688 - Marginal Structure Models to Estimate the Effect of Cytomegalovirus Infection on Hospitalization Among Children Undergoing Allogeneic Hematopoietic Cell Transplantation

Background: Children receiving an allogeneic hematopoietic cell transplant (HCT) are at risk for cytomegalovirus (CMV) infection in the post-transplant period, necessitating routine surveillance for CMV. Some patients will not have CMV detected while others will have intermittent or persistent CMV detection. Prior analyses assessing the association of CMV infection on hospitalization in the post-transplant period have been limited by methods that did not consider the time-varying nature of the exposure (CMV reactivation) and its confounders or aim to obtain causal effect estimates . We aimed to assess the causal effect of CMV reactivation on hospitalization using a causal modeling approach.The Effect of CMV Infection on Hospitalization Using Generalized Estimating Equations and Marginal Structural Models


Methods: A cohort of allogeneic HCT patients transplanted at Children’s Hospital of Philadelphia January 2004–April 2017 was assembled and followed for 100 days after transplant. Eligible patients included those under CMV surveillance, defined as having ≥2 CMV whole blood polymerase chain reaction tests in the first month after HCT. All information was abstracted from medical charts. The association of CMV reactivation on the rate of hospitalization was estimated using traditional generalized estimating equations and repeated using a marginal structural model that accounted for time-varying exposure, confounders and non-random drop-out and obtained effects with causal interpretations.

Results: The study cohort included 340 pediatric allogeneic HCT recipients under CMV surveillance testing. 46.5% were female and the median age was 9 (range: 0 to 26). The CMV infection rate was 33.9%, with a median time to CMV detection of 23.5 days (range: 4-100).  CMV infection was common in Donor+/Recipient+ (58.9%) and Donor-/Recipient+ (34.6%) patients. A traditional model estimates an additional week of CMV infection was associated with a 22% increase in average weekly hospitalization (Incidence rate ratio: 1.22 , 95%: 1.12 -1.34) . A marginal structure model estimates an additional week of CMV infection is associated with 3% increase in average weekly hospitalization incidence (Incidence rate ratio: 1.03, 95%: 0.91-1.16).

Conclusion: Our research showed the effect of CMV on hospitalization diminished after properly considering the time-varying nature of the CMV infection status and its confounders.