Joint latent class models for longitudinal and time-to-event data: a review

Area of Science:

• Statistical modeling
• Biostatistics
• Longitudinal data analysis

Background:

• Shared random-effect models are common for joint longitudinal and time-to-event data.
• These models incorporate longitudinal marker characteristics into the time-to-event model.
• An alternative, less explored approach is the joint latent class model.

Purpose of the Study:

• To provide an overview of joint latent class modeling, focusing on its application in prediction.
• To introduce the joint latent class model and compare it with shared random-effect models.
• To present dynamic predictive tools and accuracy measures for joint latent class models.

Main Methods:

• Joint latent class modeling framework.
• Estimation and goodness-of-fit procedures.
• Comparison with shared random-effect models.
• Development of dynamic predictive tools and accuracy assessment.

Main Results:

• Joint latent class models offer flexibility in dependency modeling between longitudinal markers and event time.
• These models can effectively incorporate covariates for enhanced prediction.
• The study illustrates methods using prostate cancer recurrence prediction based on Prostate Specific Antigen (PSA) levels.

Conclusions:

• Joint latent class models are well-suited for prediction problems involving longitudinal and time-to-event data.
• The approach provides a robust framework for dynamic prediction and accuracy evaluation.
• Demonstrated utility in predicting prostate cancer recurrence using PSA measurements.