Development and Validation of a Personalized Model With Transfer Learning for Acute Kidney Injury Risk Estimation Using Electronic Health Records.
Development and Validation of a Personalized Model With Transfer Learning for Acute Kidney Injury Risk Estimation Using Electronic Health Records.
复制标题
基于电子健康记录运用迁移学习开发用于急性肾损伤风险评估的个性化模型并进行验证
DOI：
10.1001/jamanetworkopen.2022.19776
复制
发表时间：
2022-07-01
影响因子：
13.8
通讯作者：
中科院分区：
文献类型：
--
作者：
本诊断性研究评估了住院患者急性肾损伤风险的评估方法，并提出了一种使用机器学习的新模型。
在评估住院患者各亚群的急性肾损伤风险方面，个性化模型是否比传统模型更准确？
在这项涉及76957例住院病例的诊断性研究中，一种带有迁移学习框架的新型个性化模型在急性肾损伤风险评估上表现更佳且更为均衡，同时还考虑了不同患者亚群中风险因素的异质性及其变化。
本研究结果表明，临床风险评估正朝着个性化评估方向发展，并凸显了灵活、个性化患者护理的必要性。
急性肾损伤（AKI）是一种在住院患者中普遍存在的异质性综合征。个性化风险评估和风险因素识别或许能实现有效干预并改善预后。
本研究旨在利用电子健康记录（EHR）开发并验证个性化AKI风险评估模型，对比个性化模型与通用模型及亚群模型，检验个性化模型是否更具优势，并评估不同亚群中风险因素及其预后的异质性。
本诊断性研究分析了一家三级医院的EHR数据，运用机器学习和逻辑回归方法开发并验证了通用、亚群及个性化风险评估模型。采用迁移学习来优化个性化模型。分析各亚群中的预测因子预后，并运用元回归探索预测因子之间的相互作用。研究分析纳入了2007年11月1日至2016年12月31日期间住院2天及以上的成年患者，排除入院时存在中度或重度肾功能障碍的患者。数据分析时间为2019年8月28日至2022年5月8日。
研究数据为EHR中的临床和实验室变量。
主要结局为任何严重程度的AKI，AKI依据“改善全球肾脏病预后组织（Kidney Disease: Improving Global Outcomes）”的血清肌酐标准进行定义。采用受试者工作特征曲线下面积（AUROC）、精确召回曲线下面积及校准度来衡量模型性能。
研究队列包含76957例住院病例。患者的平均（标准差）年龄为55.5（17.4）岁，其中男性42159例（54.8%）。在普通住院患者中，带有迁移学习的个性化模型在AKI评估的AUROC方面优于通用模型（0.78 [95%置信区间，0.77 - 0.79] 对比 0.76 [95%置信区间，0.75 - 0.76]；P <.001），在高风险亚群（0.79 [95%置信区间，0.78 - 0.80] 对比 0.75 [95%置信区间，0.74 - 0.77]；P <.001）和低风险亚群（0.74 [95%置信区间，0.73 - 0.75] 对比 0.71 [95%置信区间，0.70 - 0.72]；P <.001）中同样如此。对于肝移植和心脏手术等高风险亚群，AUROC提升幅度达到0.13。此外，带有迁移学习的个性化模型在研究充分的AKI亚群中，表现优于或与已发表的最佳模型相当。不同患者之间的预测因子预后差异显著，相互作用分析揭示了预测因子预后的调节因素。
本研究结果表明，带有迁移学习的个性化建模是一种改进的AKI风险评估方法，可应用于不同的患者亚群。个体层面的风险因素异质性和相互作用凸显了灵活、个性化护理的必要性。
This diagnostic study assesses the approaches to estimating acute kidney injury risk in hospitalized patients and proposes a novel model that uses machine learning. Are personalized models more accurate than traditional models in estimating acute kidney injury across subpopulations of hospitalized patients? In this diagnostic study involving 76 957 inpatient encounters, a new personalized model with transfer learning framework yielded improved and more equitable acute kidney injury estimation as well as accounted for the heterogeneity of risk factors and their variations in different patient subgroups. Findings of this study suggest the advancement of clinical risk estimation toward personalized estimation and highlight the need for agile, personalized patient care. Acute kidney injury (AKI) is a heterogeneous syndrome prevalent among hospitalized patients. Personalized risk estimation and risk factor identification may allow effective intervention and improved outcomes. To develop and validate personalized AKI risk estimation models using electronic health records (EHRs), examine whether personalized models were beneficial in comparison with global and subgroup models, and assess the heterogeneity of risk factors and their outcomes in different subpopulations. This diagnostic study analyzed EHR data from 1 tertiary care hospital and used machine learning and logistic regression to develop and validate global, subgroup, and personalized risk estimation models. Transfer learning was implemented to enhance the personalized model. Predictor outcomes across subpopulations were analyzed, and metaregression was used to explore predictor interactions. Adults who were hospitalized for 2 or more days from November 1, 2007, to December 31, 2016, were included in the analysis. Patients with moderate or severe kidney dysfunction at admission were excluded. Data were analyzed between August 28, 2019, and May 8, 2022. Clinical and laboratory variables in the EHR. The main outcome was AKI of any severity, and AKI was defined using the Kidney Disease: Improving Global Outcomes serum creatinine criteria. Performance of the models was measured with area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve, and calibration. The study cohort comprised 76 957 inpatient encounters. Patients had a mean (SD) age of 55.5 (17.4) years and included 42 159 men (54.8%). The personalized model with transfer learning outperformed the global model for AKI estimation in terms of AUROC among general inpatients (0.78 [95% CI, 0.77-0.79] vs 0.76 [95% CI, 0.75-0.76]; P < .001) and across the high-risk subgroups (0.79 [95% CI, 0.78-0.80] vs 0.75 [95% CI, 0.74-0.77]; P < .001) and low-risk subgroups (0.74 [95% CI, 0.73-0.75] vs 0.71 [95% CI, 0.70-0.72]; P < .001). The AUROC improvement reached 0.13 for the high-risk subgroups, such as those undergoing liver transplant and cardiac surgery. Moreover, the personalized model with transfer learning performed better than or comparably with the best published models in well-studied AKI subgroups. Predictor outcomes varied significantly between patients, and interaction analysis uncovered modifiers of the predictor outcomes. Results of this study demonstrated that a personalized modeling with transfer learning is an improved AKI risk estimation approach that can be used across diverse patient subgroups. Risk factor heterogeneity and interactions at the individual level highlighted the need for agile, personalized care.
影响因子：
4.5
作者：
Visweswaran S;Angus DC;Hsieh M;Weissfeld L;Yealy D;Cooper GF
通讯作者：
Cooper GF