[Construction and external validation of a risk prediction model for unplanned interruption during continuous renal replacement therapy]
Hongyan Xu,Qi Ren,Lihong Zhu,Juan Lin,Shangzhong Chen,Caibao Hu,Yanfei Shen,Guolong Cai
DOI: https://doi.org/10.3760/cma.j.cn121430-20231204-01045
Abstract:Objective: To identify the independent factors of unplanned interruption during continuous renal replacement therapy (CRRT) and construct a risk prediction model, and to verify the clinical application effectiveness of the model. Methods: A retrospective study was conducted on critically ill adult patients who received CRRT treatment in the intensive care unit (ICU) of Zhejiang Hospital from January 2021 to August 2022 for model construction. According to whether unplanned weaning occurred, the patients were divided into two groups. The potential influencing factors of unplanned CRRT weaning in the two groups were compared. The independent influencing factors of unplanned CRRT weaning were screened by binary Logistic regression and a risk prediction model was constructed. The goodness of fit of the model was verified by a Hosmer-Lemeshow test and its predictive validity was evaluated by receiver operator characteristic curve (ROC curve). Then embed the risk prediction model into the hospital's ICU multifunctional electronic medical record system for severe illness, critically ill patients with CRRT admitted to the ICU of Zhejiang Hospital from November 2022 to October 2023 were prospectively analyzed to verify the model's clinical application effect. Results: (1) Model construction and internal validation: a total of 331 critically ill patients with CRRT were included to be retrospectively analyzed. Among them, there were 238 patients in planned interruption group and 93 patients in unplanned interruption group. Compared with the planned interruption group, the unplanned interruption group was shown as a lower proportion of males (80.6% vs. 91.6%) and a higher proportion of chronic diseases (60.2% vs. 41.6%), poor blood purification catheter function (31.2% vs. 6.3%), as a higher platelet count (PLT) before CRRT initiation [×109/L: 137 (101, 187) vs. 109 (74, 160)], lower level of blood flow rate [mL/min: 120 (120, 150) vs. 150 (140, 180)], higher proportion of using pre-dilution (37.6% vs. 23.5%), higher filtration fraction [23.0% (17.5%, 32.9%) vs. 19.1% (15.7%, 22.6%)], and frequency of blood pump stops [times: 19 (14, 21) vs. 9 (6, 13)], the differences of the above 8 factors between the two groups were statistically significant (all P < 0.05). Binary Logistic regression analysis showed that chronic diseases [odds ratio (OR) = 3.063, 95% confidence interval (95%CI) was 1.200-7.819], blood purification catheter function (OR = 4.429, 95%CI was 1.270-15.451), blood flow rate (OR = 0.928, 95%CI was 0.900-0.957), and frequency of blood pump stops (OR = 1.339, 95%CI was 1.231-1.457) were the independent factors for the unplanned interruption of CRRT (all P < 0.05). These 4 factors were used to construct a risk prediction model, and ROC curve analysis showed that the area under the curve (AUC) predicted by the model was 0.952 (95%CI was 0.930-0.973, P = 0.003 0), with a sensitivity of 88.2%, a specificity of 89.9%, and a maximum value of 1.781 for the Youden index. (2) External validation: prospective inclusion of 110 patients, including 63 planned interruption group and 47 unplanned interruption group. ROC curve analysis showed that the AUC of the risk prediction model was 0.919 (95%CI was 0.870-0.969, P = 0.004 3), with a sensitivity of 91.5%, a specificity of 79.4%, and a maximum value of the Youden index of 1.709. Conclusions: The risk prediction model for unplanned interruption during CRRT has a high predictive efficiency, allowing for rapid and real-time identification of the high risk patients, thus providing references for preventative nursing.