Department of Pulmonary and Critical Care Medicine, East Carolina University, Greenville, NC, USA.
Department of Bioinformatics and Biostatistics, School of Public Health and Information Sciences, University of Louisville, Louisville, KY, USA.
Respir Res. 2023 Mar 30;24(1):97. doi: 10.1186/s12931-023-02407-4.
Gastroesophageal reflux disease (GERD) is a common comorbidity in patients with interstitial lung disease (ILD). We built and validated a model using the national inpatient sample (NIS) database to assess the contributory role of GERD in ILD-related hospitalizations mortality.
In this retrospective analysis, we extracted ILD-related hospitalizations data between 2007 and 2019 from the NIS database. Univariable logistic regression was used for predictor selection. Data were split into the training and validation cohorts (0.6 and 0.4, respectively). We used decision tree analysis (classification and regression tree, CART) to create a predictive model to explore the role of GERD in ILD-related hospitalizations mortality. Different metrics were used to evaluate our model. A bootstrap-based technique was implemented to balance our training data outcome to improve our model metrics in the validation cohort. We conducted a variance-based sensitivity analysis to evaluate GERD's importance in our model.
The model had a sensitivity of 73.43%, specificity of 66.15%, precision of 0.27, negative predictive value (NPV) of 93.62%, accuracy of 67.2%, Matthews Correlation Coefficient (MCC) of 0.3, F1 score of 0.4, and area under the curve (AUC) for the receiver operating characteristic (ROC) curve of 0.76. GERD did not predict survival in our cohort. GERD contribution to the model was ranked the eleventh among twenty-nine variables included in this analysis (importance of 0.003, normalized importance of 5%). GERD was the best predictor in ILD-related hospitalizations who didn't receive mechanical ventilation.
GERD is associated with mild ILD-related hospitalization. Our model-performance measures suggest overall an acceptable discrimination. Our model showed that GERD does not have a prognostic value in ILD-related hospitalization, indicating that GERD per se might not have any impact on mortality in hospitalized ILD patients.
胃食管反流病（GERD）是间质性肺疾病（ILD）患者常见的合并症。我们使用国家住院患者样本（NIS）数据库构建并验证了一个模型，以评估 GERD 在ILD 相关住院死亡率中的作用。
在这项回顾性分析中，我们从 NIS 数据库中提取了 2007 年至 2019 年间与 ILD 相关的住院数据。使用单变量逻辑回归进行预测因子选择。数据分为训练和验证队列（分别为 0.6 和 0.4）。我们使用决策树分析（分类和回归树，CART）创建预测模型，以探讨 GERD 在 ILD 相关住院死亡率中的作用。使用不同的指标来评估我们的模型。实施基于引导的技术来平衡我们的训练数据结果，以提高我们在验证队列中的模型指标。我们进行了基于方差的敏感性分析，以评估 GERD 在我们模型中的重要性。
该模型的灵敏度为 73.43%，特异性为 66.15%，精度为 0.27，阴性预测值（NPV）为 93.62%，准确性为 67.2%，马修斯相关系数（MCC）为 0.3，F1 得分为 0.4，接收器操作特征（ROC）曲线下面积（AUC）为 0.76。在我们的队列中，GERD 并不能预测生存。在这项分析中包含的 29 个变量中，GERD 对模型的贡献排名第 11（重要性为 0.003，归一化重要性为 5%）。在未接受机械通气的 ILD 相关住院患者中，GERD 是最好的预测因子。
GERD 与轻度 ILD 相关的住院有关。我们的模型性能指标表明总体上具有可接受的区分能力。我们的模型表明，GERD 在 ILD 相关住院中没有预后价值，这表明 GERD 本身可能不会对住院ILD 患者的死亡率产生任何影响。
