Survival model construction guided by fit and predictive strength

Area of Science:

• Biostatistics
• Survival Analysis
• Medical Statistics

Background:

• Survival model construction relies on goodness-of-fit and predictive strength.
• Integrating these distinct techniques into a unified framework is crucial.
• Characterizing variable effects, especially time dependencies, is key.

Purpose of the Study:

• To unify goodness-of-fit and predictive strength measures within a single framework for survival model construction.
• To guide the characterization and coding of variable effects, including time dependencies.
• To demonstrate the practical application of these integrated techniques in breast cancer studies.

Main Methods:

• Utilizing simple graphical techniques for visual goodness-of-fit assessment.
• Employing formal theorems to support the development of richer models from simpler ones.
• Integrating measures of predictive strength with goodness-of-fit techniques.

Main Results:

• Graphical techniques provide intuitive indications of model fit and suggest alternative models when assumptions are violated.
• Formal theorems underpin the intuitive graphical methods and the process of model enrichment.
• Predictive strength measures, guided by formal theorems, help identify models that best approximate the underlying survival data-generating mechanism.

Conclusions:

• The integrated framework enhances the efficiency of survival model construction.
• These methods are valuable for understanding complex variable interactions, including time dependencies.
• The approach offers practical guidance for analyzing survival data, as illustrated by breast cancer research.