Idiopathic pulmonary fibrosis (IPF) is a devastating disease of unknown etiology occurring in older adults. IPF is ultimately fatal with a median survival of 2 to 5 years, and exhibits a highly heterogeneous natural history. Broad categories of disease progression have been defined, but are not predictable at the time of diagnosis. These designations of natural history assume great importance at a time when insights from preclinical studies are beginning to translate into therapies targeted at specific key pathways of fibrosis. Stratification of disease phenotypes is important in order to decipher the effects of newly approved therapies among individuals with biologically dissimilar natural histories and to better tailor therapy to individual patients. Various prognostic tools have been developed for IPF that correlate with overall survival; most use clinical and functional variables independent of imaging findings. Prognostic determinants based on imaging features rely largely on subjective visual assessment of disease. In contrast, no good early predictive models exist that anticipate the natural history of disease in advance of significant functional decline. Given the indispensable role of high resolution computed tomography (HRCT) in the diagnosis and surveillance of IPF, we propose to mine the rich information in HRCT data sets to develop robust, quantitative features that can anticipate disease progression in advance of debilitating respiratory compromise. We propose to use the anonymized clinical data and source images on 234 patients with IPF from multicenter trials, and whose data are archived at the UCLA Computer Vision and Imaging Biomarkers Laboratory. Using an image processing pipeline developed in our laboratory for quantitative image analysis, we will train a classifier on scans annotated manually by an expert radiologist, analyzing in separate aims static image features present on baseline scans and transitional (difference) morphologic features on sequential scans that herald progressive disease. Features of anatomic distribution will be explored and reproducible imaging features will be expressed with a quantitative lung fibrosis (QLF) score. Aggregate prognostic models using Cox proportional regression models will be derived using only clinical covariates and combined clinical and imaging covariates, correlating these models with progression free survival. Finally, we will externally validate our models in an independent institutional registry of clinical and image data on patients with IPF seen in the UCLA Interstitia Lung Disease Program. Our objectives are centered on the goals of using preexisting datasets to develop clinically meaningful models that anticipate disease course in patients with IPF. We anticipate that these models can be used clinically at the individual patient level to enable more informed and timely management decisions for the choice in treatment as well as future research to define more homogeneous cohorts for testing new safe and effective therapies and to better elucidate the effects of therapies in patients with biologically heterogeneous disease progression.
Idiopathic pulmonary fibrosis (IPF) is a devastating disease of older adults that now has a few treatment options: The natural history of IPF and its rates of progression are highly variable, which hampers timely decisions about referral for lung transplantation or treatments using newer drug therapies. This research takes advantage of clinical and imaging datasets previously collected for research or clinical purposes, and will define image features using computer analysis of computed tomography (CT) images to predict disease course in advance of respiratory deterioration. The success of this research will enable us to distinguish between patients with slowly versus rapidly progressive disease, leading to more timely management decisions, and may help us to understand which patients might benefit from novel promising therapies or treatments and which may not.
