Using Trajectory Based Phenotypes to Target Treatment in Pediatric Septic Shock

/Abstract An estimated 3 million children worldwide die from sepsis each year. Of those who survive, nearly 1 in 3 are discharged with disability. Recent work has found that different subgroups or trajectory-based phenotypes (TBPs) of septic shock in adults respond differently to some treatments, suggesting that septic shock requires treatment specific to TBP rather than a single treatment approach for all patients. Although some TBPs of septic shock have been identified in children, we cannot currently identify these TBPs in real- time, and we have not verified that treatments targeted to TBPs improve clinical outcomes. In Aim 1a, I will use trajectory modeling approaches to derive and validate potential TBPs of pediatric septic shock using both static patient factors (e.g. chronic diseases) and dynamic changes in patient state (e.g. vital signs). In Aim 1b, I will test the performance of a model that uses data from only one clinically meaningful time point (i.e. time of shock diagnosis), rather than data from the whole trajectory, to assign a patient to the predicted TBP(s) found in 1a. In Aim 2, I will emulate potential interventions in two ways. First, I will conduct multiple target trial emulations to test the effect of various resuscitation strategies (e.g. steroids) on clinical outcomes (e.g. shock-free hours) in children with identified TBPs. Second, I will construct a queryable model to estimate the probability of outcomes (e.g. change blood pressure) based on the interaction between patient’s physiology (e.g. lactate), resuscitation treatments (e.g. fluid administration), and prior outcomes, in order to determine the optimal resuscitation treatment sequence to improve outcomes (e.g. volume of fluid prior to starting vasoactive medication). I will then divide this population into the TBPs identified in Aim 1b, and repeat this process to assess for differences in optimal resuscitation treatment sequence across TBPs. I hypothesize that at least one TBP will respond differently to at least one common treatment or treatment sequence. In Aim 3, I will use the algorithm found in Aim 1b to implement and assess the performance of a silent screening tool in a single center’s electronic health record (EHR) to prospectively identify patients belonging to TBPs who may benefit from an intervention modeled after an emulated intervention from Aim 2. This aim will provide me with both a real-time screening tool and critical preliminary data for future clinical trials. The training aims of this award will solidify my content expertise in the application of data science in pediatric septic shock, and add training in causal inference methods, clinical trial design, and human-centered clinical decision support. Together, these will prepare me to be an independent investigator leading a team of statisticians, informaticians, clinical trialists, and clinicians to develop and validate simple, clear, interpretable, and evidence-based tools to bring machine learning-based knowledge to the bedside to improve outcomes in pediatric septic shock Project Number: 1K23HD121858-01 | Fiscal Year: 2026 | NIH Institute/Center: Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) | Principal Investigator: Sarah Walker | Institution: UNIVERSITY OF COLORADO DENVER, Aurora, CO | Award Amount: $163,825 | Activity Code: K23 | Study Section: Special Emphasis Panel[ZRG1 HSS-X (90)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11349602