Background: The Patient Outcome Assessment and Decision Support (PADS) model is a real-time framework designed to predict both mortality and the likelihood of discharge within 48 h in critically ill patients. By combining these predictions, PADS enables clinically meaningful stratification of patient trajectories, supporting bedside decision-making and the planning of critical care resources such as nursing allocation and surgical scheduling. Methods: PADS integrates routinely collected clinical data: SOFA variables, age, gender, admission type, and comorbidities. It consists of two Long Short-Term Memory (LSTM) neural networks—one predicting the probability of death and the other the probability of discharge within 48 h. The combination places each patient into one of four states: alive/discharged within 48 h, alive/not discharged, dead within 48 h, or dead later. The model was trained using MIMIC-IV data, emphasizing ease of implementation in units with electronic health records. Out of the 76,540 stays present in MIMIC-IV (53,150 patients), 32,875 (25,555 patients) were used after excluding those with short stays (<48 h) or life support treatment limitations. The code is open, well-documented, and designed for reproducibility and external validation. Results: The model achieved strong performance: AUCROC of 0.94 (±0.03) for mortality and 0.89 (±0.07) for discharge on training data, and 0.87 (±0.02) and 0.88 (±0.03), respectively, on the test set. As a comparison, benchmark models obtain worse accuracy (−13.4% for APS III, −19% for OASIS, and −7.4% for SAPS II). Predictions are visualized in an intuitive format to support clinical interpretation. Conclusions: PADS offers a transparent, reproducible, and practical tool that supports both individual patient care and the strategic organization of intensive care resources by anticipating short-term outcomes.

Background: PADS combines two neural networks predicting ICU mortality and discharge within 48 h, placing critically ill patients into one of four clinically meaningful states. Developed on MIMIC-IV alone, it left open whether it generalizes to other ICUs, whether its models transfer across hospitals, and whether its predictions can be explained at the bedside. Methods: We evaluated PADS on four ICU databases from different hospitals and countries (MIMIC-IV, AmsterdamUMCdb, eICU-CRD, and HiRID), using the same routinely collected variables. Mortality is scored on the final 48-h window (terminal-window, not early-warning, discrimination). For each external database, we compared the MIMIC model used as-is, retrained from scratch, and retrained from the MIMIC weights, and added an explainability layer. Results: For mortality, reusing and retraining the MIMIC model gave the highest discrimination on every database (AUROC 0.955–0.986; terminal-window (near-outcome) discrimination) and stabilized training; used as-is, it ranged from chance (Amsterdam) to good (eICU, HiRID). For discharge, training fresh on local data matched or beat reusing MIMIC on every external database, consistent with discharge timing depending on local organization rather than physiology. The explainability layer produced clinically coherent, cross-checked explanations. Conclusions: Transportability was task-dependent: mortality transferred between hospitals, discharge did not. PADS demonstrated promising external transportability across heterogeneous ICU databases, particularly after local adaptation. Reusing and adapting the MIMIC-IV mortality model across hospitals improves accuracy. This approach also stabilizes training, providing a basis for potential federated deployment, whereas discharge is better trained locally. The mortality results reported here are terminal-window discrimination and do not support use of the framework as an early-warning model. A transparent explainability layer provides an interpretable representation of model predictions, addressing a key barrier to clinical adoption.