Background: It is widely acknowledged that wheezing disorders in children consist of various distinct disease phenotypes. As therapeutic and preventive strategies differ between them, identification of phenotypes early in life is a prerequisite for targeted disease management. Previous attempts to distinguish phenotypes of wheeze in children have taken into account only one or very few of the many clinical dimensions of wheezing disorders, and have usually been retrospective. Objectives: We aimed to: 1) develop statistical models of incidence and prognosis of wheezing disorders in early childhood, based on hereditary disposition, timing of exposure to environmental risk factors and previous medical history; 2) develop different definitions for separate phenotypes of wheezing disorders and compare their predictive power; 3) investigate, how phenotypes vary by gender and ethnicity; and 4) develop a simple algorithm allowing to estimate prognosis in in preschool children with wheeze. Methods: We are analysing data from three population-based childhood cohorts: a) Leicester 1990 cohort (N=1,650); b) Leicester 1998 cohort (N=8,700) (www.leicestercohorts.org); and c) Birth cohorts in Bern (N=460). We have routine data on perinatal events, immunisation, and growth; symptoms, treatment and environmental exposures were assessed repeatedly. Objective outcomes include lung function, bronchial challenge and allergy tests. Analytic methods include logistic regression, Poisson and Cox regression, and probabilistic clustering methods (latent class analysis). The accuracy and generalisability of models will be validated across the independent cohorts.Results so far: We found: 1) That ethnicity, social class and morbidity influence accuracy of parental reports and must be adjusted for. 2) That the effects of risk factors and pathophysiological pathways differ by age, ethnic group and phenotype of wheeze. For instance, the effect of prenatal tobacco smoke decreases with age, mannan-binding lectin and exhaled NO play a different role in different phenotypes, and the risk to develop multiple atopic wheeze, but not viral wheeze, is higher in South Asians. Therefore epidemiological studies on childhood wheeze need to model the complexity of interactions between age, sex, ethnicity and environment in a time-dependent fashion. 3) Using latent class analysis, a probabilistic clustering model, we distinguished three phenotypes of wheeze and two phenotypes of chronic cough which were plausible and had prognostic value. This novel multidimensional approach has thus the potential to identify clinically relevant diagnostic entities in paediatric airway disease.