Seminar Series: Dr. Osvaldo Espin-Garcia
Challenges and Opportunities for Methodological Developments in Osteoarthritis Research
Osvaldo Espin-Garcia
Assistant Professor (tenure-track)
Department of Epidemiology and Biostatistics
Western University
Assistant Professor (status-only)
Dalla Lana School of Public Health
University of Toronto
Principal Biostatistician
Department of Biostatistics
University Health Network
Toronto, ON
Short Biography:
Dr. Osvaldo Espin-Garcia is Assistant Professor at the Department of Epidemiology and Biostatistics, University of Western Ontario (UWO). He holds a PhD in Biostatistics from the Dalla Lana School of Public Health, University of Toronto (UofT) an MMath in Statistics-Biostatistics from the University of Waterloo and a BSc in Actuarial Sciences from the National Autonomous University of Mexico (UNAM). Prior to joining UWO, Dr. Espin-Garcia worked as Principal Biostatistician at the University Health Network. His research program focuses on developing statistical, machine learning and computational methods for statistical genetics, genetic epidemiology, and deep phenotyping of complex traits such as cancer, Crohn's disease, or osteoarthritis. Lastly, Dr. Espin-Garcia currently serves as a scientific member of the Ontario Cancer Research Ethics Board (OCREB) and is part of three training initiatives: first, the CANSSI Ontario Strategic Training for Advanced Genetic Epidemiology (STAGE); second, the Health Data Working Group at UofT; and third, the Collaborative Specialization in Machine Learning in Health and Biomedical Sciences as UWO.
Abstract:
Osteoarthritis (OA) is a degenerative chronic joint condition that affects 1 in 7 Canadians. Currently, there is no cure nor approved disease-modifying drug therapies for OA. It has been long recognized that OA has a complex etiology including both genetic and epigenetic factors with diagnosis and monitoring of OA typically performed via radiographic imaging, i.e., x-rays, and self-administered questionnaires. The role of physical activity (PA) as a recommended first line of treatment for OA symptoms is also undisputed. In this talk, I will highlight some of my current research, which has been motivated by several problems in the field of OA.
I will begin with recently published work aiming to categorize longitudinal response trajectories of knee joint space width measured via x-rays. The categorization is achieved through a novel Bayesian latent class linear mixed model which accounts for measurement error and monotonicity. These model features are necessary as disease characteristics are subject to measurement error at a given time due to variations in diagnostician, x-ray machine or knee positioning. Moreover, assuming a monotonic, i.e. non-decreasing or non-increasing, behaviour shows promise in reflecting the disease condition due to its chronic nature.
After that, motivated by the usage of non-pharmaceutical interventions to manage OA, I will outline applied work leveraging curve registration methods to quantify PA levels using accelerometer data. Indeed, research suggests that increased levels of PA have beneficial effects on OA, but self-reported information may be biased. Consequently, wearable devices such as accelerometers, which can accurately and objectively quantify body movements, are increasingly used. However, activity patterns obtained from accelerometers may vary substantially across individuals hindering comparability. Rooted in functional data analysis, curve registration methods aim to address this hurdle by aligning the data to a homogenized -or registered- time scale. The alignment involves an iterative two-step procedure alternating between the estimation of 1) an inverse warping function to capture horizontal variability, and 2) the functional principal components to encapsulate the vertical variability in the patterns from step 1. At last, quantification of individual PA levels is achieved by computing the area under the registered curve.
Finally, I will introduce FunColoc, a multi-trait-multi-variant approach for variant and locus-level colocalization testing between trait pairs. FunColoc is based on a generalized multivariate scalar-on-function regression model which estimates a smoothed function for a sequence of genotypes in a locus using a trait-specific variant effect function. The model is flexible and can accommodate traits with various distributions while assesses colocalization through a product of variant effect functions at the sub-region and region levels under a composite null hypothesis. Statistical significance is evaluated through permutation-based and adaptive bootstrap approaches. FunColoc was inspired in response to the need of an individual-level colocalization analysis for osteoarthritis progression and micro-RNA expression levels. I will conclude the presentation by commenting on additional pressing challenges and future work.
Keywords:
Biostatistics, Statistical genetics and genomics, Cost-efficient Study design, Longitudinal data, Latent variable models, Mixture modelling, Missing data
Date: Friday, October 25
Time: 1:30 pm - 2:30 pm
Location: PHFM 3015 (Western Centre for Public Health and Family Medicine) or Zoom (link may be requested at EpiBio@uwo.ca)
