcal/kg, encompassing both underfeeding and adequate nutrition3,4. This variation reflects the dynamic nature of critical illness and the challenges in determining optimal nutrition within this highly heterogenous patient population5,6,7,8. Delivering adequate nutrition in this population is further complicated by challenges such as gastrointestinal dysfunction, hemodynamic instability, and frequent interruptions for procedures9,10,11.The first week of critical illness is divided into two distinct phases: the early acute period and the late acute period1,12. The early acute period spans the 48 h of critical illness and is marked by hemodynamic instability and acute illness response. The late acute period, spanning days 3-7, is characterized by muscle wasting and evolving nutritional needs though the timing of this transition likely varies amongst patients. Common clinical practice is to start a form of restrictive dose EN, such as trophic dose feeding, during the first 48 h and progressively increase nutritional support during the late acute period to meet the evolving metabolic demands of critically ill patients.However, these strategies are not persona Read More