Statistical methods that are applied to trials of early or prophylactic interventions in childhood asthma may differ in important respects from those currently used for therapeutic trials in established asthma, which typically involve randomization of individuals and measurement of within-individual changes in continuous measures of disease severity. Randomization of small numbers of larger units (e.g., health centres) is less effective for controlling confounding than randomization of individuals. Because early interventions are by definition targeted at healthy children or those with mild disease, outcomes cannot usually be assessed as within-subject changes, but are analysed as between-subject comparisons of traits, states or events. Methods derived from observational epidemiology, including multiple regression, logistic regression and proportional hazards regression, are appropriate here. When there are repeated measures of outcome (such as lung function monitored at intervals through childhood), random effects models or mixed longitudinal models may be used. Calculations of sample size need to take into account the proposed form of randomization and analysis, and also the efficacy of the intervention and degree of compliance. Multiple end-points should be ranked for importance at the start of a trial and greater weight attached to significant results for primary outcomes.
|Original language||English (US)|
|Journal||European Respiratory Journal, Supplement|
|State||Published - 1998|
ASJC Scopus subject areas
- Pulmonary and Respiratory Medicine