To predict the likelihood of bloodstream infection (BSI) in febrile pediatric cancer patients, oncology researchers at Monroe Carell Jr. Children’s Hospital at Vanderbilt developed a clinical risk prediction model for patient evaluation. Refined versions of the model have been validated at six independent centers.
A new study published in the Journal of Clinical Oncology demonstrates the safety of implementing the Esbenshade Vanderbilt (EsVan) model in clinical practice.
“This study confirms that our model accurately predicts BSI risk in febrile pediatric cancer patients,” said Adam Esbenshade, M.D., an associate professor of pediatric hematology and oncology at Vanderbilt and first author of the paper. “Additionally, with real-time clinical utilization, we’ve been able to refine the model’s variables.”
“This study confirms that our model accurately predicts BSI risk.”
Stratifying BSI Risk
Pediatric cancer patients presenting with fever are typically treated with broad-spectrum antibiotics and discharged home. However, the need for empirical antibiotics in patients with non-neutropenic fever (absolute neutrophil count ≥500/µl) has been unclear.
The EsVan model uses 12 clinical variables in these patients to predict BSI risk and inform clinical management. Refined, modified versions – EsVan2 and EsVan2b – were developed and validated in 2017 and have demonstrated even better performance in classifying patients across multiple datasets, Esbenshade said.
The most recent, prospective study evaluated the models using a publicly available web-based capture tool developed by the researchers.
The study included 937 febrile, non-severe neutropenia episodes at Children’s Hospital prospectively occurring between April 2015 and August 2019, stratified into low- (88.9 percent), intermediate- (8.6 percent), and high-risk (2.3 percent) episodes. Overall BSI incidence was 4.2 percent (39 of 937 episodes).
To prospectively implement the models, clinical data were entered into the web module. A minimum of seven days after the initial fever, patients’ medical charts were assessed to record outcome data.
Within the low-risk group, BSI incidence was 1.9 percent (16 of 834 episodes). Empirical antibiotics were administered in 21.1 percent of cases and 72.3 percent required no intravenous antibiotics within seven days of presentation.
“No modeling strategy will predict all BSI with 100 percent accuracy,” wrote the authors. “However, all patients who were low risk and discharged home without antibiotics were still well appearing and, in many cases, afebrile when they returned after infection.”
Episodes with a predicted risk of 10 to 39.9 percent were classified as intermediate risk and antibiotics recommended. Patients with a predicted risk ≥40 percent were considered high risk and were recommended admission with intravenous antibiotics until cultures were negative for 48 hours. Actual BSI incidence was 13.6 percent for intermediate-risk episodes and 54.5 percent for high-risk episodes.
Continuing the Evaluation
The EsVan web research tool is available for use by other pediatric oncology sites and is designed to predict BSI in real time. Deidentified data are recorded directly into an institution-specific REDcap® database.
The researchers will incorporate the new data as they continue to critically reevaluate variables to further simplify and improve the model. The model is now undergoing external prospective validation at several centers.
“We need a systematic approach to the management of fever in pediatric oncology patients.”
Esbenshade says this is just one step toward reducing the burden of pediatric cancer care. “Antibiotics increase health care costs and can be associated with adverse effects,” he said. “We need a systematic approach to the management of fever in pediatric oncology patients to balance patient safety, cost of treatment, and antibiotic stewardship.”