In Physiological measurement ; h5-index 36.0
Objective Sepsis has seriously threatened the human life. Early identification of patient status risk and appropriate treatment can reduce the septic shock risk and mortality. Our purpose is to design and validate an adjuvant therapy system based on deep reinforcement learning (DRL), which can provide treatment recommendations with provident and assess the risks of patient status and treatment options in early stage. Approach The data is from Beth Israel Deaconess Medical Center. The raw data included 53,423 patients from MIMIC-III. 19,620 eligible samples were screened to form the final cohort. First, the patient's physiological parameters are fed into the DRL therapy strategy recommendation module (TSRM), which provides a forward-looking recommendation for treatment strategy. Then, the recommended strategies above are fed into the reinforcement learning risk assessment module (RAM), which predicts risk of patient status and treatment strategy from a long-term perspective. The deep reinforcement learning model designed in this paper assists in formulating treatment plans and evaluating treatment risks and patient status through continuous interaction with patient trajectory. So that this model has the foresight that the supervising-deep- learning model does not have. Main Results The experiment shows that, in the test set, for the TSRM, the mortality is the lowest when actually implemented treatment strategy is the same as the AI-recommended. For the RAM, it can accurately grasp the deterioration trend of the patients, reasonably assess the risk of the patient's status and the treatment plans at an early stage. The assessment results of the model were matched with the actual clinical records. Significance A DRL-based sepsis adjuvant therapy model is proposed. It can prospectively assist physicians in proposing treatment strategies, early assessment the risk of patient status and treatment methods, and detecting deterioration trends in advance.
Zhang Quan, Wang Jianqi, Liu Guohua, Zhang Wenjia
2023-Jan-04
deep learning, reinforcement learning, risk assessment, safety analysis of treatment, septic shock, treatment model
