New weighting methods when cases are only a subset of events in a nested case-control study

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trials

Background:

• Nested case-control (NCC) studies are vital for risk model development using costly biomarkers in large cohorts.
• Traditional NCC designs obtain biomarker data from all events (cases) and a subcohort of controls.
• Untypical NCC designs, selecting a subset of cases due to resource constraints, face challenges with existing biased estimators.

Purpose of the Study:

• To develop novel inverse probability weighted (IPW) estimators for untypical NCC designs.
• To provide consistent estimation for risk model parameters and predictive performance metrics.
• To extend applicability beyond proportional hazards models to time-specific generalized linear models.

Main Methods:

• Proposed new weighting methods for untypical NCC designs.
• Developed an inference procedure using perturbation resampling to account for sampling-induced variance and covariance.
• Validated methods for both typical and untypical NCC designs, and for proportional hazards and time-specific generalized linear models.

Main Results:

• The new weighting methods yield consistent IPW estimators for relative risk and predictive performance metrics.
• The proposed methods demonstrate reduced bias and variance compared to existing approaches in untypical NCC designs.
• The perturbation resampling procedure accurately captures variance and covariance from sampling processes.

Conclusions:

• The novel weighting methods provide a robust and accurate approach for risk model evaluation in untypical NCC studies.
• These methods enhance the reliability of risk prediction by offering consistent estimation of key performance metrics.
• The approach is flexible, applicable to various statistical models and NCC design variations.