Nonparametric tests for continuous covariate effects with multistate survival data

Area of Science:

• Biostatistics
• Survival Analysis
• Clinical Trials

Background:

• Evaluating covariate effects on multistate event probabilities is crucial in clinical and observational studies.
• Current methods like covariate discretization or regression models have limitations, including information loss and potential bias.

Purpose of the Study:

• To propose novel nonparametric tests for assessing covariate effects on complex multistate event probabilities.
• To overcome limitations of existing methods, such as arbitrary discretization and model misspecification.

Main Methods:

• Developed nonparametric tests using integrals of estimates continuously indexed by covariate dichotomizations.
• Derived general asymptotic results under null and alternative hypotheses.
• Verified results using empirical process theory and demonstrated consistency under stochastic ordering.

Main Results:

• The proposed nonparametric tests avoid arbitrary discretization of continuous covariates.
• The tests are consistent under stochastic ordering, a common feature in multistate data.
• A new nonparametric measure of covariate effect was developed.

Conclusions:

• The novel nonparametric testing procedure offers significant gains over traditional categorization or regression-based methods.
• These methods are robust and provide accurate assessments of covariate effects in multistate event analysis.
• The findings are supported by simulation studies and real-world data analyses.