Machine 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 Sepsis

Abstract

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.

The past, current, and future of neonatal intensive care units with artificial intelligence: a systematic review

“…Only Apgar scores could be changed. Hsu et al, 2020 71 RF KNN ANN XGBoost Elastic-net To predict mortality of neonates when they were on mechanical intubation 1734 neonates 70% training 30% test Mortality scores Patient demographics Lab results Blood gas analysis Respirator parameters Cardiac inotrop agents from onset of respiratory failure to 48 hours 93.9% (AUC) RF has achieved the highest prediction of mortality + Employed several ML and statistics + Explained the feature analysis and importance into analysis - Two center study - Algorithmic bias - Inability to real time prediction Stocker et al, 2022 75 RF To predict blood culture test positivity according to the all variables, all variables without biomarkers, only biomarkers, only risk factors, and only clinical signs 1710 neonates from 17 centers Secondary analysis of NeoPInS data Biomarkers(4 variables) Risk factors (4 variables) Clinical signs(6 variables) Other variables(14) All variables (28) They included to RF analysis to predict culture positive early onset sepsis Only biomarkers 73.3% (AUC) All variables 83.4% (AUC) Biomarkers are the most important contributor + CRP and WBC are the most important variables in the model + Decrease the overtreatment + Multi-center data - Overfitting of the model due to the discrepancy with currently known clinical practice - Seemed not evaluated the clinical signs and risk factors which are really important in daily practice Temple et al, 2016 229 supervised ML and NLP To identify patients that will be medically ready for discharge in the subsequent 2–10 days. 4693 patients (103,206 patient-days 178 NLP using a bag of words (BOW) surgical diagnoses, pulmonary hypertension, retinopathy of prematurity, and psychosocial issues 63.3% (AUC) 67.7% (AUC) 75.2% (AUC) 83.7% (AUC) + Could potentially avoid over 900 (0.9%) hospital days …”

Early initiation of antibiotic therapy and short-term outcomes in preterm infants: a single-centre retrospective cohort analysis

“…If it is indeed assumed that disturbances of the microbiome at a very early stage are potentially harmful, then everyday antibiotics started later may be beneficial. The big problem is that despite continuous improvement in clinical care and intensive research, the diagnosis of EOS is still uncertain, and currently, it seems that about 100 newborns are treated with antibiotics to catch one with culture-proven EOS 34. New approaches to develop better tools for diagnosing/predicting EOS include gene profiling, transcriptome analysis and machine learning approaches and their combination with clinical and laboratory chemistry parameters 34–36.…”

“…The big problem is that despite continuous improvement in clinical care and intensive research, the diagnosis of EOS is still uncertain, and currently, it seems that about 100 newborns are treated with antibiotics to catch one with culture-proven EOS. 34 New approaches to develop better tools for diagnosing/predicting EOS include gene profiling, transcriptome analysis and machine learning approaches and their combination with clinical and laboratory chemistry parameters. [34][35][36] Further development of such approaches could help identify premature infants at high risk of EOS and enable more targeted use of antibiotics immediately after birth.…”

“…34 New approaches to develop better tools for diagnosing/predicting EOS include gene profiling, transcriptome analysis and machine learning approaches and their combination with clinical and laboratory chemistry parameters. [34][35][36] Further development of such approaches could help identify premature infants at high risk of EOS and enable more targeted use of antibiotics immediately after birth. Factors associated with very early initiation of antibiotic treatment were GA and low birth weight, low Apgar scores at 5 and 10 min, IAI, pPROM and spontaneous birth, as well as invasive procedures.…”

Prediction modelling in the early detection of neonatal sepsis

“…Similarly, clinical and laboratory biomarkers such as maternal white blood cells, absolute neutrophil count was also used for the diagnosis of the disease [ 13 ]. Incorporation of laboratory diagnostic markers such as WBC along with standard biomarker such as CRP in the prediction model has considerably reduced the applicability of antibiotics in early-onset neonatal sepsis [ 14 ]. Sepsis risk calculator, scoring system generation were some of the desired results of prediction modelling implementation [ 15 , 16 ].…”