Development of a Prediction Model of Early Acute Kidney Injury in Critically Ill Children Using Electronic Health Record Data
Development of a Prediction Model of Early Acute Kidney Injury in Critically Ill Children Using Electronic Health Record Data
复制标题
DOI：
10.1097/pcc.0000000000000750
复制
发表时间：
2016-06-01
影响因子：
4.1
通讯作者：
Khemani, Robinder G.
中科院分区：
文献类型：
--
作者：
Sanchez-Pinto, L. Nelson;Khemani, Robinder G.
目的：急性肾损伤与危重患儿预后不良独立相关。然而，急性肾损伤的主要生物标志物血清肌酐是损伤的晚期标志物，可导致诊断延迟。我们的目标是开发和验证一个数据驱动的多变量临床预测模型的急性肾损伤在一个普通的PICU使用电子健康记录数据。设计：推导和验证的预测模型使用回顾性数据。患者：2003年5月至2015年3月期间入院的所有1个月至21岁的患者，入院时无急性肾损伤，存活并在ICU至少24 hours.Setting:多学科三级PICU干预：主要结局是早期急性肾损伤，其定义为入院后72小时内在ICU发生的新发急性肾损伤。多变量逻辑回归分析使用ICU住院前12小时的电子健康记录数据推导出儿科早期阿基风险评分。测量和主要结果：共有9，396例患者纳入分析，其中4%有早期急性肾损伤，这些患者的死亡率显著高于无早期急性肾损伤的患者。（26% vs 3.3%; p < 0.001）。对33个候选变量进行了测试。最终模型有7个预测因子，具有良好的区分度（曲线下面积0.84）和适当的校准。该模型在两个验证集中进行了验证，并保持了良好的区分度（曲线下面积，0.81和0.86）。我们开发并验证了儿科早期阿基风险评分，这是一种数据驱动的急性肾损伤临床预测模型，仅使用客观的电子健康记录数据，在ICU护理的前12小时内以真实的时间提供，并可在PICU中推广。该预测模型旨在以自动化临床决策支持系统的形式实现，并可用于指导预防，治疗和研究策略。
Objective: Acute kidney injury is independently associated with poor outcomes in critically ill children. However, the main biomarker of acute kidney injury, serum creatinine, is a late marker of injury and can cause a delay in diagnosis. Our goal was to develop and validate a data-driven multivariable clinical prediction model of acute kidney injury in a general PICU using electronic health record data.Design: Derivation and validation of a prediction model using retrospective data.Patients: All patients 1 month to 21 years old admitted between May 2003 and March 2015 without acute kidney injury at admission and alive and in the ICU for at least 24 hours.Setting: A multidisciplinary, tertiary PICU.Intervention: The primary outcome was early acute kidney injury, which was defined as new acute kidney injury developed in the ICU within 72 hours of admission. Multivariable logistic regression was performed to derive the Pediatric Early AKI Risk Score using electronic health record data from the first 12 hours of ICU stay.Measurements and Main Results: A total of 9,396 patients were included in the analysis, of whom 4% had early acute kidney injury, and these had significantly higher mortality than those without early acute kidney injury ( 26% vs 3.3%; p < 0.001). Thirty-three candidate variables were tested. The final model had seven predictors and had good discrimination ( area under the curve 0.84) and appropriate calibration. The model was validated in two validation sets and maintained good discrimination ( area under the curves, 0.81 and 0.86).Conclusion: We developed and validated the Pediatric Early AKI Risk Score, a data-driven acute kidney injury clinical prediction model that has good discrimination and calibration in a general PICU population using only electronic health record data that is objective, available in real time during the first 12 hours of ICU care and generalizable across PICUs. This prediction model was designed to be implemented in the form of an automated clinical decision support system and could be used to guide preventive, therapeutic, and research strategies.