Medicines Evaluation Unit, University of Manchester, Manchester University NHS Foundation Hospitals Trust, Manchester M23 9QZ, UK.
UCL Respiratory, University College London, London, UK.
Ther Adv Respir Dis. 2022 Jan-Dec;16:17534666221107314. doi: 10.1177/17534666221107314.
Demographic and disease characteristics have been associated with the risk of chronic obstructive pulmonary disease (COPD) exacerbations. Using previously collected multinational clinical trial data, we developed models that use baseline risk factors to predict an individual's rate of moderate/severe exacerbations in the next year on various pharmacological treatments for COPD.
Exacerbation data from 20,054 patients in the ETHOS, KRONOS, TELOS, SOPHOS, and PINNACLE-1, PINNACLE-2, and PINNACLE-4 studies were pooled. Machine learning was used to identify predictors of moderate/severe exacerbation rates. Important factors were selected for generalized linear modeling, further informed by backward variable selection. An independent test set was held back for validation.
Prior exacerbations, eosinophil count, forced expiratory volume in 1 s percent predicted, prior maintenance treatments, reliever medication use, sex, COPD Assessment Test score, smoking status, and region were significant predictors of exacerbation risk, with response to inhaled corticosteroids (ICSs) increasing with higher eosinophil counts, more prior exacerbations, or additional prior treatments. Model fit was similar in the training and test set. Prediction metrics were ~10% better in the full model than in a simplified model based only on eosinophil count, prior exacerbations, and ICS use.
These models predicting rates of moderate/severe exacerbations can be applied to a broad range of patients with COPD in terms of airway obstruction, eosinophil counts, exacerbation history, symptoms, and treatment history. Understanding the relative and absolute risks related to these factors may be useful for clinicians in evaluating the benefit: risk ratio of various treatment decisions for individual patients.Clinical trials registered with www.clinicaltrials.gov (NCT02465567, NCT02497001, NCT02766608, NCT02727660, NCT01854645, NCT01854658, NCT02343458, NCT03262012, NCT02536508, and NCT01970878).
人口统计学和疾病特征与慢性阻塞性肺疾病（COPD）加重的风险相关。我们使用先前收集的多国临床试验数据，开发了使用基线风险因素预测个体在接受 COPD 各种药物治疗后下一年中中重度/重度加重的风险模型。
汇总了来自 ETHOS、KRONOS、TELOS、SOPHOS 以及 PINNACLE-1、PINNACLE-2 和 PINNACLE-4 研究的 20,054 例患者的加重数据。使用机器学习识别中重度加重率的预测因素。通过向后变量选择，为广义线性模型选择重要因素。保留一个独立的测试集进行验证。
既往加重次数、嗜酸性粒细胞计数、1 秒用力呼气量占预计值的百分比、既往维持治疗、缓解药物使用、性别、COPD 评估测试评分、吸烟状态和地区是加重风险的重要预测因素，吸入皮质激素（ICS）的反应随着嗜酸性粒细胞计数的增加、既往加重次数的增加或增加的既往治疗而增加。训练集和测试集的模型拟合度相似。在完整模型中的预测指标比仅基于嗜酸性粒细胞计数、既往加重次数和 ICS 使用的简化模型要好约 10%。
这些预测中重度加重率的模型可应用于广泛的 COPD 患者，无论气道阻塞程度、嗜酸性粒细胞计数、加重史、症状和治疗史如何。了解这些因素的相对和绝对风险可能对评估各种治疗决策对个体患者的获益风险比的临床医生有用。
临床试验在 www.clinicaltrials.gov 注册（NCT02465567、NCT02497001、NCT02766608、NCT02727660、NCT01854645、NCT01854658、NCT02343458、NCT03262012、NCT02536508 和 NCT01970878）。
