ABSTRACT Background: Increased extravascular lung water (EVLW) induced by systemic inflammatory response under cardiopulmonary bypass (CPB) and diaphragmatic dysfunction due to phrenic nerve injury during cardiac surgery leads to weaning failure from mechanical ventilation (MV) in pediatric patients undergoing cardiac surgery. We hypothesized that ultrasound measurement of EVLW shown by B-lines and diaphragm function in the form of thickening and excursion will be able to predict weaning failure defined as reintubation within 48 hours of endotracheal extubation in such patients. Methods: Fifty patients aged (1 month to 18 years) undergoing congenital cardiac surgeries were enrolled in the study. The ultrasound measurement of B-line of lung, diaphragm excursion (DE) and diaphragm thickness (DT) were done preoperatively, on pressure support ventilation (PSV) during weaning from mechanical ventilation (MV) and 4 hours after extubation. Results: 7 out of 50 patients had weaning failure. The patients with weaning failure (group 1, n=7) were younger, with median age of 1 year (0.25-7) compared to those who tolerated weaning (group 2, n=43), median age of 3 years (0.25-17), p=0.040. The B-line score in group 1 increased from a preoperative score of 0 to post-extubation period score of 2, the score being significantly higher than the patients of group 2 (p=0.024). The left diaphragm thickening fraction of <17.15% predicted weaning failure with a sensitivity of 85%, specificity of 51.4%, (AUC ROC 0.75, p= 0.032). Conclusion: LUS cannot predict weaning failure. The diaphragmatic thickening fraction <17.15% was found to be a predictor of weaning failure in our patients.

The quality of machine learning (ML) models deployed in dynamic environments tends to decline over time due to disparities between the data used for training and the upcoming data available for prediction, which is commonly known as drift. Therefore, it is important for ML models to be capable of detecting any changes or drift in the data distribution and updating the ML model accordingly. This study presents various drift detection techniques to identify drift in the survival outcomes of patients who underwent cardiac surgery. Additionally, this study proposes several drift adaptation strategies, such as adaptive learning, incremental learning, and ensemble learning. Through a detailed analysis of the results, the study confirms the superior performance of ensemble model, achieving a minimum mean absolute error (MAE) of 10.684 and 2.827 for predicting hospital stay and ICU stay, respectively. Furthermore, the models that incorporate a drift adaptive framework exhibit superior performance compared to the models that do not include such a framework.