Pseudo maximum likelihood approach for the analysis of multivariate left-censored longitudinal data

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Biomarker Research

Background:

• Longitudinal data with left censoring presents analytical challenges for traditional models.
• Full likelihood methods for mixed models become computationally intensive with high censoring and complex random effects.
• Accurate modeling is crucial for understanding biomarker associations with clinical outcomes.

Purpose of the Study:

• To develop a computationally efficient method for analyzing longitudinal data with left censoring.
• To propose a pseudo-likelihood approach for joint modeling of correlated biomarkers and mortality.
• To assess the association between pro-inflammatory (interleukin-6) and anti-inflammatory (interleukin-10) markers and 30-day mortality in sepsis patients.

Main Methods:

• Developed a pseudo-likelihood method to overcome computational complexities of full likelihood in mixed models.
• Applied the method to model longitudinal, left-censored data of interleukin-6 and interleukin-10.
• Jointly analyzed biomarker data with 30-day mortality, accounting for marker correlation and censoring.

Main Results:

• The proposed pseudo-likelihood method simplifies computations for longitudinal data with high levels of left censoring.
• The method accommodates a wide range of multivariate models and data structures.
• Demonstrated feasibility in analyzing correlated, left-censored inflammatory markers (interleukin-6, interleukin-10) in relation to sepsis mortality.

Conclusions:

• Pseudo-likelihood offers a computationally feasible alternative to full likelihood for complex longitudinal data.
• This method enables robust joint analysis of correlated, left-censored biomarkers and clinical outcomes.
• Facilitates a better understanding of inflammatory marker roles in sepsis prognosis.