The most commonly used statistical methods for the design and analysis of cancer clinical trials today rely heavily on the idea of proportional hazards or its generalizations; i.e., that the important differences between treatments are reflected in the differences between failure rates. Implicit in the use of these methods is the assumption that early failure rate is predictive of long-term outcome. Today, with modern therapy, this assumption increasingly is hard to justify, since in many diagnoses a large fraction of patients are cured by treatment, and the size of the fraction cured can be unrelated to the rapidity with which patients fail. The objective of this research is to develop a useful software product that provides all of the statistical methodological tools and facilities necessary for statistical analysis using sophisticated parametric and semi-parametric models for survival analysis. Particular emphasis will be given to models that explicitly admit a cured fraction. These models are uniquely suited to the problem of distinguishing between the effects of treatment on the fraction of patients who are cured, and the effects of treatment on the distribution of failure times in patients who are not cured. PROPOSED COMMERCIAL APPLICATION: Statistical methods for survival analysis are used in research hospitals, academic departments, cancer centers, and pharmaceutical companies throughout the world by non-statisticians and by statisticians. Cure model analysis is well developed methodologically and is applicable, and in many cases essential, for the correct interpretation of survival data. Ours will be the first product that provides these methods in a comprehensive and useful software package
