Maternal-Neonatal ReportsMachine Learning Used to Compare the Diagnostic Accuracy of Risk Factors, Clinical Signs and Biomarkers and to Develop a New Prediction Model for Neonatal Early-onset SepsisStocker, Martin MD*; Daunhawer, Imant†; van Herk, Wendy MD‡; el Helou, Salhab MD§; Dutta, Sourabh MD§; Schuerman, Frank A. B. A. MD¶; van den Tooren-de Groot, Rita K. MD‖; Wieringa, Jantien W. MD‖; Janota, Jan MD**; van der Meer-Kappelle, Laura H. MD††; Moonen, Rob MD‡‡; Sie, Sintha D. MD§§; de Vries, Esther MD¶¶,‖‖; Donker, Albertine E. MD***; Zimmerman, Urs MD†††; Schlapbach, Luregn J. MD‡‡‡; de Mol, Amerik C. MD§§§; Hoffmann-Haringsma, Angelique MD¶¶¶; Roy, Madan MD‖‖‖; Tomaske, Maren MD****; Kornelisse, René F. PhD††††; van Gijsel, Juliette MD‡‡‡‡; Plötz, Frans B. PhD§§§§; Wellmann, Sven MD¶¶¶¶; Achten, Niek B. MD§§§§; Lehnick, Dirk PhD‖‖‖‖; van Rossum, Annemarie M. C. MD‡; Vogt, Julia E. PhD* Author Information From the *Department of Paediatrics, Neonatal and Paediatric Intensive Care Unit, Children’s Hospital Lucerne, Lucerne †Department of Computer Science, ETH Zurich, Switzerland ‡Department of Paediatrics, Division of Paediatric Infectious Diseases and Immunology, Erasmus MC University Medical Centre-Sophia Children’s Hospital, Rotterdam, The Netherlands §Division of Neonatology, McMaster University Children’s Hospital, Hamilton Health Sciences, Hamilton, ON, Canada ¶Department of Neonatal Intensive Care Unit, Isala Women and Children’s Hospital, Zwolle ‖Department of Paediatrics, Haaglanden Medical Centre, ‘s Gravenhage, The Netherlands **Department of Obstetrics and Gynecology, Motol University Hospital, Second Medical Faculty, Prague, Czech Republic ††Department of Neonatology, Reinier de Graaf Gasthuis, Delft ‡‡Department of Neonatology, Zuyderland Medical Centre, Heerlen §§Department of Neonatology, Amsterdam UMC, Vrije Universiteit Amsterdam, Amsterdam ¶¶Department of Jeroen Bosch Academy Research, Jeroen Bosch Hospital, ‘s-Hertogenbosch ‖‖Department of Tranzo, Tilburg University, Tilburg ***Department of Paediatrics, Maxima Medical Centre, Veldhoven, The Netherlands †††Department of Paediatrics, Kantonsspital Winterthur, Winterthur ‡‡‡Neonatal and Pediatric Intensive Care Unit, Children`s Research Center, University Children’s Hospital Zurich, Zurich, Switzerland §§§Department of Neonatology, Albert Schweitzer Hospital, Dordrecht ¶¶¶Department of Neonatology, Sint Franciscus Gasthuis, Rotterdam, The Netherlands ‖‖‖Department of Neonatology, St. Josephs Healthcare, Hamilton Health Sciences, Hamilton, ON, Canada ****Department of Paediatrics, Stadtspital Triemli, Zürich, Switzerland ††††Department of Paediatrics, Division of Neonatology, Erasmus MC University Medical Centre-Sophia Children’s Hospital, Rotterdam ‡‡‡‡Therapeuticum Utrecht, Utrecht §§§§Department of Pediatrics, Tergooi Hospital, Blaricum, the Netherlands and Amsterdam University Medical Center, Department of Pediatrics, Amsterdam, The Netherlands ¶¶¶¶Department of Neonatology, University Children’s Hospital Regensburg (KUNO), University of Regensburg, Regensburg, Germany ‖‖‖‖Department of Health Sciences and Medicine, Head Biostatistics and Methodology, University of Lucerne, Lucerne, Switzerland. Accepted for publication August 22, 2021 This study was supported by The Thrasher Foundation (9143) to [M.S.]; The NutsOhra Foundation (1101-059) to [A.M.C.v.R.]; The Sophia Foundation for Scientific research (681) to [W.v.H.]; and the Swiss National Science Foundation (200021_188466) to [I.D.]. In addition, Thermofisher provided procalcitonin kits and provided an unrestricted grant for the organization of 4 investigator meetings (2008, 2009, 2013 and 2015). The authors have no conflicts of interest to disclose. M.S. and I.D. contributed equally as co-first authors. A.M.C.v.R. and J.E.V. contributed equally as co-senior authors. Address for correspondence: Martin Stocker, MD, Department of Paediatrics, Division of Neonatal and Paediatric Intensive Care, Children’s Hospital, Lucerne, Switzerland. E-mail: [email protected]. The Pediatric Infectious Disease Journal: March 2022 - Volume 41 - Issue 3 - p 248-254 doi: 10.1097/INF.0000000000003344 Buy Metrics Abstract Background: Current strategies for risk stratification and prediction of neonatal early-onset sepsis (EOS) are inefficient and lack diagnostic performance. The aim of this study was to use machine learning to analyze the diagnostic accuracy of risk factors (RFs), clinical signs and biomarkers and to develop a prediction model for culture-proven EOS. We hypothesized that the contribution to diagnostic accuracy of biomarkers is higher than of RFs or clinical signs. Study Design: Secondary analysis of the prospective international multicenter NeoPInS study. Neonates born after completed 34 weeks of gestation with antibiotic therapy due to suspected EOS within the first 72 hours of life participated. Primary outcome was defined as predictive performance for culture-proven EOS with variables known at the start of antibiotic therapy. Machine learning was used in form of a random forest classifier. Results: One thousand six hundred eighty-five neonates treated for suspected infection were analyzed. Biomarkers were superior to clinical signs and RFs for prediction of culture-proven EOS. C-reactive protein and white blood cells were most important for the prediction of the culture result. Our full model achieved an area-under-the-receiver-operating-characteristic-curve of 83.41% (±8.8%) and an area-under-the-precision-recall-curve of 28.42% (±11.5%). The predictive performance of the model with RFs alone was comparable with random. Conclusions: Biomarkers have to be considered in algorithms for the management of neonates suspected of EOS. A 2-step approach with a screening tool for all neonates in combination with our model in the preselected population with an increased risk for EOS may have the potential to reduce the start of unnecessary antibiotics. Copyright © 2021 Wolters Kluwer Health, Inc. All rights reserved.