Individual prediction regions for multivariate longitudinal data with small samples.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Veterinary Medicine

Background:

• Individualized follow-up in medicine and doping control uses reference intervals for single variables.
• Current methods do not account for correlations between multiple health variables over time.
• Personalized health monitoring requires methods that consider inter-variable relationships.

Purpose of the Study:

• To develop a general method for jointly following up several correlated variables over time.
• To provide a robust methodology for individualized health monitoring using multivariate data.
• To address limitations of variable-by-variable follow-up in longitudinal health assessments.

Main Methods:

• Utilized a multivariate linear mixed-effects model for joint follow-up of correlated variables.
• Developed a method for estimating model parameters.
• Derived asymptotic and small-sample individualized prediction regions for robust monitoring.

Main Results:

• The proposed method allows for the joint follow-up of multiple correlated variables.
• Asymptotic prediction regions are derived for large sample sizes.
• Three alternative prediction regions are proposed and compared for improved performance with small sample sizes.

Conclusions:

• The new methodology offers a significant advancement in personalized health monitoring by analyzing multiple correlated variables simultaneously.
• The approach is applicable to various fields, including medical and doping controls.
• The study demonstrates the method's utility through an illustration of kidney insufficiency follow-up in cats.