Dynamic predictions using flexible joint models of longitudinal and time-to-event data

Area of Science:

• Biostatistics
• Clinical Research Methodology
• Longitudinal Data Analysis

Background:

• Joint models for longitudinal and time-to-event data are vital in clinical studies, linking biomarker trajectories to patient outcomes.
• Dynamic prediction of prognosis, especially survival probabilities, is a key area in personalized medicine.
• Existing models may lack flexibility in characterizing individual patient data for accurate dynamic predictions.

Purpose of the Study:

• To propose a novel joint model utilizing individual-level penalized splines (P-splines).
• To flexibly model the coevolution of longitudinal biomarkers and time-to-event outcomes.
• To facilitate straightforward dynamic predictions of survival probabilities.

Main Methods:

• Development of a joint model incorporating individual-level penalized splines (P-splines).
• Characterization of the longitudinal and time-to-event processes.
• Utilizing the multivariate skew-normal distribution of random P-spline coefficients for prediction.

Main Results:

• The proposed P-spline joint model demonstrated superior dynamic prediction performance in simulations.
• The model effectively characterized individual longitudinal trajectories and their association with event times.
• Dynamic survival probability predictions were found to be straightforward with the proposed method.

Conclusions:

• The novel P-spline joint model offers enhanced dynamic prediction capabilities for clinical research.
• This approach provides a flexible and effective tool for personalized medicine and prognosis.
• The model's performance surpasses existing methods in dynamic prediction accuracy.