Nonparametric models and methods for designs with dependent censored data: part I

Area of Science:

• Statistics
• Biostatistics
• Nonparametric methods

Background:

• Repeated measures designs are common in various scientific fields.
• Censored data presents unique challenges in statistical analysis.
• Existing methods often rely on strong assumptions like proportional hazards or location shift.

Purpose of the Study:

• To develop a nonparametric (NP) approach for analyzing repeated measures designs with censored data.
• To extend existing NP methodology for hypothesis testing in such designs.
• To provide a flexible analytical tool when standard assumptions are not met.

Main Methods:

• Utilized the nonparametric model of Akritas and Arnold for marginal distributions.
• Developed test procedures for hypotheses of no main effects, no interaction, and no simple effects.
• Derived large-sample distributions based on an i.i.d. representation for Kaplan-Meier integrals.

Main Results:

• The proposed NP procedures do not require modeling assumptions, offering wider applicability.
• The methodology is effective for censored, ordinal, and tied data.
• Small-sample approximations were developed and validated through simulation studies.

Conclusions:

• The NP approach provides a valuable alternative for analyzing repeated measures with censored data.
• It offers robust testing even when data violates assumptions of other methods.
• The methodology proved useful in real-world examples, including censored and missing data scenarios.