Reflection on modern methods: Dynamic prediction using joint models of longitudinal and time-to-event data

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Health Informatics

Background:

• Individualized prediction is crucial in clinical medicine.
• Current models often use single-time point data, not reflecting dynamic health progression.
• Physician decision-making involves continuously updating prognosis based on new information.

Purpose of the Study:

• To introduce the development of dynamic risk predictions.
• To provide resources for implementation and assessment of dynamic predictions.
• To explain the theory and methodology behind dynamic prediction models.

Main Methods:

• Utilizing joint models for longitudinal and survival data.
• Developing adaptable R code for practical application.
• Presenting measures for assessing predictive performance.

Main Results:

• Demonstrated dynamic risk predictions using a chronic liver disease dataset.
• Provided a framework for dynamically adjusting individual patient prognosis.
• Illustrated the application of joint models in a real-world clinical scenario.

Conclusions:

• Dynamic risk predictions offer a more realistic approach to patient prognosis than static models.
• Joint models enable continuous updating of predictions as new data becomes available.
• This methodology supports improved clinical decision-making and personalized patient care.