Theses Doctoral

Antimicrobial Stewardship in the Neonatal Population

Duchon, Jennifer

Antimicrobials are the most frequently used medications in the Neonatal Intensive Care Unit (NICU). Antimicrobial Stewardship (AMS) efforts may be used to mitigate the consequences of antimicrobial overuse while optimizing clinical outcomes through the safe, judicious use of antimicrobials. One target of AMS efforts is to reduce the incidence of necrotizing enterocolitis (NEC), a serious intestinal infection in neonates of which a necessary component is dysbiosis, the development of aberrant intestinal microbiota typically associated with prior antibiotic use. The goal of this ILE is to implement and enhance AMS efforts in the neonatal population with a focus on preventing NEC. The specific aims progress through three relevant, practical examples of AMS in a stepwise manner.

Methods: In Aim 1, a systematic review of the literature evaluating the relationship between antimicrobial therapy and subsequent development of NEC and a meta-analysis including non-interventional studies was performed. Data were pooled on adjusted odds ratios (OR) and analyzed using the generic inverse variance method. All analyses were random effects models. A sensitivity analysis was performed based on a range (0-40%) of credibility ceilings. In Aim 2, institutional guidelines for early and late onset neonatal sepsis using the principles of AMS and the evidence for safe restriction of antimicrobials targeted for reduction in use in neonates by the National Healthcare Safety Network (NHSN) were created and implemented .

In Aim 3, a reproducible 2-class latent variable model to extract a date-stamped diagnosis of NEC from the Pediatric Health Information System (PHIS) database was created as a tool to enhance research evaluating antibiotic use and NEC from large databases. This model was created using a subset of infants at two PHIS sites that were able to be validated. M plus software was used.

Conclusions: For Aim 1, 36 studies met inclusion criteria for the systematic review, with 33 proceeding to quantitative analysis. There were 10 RCTs, the remaining being observational studies. Using the ROBINS-I or RoB 2.0 tools as appropriate, all studies including the RCTs had a least a moderate or high risk of bias respectively. The overall analysis failed to provide evidence of an association between prior antimicrobial use and NEC when all 33 studies were included, with a summary OR of 1.13, CI95 (0.88, 1.45) and significant heterogeneity, I2 = 77%. Multiple subgroup analyses were performed: “intent” of antibiotic use (prophylaxis versus not) drug delivery method (oral versus parenteral) and study type.
Subgroup analysis of prophylactic enteral antibiotics showed a reduction in NEC: OR 0.2 CI95 (0.08, 0.54), I2 = 35% while prior use of parental antibiotics showed a positive association with NEC OR 1.48, CI95 (1.18, 1.86), I2 72%; for this subgroup, using a c% shows heterogeneity first reaching an estimate of 0% at a ceiling of 10% with nominal statistical significance is maintained starting at a ceiling of 10%. This shows that consideration of the biologic mechanism of the exposure-disease association, as indicated by the subgroup analyses in this study, must be considered when performing further dataset evaluations lest biased conclusions will be reached.

For Aim 2, Four guidelines were created and implemented and are being validated:
• The evaluation and management of infants ≥ 35 weeks gestational age at risk for early onset sepsis at Tufts Medical Center
• The evaluation and management of infants ≥ 36 weeks gestational age at risk for early onset sepsis at BronxCare Hospital Center
• The evaluation and management of infants < 36 weeks gestational age at risk for early onset sepsis at BronxCare Hospital Center
• The evaluation and management of infants at risk for late onset sepsis at BronxCare Hospital Center

For Aim 3 a model was successfully created that can be used to add an important layer of detail, time-of-event, to patient level variables in a large data set. This model can also be used to tabulate the sensitivity of a disease in the absence of a gold standard. The model is portable and could serve as a template for the PHIS or other large databases where certain important exposures may not be date stamped. The model may be adapted to not only allow for appropriate extraction of variables, but also allow the correct modelling of time-dependent co-variables.


  • thumnail for Duchon_cumc.columbia_0054E_10079.pdf Duchon_cumc.columbia_0054E_10079.pdf application/pdf 4.42 MB Download File

More About This Work

Academic Units
Thesis Advisors
Larson, Elaine L.
Davidson, Leslie L.
Dr.P.H., Mailman School of Public Health, Columbia University
Published Here
October 27, 2021