The invention discloses a prediction model for organ failure induced by acute pancreatitis, and the model comprises the following steps: S100, preprocessing patient information, and recording events and time nodes by means of {Variables, Time}; s200, sorting the events according to a time sequence, and filling missing values by adopting a Decay mechanism; and S300, carrying out one-hot coding on the data by using an Embedded mechanism, mapping the data to a real vector space, normalizing the data and then inputting the normalized data into a Phased LSTM model. Time gate output is calculated according to the interval time of a patient from admission to the time node of the event, the model training process is accelerated by utilizing the output result of the time gate, the neuron of the output layer is 2, and a softmax function is adopted as an activation function. According to the method, heterogeneous multi-dimensional data can be processed, time information can be flexibly used, andmeanwhile, the judgment of the model is closer to a description of a disease natural process in the real world.