Predicting pediatric asthma admissions for Barbados

Abstract

Rationale To predict future admissions of pediatric asthma using temporal patterns embedded in 6 years of historical data. Methods Data from 1997 to 2002 on asthmatic admissions to the Accident and Emergency Department (A&ED) of the Queen Elizabeth Hospital, Barbados, were analyzed using neural network models. Weekly admissions were modeled using the previous ten weeks of admissions and knowledge of the week being predicted (1-52). Three hundred and two data sets were produced for modeling (52 weeks × 6 years) minus 10 lead weeks. Results Pediatric admissions to A&ED for asthma over the study period averaged 63% of total asthmatics attending. 1997 had the highest number of admissions and 2001 the lowest. The major peak in admissions for all years occurs between weeks 34 to 48 (mid September to end of November). The best neural network model had a high prediction accuracy, with root mean square errors of 22 admissions per week (average admissions per week were 122) and an R 2 of 0.78. Conclusion Models developed to predict asthma admissions can accurately predict the level of asthma admissions for the next week. This information can be used to aid in identifying future peak periods of asthma admissions, alert health officials and educate individual patients of increased risk and mitigate asthma events that lead to A&ED admissions.