Development and validation of parsimonious algorithms to classify acute respiratory distress syndrome phenotypes: a secondary analysis of randomised controlled trials.
Development and validation of parsimonious algorithms to classify acute respiratory distress syndrome phenotypes: a secondary analysis of randomised controlled trials.
复制标题
DOI：
10.1016/s2213-2600(19)30369-8
复制
发表时间：
2020-03
期刊：
影响因子：
--
通讯作者：
Calfee CS
中科院分区：
文献类型：
--
作者：
Sinha P;Delucchi KL;McAuley DF;O'Kane CM;Matthay MA;Calfee CS
使用潜在类别分析 (LCA)，在五个随机对照试验 (RCT) 队列中，确定了 ARDS 的两种不同表型（低炎症和高炎症）。表型与不同的结果和治疗反应相关。该项目的目标是开发用于表型识别的简约分类器模型，该模型在临床环境中使用准确且可行。在这项回顾性研究中，三个 ARDS 网络 RCT 队列（ARMA、ALVEOLI 和 FACTT）被用作推导数据集（N=2022），第四个（SAILS）被用作验证数据集（N=715）。所有这些队列中 LCA 衍生的表型均作为参考标准。机器学习算法用于选择重要的分类器变量，然后用于开发嵌套逻辑回归模型。然后在验证数据集中评估基于简约性和预测准确性的最佳逻辑回归模型。最后，在两个外部 ARDS 临床试验数据集（START 和 HARP-2）中测试了模型的预后有效性。六个最重要的分类变量是 IL-8、IL-6、蛋白 C、可溶性 TNF-receptor-1、碳酸氢盐和血管加压药的使用。从嵌套模型中，3 变量（IL-8、碳酸氢盐和蛋白 C）和 4 变量模型（3 变量加血管加压药的使用）被判定为表现最佳。在验证队列中，两种模型均显示出良好的准确性（AUC 0·94；95% CI：0·92–0·95 和 0·95；95% CI：0·93–0·96）。在外部数据集中，在推导数据集中开发的 3 变量模型确定了两种具有不同临床特征的表型和与先前发现一致的结果，包括 HARP-2 中辛伐他汀的差异生存率。使用 3-4 个变量的简单分类器模型可以准确识别 ARDS 表型。在开发关键生物标志物的实时测试和前瞻性验证之前，这些模型可以促进 ARDS 表型的识别，从而使其能够在临床试验和实践中应用。
Using latent class analysis (LCA), in five randomized control trial (RCT) cohorts, two distinct phenotypes of ARDS have been identified (hypo-inflammatory and hyper-inflammatory). The phenotypes are associated with differential outcomes and treatment response. The objective of this project was to develop parsimonious classifier models for phenotype identification that could be accurate and feasible to use in the clinical setting. In this retrospective study, three ARDS network RCT cohorts (ARMA, ALVEOLI, and FACTT) were used as the derivation dataset (N=2022), and a fourth (SAILS) was used as the validation dataset (N=715). LCA-derived phenotypes in all of these cohorts served as the reference standard. Machine-learning algorithms were used to select important classifier variables, which were then used to develop nested logistic regression models. The best logistic regression models based on parsimony and predictive accuracy were then evaluated in the validation dataset. Finally, the models’ prognostic validity was tested in two external ARDS clinical trial datasets (START and HARP-2). The six most important classifier variables were IL-8, IL-6, protein C, soluble TNF-receptor-1, bicarbonate, and vasopressor-use. From the nested models, 3-variable (IL-8, bicarbonate, and protein C) and 4-variable models (3-variable plus vasopressor use) were adjudicated to be the best performing. In the validation cohort, both models showed good accuracy (AUC 0·94; 95% CI: 0·92–0·95 and 0·95; 95% CI: 0·93–0·96 respectively). In the external datasets, 3-variable models developed in the derivation dataset identified two phenotypes with distinct clinical features and outcomes consistent with prior findings, including differential survival with simvastatin in HARP-2. ARDS phenotypes can be accurately identified with simple classifier models using 3–4 variables. Pending the development of real-time testing for key biomarkers and prospective validation, these models could facilitate identification of ARDS phenotypes to enable their application in clinical trials and practice.