On the Identification of Associations between Flow Cytometry Data, Systemic Sclerosis and Cancer
MetadataShow full item record
This work seeks to develop reliable biomarkers of disease activity, progression and outcomes through the identification of significant associations between high-throughput flow cytometry data and a scleroderma clinical phenotype – initially, interstitial lung disease (ILD) - which is the leading cause of morbidity and mortality in Systemic Sclerosis (SSc). A specific aim of the work involves developing a clinically useful screening tool (hereafter a filter). Such a filter could yield accurate assessments of disease state such as the risk or presence of SSc-ILD, the activity of lung involvement and the possibility to respond to therapeutic intervention. Ultimately this instrument should facilitate a refined stratification of SSc patients into clinically relevant subsets at the time of diagnosis and subsequently during the course of the disease, preventing bad outcomes from disease progression or unnecessary treatment side effects. This role could involve a scenario in which an SSc patient passes the presumptive (FVCstpp) test for ILD, but the filter indicates that their flow cytometry (FC) profile is consistent with ILD. In such a case, a physician might: 1) increase frequency of testing to detect early development of ILD; 2) implement more sophisticated diagnostic procedures (e.g., high resolution chest CT scan - HRCT) to confirm the presence of ILD; and 3) consider prophylactic disease modifying treatments. Note that the intention of this research is not to develop screening tools that merely aim at predictive accuracy, but to produce methods that also contribute to the understanding of disease mechanisms. Having used ILD as phenotype, subsequent analyses in this thesis used different phenotypes: antiTopoisomerase (ATA), antiCentromere Anti Nuclear Antibodies (these antibodies are most strongly associated with diffuse and limited systemic sclerosis respectively) and cancer. This research was based on clinical and peripheral blood flow cytometry data (Immune Response In Scleroderma, IRIS) from consented patients followed at the Johns Hopkins Scleroderma Center. Methods. The methods utilized in the work involve: (1) data mining (Conditional Random Forests - CRF) to identify subsets of FC variables that are highly effective in classifying ILD patients; (2) Gene Set Enrichment Analysis (GSEA) to further refine FC subsets; (3) stochastic simulation and Classification and Regression Trees (CART) to design, test and validate ILD filters; and (4) Stepwise Generalized Linear Model (GLM) regression and Drop-in-Deviance testing to identify minimal size, best performing models for predicting ILD status from both FC and selected clinical variables. Results. IRIS flow cytometry data provides useful information in assessing the ILD status of SSc patients. Our hybrid analysis approach proved successful in predicting SSc patient ILD status with a high degree of success (out-of-sample > 82%; training data set 79 patients, validation data set 40 patients). Pre-partitioning patients into groups using CART significantly increased validation performance to 95% successful ILD identification. When the phenotype was Cancer, FC subsets, created through ranked Student t Test scores and point-wise GLM were statistically significant (p < 0.05) using GSEA. After applying Stepwise GLM on the CRF FC subsets, four FC variables were observed to be highly associated with Cancer in SSc patients. An ILD-Cancer GSEA intercomparison was made (use the best ILD FC set with cancer as the phenotype, and vice-versa) showed that GSEA results were highly phenotype-specific. Other phenotypes including ATA and ACA were also analyzed and found to be statistically significantly associated with certain subset of FC variables, but with different FC set sizes (38 and 6 respectively) based on the CRF-GSEA-Stepwise GLM algorithm. In future research, HRCT confirmation of patient ILD status will be a critical next step in developing additional confidence with our approach (and the appropriateness of an 80% FVCstpp threshold for presumptive ILD determination).