1533. Predictive Model for Development of Adenoviremia among Hematopoietic Pediatric Stem Cell Transplant Recipients Undergoing Surveillance Testing
Session: Poster Abstract Session: Infections and Transplantation
Saturday, October 5, 2013
Room: The Moscone Center: Poster Hall C
  • 2013.9.27_HAdV.PPM.pdf (118.9 kB)
  • Background:

    Adenovirus (HAdV) infections are associated with significant morbidity and mortality in pediatric hematopoietic stem cell transplant (HSCT) recipients. Serial HAdV polymerase chain reaction (PCR) testing is sometimes used to identify adenoviremia with a goal of early intervention. It is laborious and costly to serially test all HSCT recipients. We aimed to derive a prediction rule using data available at the time of HSCT to identify those who did and did not develop adenoviremia. An accurate prediction model could allow for more focused HAdV testing. 


    We retrospectively identified 126 pediatric patients that: (1) received HSCT at The Children’s Hospital of Philadelphia between 2006 and 2012; and, (2) underwent surveillance HAdV PCR testing. Using stochastic gradient boosting (SGB) methodology we derived a prediction model to discriminate between those patients who did and did not develop adenoviremia in a 6-month post transplant period. To establish a balanced dataset, adenoviremic patients were compared to randomly selected controls. The model was repeated 100 times to achieve an average confusion matrix. Predictors included age, race, gender, previous HSCT, indication for HSCT, allogeneic vs. autologous HSCT, match status, stem cell source, T-cell depletion, receipt of conditioning, total body irradiation and presence of chronic comorbid conditions at time of HSCT.


    Twenty-nine (23%) patients developed adenoviremia; median time to adenoviremia was 37 days (IQR: 20 to 69 days). No single baseline characteristic accurately discriminated those with and without adenoviremia. The best-fit SGB predictive model had a sensitivity of 75% (95% CI: 0.55 – 0.88) and specificity of 77% (95% CI: 0.57 – 0.90). The false positive and negative rates were 0.23 (95% CI: 0.10 – 0.43) and 0.25 (95% CI: 0.11 – 0.44), respectively with an average out of bag error of 0.21. 


    PCR surveillance testing in pediatric HSCT recipients commonly identified adenoviremia. The derived model for predicting adenoviremia had reasonable sensitivity, specificity and out of bag error. Refining the model to include additional baseline variables such as specific conditioning medications and more adenoviremia events is necessary to reduce the false positive and false negative rates.

    Sarah B. Klieger, MPH1, Richard Hodinka, PhD2, Yushi Liu, PhD3, Hans Petersen, MS4, Adriana Kajon, PhD5, Zacharoula Oikonomopoulou, MD1, Roshni Patel, BA1, Charalampos Gousis, MD1, Brian Fisher, DO, MPH, MSCE6 and Immunocompromised Pediatric Adenovirus Collaborative supported by HHSN2722011000040C, (1)Division of Infectious Diseases, Center for Pediatric Clinical Effectiveness, The Children's Hospital of Philadelphia, Philadelphia, PA, (2)Children's Hospital of Philadelphia, Philadelphia, PA, (3)Molecular Biology and Lung Cancer Program, Lovelace Respiratory Research Institute, Albuquerque, NM, (4)Infectious Disease Program, Lovelace Respiratory Research Institute, Albuquerque, NM, (5)Lovelace Respiratory Research Institute, Albuquerque, NM, (6)The Children's Hospital of Philadelphia, Philadelphia, PA


    S. B. Klieger, None

    R. Hodinka, None

    Y. Liu, None

    H. Petersen, None

    A. Kajon, None

    Z. Oikonomopoulou, None

    R. Patel, None

    C. Gousis, None

    B. Fisher, None

    Findings in the abstracts are embargoed until 12:01 a.m. PST, Oct. 2nd with the exception of research findings presented at the IDWeek press conferences.