Estimating Mean Viral Load Trajectory From Intermittent Longitudinal Data and Unknown Time Origins

Area of Science:

• Epidemiology and Biostatistics
• Infectious Disease Modeling

Background:

• Viral load (VL) in respiratory infections is a key indicator of infectiousness.
• Current methods often lack longitudinal data, with VL measured only once per individual.
• Estimating the typical VL trajectory is vital for public health policy and recommendations.

Purpose of the Study:

• To develop statistical approaches for estimating the mean viral load (VL) trajectory over time.
• To accurately reconstruct daily mean VL curves using limited, partially observed longitudinal data.
• To address challenges posed by unknown infection dates and missing VL measurements.

Main Methods:

• A discrete-time, likelihood-based statistical model for partially observed longitudinal data.
• Utilized a multivariate normal model to account for within-individual measurement correlations.
• Developed an expectation-maximization (EM) algorithm to handle latent variables (unknown time origins and missing data).

Main Results:

• Demonstrated that two VL measurements per individual can accurately estimate the mean VL function.
• Successfully reconstructed daily mean VL dynamics using the proposed statistical approach.
• Applied the method to SARS-CoV-2 cycle-threshold data, validating its practical utility.

Conclusions:

• The developed statistical method effectively estimates viral load trajectories from limited data.
• This approach is valuable for understanding disease dynamics, especially at the onset of a pandemic.
• Accurate VL reconstruction aids in informing public health strategies and interventions.