TY - JOUR T1 - Robust prediction of later asthma in symptomatic toddlers: A novel approach JF - European Respiratory Journal JO - Eur Respir J VL - 38 IS - Suppl 55 SP - p1496 AU - Anina Pescatore AU - Ben Spycher AU - Lutz Duembgen AU - Marie-Pierre Strippoli AU - Cristian Dogaru AU - Michael Silverman AU - Claudia Kuehni Y1 - 2011/09/01 UR - http://erj.ersjournals.com/content/38/Suppl_55/p1496.abstract N2 - Aim: Many children have asthma-like symptoms in early life, but few develop asthma. Several models for predicting later asthma in symptomatic toddlers have been built, but some included factors that are difficult to assess, and methods used were prone to overfitting, leading to selection or exaggeration of irrelevant factors. We aimed to identify predictors for later asthma avoiding previous limitations.Methods: In a population-based cohort, we selected 1-3 year-olds with respiratory symptoms (current wheeze or recurrent cough) and related healthcare visits. Asthma (current wheeze and treatment) was assessed 5 years (N=1226) and 8 years (N=866) later.The included factors are easy to assess in clinical practice: family history, symptoms at baseline, demographic and perinatal data. We used lasso penalized logistic regression to select predictors. This minimizes the number of included predictors while maximizing area under ROC curve (AUC).Results: Main predictors selected in the model for asthma 5 yrs later (AUC=0.76) were ≥4 wheezing attacks in the past 12mo (OR=1.65), wheeze causing breathlessness (3.1) and activity disturbance (2.4), eczema (1.5) and male sex (1.5). Other predictors (OR<1.5) were: non-viral triggers for wheeze or cough, parental history of asthma, older age at baseline and low birth weight. The results for asthma 8yrs later (AUC=0.72) were similar.Conclusion: Among factors easy to assess in symptomatic toddlers, wheeze severity, eczema and male sex are main predictors of asthma in mid-childhood. Because our approach for variable selection avoids overfitting, the resulting prediction models should perform well with new data. However, external validation is needed.Funding: SNF PDFMP3-123162. ER -