Primary Submission Category: Mediation Analysis, Mechanisms
A Causal Mediation Model for Continuous Time Markov State Processes with application to Microbiome Data
Authors: Debarghya Nandi, Soumya Sahu,
Presenting Author: Debarghya Nandi*
Chronic stress affects millions globally and is linked to elevated cortisol levels, which regulate essential functions like immune response, metabolism, and inflammation. Prolonged cortisol elevation disrupts systems, including the vaginal microbiome, vital for women’s reproductive health. Categorized into community state types (CSTs), this microbiome can shift due to factors like hormonal imbalances or stress. Research indicates stress-induced cortisol destabilizes CSTs, triggering transitions from healthy to dysbiotic states linked to adverse outcomes such as bacterial vaginosis (BV). This study investigates how stress and cortisol influence CST transitions, aiming to understand stress-induced dysbiosis and its implications for reproductive health.
We developed a longitudinal causal mediation model with stress as the exposure, cortisol as the mediator, and CST transitions as the outcome. Our approach models the probability of transitioning between CSTs, estimating shifts from healthy to dysbiotic states based on stress and cortisol over time. Using a joint modeling framework, we capture correlations between cortisol dynamics and CST transitions, defining direct effects (stress impact independent of cortisol) and indirect effects (mediated via cortisol).Extensive simulations demonstrated high performance in terms of bias and coverage, validating the model for real-world application. We plan to apply it to our microbiome dataset to uncover the causal pathways.