Novel statistical methods for determination of patient classification in personalized medicine: illustrations in cystic fibrosis
Presenter: Barbara Bailey
When: Friday, July 15, 2016 Time: 9:00 AM - 10:30 AM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
Characterizing and Clustering an Adult Cystic Fibrosis Patient Population using Longitudinal Lung Function Measurements
Cystic Fibrosis (CF) is a multi-systemic disease resulting from mutations in the Cystic Fibrosis Transmembrane Regulator (CFTR) gene and has major clinical manifestations in the sino-pulmonary and gastro-intestinal tracts. Adult CF patient longitudinal lung function data are used to describe and characterize the dynamics of lung function over time. We fit quantile splines and estimate the rate of change of the lung function over time. The estimated derivatives, along with corresponding summary statistics are used for patient clustering. Informative groupings are identified using a proximity matrix generated by unsupervised Random Forests and clustering by Partitioning around Medoids (PAM).
Novel statistical methods for determination of patient classification in personalized medicine: illustrations in cystic fibrosis
Presenter: Sonya Heltshe
When: Friday, July 15, 2016 Time: 9:00 AM - 10:30 AM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
Distinguishing disease severity using high dimensional protein markers �a pragmatic approach
Serum proteomics by mass spectrometry offers identification and relative quantification of thousands of expressed protein isoforms representative of molecular processes; however, it presents the classic big data problem. Our aim is to reduce an exploratory assemblage of proteins to a panel that discriminates cystic fibrosis (CF) disease severity. Eighty-eight CF patient samples were assayed: 44 mild and 44 severe CF (matched on clinical characteristics). Following mass spectrometry, 61,941 isoforms were normalized to relative abundance (RA: 0 to 1). For each isoform, RA was summarized and a battery of paired univariate statistical tests was performed. Dimensionality reduction methods included principal component analysis, logistic regression with lasso, and random forests on the set with p?0.1 from the battery. Initial screening reduced the set to 19,682 isoforms (present in ?8 samples). Thirty-seven isoforms had p?0.01 across the battery of four tests (744 had p?0.1). The variability captured by the 1st and 2nd principal components was 12%, and the degree to which each component predicts mild and severe was sensitive to the number of isoforms in the starting model. The lasso yielded 74 isoforms, of which 21 were significant in the battery of tests at p?0.01. Forest methods performed with a 14.8% classification error rate and identified isoforms that overlapped with the lasso (36/74) and battery (22/37) results. Small sample size, skewness, and the dominating effect of zero detection impede cross-validation and limit statistical power using standard methods. Further refinement and proteomic pathway analysis may support transition from �discovery� to �validation� of protein markers in CF. Supported by The Cystic Fibrosis Foundation Therapeutics, Emory University Children's Fund, and The Children's Foundation Fund of Cincinnati Children's Hospital.
Novel statistical methods for determination of patient classification in personalized medicine: illustrations in cystic fibrosis
Presenter: Rhonda Szczesniak
When: Friday, July 15, 2016 Time: 9:00 AM - 10:30 AM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
A Flexible Mixture Model to Characterize Clinical Personas of Parents of Children with CF
Cystic Fibrosis (CF) is a genetic disease that dramatically decreases life expectancy and quality. Despite the single disease causing gene in CF, there remains large unexplained variability in disease progression and patient outcomes. We introduce methods for identifying classes of patients to better understand these differences and move toward a more personalized approach to treatment.
Novel statistical methods for determination of patient classification in personalized medicine: illustrations in cystic fibrosis
Presenter: Brandie Wagner
When: Friday, July 15, 2016 Time: 9:00 AM - 10:30 AM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
Identification of Pediatric Subjects with Cystic Fibrosis at High Risk of Early Chronic Pseudomonas aeruginosa Infection using a Hidden Markov Multi-state Model with a Frailty Term
Chronic Pseudomonas aeruginosa (Pa) infection in cystic fibrosis (CF) is associated with increased morbidity and mortality. Pa infection is a dynamic process that includes an intermediate subtype where Pa is more easily eradicated compared to the established chronic Pa infection. Thus, it is reasonable to assume that Pa infection, including different strains and diverse virulence factors governs the transition of the initial Pa acquisition to the intermittent subtype and ultimately the chronic infection. Respiratory cultures are commonly performed at quarterly outpatient visits, so there are periods of time where infection is unobserved or interval censored. Moreover, due to the use of oropharyngeal swabs in patients who do not produce sputum, especially in early CF, sampling error may be influencing culture results, thus resulting in misclassification of infection state. Hidden Markov Multi-state models offer an innovative and useful methodology to analyze the course of Pa infection in CF and may be useful for finding factors that influence disease progression. We will utilize data from the ongoing Early Pseudomonas Infection Control (EPIC) Observational study which collected information from young children with CF. This analysis will estimate important quantities such as the length of time until a subject transitions to chronic Pa infection, the ability to assess the effects of the risk factors on this disease process, enabling identification of high risk patients, as well as, the misspecification of the observed states associated with different sample types. This work was funded by Cystic Fibrosis Foundation Therapeutics (OBSERV13K0 � Rosenfeld; WAGNER15A0 -Wagner and Juarez-Colunga).
