83 A multivariable composite outcome to define disease severity in children with cystic fibrosis. (30th November 2020)

• Record Type: Journal Article
• Authors: Filipow, Nicole; Davies, Gwyneth; Main, Eleanor; Stanojevic, Sanja
• Is Part Of: Archives of disease in childhood. Volume 105(2020)Supplement 2
• Page Start: A28
• Page End: A29
• Publication Date: 2020-11-30
• Subjects: Children -- Diseases -- Periodicals; Infants -- Diseases -- Periodicals; 618.920005
• DOI: 10.1136/archdischild-2020-gosh.83
• Languages: English
• ISSNs: 0003-9888

Abstract

Abstract : Introduction: Improvements in cystic fibrosis (CF) care have resulted in improved outcomes, specifically many children maintain lung function in the normal range. Nonetheless there are children with poor outcomes, and there is a need for a more comprehensive multi-factorial measure that summarises the overall health status in this new era of CF care. Objective: To define phenotypically distinct clusters of pediatric CF patients that are linked to different health outcomes by using basic machine learning algorithms. Method: Data from the Toronto CF Clinical Database were used to define phenotypic clusters based on a broad variety of patient-descriptive variables. A Partitioning Around Medoids (PAM) clustering method was iteratively carried out on different combinations of the variables until a maximum distinction between outcome measures could be identified, which included time to recurrent event analyses for both pulmonary exacerbations and hospital admissions. The results were validated in GOSH CF clinical data housed within the GOSH-DRIVE DRE. Results: We were able to define 4 discrete clusters of patients based on 9 routinely collected clinical variables using data from 537 patients and 12200 encounters. Lung function outcomes were not used to define the clusters, however, there was a distinction between the different clusters, such that the cluster with the poorest outcomes also had the worst lung function. The cluster with the poorest outcomes also had the greatest risk of hospitalization and pulmonary exacerbation, which suggests that the approach correctly identifies patients with a more severe disease phenotype. The results were consistent in the GOSH clinical data. Conclusion: Four clusters of pediatric CF patients were identified with corresponding differences in clinical characteristics and outcomes. Future work will identify risk factors for transitioning to a severe disease cluster, and those factors that may improve health outcomes.