Hierarchical Nonlinear Mixed Effect Modeling: Defining Post-radiation Therapy Relapse in Prostate Cancer Patients
Abstract
The research accomplished and described here validates and extends a model to classify prostate cancer patients according to disease relapse following definitive radiation therapy. The original model was developed within a hierarchical nonlinear mixed effect modeling framework with likelihood based estimation incorporating the EM algorithm. The model was tested statistically using a subset of 35 patients with relatively homogenous tumor and treatment characteristics. The research described in this report successfully applied the methodology to a larger population of men (>600 patients) representing all stages of disease via the modeling of covariates, including tumor differentiation, stage, and pre-treatment PSA. The success of the modeling was dependent upon a Bayesian framework with Markov chain Monte Carlo methodology for estimating mixture distribution parameters. Poor mixing and slow convergence were encountered and required various re-parameterizations and creative initialization techniques. The analysis includes an assessment of predictors of post-nadir rise, as salvage therapy strategies are often designed around the rate of increase in PSA levels post-nadir, as well as an analysis of predictors of initial decline and its relationship to outcome. The modeling was compared to biochemical classification using a clinical definition of relapse and also to clinical results as obtained from imaging and/or biopsy.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2004
- Accession Number
- ADA432448
Entities
People
- Alexandra L. Hanlon
Organizations
- Fox Chase Cancer Center