Victor Fossaluza (IME-USP)

Title: Bayesian Inference in Stochastic Processes to Identify Mortality Attributed to Sepsis

We introduce a novel method for calculating attributable population fractions (PAF) and attributable hazard functions (AHF) within the framework of stochastic processes and non-homogeneous Markov chains. This approach is designed to align with existing literature while offering enhanced flexibility for diverse study designs. Motivated by a Brazilian study of over 3800 hospitalized patients across 38 medical centers, which explored the relationship between sepsis exposure and patient outcomes (death and discharge), our method provides a dynamic measure that accounts for time-dependent variations and the impact of covariates. Our Adapted Attributable Hazard Fraction (AAHF) incorporates elements from competing risks analysis and survival analysis, allowing for the calculation of metrics both for specific subpopulations and general measures. 

We apply this new approach to filtered data from the motivating study, analyzing transitions in patient outcomes and risk factors over time. Our findings highlight the delayed impact of sepsis on mortality. Early in hospitalization (days 1-13), no significant difference in mortality due to sepsis is observed, possibly due to effective interventions or undetected high-risk patients. However, from day 14 onwards, sepsis-related mortality begins to increase, peaking at around 2% by day 18, underscoring the importance of continuous monitoring and aggressive sepsis management in long-term hospitalized patients (joint work with E. Wagner).