The paper introduces the design and building of clinical critical disease early warning model based on multi-scale Convolution Neural Networks(CNN) and Long Short Term Memory (LSTM) neural network. Taking 2 198 critically ill patients as the object of study, it conducts an experiment towards the model. The results show that the prediction accuracy of the model is higher, which is helpful to explore the hidden correlation between patient data and changes in patients' conditions, so as to offer assistance in medical decision-making.