Clinical Decision Support for Improved Neonatal Care: The Development of a Machine Learning Model for the Prediction of Late-onset Sepsis and Necrotizing Enterocolitis

OBJECTIVE

To develop an artificial intelligence-based software system for predicting late-onset sepsis (LOS) and necrotizing enterocolitis (NEC) in infants admitted to the neonatal intensive care unit (NICU).

STUDY DESIGN

Single-center, retrospective cohort study, conducted in the NICU of the Antwerp University Hospital. Continuous monitoring data of 865 preterm infants born at <32 weeks gestational age, admitted to the NICU in the first week of life, were used to train an XGBoost machine learning (ML) algorithm for LOS and NEC prediction in a cross-validated setup. Afterward, the model's performance was assessed on an independent test set of 148 patients (internal validation).

RESULTS

The ML model delivered hourly risk predictions with an overall sensitivity of 69% (142/206) for all LOS/NEC episodes and 81% (67/83) for severe LOS/NEC episodes. The model showed a median time gain of ≤10 hours (IQR, 3.1-21.0 hours), compared with historical clinical diagnosis. On the complete retrospective dataset, the ML model made 721 069 predictions, of which 9805 (1.3%) depicted a LOS/NEC probability of ≥0.15, resulting in a total alarm rate of <1 patient alarm-day per week. The model reached a similar performance on the internal validation set.

CONCLUSIONS

Artificial intelligence technology can assist clinicians in the early detection of LOS and NEC in the NICU, which potentially can result in clinical and socioeconomic benefits. Additional studies are required to quantify further the effect of combining artificial and human intelligence on patient outcomes in the NICU.

Machine Learning Detects Intraventricular Haemorrhage in Extremely Preterm Infants

OBJECTIVE

To test the potential utility of applying machine learning methods to regional cerebral (rcSO₂) and peripheral oxygen saturation (SpO₂) signals to detect brain injury in extremely preterm infants.

STUDY DESIGN

A subset of infants enrolled in the Management of Hypotension in Preterm infants (HIP) trial were analysed (n = 46). All eligible infants were <28 weeks' gestational age and had continuous rcSO₂ measurements performed over the first 72 h and cranial ultrasounds performed during the first week after birth. SpO₂ data were available for 32 infants. The rcSO₂ and SpO₂ signals were preprocessed, and prolonged relative desaturations (PRDs; data-driven desaturation in the 2-to-15-min range) were extracted. Numerous quantitative features were extracted from the biosignals before and after the exclusion of the PRDs within the signals. PRDs were also evaluated as a stand-alone feature. A machine learning model was used to detect brain injury (intraventricular haemorrhage-IVH grade II-IV) using a leave-one-out cross-validation approach.

RESULTS

The area under the receiver operating characteristic curve (AUC) for the PRD rcSO₂ was 0.846 (95% CI: 0.720-0.948), outperforming the rcSO₂ threshold approach (AUC 0.593 95% CI 0.399-0.775). Neither the clinical model nor any of the SpO₂ models were significantly associated with brain injury.

CONCLUSION

There was a significant association between the data-driven definition of PRDs in rcSO₂ and brain injury. Automated analysis of PRDs of the cerebral NIRS signal in extremely preterm infants may aid in better prediction of IVH compared with a threshold-based approach. Further investigation of the definition of the extracted PRDs and an understanding of the physiology underlying these events are required.

Hypotension in Preterm Infants (HIP) randomised trial

OBJECTIVE

To determine whether restricting the use of inotrope after diagnosis of low blood pressure (BP) in the first 72 hours of life affects survival without significant brain injury at 36 weeks of postmenstrual age (PMA) in infants born before 28 weeks of gestation.

DESIGN

Double-blind, placebo-controlled randomised trial. Caregivers were masked to group assignment.

SETTING

10 sites across Europe and Canada.

PARTICIPANTS

Infants born before 28 weeks of gestation were eligible if they had an invasive mean BP less than their gestational age that persisted for ≥15 min in the first 72 hours of life and a cerebral ultrasound free of significant (≥ grade 3) intraventricular haemorrhage.

INTERVENTION

Participants were randomly assigned to saline bolus followed by either a dopamine infusion (standard management) or placebo (5% dextrose) infusion (restrictive management).

PRIMARY OUTCOME

Survival to 36 weeks of PMA without severe brain injury.

RESULTS

The trial terminated early due to significant enrolment issues (7.7% of planned recruitment). 58 infants were enrolled between February 2015 and September 2017. The two groups were well matched for baseline variables. In the standard group, 18/29 (62%) achieved the primary outcome compared with 20/29 (69%) in the restrictive group (p=0.58). Additional treatments for low BP were used less frequently in the standard arm (11/29 (38%) vs 19/29 (66%), p=0.038).

CONCLUSION

Though this study lacked power, we did not detect major differences in clinical outcomes between standard or restrictive approach to treatment. These results will inform future studies in this area.

TRIAL REGISTRATION NUMBER

NCT01482559, EudraCT 2010-023988-17.

Nonstationary coupling between heart rate and perfusion index in extremely preterm infants in the first day of life

OBJECTIVE

Adaptation to the extra-uterine environment presents many challenges for infants born less than 28 weeks of gestation. Quantitative analysis of readily-available physiological signals at the cotside could provide valuable information during this critical time. We aim to assess the time-varying coupling between heart rate (HR) and perfusion index (PI) over the first 24 hours after birth and relate this coupling to gestational age, inotropic therapy, and short-term clinical outcome.

APPROACH

We develop new nonstationary measures of coupling to summarise both frequency- and direction-dependent coupling. These measures employ a coherence measure capable of measuring time-varying Granger casuality using a short-time information partial directed coherence function. Measures are correlated with gestational age, inotropic therapy (yes/no), and outcome (adverse/normal).

MAIN RESULTS

In a cohort of 99 extremely preterm infants (<28 weeks of gestation), we find weak but significant coupling in both the HR-to-PI and PI-to-HR directions (P<0.05). HR-to-PI coupling increases with maturation (correlation r=0.26; P=0.011); PI-to-HR coupling increases with inotrope administration (r=0.27; P=0.007). And nonstationary features of PI-to-HR coupling are associated with (r=0.27; P=0.009).

SIGNIFICANCE

Nonstationary features are necessary to distinguish different coupling types for complex biomedical systems. Time-varying directional coupling between PI and HR provides objective and independent biomarkers of adverse outcome in extremely preterm infants.

Machine Learning to Support Hemodynamic Intervention in the Neonatal Intensive Care Unit

Hemodynamic support in neonatal intensive care is directed at maintaining cardiovascular wellbeing. At present, monitoring of vital signs plays an essential role in augmenting care in a reactive manner. By applying machine learning techniques, a model can be trained to learn patterns in time series data, allowing the detection of adverse outcomes before they become clinically apparent. In this review we provide an overview of the different machine learning techniques that have been used to develop models in hemodynamic care for newborn infants. We focus on their potential benefits, research pitfalls, and challenges related to their implementation in clinical care.

Monitoring Circulation During Transition in Extreme Low Gestational Age Newborns: What's on the Horizon?

Echocardiography and near-infrared spectroscopy have significantly changed our view on hemodynamic transition of the extreme preterm infant. Instead of focusing on maintaining an arbitrary target value of blood pressure, we aim for circulatory well-being by a comprehensive holistic assessment of markers of cardiovascular instability. Most of these clinical and biochemical indices are influenced by transition itself and remain poor discriminators to identify patients with a potential need for therapeutic intervention. At the same time, the evolution in data capturing and storage has led to a change in our approach to monitor vital parameters. Continuous trend monitoring has become more and more relevant. By using signal extraction methods, changes in trends over time can be quantified. In this review, we will discuss the impact of these innovations on the current monitoring practices and explore some of the potential benefits these techniques may have in improving real-time detection of extreme low birth weight infants at risk for morbidity related to impaired hemodynamic transition.

Decreased Variability and Low Values of Perfusion Index on Day One Are Associated with Adverse Outcome in Extremely Preterm Infants

OBJECTIVE

To develop new quantitative features for the Perfusion Index signal recorded continuously over the first 24 hours of life in a cohort of extremely low gestational age newborns and to assess the association of these features with normal and adverse short-term outcome.

STUDY DESIGN

A cohort study of extremely low gestational age newborns. Adverse outcome was defined as early mortality before 72 hours of life, acquired severe periventricular-intraventricular hemorrhage, or severe cystic leukomalacia. Perfusion Index values were obtained from the plethysmographic signal of a pulse oximeter. Perfusion Index signals were separated into low-frequency (trend) and high-frequency (detrend) components. Three features were extracted during four 6-hour epochs: mean of the trend component (mean-trend), SD of the trend component (SD-trend), and SD of the detrend component (SD-detrend). The SD features represent long-term variability (SD-trend) and short-term variability (SD-detrend) of the Perfusion Index. A mixed-effects model was fitted to each feature.

RESULTS

Ninety-nine infants were included in the analysis. Quadratic-time mixed-effects models provided the best fit for all 3 features. The mean-trend component was lower for the adverse outcome compared with the normal outcome group with a difference of 0.142 Perfusion Index (P = .001). SD-detrend component was also lower for the adverse compared with the normal outcome group, although this difference of 0.031 Perfusion Index/days² was dependent on time (P < .001).

CONCLUSION

Low values and reduced short-term variability of Perfusion Index on day 1 are associated with adverse outcome.