Joint modelling of survival data

Challenge

Modelling the association between a longitudinal biomarker (tumour size) and time-to-event outcome (overall survival) to improve outcome predictions:

  • Need to estimate the biomarker profile over time while adjusting for informative dropout due to death

  • Aim is to predict survival outcome for censored patients using biomarker profile up to a specific timepoint

Request

Predict the survival time of patients with solid tumours in an oncology treatment trial using longitudinal data on sum of diameters of target lesions

Response

  • Used joint modelling within a Bayesian framework to model tumour size and overall survival simultaneously
  • Estimated the association between tumour size and mortality and obtained survival predictions for individual patients who were still alive
  • Used patient-specific predictions to extrapolate the study results into the future and compared to a standard parametric survival model
  • Used hierarchical modelling to provide subgroup-specific estimates of association and predicted survival by primary tumour site

Statistical analysis