Page 73 - ADULT-ONSET ASTHMA PREDICTORS OF CLINICAL COURSE AND SEVERITY

DIAGNOSTIC ACCURACY OF MINIMALLY INVASIVE MARKERS FOR DETECTION OF AIRWAY EOSINOPHILIA IN ASTHMA
dis nguished eosinophilic from non-eosinophilic pa ents, included at least an arbitrary number of 50 asthma c pa ents, and were published before January 2014. We contacted corresponding authors through email, and asked whether they were willing to calculate and share es mates of accuracy or to send their (blinded) dataset.
Whenever we obtained datasets, we evaluated diagnos c accuracy as follows. First, we es mated the ability of each index test to discriminate between pa ents with and without airway eosinophilia by calcula ng the Area Under the Receiver Opera ng Characteris c Curve (AUC-ROC). Then we selected the “op mal cutpoint” of sensi vity and speci city on the ROC curve using the Youden index, as had been done by almost all included diagnos c accuracy studies. Depending on the reference standard available, we repeated this analysis for each de ni on of airway eosinophilia used in the included studies. Pa ents with missing data on the index test or reference standard were excluded from the analysis for that speci c marker. Datasets were analysed using R v3.0.
META-ANALYSIS
We analysed studies in children and adults separately. To get a view of the overall diagnos c performance of each marker, we performed random e ects meta-analysis of AUC es mates11, independent of which reference standard or de ni on of airway eosinophilia had been used. Whenever a study reported more than one AUC es mate for one marker in the same group of pa ents, for example because the study relied on mul ple de ni ons of airway eosinophilia, we included the highest AUC reported. If a study reported an AUC es mate in the total study group as well as in subgroups, we only included the es mate for the total study group. However, if a study reported on these es mates in subgroups only and not in the total study group, we included the AUCs of all subgroups. If su cient data were available (≥3 studies), we repeated this meta-analysis for studies that had used the same reference standard and airway eosinophilia de ni on. We assessed sta s cal heterogeneity using the I2 sta s c12.
From each collected or reconstructed 2x2 table, we calculated es mates of sensi vity and speci city and 95% CIs. We used a hierarchical random e ects model8 to obtain summary es mates of sensi vity and speci city for studies that had used the same reference standard and airway eosinophilia de ni on. We did so whenever four or more tables were available. If ar cles provided data on direct, head-to-head comparisons of two or more markers, we evaluated whether there were signi cant di erences in accuracy between markers. Such direct comparisons ensure that di erences in accuracy are not caused by heterogeneity across study popula ons. We used Deeks’ funnel plot asymmetry test to assess risk of publica on bias13. SAS v9.2 was used to t the models.
71