Study on the construction of a association rule prediction model for transfer rate of intensive care unit patients and the comparison of the predictive performance

Zhang Caifeng; Wen Hongyi; Tian Long

doi:10.3969/j.issn.1002-1949.2025.02.006

Abstract：Objective To create a new model to predict the transfer rate of the intensive care unit(ICU)patients and to evaluate and compare the predictive performance of the new model based on the traditional medical statistical model as the reference.Methods A retrospective study was conducted on the ICU patients(n=4 000)in the First Affiliated Hospital of Hebei North University from February 2011 to February 2023.After screening,they were randomly divided into a modeling group(n=2 370)and a validation group(n=1 630).A new association rule prediction model for transfer rate of the ICU patients was created based on the baseline data from the modeling group and FP-Growth algorithm.The new model was validated internally and externally by using calibration curve,clinical decision curve and receiver operating characteristic(ROC)curve,respectively.A traditional medical statistical prediction model was created based on the baseline data from the modeling group and multiple Logistic regression analysis.The predictive performance of the new model was evaluated and compared based on the traditional medical statistical model as the reference.Results The new association rule prediction model for transfer rate of the ICU patients showed that when the baseline data combination of the patients met the transfer criteria within 1 week,1-2 weeks,2-3 weeks and 3-4 weeks,the transfer rates were 70％,41％,19％and 12％,respectively.The internal and external validation results of the new model showed good consistency and could provide clinical net benefits.There was no statistically significant difference in the area under ROC curve(AUC)for the transfer rate predicted by the new model at specific time between the patients from modeling group and the validation group(P＞0.05).Compared with the independent risk factors of the traditional medical statistical prediction model,the composition of preceding item of the new model was more complex,and its AUC for the transfer rate of the patients from the modeling group at specific time was higher(all P＜0.05).Conclusions The predictive performance of the new association rule prediction model for the transfer rate of ICU patients can meet clinical requirements,and is more suitable for optimizing the management and resource allocation for ICU patients.