Data-Driven Clustering Distinguishes Infection from Rejection During Hospital Admissions Following Pediatric Lung Transplantation
1Department of Pediatrics, Boston Children's Hospital, Boston, MA, 2Division of Pulmonary Medicine, Nationwide Children's Hospital, Ohio, MA, 3Division of Pulmonary Medicine, Boston Children's Hospital, Boston, MA, 4Department of Anesthesiology, Critical Care and Pain Medicine, Division of Infectious Disease, Boston Children's Hospital, Boston, MA
Meeting: 2022 American Transplant Congress
Abstract number: 1481
Keywords: Infection, Lung, Pediatric, Rejection
Topic: Clinical Science » Lung » 64 - Lung: All Topics
Session Information
Session Time: 7:00pm-8:00pm
Presentation Time: 7:00pm-8:00pm
Location: Hynes Halls C & D
*Purpose: When lung transplantation patients are admitted with respiratory symptoms, distinguishing between episodes of infection and rejection is central to determining immunosuppressive management.
*Methods: Data comprising demographic, clinical, laboratory, imaging, and pulmonary function variables during hospitalizations for features of undifferentiated respiratory illness or infectious symptoms were included. Unsupervised clustering and principal component analysis were performed to delineate variables that could discriminate between infectious events and episodes of rejection (acute cellular rejection, antibody mediated rejection and chronic lung allograft dysfunction). A random forest machine learning (ML) algorithm was used to create decision trees that discriminate between episodes of infection and rejection using a discovery cohort (67%) and a validation cohort (33%). A sensitivity analysis was conducted by attempting to differentiate between cystic fibrosis (CF) and non-cystic fibrosis patients.
*Results: There were 4267 days of hospital admission ranging between 2 and 156 days. Seventy-eight admissions (70 were non-EBV related infection episodes and 8 rejection exacerbations) met clinical criteria as being either an infection or rejection episode among 23 patients (median age – 19.7 years and gender M:F 14:9) and across 18,844 data points. Twenty-eight infections were non-focal and 42 were localized. A combination of an abnormal respiratory system exam and absolute immature granulocyte count (3.91 vs. 32.01 cells/ul; p = 0.017) satisfactorily discriminated between infection episodes and rejection. The sensitivity analysis revealed good discrimination between the 37 CF admissions and the 41 non-CF admissions with primarily hematological parameters.
*Conclusions: Using a data-driven clustering approach, we identified features that cluster patients into a group with a high likelihood of infection vs rejection. These data-driven phenotypes may help refine the positive predictive value of a diagnosis when a lung-transplant patient presents with respiratory symptoms.
To cite this abstract in AMA style:
Britto CD, Boyer D, Visner G, Priebe G, Midyat L. Data-Driven Clustering Distinguishes Infection from Rejection During Hospital Admissions Following Pediatric Lung Transplantation [abstract]. Am J Transplant. 2022; 22 (suppl 3). https://atcmeetingabstracts.com/abstract/data-driven-clustering-distinguishes-infection-from-rejection-during-hospital-admissions-following-pediatric-lung-transplantation/. Accessed May 16, 2025.« Back to 2022 American Transplant Congress