目的:肺部感染(PI),免疫抑制治疗的严重并发症,影响患者预后。作为这项研究的一部分,我们旨在构建肺部感染预测(PIP)模型,并在接受免疫抑制药物(ISDs)的患者中进行验证.方法:完全,7,977名接受ISD治疗的患者被随机分为7:3至发展中(n=5,583)与验证数据集(n=2,394)。我们的预测列线图是使用最小绝对收缩和选择算子(LASSO)和多变量COX回归分析建立的。使用一致性指数(C指数)和校准曲线,对最终模型的预测性能进行了评价。结果:服用免疫抑制药物的患者中,在548(6.9%)中观察到PI。免疫抑制治疗后PI发生的中位时间为123.0(四分位距:63.0,436.0)天。13个具有统计学意义的独立预测因子(性别,年龄,高血压,DM,恶性肿瘤,使用生物制剂,使用CNI,使用甲基强的松龙500毫克,使用甲基强的松龙40毫克,使用40毫克总剂量的甲基强的松龙,使用口服糖皮质激素,白蛋白水平,和血红蛋白水平)使用LASSO算法和多变量COX回归分析进行筛选。建立在这些特征上的PIP模型表现得相当好,发展C指数为0.87(灵敏度:85.4%;特异性:81.0%),验证C指数为0.837、0.829、0.832和0.830,用于预测90-,180-,270天和360天PI概率,分别。决策曲线分析(DCA)和校准曲线显示出列线图的出色临床实用性和校准性能。结论:本文提出的PIP模型可以帮助预测接受免疫抑制治疗的个体患者的PI风险,并有助于个性化临床决策。
Objective: Pulmonary infection (PI), a severe complication of immunosuppressive therapy, affects patients\' prognosis. As part of this study, we aimed to construct a pulmonary infection prediction (PIP) model and validate it in patients receiving immunosuppressive drugs (ISDs). Methods: Totally, 7,977 patients being treated with ISDs were randomised 7:3 to the developing (n = 5,583) versus validation datasets (n = 2,394). Our predictive nomogram was established using the least absolute shrinkage and selection operator (LASSO) and multivariate COX regression analyses. With the use of the concordance index (C-index) and calibration curve, the prediction performance of the final model was evaluated. Results: Among the patients taking immunosuppressive medication, PI was observed in 548 (6.9%). The median time of PI occurrence after immunosuppressive therapy was 123.0 (interquartile range: 63.0, 436.0) days. Thirteen statistically significant independent predictors (sex, age, hypertension, DM, malignant tumour, use of biologics, use of CNIs, use of methylprednisolone at 500 mg, use of methylprednisolone at 40 mg, use of methylprednisolone at 40 mg total dose, use of oral glucocorticoids, albumin level, and haemoglobin level) were screened using the LASSO algorithm and multivariate COX regression analysis. The PIP model built on these features performed reasonably well, with the developing C-index of 0.87 (sensitivity: 85.4%; specificity: 81.0%) and validation C-indices of 0.837, 0.829, 0.832 and 0.830 for predicting 90-, 180-, 270- and 360-day PI probability, respectively. The decision curve analysis (DCA) and calibration curves displayed excellent clinical utility and calibration performance of the nomogram. Conclusion: The PIP model presented herein could aid in the prediction of PI risk in individual patients who receive immunosuppressive treatment and help personalise clinical decision-making.