Statistical Inference on the Cure Time

TL;DR

This paper introduces a new statistical framework for estimating cure time in cancer survival analysis, addressing limitations of previous methods that assumed infinite cure time, and applies it to real-world data for improved health policy insights.

Contribution

It defines a generalized concept of statistical cure via conditional survival and develops new models to estimate cure time, enabling practical inference and application in health care.

Findings

Cure times estimated for 22 major cancers in Taiwan.

Statistical inference on cure time demonstrated with covariate analysis.

Methodology validated through simulation studies.

Abstract

In population-based cancer survival analysis, the net survival is important for government to assess health care programs. For decades, it is observed that the net survival reaches a plateau after long-term follow-up, this is so called ``statistical cure''. Several methods were proposed to address the statistical cure. Besides, the cure time can be used to evaluate the time period of a health care program for a specific patient population, and it also can be helpful for a clinician to explain the prognosis for patients, therefore the cure time is an important health care index. However, those proposed methods assume the cure time to be infinity, thus it is inconvenient to make inference on the cure time. In this dissertation, we define a more general concept of statistical cure via conditional survival. Based on the newly defined statistical cure, the cure time is well defined. We develop cure time model methodologies and show a variety of properties through simulation. In data analysis, cure times are estimated for 22 major cancers in Taiwan, we further use colorectal cancer data as an example to conduct statistical inference via cure time model with covariate sex, age group, and stage. This dissertation provides a methodology to obtain cure time estimate, which can contribute to public health policy making.