Background: Cystic Fibrosis (CF) is a multi-systemic disorder resulting from genetic variation in the Cystic Fibrosis Transmembrane Conductance Regulator (CFTR) gene which can result in bronchiectasis, chronic sinusitis, pancreatic malabsorption, cholestatic liver disease and distal intestinal obstructive syndrome. This study generates multi-dimensional clinical phenotypes that capture the complexity and spectrum of the disease manifestations seen in adult CF patients using statistically robust techniques. Methods: Pre-transplant clinical data from adult (age ≥18 years) CF patients (n = 992) seen in six regionally distinct US CF centers between 1/1/2014 and 6/30/2015 were included. Demographic, spirometry, nutritional, microbiological and therapy data were used to generate clusters using the Random Forests statistical-learning and Partitioning around Medoids (PAM) clustering algorithms. Five commonly measured demographic, physiological and nutritional parameters were needed to create the final phenotypes that are highly similar to a regionally matched group of patients from the CF Foundation Patient Registry Results: This approach identified high-risk phenotypes with expected characteristics including high rates of pancreatic insufficiency, diabetes and Pseudomonas aeruginosa colonization. It also identified unexpected populations including a) a male-dominated, well-nourished group with good lung function with a high prevalence of severe genotypes (i.e. 60% subjects had two minimal function CFTR variations), b) and an older, “survivor” phenotype that had high rates of chronic P. aeruginosa infection. Conclusions: This study identified recognizable phenotypes that capture the clinical complexity in a statistically robust manner and which may aide in the identification of specific genetic and environmental factors responsible for these disease manifestation patterns.
PubMed: MeSH publication types
- Journal Article