A penalized reduced rank proportional hazards model for age-related multi-morbidity event data

M. Rodríguez-Girondo

The recent incorporation of detailed age-at-disease-onset profiles from electronic health records in large epidemiological studies like the Leiden Longevity Study and the UK Biobank provides valuable opportunities to investigate the factors influencing age-related multi-morbidity. However, the complexity of the data, including multi-dimensional time-to-event outcomes and high-dimensional covariates is challenging. We propose a novel methodological framework for analyzing this type of data using a novel Lasso-penalized reduced-rank proportional hazards model. This model enables simultaneous fitting on the age-at-disease-onset of multiple age-related diseases, assuming the existence of shared latent factors underlying all considered age-related diseases. To deal with high-dimensional omics covariates, we propose incorporating a Lasso-type penalization. The performance of the new method is illustrated using UK Biobank data, utilizing metabolomics data as predictor variables.

Keywords: Reduced-rank regression, survival analysis, penalization, multivariate outcome

Scheduled

GT04.BIO1 Invited Session. Complex Data
November 7, 2023  3:30 PM
CC3: Room 1

Graphical model inference with external network data

J. Jewson

Youden index-based methods to select and combine biomarkers and covariates.

C. J. Salaroli, M. C. Pardo