An efficient joint model for high dimensional longitudinal and survival data via generic association features.

Area of Science:

• Biostatistics
• Machine Learning
• Personalized Medicine

Background:

• Joint modeling of longitudinal data and censored durations is crucial for accurate prognostication.
• High-dimensional data presents challenges for standard joint models.
• Existing methods like shared random effect and joint latent class models have limitations.

Purpose of the Study:

• Introduce FLASH, a novel prognostic method for joint modeling.
• Address the challenge of high-dimensional longitudinal and time-independent features.
• Improve prediction accuracy and model interpretability in prognostic settings.

Main Methods:

• Developed a new joint model combining shared random effects and latent class approaches.
• Incorporated regularization techniques for feature selection in high-dimensional contexts.
• Utilized an expectation-maximization algorithm for efficient model estimation.

Main Results:

• FLASH significantly outperforms state-of-the-art joint models in C-index for real-time prediction.
• Demonstrated superior computational speed, orders of magnitude faster than competing methods.
• Successfully identified practically relevant and interpretable prognostic features.

Conclusions:

• FLASH offers a powerful and efficient solution for joint modeling in high-dimensional settings.
• The method enhances prognostic accuracy and interpretability, crucial for healthcare applications.
• FLASH advances personalized medicine and churn prediction through improved feature identification.