Prediction models for childhood asthma: A systematic review.

Area of Science:

• Pediatric Pulmonology
• Epidemiology
• Biostatistics

Background:

• Objective diagnosis of childhood asthma before age five is difficult, leading to undertreatment or overtreatment.
• Predictive tools for school-age asthma risk can aid early asthma care in preschoolers.
• This review systematically appraises models predicting school-age asthma from data in children aged five and under.

Purpose of the Study:

• To systematically identify and critically appraise studies developing or updating prediction models for school-age asthma.
• To evaluate the performance and generalizability of existing asthma prediction models.
• To explore potential improvements for future asthma prediction models.

Main Methods:

• Searched MEDLINE, Embase, and Web of Science Core Collection up to July 2019.
• Included studies using data from children ≤5 years to predict asthma in school-age children (6-13 years).
• Evaluated development and validation of predictive models, including regression-based and machine learning approaches.

Main Results:

• Identified 24 studies with 26 predictive models (21 regression-based, 5 machine learning).
• Model performance (AUC) ranged from 0.66-0.87; external validation showed modest generalizability (AUC 0.62-0.83).
• Fifteen models required additional clinical tests; models showed moderate ability to rule in or rule out asthma, but not both.

Conclusions:

• Existing asthma prediction models offer moderate performance and generalizability.
• Traditional methods have limitations impacting predictive accuracy.
• Machine learning approaches show promise for enhancing future school-age asthma prediction.