Background: Cardiac surgery is a high-risk procedure that leaves little margin for error and can result in severe morbidity or mortality. During the surgery, detecting and understanding the trends hidden in physiological parameters, such as heart rate (HR) and invasive blood pressure (IBP), is vital. These trends can provide insight into future adverse outcomes and guide the clinician to perform appropriate interventions. Objectives: This study aims to introduce a real-time time-series forecasting system to predict future HR values (in beats per minute) and IBP values (in mm Hg) in a defined interval. Methods: The 74715 HR time series samples and 98278 IBP time series samples from 1056 patients undergoing cardiac surgeries at a single tertiary care hospital were recorded per minute. The raw data was processed to remove artefacts, and various deep learning-based models were trained with the optimized hyperparameters. The best-performing model was deployed in a graphical user interface, allowing hospital staff to visualize and analyze the predicted data. Results: The Long Short Term Memory (LSTM) model offers the best performance for HR time series prediction with RMSE (5.26) and MAE (2.57). The Bidirectional Short Term Memory (BiLSTM) achieved RMSE (10.22) and MAE (6.28), making it the best-performing model for IBP time series forecasting. Conclusion: Deep neural network models were trained and deployed on non-linear intra-operative time series data to accurately predict the future HR and IBP values. These models may be used in cardiac surgeries to forecast future physiological trends based on short-term physiological history.

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.