Simple tool predicts future asthma in preschoolers
medwireNews: Researchers have developed a tool that can predict future asthma in preschool children presenting with wheeze and cough on the basis of patient history alone.
They say that it could provide a simple, noninvasive means of asthma prediction in primary care and an alternative to previously proposed tools, which may be less accurate or require blood or skin prick tests.
“The proposed asthma prediction tool is simple and uses information that is noninvasive and easy to assess,” the team, led by Claudia Kuehni (University of Bern, Switzerland), comment. “This makes it an ideal instrument for use in clinical practice and research.”
They developed a 10-item questionnaire using data from a group of 5878 children from the UK aged 1–3 years old, 2444 of whom had reported respiratory symptoms. Among the 1226 children who were followed up to school age (mean duration, 4.5 years), 345 (28.1%) developed asthma, they report in the Journal of Allergy & Clinical Immunology.
Kuehni and team used a method known as LASSO-penalized logistic regression to narrow down the 22 most important predictors for future asthma, the five most significant of which were shortness of breath, frequent wheeze, wheeze without colds, activity disturbance by wheeze, and wheeze/cough triggered by exercise.
This method helps to avoid overfitting, which the authors explain is an issue that has afflicted the creation of other models.
Other factors in the model included age, birthweight, and parental history, but not socioeconomic status, which could limit the generalizability of the tool.
Applying this model to the population data resulted in an area under the curve (AUC) of 0.78 and a scaled Brier score of 0.23, where 1 equals perfect prediction. And, when the authors revised the model to use only 10 variables, it demonstrated an almost comparable discriminative ability.
While a maximum score of 15 is associated with a 95% probability of developing asthma 5 years later, the authors found that a score of 5 had a sensitivity and specificity of 72% and 71%, respectively, and corresponding values for a score of 10 were 22% and 98%.
Noting that the model stood up well to internal validation, with only a minor decrease in performance (Brier score, 0.16; AUC, 0.74), Kuehni and team say that it must now be validated in external populations.
They write: “We encourage the application and validation of this tool in ongoing epidemiologic studies and clinical care (particularly primary care).”
medwireNews (www.medwirenews.com) is an independent clinical news service provided by Springer Healthcare Limited. © Springer Healthcare Ltd; 2013
By Kirsty Oswald, medwireNews Reporter