The CCOS Communications Team interviewed researchers at the 2023 CTSA Fall Program Annual Meeting poster session in November as part of a series to feature ongoing projects across the CTSA hubs. In this article, we’re featuring Dr. Sarah Walker, TL1/32, Attending, and Advanced Research Fellow at Northwestern University who presented her work on a supervised machine learning model to predict fluid response in hypotensive children.
Research Question
More than 600,000 children worldwide are diagnosed with shock, which has a mortality rate as high as 50% and causes one in three to experience long-term morbidities. The first step in shock resolution is the restoration of normal blood pressure through early and rapid fluid administration. However, fluid administration does not universally reverse low blood pressure in children, and both prolonged shock and excessive fluid administration contribute to morbidity and mortality. Hypotension in children is clinically difficult to manage, which inspired Dr. Walker and her team to use a machine learning-based model to predict which children would have a sustained positive clinical response to a fluid bolus to restore a normal blood pressure.
Research Plan
Dr. Walker plans to conduct a single-center retrospective study of critically ill children with low blood pressure who received at least 10 ml/kg of intravenous (IV) normal saline within 72 hours of pediatric intensive care unit admission between 2013 and 2023. Physiological variables will be extracted from data stored in bedside monitors and clinical variables will be extracted from electronic health records. Elastic net, random forest, and a long short-term memory model will be used to predict fluid responsiveness, which will be defined as an increase of at least 10% in the patient’s mean arterial pressure (MAP). Multiple organ dysfunction on day seven or death by day 28 will be compared between four groups: 1) Fluid Responsive (FR) and non-FRs, 2) predicted FRs and non-FRs, 3), FRs and non-FRs stratified by race/ethnicity, and 4) FRs and non-FRs stratified by sex as a biologic variable.
Next Steps
Dr. Walker and her team anticipate that approximately 800 critically ill children will receive 2,000 IV saline fluid boluses with a 60% rate of FR. The researchers hypothesize that non-FRs will have a higher complicated course than FRs. They also predict, based on previous adult studies, that patients of Black race and/or Hispanic ethnicity will have a higher rate of complicated course than those who are non-Hispanic and white, with no difference in complicated course when comparing biological sex.
Reflection with the Researcher: What real-world benefits do you envision might come from work like this, and how would patients or community members benefit?
Dr. Walker stated that “...in current practice, it is hard to determine which hypotensive children will improve after receiving IV fluids.” As such, Dr. Walker and her research team are also using machine learning to help predict which children might benefit from additional therapies, like targeted vasoactive medicines, in order to decrease overall hypotension-related morbidity and mortality in children.
