A model-informed rank test for right-censored data with intermediate states.

Area of Science:

• Biostatistics
• Survival Analysis
• Clinical Trials

Background:

• Traditional survival analysis methods like log-rank and generalized Wilcoxon tests are standard for comparing survival distributions.
• These methods may not fully utilize available data when subjects transition through intermediate disease states with unknown exact times.
• Auxiliary information on disease progression can enhance the accuracy of survival comparisons.

Purpose of the Study:

• To develop a modified Wilcoxon test that incorporates auxiliary information from intermediate disease states.
• To improve the power of survival distribution comparisons when exact transition times are unknown.
• To provide a robust statistical method for analyzing diseases with multiple stages, such as cancer staging.

Main Methods:

• Fitting a multi-state Markov model to the complete dataset to capture disease progression.
• Estimating the probability of one subject surviving longer than another, given censoring times and last observed status.
• Calculating expected ranks based on these probabilities to form the basis of a new test statistic.

Main Results:

• Simulations show the proposed test can increase statistical power compared to standard log-rank and generalized Wilcoxon tests.
• The modified test maintains the nominal type 1 error rate, ensuring reliability.
• The method's utility is demonstrated using a real-world dataset from amyotrophic lateral sclerosis patients.

Conclusions:

• The modified Wilcoxon test offers a powerful alternative for survival analysis in the presence of intermediate disease states.
• This approach enhances the utilization of longitudinal data in clinical studies.
• The method is applicable to various diseases with complex progression pathways, improving clinical trial efficiency.