Prediction of transition probabilities in multi-state models with nested case-control data.

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Multi-state models are crucial for analyzing complex, interrelated life events.
• Nested case-control (NCC) sampling is used in resource-limited settings but limits data reuse for multiple events.
• Inverse probability weighting (IPW) offers an alternative for inference with NCC data, primarily for relative risk estimation.

Purpose of the Study:

• To extend IPW-based pseudolikelihood methods for predicting transition probabilities in general multi-state models.
• To evaluate and improve the efficiency of IPW methods for transition probability prediction.
• To propose and validate novel, more efficient IPW approaches.

Main Methods:

• Developed two novel IPW-based pseudolikelihood approaches for enhanced efficiency in transition probability prediction.
• The first approach calibrates design weights using cohort-level information.
• The second approach jointly models transitions originating from the same state, deriving explicit variance estimates.

Main Results:

• Simulation studies confirmed substantial efficiency gains from both proposed IPW methods.
• The combined application of both novel approaches yielded further significant improvements in efficiency.
• Methods were illustrated using real-world data from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial.

Conclusions:

• The proposed IPW methods significantly enhance the efficiency of transition probability prediction in multi-state models using NCC data.
• These novel approaches overcome limitations of standard IPW methods, enabling more robust analysis in resource-constrained environments.
• The findings provide valuable tools for epidemiological research and public health studies involving complex event trajectories.