The Study Of Risk Factors Of Prolonged Length Of Stay In Emergency Department

ObjectTo investigate prolonged length of stay of patients in emergency department(ED)in one “AAA” hospital in Guangzhou,and to analyze the influence of different patient factors,medical factors and organizational factors of prolonged length of stay.Based on the risk factors,we looked for measures to reduce the length of stay to alleviate the ED crowding.MethodData of patients from the ED in a “AAA” hospital in Guangzhou from March2019 to November 2020 were retrospectively studied.With 6 hours as the cut-off point of prolonged length of stay,based on the literature and the database of the choice may prolong length of stay risk factors on the patients,through the logistic regression method to explore the possible risk factors.Then according to the risk factors of the hospital to develop effective measures to alleviate the ED crowding,including the use of machine learning algorithm to build an efficient triage system.ResultFrom March 2019 to November 2020 a total of 116,474 patients were seen in the ED,the median length of stay was 2.2(0.6-4.1)h,a number of 14,790(12.7%)patients stayed in ED more than 6 h.Binary logistic regression analysis showed that the independent factors for patients who stayed more than 6 hours were age,sex,charge type,triage level,imaging examination,laboratory test,diagnosis at ED visit,mode of presentation,visit by green channel,wait time,destination after ED visit,time of presentation and holiday visit(p<0.05).The risk for patients admitted to the hospital stayed more than 6 hours was 5.494 times more than discharge patients.The risk for Ⅱ level patients stayed more than 6 hours was 4.872 times more than Ⅳa level patients.The risk for who came to ED by means of holding by others,ambulance and wheelchair stayed more than 6 hours was 3.162,2.823,2.255 times respectively more than patients who came to ED walk by themselves.The risk for patients with circulatory system disease stayed more than 6 hours was 2.272 times more than patients with disease of skin.Among the five machine learning models,the accuracy(0.789),AUC(0.940)and macro-F1(0.538)of the FCNN were the best,while the sensitivity,specificity,positive predictive value and negative predictive value of the MLR were better than the other four machine learning models.There was room for improvement in the overall performance of the five machine learning models.ConclusionFirstly,the ED crowding was still severe,and measures should be taken to alleviate the crowding.Patients of admission to the general ward,deceased,triage level of Ⅰ and Ⅱ,having test items,cardiovascular disease,the way of admission to the hospital is embraced,ambulance and wheelchair,aged ≥80 years old were all high risk groups of prolonged length of stay.Secondly,the machine learning-based model which was used to support decision-making in emergency department triage for patients with suspected cardiovascular disease,it may be able to assist nurses in clinical triage to efficiently allocate the medical resource and alleviate the ED crowding.