Dynamic prediction of interval-censored failure time data with longitudinal marker

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Longitudinal Data Analysis

Background:

• Accurate patient prognosis prediction is crucial for clinical decision-making.
• Existing prediction models need to dynamically update with evolving patient conditions.
• Joint models for longitudinal markers and time-to-event data are common but require methods for dynamic evaluation.

Purpose of the Study:

• To develop dynamic measures for assessing the predictive accuracy of longitudinal markers.
• To address the challenge of interval-censored failure time data in prognostic modeling.
• To evaluate prediction performance in a real-world dementia prediction scenario.

Main Methods:

• Proposed novel dynamic area under the curve and Brier score metrics.
• Incorporated the specific data structure of interval-censored failure time data.
• Conducted simulation studies comparing joint modeling and landmarking approaches.

Main Results:

• The proposed dynamic measures effectively reflect the incomplete data structure.
• Simulation results demonstrated the performance of the joint model compared to landmarking.
• The methods were successfully applied to predict dementia using cognitive scores.

Conclusions:

• Dynamic evaluation metrics are essential for updating prognostic models.
• The proposed methods offer a robust way to assess predictive accuracy with interval-censored data.
• This approach enhances the prediction of clinical events, such as dementia onset.