Prognostication in comatose patients after cardiac arrest (CA) remains one of the biggest challenges for a neurologist in the intensive care unit, where accurate and reliable outcome prediction is crucial as it informs decisions regarding continued care or withdraw of life support. Electroencephalogram (EEG) provides valuable and non-invasive insights into patients' neurological status and has been used in many prediction models. However, traditional models often view EEG as stationary data, neglecting the dynamic patterns of EEG signals in response to internal and external perturbations, which may be critical for outcome prediction. In this study, we introduce a novel mixed nonstationary outcome prediction model (MNOPM) of EEG for comatose patients after a heart attack. The "mixed" aspect of the model refers to the combination of various machine learning frameworks, while "nonstationary" denotes the consideration of both time-varying dynamic patterns and nonstationary features of EEG. The MNOPM comprises three parts: a recurrent neural network (RNNs) to capture dynamic features, a convolutional neural network (CNNs) to capture interactions between EEG channels, and an eXtreme Gradient Boosting (XGBoost) algorithm to combine time-invariant features, time-varying features, and outputs from RNNs and CNNs for the final prediction. The proposed model addresses the time-varying nonlinearity of EEG signals, demonstrating the potential to improve outcome prediction accuracy and reliability for comatose patients following cardiac arrest. The MNOPM model has been validated using the I-CARE (International Cardiac Arrest Research Consortium Database). The model demonstrates superior performance during K-fold validation on the training set and achieves a score of 0.119 in the unofficial phase. Due to time constraints, we were unable to submit the complete version as calculating nonstationary features takes an exceptionally long time. We continue to work on optimizing the algorithms to expedite the process and incorporate additional features to further enhance the model's performance.