Dynamic prediction of the need for renal replacement therapy in intensive care unit patients using a simple and robust model

详细信息    查看全文

• 作者：Felix Erdfelder ; Daniel Grigutsch…
• 关键词：Acute kidney injury ; Renal replacement therapy ; Decision support techniques ; Online prediction
• 刊名：Journal of Clinical Monitoring and Computing
• 出版年：2017
• 出版时间：February 2017
• 年：2017
• 卷：31
• 期：1
• 页码：195-204
• 全文大小：
• 刊物类别：Medicine
• 刊物主题：Anesthesiology; Intensive / Critical Care Medicine; Statistics for Life Sciences, Medicine, Health Sciences;
• 出版者：Springer Netherlands
• ISSN：1573-2614
• 卷排序：31

文摘

We aimed at identifying a model that dynamically predicts future need for renal replacement therapy (RRT) in intensive care unit (ICU) patients and can easily be implemented for online monitoring at the bedside. 7290 interdisciplinary ICU admissions were investigated. Patients with <3 days of stay or RRT in the first 2 days were excluded. 1624 of the remaining 2625 patients had a normal serum creatinine at admission. Every second of these 1624 patients was used for model calibration whereas the other half and, in addition, the 1001 patients with elevated serum creatinine were exclusively used for validation. Discriminant analysis was used to determine and validate a combination of clinical parameters that predicts the need for RRT 72 h ahead. Based on the calibration sample, stepwise discriminant analysis selected the serum values of (1) current urea, (2) current lactate, (3) the ratio of current and admission serum creatinine, and (4) the mean urine output of the previous 24 h. In the validation datasets, the model reached areas under the receiver operating characteristic curve of 0.866 and 0.833 in patients with normal and elevated serum creatinine at admission, respectively. Moreover, the model’s predictive value extended to at least 5 days prior to initiation of RRT and exceeded that of the RIFLE classification at all investigated prediction intervals. We identified a robust model that dynamically predicts the future need for RRT successfully. This tool may help improve timing of therapy and prognosis in ICU patients.