Random Change-Point Non-linear Mixed Effects Model for left-censored longitudinal data: An application to HIV...

Area of Science:

• Biostatistics
• Epidemiology
• Longitudinal Data Analysis

Background:

• Change-point models are crucial for analyzing longitudinal data to detect trend shifts.
• Identifying the exact time of change is challenging, especially with unknown change points in population data.
• Existing models often struggle with non-linear trends and left-censored observations.

Purpose of the Study:

• To develop and validate an unknown change-point model for longitudinal data.
• To accommodate both linear and non-linear mixed effects before and after a change point.
• To handle left-censored data within a random change-point non-linear mixed effects framework.

Main Methods:

• Utilized the stochastic approximation expectation maximization (SAEM) algorithm.
• Incorporated the Metropolis-Hasting sampler for parameter estimation.
• Applied the model to longitudinal viral load (VL) data from the New York City HIV surveillance registry.

Main Results:

• Successfully fitted a random change-point non-linear mixed effects model to the VL data.
• The model effectively estimated individual-specific trends and change points.
• Demonstrated the model's capability in handling left-censored observations in real-world epidemiological data.

Conclusions:

• The proposed unknown change-point model provides a robust framework for analyzing longitudinal data with trend shifts.
• The SAEM algorithm with Metropolis-Hasting sampler is effective for fitting complex mixed-effects models.
• This methodology offers valuable insights for understanding disease progression and intervention effects using HIV viral load data.