Predicting bacterial infections among pediatric cancer patients with febrile neutropenia: External validation of the PICNICC model.

ABSTRACT

INTRODUCTION: The Predicting Infectious Complications in Neutropenic Children and Young People with Cancer (PICNICC) model was recently developed for antibiotic stewardship among pediatric cancer patients, but limited information is available about its clinical usefulness. We aimed to assess the performance of the PICNICC model for predicting microbiologically documented bacterial infections among pediatric cancer patients with febrile neutropenia. MATERIALS AND METHODS: We used data for febrile neutropenia episodes at a pediatric cancer center in Aarhus, Denmark between 2000 and 2016. We assessed the area under the receiver operating characteristic curve (AUC), calibration, and clinical usefulness (i.e., net benefit). We also recalibrated the model using statistical updating methods. RESULTS: We observed 306 microbiologically documented bacterial infections among 1,892 episodes of febrile neutropenia. The AUC of the model was 0.73 (95% confidence limits [CL] 0.71-0.75). The calibration intercept (calibration-in-the-large) was -0.69 (95% CL -0.86 to -0.51) and the slope was 0.77 (95% CL 0.65-0.89). Modest net benefit was observed at a decision threshold of 5%. Recalibration improved calibration but did not improve net benefit. CONCLUSIONS: The PICNICC model has potential for reducing unnecessary antibiotic exposure for pediatric cancer patients with febrile neutropenia, but continued validation and refinement is necessary to optimize clinical usefulness.