Multivariate models for concurrent outcomes and APE-associated predictions.

aData from study group 1, n = 56. We found no evidence of two-way interactions or non-linear effects using squared terms for these models. Age, gender, CF-related diabetes, airway infection with either Pseudomonas aeruginosa or Staphylococcus aureus and chronic azithromycin, oral or inhaled steroid use had no significant interactions with any inflammatory marker terms in any multivariate model. Log transformed values of biomarkers were used for modeling outcomes. Concurrent FEV1% and Weight-for-age z-score models used linear regression. The model for the number of APE occurring in the year prior to initial sputum collection used quasi-Poisson regression.bData from study group 2, n = 26. Additional adjustment for the stable FEV1% measurement, sequence of stable and APE time point collections, airway infection with either Pseudomonas aeruginosa or Staphylococcus aureus, use of azithromycin or steroids had no significant effect in these models.cEstimates of the mean change in FEV1% per unit change in log scale biomarkers. Results from a linear regression model for the associations between difference in FEV1% between stable and APE time points and GM-CSF (log scale) measured at the APE onset time point. Each univariate representing measurements obtained during clinically stable and APE time points were added in turn to a model containing GM-CSF measured at the APE time point, the only statistically significant univariate. IL-5 (p = 0.006) and IL-10 (p = 0.015) measured at the APE time point and TCC (p = 0.012) measured at the stable time point were found to be positively associated with FEV1% decline independently of GM-CSF. Backward selection of a multivariate model containing GM-CSF (APE), IL-5 (APE), IL-10 (APE), and TCC (Stable) produced the final model presented here.dEstimates of the predicted total number of APE during 5 years of follow up per unit change in log scale biomarkers measured during clinical stability. Results show a quasi-Poisson regression model for the association with number of APE during 5 years of follow-up. HMGB-1 (log scale) was the only significant univariate (pp1) as an indicator of baseline inflammation, retained only HMBG-1. A 1 unit change in log scale HMGB-1 is associated with a mean change in number of APE of 0.34.a. 研究队列1的数据，样本量n=56。本研究针对上述模型采用平方项分析，未发现双向交互作用或非线性效应的证据。年龄、性别、囊性纤维化相关糖尿病（CF-related diabetes）、感染铜绿假单胞菌（Pseudomonas aeruginosa）或金黄色葡萄球菌（Staphylococcus aureus）的气道感染、长期阿奇霉素（azithromycin）治疗以及口服/吸入糖皮质激素（steroid）的使用，在所有多变量模型中均未与任何炎症标志物（inflammatory marker）项存在显著交互作用。建模时采用生物标志物（biomarkers）的对数转换值作为结局变量。针对同步测量的一秒用力呼气容积占预计值百分比（FEV1%）与年龄别体重Z评分（Weight-for-age z-score）的模型，采用线性回归（linear regression）分析。针对初始痰液采集前1年内发生的急性加重（APE）次数的模型，采用拟泊松回归（quasi-Poisson regression）分析。 b. 研究队列2的数据，样本量n=26。针对稳定期FEV1%测量值、稳定期与APE时点的采集顺序、感染铜绿假单胞菌或金黄色葡萄球菌的气道感染以及阿奇霉素/糖皮质激素使用情况进行额外校正后，上述模型未发现显著效应。 c. 生物标志物对数转换后每单位变化对应的FEV1%平均变化估值。针对稳定期与APE时点的FEV1%差值与APE发作时点测量的粒细胞-巨噬细胞集落刺激因子（GM-CSF，对数转换）之间的关联，采用线性回归模型分析所得结果。将代表临床稳定期与APE时点测量值的单变量依次加入仅包含APE时点GM-CSF的模型中，后者是唯一具有统计学显著性的单变量。研究发现，APE时点测量的白细胞介素5（IL-5，p=0.006）、白细胞介素10（IL-10，p=0.015），以及稳定期时点测量的TCC（p=0.012）均与FEV1%下降呈正相关，且该关联独立于GM-CSF。本研究对包含APE时点GM-CSF、APE时点IL-5、APE时点IL-10以及稳定期TCC的多变量模型进行向后筛选（backward selection），得到了本文展示的最终模型。 d. 临床稳定期测量的生物标志物对数转换后每单位变化对应的5年随访期间APE总预测次数估值。分析结果显示，针对5年随访期间APE次数的关联分析采用拟泊松回归模型。高迁移率族蛋白B1（HMGB-1，对数转换）作为基线炎症标志物是唯一具有统计学显著性的单变量（标注为pp1），最终模型仅保留HMGB-1。对数转换后的HMGB-1每增加1个单位，APE总次数的平均变化量为0.34。