Key points are not available for this paper at this time.
To the Editor: We read with interest the report by Assies et al (1) on the feasibility and reliability of pediatric Rapid Ultrasound for Shock and Hypotension (p-RUSH) in Malawian children with undifferentiated shock. Of note, the rapid median time to p-RUSH with mainly interpretable video clips and reliable cardiac assessments addresses an important clinical gap in phenotyping shock states. However, the work raises three questions that we would welcome the authors to consider. First, the study highlighted challenges in interpreting inferior vena cava (IVC) collapsibility (fair interobserver agreement, kappa = 0.275), and we wondered whether the authors had experience using artificial intelligence (AI) enhancement to help? For example, Shokoohi et al (2) described using deep learning algorithms to enhance point-of-care ultrasound (POCUS) image analysis by automating feature detection and reducing observer variability. Applying such tools to IVC assessments in p-RUSH could help quantify collapsibility with greater precision, particularly in settings where expert sonographers are scarce. For instance, AI models could be trained on pediatric IVC datasets like yours (1) (and others from low- and middle-income LMIC settings as recently reviewed 3), with the aim of refining shock phenotyping. This integration would not only improve reliability but also facilitate task-shifting to non-specialist clinicians, a key consideration for scalability. Second, have the authors considered expanding the p-RUSH framework to include dynamic treatment monitoring, which may amplify its clinical impact. The current study focused on initial shock phenotyping, but serial p-RUSH assessments may better guide fluid resuscitation. For example, tracking IVC diameter changes post-fluid bolus or monitoring cardiac contractility in response to inotropes could help to better titrate therapy. A systematic review of literature up to mid-2022, of tools and measures to predict fluid responsiveness in pediatric shock and critical illness, noted that dynamic ultrasound markers outperform static measures (4). Hence, suggesting that p-RUSH’s utility extends beyond diagnosis to therapeutic adjustment. Third, have the authors considered widening the validation of the p-RUSH algorithm across more diverse LMIC settings? The p-RUSH study was a 2019 single-center design in Malawi, and the conclusions about contemporary applicability and generalizability need to be considered. A multicenter validation spanning rural and urban LMIC settings is now needed. Such an approach would align with the ideas presented in a 2024 editorial in the Journal (5). Here, the emphasis was on contextualizing POCUS protocols to local needs and, like the recent p-RUSH report (1), ensures a practical patient-centered approach to early resuscitation. In summary, the recent report by Assies et al (1) provides valuable insight into the use of p-RUSH assessment in patients managed in LMIC setting in Malawi. Our questions are about enhancing IVC assessment with AI, integrating dynamic monitoring, and validating across diverse, international settings. We believe that these developments would strengthen the role of p-RUCH in personalized shock care. Such advances would be timely and could ensure that practitioners are guided to provide the right treatment—whether fluid, transfusion, or inotropes—to the right patient, at the right time (6).
Chen et al. (Fri,) studied this question.