Vital sign metrics of VLBW infants in three NICUs: implications for predictive algorithms

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.

Early prediction of severe retinopathy of prematurity requiring laser treatment using physiological data

BACKGROUND

Early risk stratification for developing retinopathy of prematurity (ROP) is essential for tailoring screening strategies and preventing abnormal retinal development. This study aims to examine the ability of physiological data during the first postnatal month to distinguish preterm infants with and without ROP requiring laser treatment.

METHODS

In this cohort study, preterm infants with a gestational age <32 weeks and/or birth weight <1500 g, who were screened for ROP were included. Differences in the physiological data between the laser and non-laser group were identified, and tree-based classification models were trained and independently tested to predict ROP requiring laser treatment.

RESULTS

In total, 208 preterm infants were included in the analysis of whom 30 infants (14%) required laser treatment. Significant differences were identified in the level of hypoxia and hyperoxia, oxygen requirement, and skewness of heart rate. The best model had a balanced accuracy of 0.81 (0.72-0.87), a sensitivity of 0.73 (0.64-0.81), and a specificity of 0.88 (0.80-0.93) and included the SpO₂/FiO₂ ratio and baseline demographics (including gestational age and birth weight).

CONCLUSIONS

Routinely monitored physiological data from preterm infants in the first postnatal month are already predictive of later development of ROP requiring laser treatment, although validation is required in larger cohorts.

IMPACT

Routinely monitored physiological data from the first postnatal month are predictive of later development of ROP requiring laser treatment, although model performance was not significantly better than baseline characteristics (gestational age, birth weight, sex, multiple birth, prenatal glucocorticosteroids, route of delivery, and Apgar scores) alone. A balanced accuracy of 0.81 (0.72-0.87), a sensitivity of 0.73 (0.64-0.81), and a specificity of 0.88 (0.80-0.93) was achieved with a model including the SpO₂/FiO₂ ratio and baseline characteristics. Physiological data have potential to play a significant role for future ROP prediction and provide opportunities for early interventions to protect infants from abnormal retinal development.

Cardiorespiratory signature of neonatal sepsis: development and validation of prediction models in 3 NICUs

BACKGROUND

Heart rate characteristics aid early detection of late-onset sepsis (LOS), but respiratory data contain additional signatures of illness due to infection. Predictive models using cardiorespiratory data may improve early sepsis detection. We hypothesized that heart rate (HR) and oxygenation (SpO₂) data contain signatures that improve sepsis risk prediction over HR or demographics alone.

METHODS

We analyzed cardiorespiratory data from very low birth weight (VLBW, <1500 g) infants admitted to three NICUs. We developed and externally validated four machine learning models to predict LOS using features calculated every 10 m: mean, standard deviation, skewness, kurtosis of HR and SpO₂, and cross-correlation. We compared feature importance, discrimination, calibration, and dynamic prediction across models and cohorts. We built models of demographics and HR or SpO₂ features alone for comparison with HR-SpO₂ models.

RESULTS

Performance, feature importance, and calibration were similar among modeling methods. All models had favorable external validation performance. The HR-SpO₂ model performed better than models using either HR or SpO₂ alone. Demographics improved the discrimination of all physiologic data models but dampened dynamic performance.

CONCLUSIONS

Cardiorespiratory signatures detect LOS in VLBW infants at 3 NICUs. Demographics risk-stratify, but predictive modeling with both HR and SpO₂ features provides the best dynamic risk prediction.

IMPACT

Heart rate characteristics aid early detection of late-onset sepsis, but respiratory data contain signatures of illness due to infection. Predictive models using both heart rate and respiratory data may improve early sepsis detection. A cardiorespiratory early warning score, analyzing heart rate from electrocardiogram or pulse oximetry with SpO₂, predicts late-onset sepsis within 24 h across multiple NICUs and detects sepsis better than heart rate characteristics or demographics alone. Demographics risk-stratify, but predictive modeling with both HR and SpO₂ features provides the best dynamic risk prediction. The results increase understanding of physiologic signatures of neonatal sepsis.

Application and potential of artificial intelligence in neonatal medicine

Neonatal care is becoming increasingly complex with large amounts of rich, routinely recorded physiological, diagnostic and outcome data. Artificial intelligence (AI) has the potential to harness this vast quantity and range of information and become a powerful tool to support clinical decision making, personalised care, precise prognostics, and enhance patient safety. Current AI approaches in neonatal medicine include tools for disease prediction and risk stratification, neurological diagnostic support and novel image recognition technologies. Key to the integration of AI in neonatal medicine is the understanding of its limitations and a standardised critical appraisal of AI tools. Barriers and challenges to this include the quality of datasets used, performance assessment, and appropriate external validation and clinical impact studies. Improving digital literacy amongst healthcare professionals and cross-disciplinary collaborations are needed to harness the full potential of AI to help take the next significant steps in improving neonatal outcomes for high-risk infants.

The Impact of Dexmedetomidine Initiation on Cardiovascular Status and Oxygenation in Critically ill Neonates

Dexmedetomidine is being increasingly used as a primary or adjunctive sedative agent in neonates. There are a paucity of high-quality, high-resolution physiologic data during administration, despite significant potential cardiorespiratory effects. Term and preterm infants admitted between January 2018 and July 2020 were screened for dexmedetomidine exposure. Prospectively recorded vital signs (heart rate, oxygenation, arterial blood pressure) were cross-matched with pharmacy records to identify infants with data available 24 h before and 48 h after drug initiation. Vital sign data were processed via a standardized pipeline to (a) remove missing data, (b) obtain baseline averages of vital signs for 24 h preceding dexmedetomidine, and (c) calculate the hourly mean deviation from the baseline for the 48 h following initiation of dexmedetomidine. Infants were clustered by postmenstrual age (preterm ≤ 35 weeks; term > 35 weeks). 72 infants were identified with mean gestational age of 32 weeks and mean ± SD birth weight of 1976 ± 1341 g. Although both groups of infants experienced bradycardia, heart rate in term infants dropped faster and reached a nadir 5 beats per minute lower, before converging at a common deviation of - 10 beats per minute. No hypo- or hypertension was noted in either group. Unexpected instability of oxygenation occurred in a subset of preterm infants, requiring escalation of respiratory support. Administration of dexmedetomidine results in differential timing and magnitude of bradycardia in term and preterm infants, no major impact on blood pressure, and a surprising instability of oxygenation in preterm infants, requiring increased ventilatory support. Further investigation is warranted.

Discovery of signatures of fatal neonatal illness in vital signs using highly comparative time-series analysis

To seek new signatures of illness in heart rate and oxygen saturation vital signs from Neonatal Intensive Care Unit (NICU) patients, we implemented highly comparative time-series analysis to discover features of all-cause mortality in the next 7 days. We collected 0.5 Hz heart rate and oxygen saturation vital signs of infants in the University of Virginia NICU from 2009 to 2019. We applied 4998 algorithmic operations from 11 mathematical families to random daily 10 min segments from 5957 NICU infants, 205 of whom died. We clustered the results and selected a representative from each, and examined multivariable logistic regression models. 3555 operations were usable; 20 cluster medoids held more than 81% of the information, and a multivariable model had AUC 0.83. New algorithms outperformed others: moving threshold, successive increases, surprise, and random walk. We computed provenance of the computations and constructed a software library with links to the data. We conclude that highly comparative time-series analysis revealed new vital sign measures to identify NICU patients at the highest risk of death in the next week.

Continuous cardiorespiratory monitoring is a dominant source of predictive signal in machine learning for risk stratification and clinical decision support. Physiol Meas 2021;42. [PMID: 34580243 DOI: 10.1088/1361-6579/ac2130]

Beaulieu-Jones and coworkers propose a litmus test for the field of predictive analytics-performance improvements must be demonstrated to be the result of non-clinician-initiated data, otherwise, there should be caution in assuming that predictive models could improve clinical decision-making (Beaulieu-Joneset al2021). They demonstrate substantial prognostic information in unsorted physician orders made before the first midnight of hospital admission, and we are persuaded that it is fair to ask-if the physician thought of it first, what exactly is machine learning for in-patient risk stratification learning about? While we want predictive analytics to represent the leading indicators of a patient's illness, does it instead merely reflect the lagging indicators of clinicians' actions? We propose that continuous cardiorespiratory monitoring-'routine telemetry data,' in Beaulieu-Jones' terms-represents the most valuable non-clinician-initiated predictive signal present in patient data, and the value added to patient care justifies the efforts and expense required. Here, we present a clinical and a physiological point of view to support our contention.