Seminar Series: Dr. Aya Mitani
Dynamic risk prediction for clustered data: recent advances in joint models and landmark models with application to periodontitis
Aya Mitani
, PhD, MPH
Associate Professor
Division of Biostatistics
Dalla Lana School of Public Health
University of Toronto
Short Biography:
Dr. Mitani's research focuses on developing statistical methods for analyzing correlated data and reducing biases that can arise from informative cluster size, study design, missing data, and misclassification in multilevel observational studies and complex surveys. She is particularly interested in applying these methods to real-world health problems through interdisciplinary collaborations. Her recent work has focused on oral health and substance use research, although she has collaborated with researchers across a wide range of disciplines.
Abstract:
Modern biomedical studies increasingly collect multilevel longitudinal data alongside time-to-event outcomes, creating both opportunities and challenges for individualized risk prediction. Motivated by tooth loss in periodontitis, this talk presents recent advances in dynamic survival prediction for clustered data where repeated measurements are nested within units (teeth) and units are nested within clusters (patients). I will describe (i) a clustered dynamic landmarking framework that updates predictions at prespecified times using each tooth’s history while borrowing strength across teeth through hierarchical mixed-effects and Gaussian-process models, and (ii) multilevel joint models that link continuous and binary biomarkers (e.g., probing pocket depth and mobility) to time to tooth loss while accounting for intra-tooth and intra-patient correlation. I will demonstrate the methods using electronic dental records of the members of the Canadian Armed Forces.
Area of Research:
Correlated data, dynamic prediction, complex surveys, oral health
Date: Friday, November 27
Time: 1:30 pm - 2:30 pm
Location: PHFM 3015 (Western Centre for Public Health and Family Medicine) or Zoom (request link by email epibio@uwo.ca )