Developing and evaluating risk prediction models with panel current status data.

Area of Science:

• Biostatistics
• Epidemiology
• Biomedical Data Analysis

Background:

• Panel current status data are common in biomedical studies, collected at scheduled visits.
• Existing survival analysis methods (proportional hazards, accelerated failure time) have implementation and small-sample limitations.
• Current methods lack ways to assess prediction performance for risk models with this data type.

Purpose of the Study:

• To propose a simple, consistent estimator for nonparametric transformation (NPT) models using panel current status data.
• To develop nonparametric estimators for evaluating risk score prediction performance, independent of model fit adequacy.
• To address limitations of existing methods in terms of implementation, sample size, and model misspecification.

Main Methods:

• Fitting a logistic regression working model within a general class of NPT models.
• Developing nonparametric estimators for prediction performance evaluation.
• Utilizing simulation studies to assess estimator performance.
• Applying the methods to data from the Framingham Offspring Study.

Main Results:

• The proposed estimator is consistent for the NPT model parameter (up to a scale multiplier).
• Nonparametric prediction performance estimators are valid regardless of model adequacy.
• Simulation results show good finite sample performance and superiority over existing estimators.
• The methods are demonstrated effectively on real-world study data.

Conclusions:

• The proposed methods offer a robust and practical approach for analyzing panel current status data.
• The new estimators enhance the evaluation of risk models and prediction performance in biomedical research.
• These advancements are particularly beneficial for small sample sizes and potential model misspecification scenarios.