Project Summary/Abstract Patients with hospital-onset sepsis (HOS) frequently receive delayed treatment and have nearly 3-fold greater risk of acute respiratory failure (ARF) and death than patients with community-onset sepsis, yet remain an understudied population. The current absence of data characterizing heterogeneity in clinical presentation, ARF risk, and treatment benefit, presents a critical barrier to designing interventions that improve treatment timing and prevent ARF for the most vulnerable patients. The candidate’s long-term career goal is to build an independently funded research program advancing precision approaches to HOS and ARF treatment. As an essential next step, the overall objective of this project is to create new knowledge about HOS heterogeneity and progression to ARF to inform the future development of treatment strategies tailored to patients’ individual organ failure risks and anticipated benefits from specific interventions. In preliminary studies, the candidate has built a retrospective cohort of HOS patients admitted to five organizationally diverse hospitals of an academic health system, and preliminary analyses have demonstrated an association between timing of antimicrobial administration for HOS and progression to ARF. This project will expand this cohort in order to complete the following aims: 1) identify latent phenotypes of HOS illness presentation using consensus k-means clustering accounting for informative missingness and including baseline patient factors, measures of organ dysfunction and severity of illness, and measures of degree and rapidity of clinical deterioration, 2) develop a novel model to predict ARF among patients with HOS using super learner machine learning predictive modeling including patient factors and physiologic data at sepsis onset; and 3) assess the association of antimicrobial treatment delays with progression to ARF among patients with HOS using inverse probability of treatment weighting to account for confounding by indication, and estimate heterogeneity of this effect by underlying ARF risk. To successfully complete these aims and acquire the necessary skills to transition to research independence, the candidate has developed a training plan that includes formal didactic training, research seminars, works in progress, and scientific meeting presentations focused on methods in machine learning, predictive modeling, and causal inference. Her career development will be supported by close guidance from her mentorship team who have deep content and methodologic expertise in relevant fields for this work.