Prediction of Return of Spontaneous Circulation in a Pediatric Swine Model of Cardiac Arrest Using Low-Resolution Multimodal Physiological Waveforms.

Monitoring physiological waveforms, specifically hemodynamic variables (e.g., blood pressure waveforms) and end-tidal CO2 (EtCO2), during pediatric cardiopulmonary resuscitation (CPR) has been demonstrated to improve survival rates and outcomes when compared to standard depth-guided CPR. However, waveform guidance has largely been based on thresholds for single parameters and therefore does not leverage all the information contained in multimodal data. We hypothesize that the combination of multimodal physiological features improves the prediction of the return of spontaneous circulation (ROSC), the clinical indicator of short-term CPR success. We used machine learning algorithms to evaluate features extracted from eight low-resolution (4 samples per minute) physiological waveforms to predict ROSC. The waveforms were acquired from the 2nd to 10th minute of CPR in pediatric swine models of cardiac arrest (N = 89, 8-12 kg). The waveforms were divided into segments with increasing length (both forward and backward) for feature extraction, and machine learning algorithms were trained for ROSC prediction. For the full CPR period (2nd to 10th minute), the area under the receiver operating characteristics curve (AUC) was 0.93 (95% CI: 0.87-0.99) for the multivariate model, 0.70 (0.55-0.85) for EtCO2 and 0.80 (0.67-0.93) for coronary perfusion pressure. The best prediction performances were achieved when the period from the 6th to the 10th minute was included. Poor predictions were observed for some individual waveforms, e.g., right atrial pressure. In conclusion, multimodal waveform features carry relevant information for ROSC prediction. Using multimodal waveform features in CPR guidance has the potential to improve resuscitation success and reduce mortality.

Using Data-Driven Machine Learning to Predict Unplanned ICU Transfers with Critical Deterioration from Electronic Health Records.

OBJECTIVE: We aimed to develop a data-driven machine learning model for predicting critical deterioration events from routinely collected EHR data in hospitalized children. MATERIALS: This retrospective cohort study included all pediatric inpatients hospitalized on a medical or surgical ward between 2014-2018 at a quaternary children's hospital. METHODS: We developed a large data-driven approach and evaluated three machine learning models to predict pediatric critical deterioration events. We evaluated the models using a nested, stratified 10-fold cross-validation. The evaluation metrics included C-statistic, sensitivity, and positive predictive value. We also compared the machine learning models with patients identified as high-risk Watchers by bedside clinicians. RESULTS: The study included 57,233 inpatient admissions from 34,976 unique patients. 3,943 variables were identified from the EHR data. The XGBoost model performed best (C-statistic=0.951, CI: 0.946 ∼ 0.956). CONCLUSIONS: Our data-driven machine learning models accurately predicted patient deterioration. Future sociotechnical analysis will inform deployment within the clinical setting.

Classifying social determinants of health from unstructured electronic health records using deep learning-based natural language processing.

OBJECTIVE: Social determinants of health (SDOH) are non-medical factors that can profoundly impact patient health outcomes. However, SDOH are rarely available in structured electronic health record (EHR) data such as diagnosis codes, and more commonly found in unstructured narrative clinical notes. Hence, identifying social context from unstructured EHR data has become increasingly important. Yet, previous work on using natural language processing to automate extraction of SDOH from text (a) usually focuses on an ad hoc selection of SDOH, and (b) does not use the latest advances in deep learning. Our objective was to advance automatic extraction of SDOH from clinical text by (a) systematically creating a set of SDOH based on standard biomedical and psychiatric ontologies, and (b) training state-of-the-art deep neural networks to extract mentions of these SDOH from clinical notes. DESIGN: A retrospective cohort study. SETTING AND PARTICIPANTS: Data were extracted from the Medical Information Mart for Intensive Care (MIMIC-III) database. The corpus comprised 3,504 social related sentences from 2,670 clinical notes. METHODS: We developed a framework for automated classification of multiple SDOH categories. Our dataset comprised narrative clinical notes under the "Social Work" category in the MIMIC-III Clinical Database. Using standard terminologies, SNOMED-CT and DSM-IV, we systematically curated a set of 13 SDOH categories and created annotation guidelines for these. After manually annotating the 3,504 sentences, we developed and tested three deep neural network (DNN) architectures - convolutional neural network (CNN), long short-term memory (LSTM) network, and the Bidirectional Encoder Representations from Transformers (BERT) - for automated detection of eight SDOH categories. We also compared these DNNs to three baselines models: (1) cTAKES, as well as (2) L2-regularized logistic regression and (3) random forests on bags-of-words. Model evaluation metrics included micro- and macro- F1, and area under the receiver operating characteristic curve (AUC). RESULTS: All three DNN models accurately classified all SDOH categories (minimum micro-F1 = 0.632, minimum macro-AUC = 0.854). Compared to the CNN and LSTM, BERT performed best in most key metrics (micro-F1 = 0.690, macro-AUC = 0.907). The BERT model most effectively identified the "occupational" category (F1 = 0.774, AUC = 0.965) and least effectively identified the "non-SDOH" category (F = 0.491, AUC = 0.788). BERT outperformed cTAKES in distinguishing social vs non-social sentences (BERT F1 = 0.87 vs. cTAKES F1 = 0.06), and outperformed logistic regression (micro-F1 = 0.649, macro-AUC = 0.696) and random forest (micro-F1 = 0.502, macro-AUC = 0.523) trained on bag-of-words. CONCLUSIONS: Our study framework with DNN models demonstrated improved performance for efficiently identifying a systematic range of SDOH categories from clinical notes in the EHR. Improved identification of patient SDOH may further improve healthcare outcomes.

Natural language processing and machine learning of electronic health records for prediction of first-time suicide attempts.

OBJECTIVE: Limited research exists in predicting first-time suicide attempts that account for two-thirds of suicide decedents. We aimed to predict first-time suicide attempts using a large data-driven approach that applies natural language processing (NLP) and machine learning (ML) to unstructured (narrative) clinical notes and structured electronic health record (EHR) data. METHODS: This case-control study included patients aged 10-75 years who were seen between 2007 and 2016 from emergency departments and inpatient units. Cases were first-time suicide attempts from coded diagnosis; controls were randomly selected without suicide attempts regardless of demographics, following a ratio of nine controls per case. Four data-driven ML models were evaluated using 2-year historical EHR data prior to suicide attempt or control index visits, with prediction windows from 7 to 730 days. Patients without any historical notes were excluded. Model evaluation on accuracy and robustness was performed on a blind dataset (30% cohort). RESULTS: The study cohort included 45 238 patients (5099 cases, 40 139 controls) comprising 54 651 variables from 5.7 million structured records and 798 665 notes. Using both unstructured and structured data resulted in significantly greater accuracy compared to structured data alone (area-under-the-curve [AUC]: 0.932 vs. 0.901 P < .001). The best-predicting model utilized 1726 variables with AUC = 0.932 (95% CI, 0.922-0.941). The model was robust across multiple prediction windows and subgroups by demographics, points of historical most recent clinical contact, and depression diagnosis history. CONCLUSIONS: Our large data-driven approach using both structured and unstructured EHR data demonstrated accurate and robust first-time suicide attempt prediction, and has the potential to be deployed across various populations and clinical settings.

Early prediction of critical events for infants with single-ventricle physiology in critical care using routinely collected data.

OBJECTIVE: Critical events are common and difficult to predict among infants with congenital heart disease and are associated with mortality and long-term sequelae. We aimed to achieve early prediction of critical events, that is, cardiopulmonary resuscitation, emergency endotracheal intubation, and extracorporeal membrane oxygenation in infants with single-ventricle physiology before second-stage surgery. We hypothesized that naïve Bayesian models learned from expert knowledge and clinical data can predict critical events early and accurately. METHODS: We collected 93 patients with single-ventricle physiology admitted to intensive care units in a single tertiary pediatric hospital between 2014 and 2017. Using knowledge elicited from experienced cardiac-intensive-care-unit providers and machine-learning techniques, we developed and evaluated the Cardiac-intensive-care Warning INdex (C-WIN) system, consisting of a set of naïve Bayesian models that leverage routinely collected data. We evaluated predictive performance using the area under the receiver operating characteristic curve, sensitivity, and specificity. We performed the evaluation at 5 different prediction horizons: 1, 2, 4, 6, and 8 hours before the onset of critical events. RESULTS: The area under the receiver operating characteristic curves of the C-WIN models ranged between 0.73 and 0.88 at different prediction horizons. At 1 hour before critical events, C-WIN was able to detect events with an area under the receiver operating characteristic curve of 0.88 (95% confidence interval, 0.84-0.92) and a sensitivity of 84% at the 81% specificity level. CONCLUSIONS: Predictive models may enhance clinicians' ability to identify infants with single-ventricle physiology at high risk of critical events. Early prediction of critical events may indicate the need to perform timely interventions, potentially reducing morbidity, mortality, and health care costs.