A semiparametric accelerated failure time-based mixture cure tree.

Area of Science:

• Biostatistics
• Survival Analysis
• Machine Learning in Healthcare

Background:

• The mixture cure rate model (MCM) is standard for survival data with a cured subgroup.
• Traditional MCMs often use generalized linear models (e.g., logit) for cure probability, limiting covariate effect modeling.
• Existing methods struggle with nonlinear relationships in cure probability prediction.

Purpose of the Study:

• To propose a novel mixture cure rate model (MCM) incorporating decision trees for cure probability.
• To enhance the modeling of cure probability by capturing complex, nonlinear covariate effects.
• To improve the accuracy and precision of cure probability estimates and overall predictive accuracy.

Main Methods:

• Developed a new MCM where cure probability is modeled via a decision tree classifier.
• Modeled the survival distribution of the uncured subgroup using an accelerated failure time (AFT) structure.
• Implemented an expectation-maximization (EM) algorithm for parameter estimation.

Main Results:

• The proposed decision tree-based MCM demonstrated superior performance in capturing nonlinear classification boundaries compared to logit and spline-based MCMs.
• Achieved more accurate and precise estimates of cured probabilities, leading to improved predictive accuracy.
• Showcased enhanced estimation for the survival distribution of uncured subjects due to better nonlinear boundary capture.

Conclusions:

• The novel MCM effectively models nonlinear relationships in cure probability, outperforming traditional methods.
• This approach significantly improves the accuracy of cure probability estimation and survival prediction.
• The proposed method offers a powerful tool for analyzing complex survival data, as demonstrated with bone marrow transplant data.