More than just a number: modelling biological aging and vulnerability

Abstract

Aging is a complex process, and chronological age alone does not reflect how far someone has progressed in their individual aging trajectory. Biological age aims to capture this underlying state, but estimating it accurately is difficult. This dissertation, More than just a number: modelling biological aging and vulnerability, evaluates statistical methods for biological age prediction and stresses the need for rigorous methodology.

Many current predictors use cross-sectional data and regress chronological age on a set of biomarkers. Since biological age is latent, these models rely on the strong, untestable assumption that associations with chronological age reflect the underlying true aging rate. Chapters 2 and 3 of this thesis show that this leads to biased estimates and systematic distortions from regression to the mean, particularly when comparing different age groups.

Chapter 4 introduces the AccelerAge framework, which uses an Accelerated Failure Time model to estimate how biomarkers influence remaining lifespan. Chapter 5 presents a penalized reduced-rank regression model for multi-outcome survival data, revealing latent factors linked to diseases and lifespan. Chapter 6 develops a framework to disentangle components of excess mortality in vulnerable groups during crises, improving insight into disproportionate impacts on such groups.