Readmission prediction model based on multi-modal data

Abstract

The invention discloses a re-admission prediction method based on multi-modal data. The re-admission prediction method is realized by adopting a constructed re-admission prediction model based on the multi-modal data. After target data is prepared, the method comprises the following steps: processing unstructured data through an unstructured datamation processing module of a model to obtain a document-level embedding vector of the unstructured data; processing the structured data through a structured data processing module of the model to form structured data vectors, including a time sequence vector and a static information vector; fusing the document-level embedded vector, the time sequence vector and the static information vector through a data fusion module of the model to obtain final data represented by the patient; and inputting the data represented by the patient into an Adaboost-LR model for classification prediction to obtain a prediction result. According to the invention, the defect of insufficient data utilization in the current re-admission prediction technology is overcome.

Description

technical field [0001] The invention relates to the technical field of admission prediction, in particular to a readmission prediction model based on multimodal data. Background technique [0002] Readmission usually refers to a patient being readmitted within 30 days of discharge. During hospitalization, especially in the intensive care unit (ICU), identifying high-risk readmission patients can help reduce the risk of readmission and is critical for preventing serious life-threatening events and reducing healthcare costs. The unplanned readmission rate is not only a reflection of the quality of medical services, but also an important indicator to measure the quality of hospital care, which is related to the clinical and economic burden of patients and society. Intensive care unit (ICU) readmission brings psychological stress to patients in addition to further economic and mortality risks. [0003] An electronic health record (ehr) is a longitudinal electronic record of pa...

AI Technical Summary

Problems solved by technology

However, most of the previous work focused on predictive modeling using structured data or unstructured clinical annotations, and few people paid enough attention to combining structured data and unstructured clinical annotations.

Images