RESUMEN
BACKGROUND: Robust risk assessment is crucial for the growing repaired tetralogy of Fallot population at risk of major adverse clinical outcomes; however, current tools are hindered by lack of validation. This study aims to develop and validate a risk prediction model for death in the repaired tetralogy of Fallot population. METHODS AND RESULTS: Patients with repaired tetralogy of Fallot enrolled in the INDICATOR (International Multicenter Tetralogy of Fallot Registry) cohort with clinical, arrhythmia, cardiac magnetic resonance, and outcome data were included. Patients from London, Amsterdam, and Boston sites were placed in the development cohort; patients from the Toronto site were used for external validation. Multivariable Cox regression was used to evaluate factors associated with time from cardiac magnetic resonance until the primary outcome: all-cause death. Of 1552 eligible patients (n=1221 in development, n=331 in validation; median age at cardiac magnetic resonance 23.4 [interquartile range, 15.6-35.6] years; median follow up 9.5 years), 102 (6.6%) experienced the primary outcome. The multivariable Cox model performed similarly during development (concordance index, 0.83 [95% CI, 0.78-0.88]) and external validation (concordance index, 0.80 [95% CI, 0.71-0.90]) and identified older age at cardiac magnetic resonance, obesity, type of tetralogy of Fallot repair, higher right ventricular end-systolic volume index, and lower biventricular global function index as independent predictors of death. A risk-scoring algorithm dividing patients into low-risk (score ≤4) versus high-risk (score >4) groups was validated to effectively discriminate risk of death (15-year survival of 95% versus 74%, respectively; P<0.001). CONCLUSIONS: This externally validated mortality risk prediction algorithm can help identify vulnerable patients with repaired tetralogy of Fallot who may benefit from targeted interventions.
Asunto(s)
Sistema de Registros , Tetralogía de Fallot , Humanos , Tetralogía de Fallot/cirugía , Tetralogía de Fallot/mortalidad , Masculino , Femenino , Medición de Riesgo/métodos , Adulto , Adolescente , Adulto Joven , Factores de Riesgo , Procedimientos Quirúrgicos Cardíacos/efectos adversos , Procedimientos Quirúrgicos Cardíacos/mortalidad , Reproducibilidad de los Resultados , Factores de Tiempo , Valor Predictivo de las Pruebas , Causas de MuerteRESUMEN
OBJECTIVES: The pediatric Sequential Organ Failure Assessment (pSOFA) score summarizes severity of organ dysfunction and can be used to predict in-hospital mortality. Manual calculation of the pSOFA score is time-consuming and prone to human error. An automated method that is open-source, flexible, and scalable for calculating the pSOFA score directly from electronic health record data is desirable. DESIGN: Single-center, retrospective cohort study. SETTING: Quaternary 40-bed PICU. PATIENTS: All patients admitted to the PICU between 2015 and 2021 with ICU stay of at least 24 hours. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We used 77 records to evaluate the automated score. The automated algorithm had an overall accuracy of 97%. The algorithm calculated the respiratory component of two cases incorrectly. An expert human annotator had an initial accuracy of 75% at the patient level and 95% at the component level. An untrained human annotator with general clinical research experience had an overall accuracy of 16% and component-wise accuracy of 67%. Weighted kappa for agreement between the automated method and the expert annotator's initial score was 0.92 (95% CI, 0.88-0.95), and between the untrained human annotator and the automated score was 0.50 (95% CI, 0.36-0.61). Data from 9146 patients (in-hospital mortality 3.6%) were included to validate externally the discriminability of the automated pSOFA score. The admission-day pSOFA score had an area under the receiver operating characteristic curve of 0.79 (95% CI, 0.77-0.82). CONCLUSIONS: The developed automated algorithm calculates pSOFA score with high accuracy and is more accurate than a trained expert rater and nontrained data abstracter. pSOFA's performance for predicting in-hospital mortality was lower in our cohort than it was for the originally derived score.
Asunto(s)
Algoritmos , Mortalidad Hospitalaria , Unidades de Cuidado Intensivo Pediátrico , Puntuaciones en la Disfunción de Órganos , Humanos , Estudios Retrospectivos , Masculino , Femenino , Niño , Preescolar , Lactante , Adolescente , Registros Electrónicos de Salud , Insuficiencia Multiorgánica/diagnóstico , Insuficiencia Multiorgánica/mortalidad , Reproducibilidad de los ResultadosRESUMEN
OBJECTIVES: The pediatric Sequential Organ Failure Assessment (pSOFA) score was designed to track illness severity and predict mortality in critically ill children. Most commonly, pSOFA at a point in time is used to assess a static patient condition. However, this approach has a significant drawback because it fails to consider any changes in a patients' condition during their PICU stay and, especially, their response to initial critical care treatment. We aimed to evaluate the performance of longitudinal pSOFA scores for predicting mortality. DESIGN: Single-center, retrospective cohort study. SETTING: Quaternary 40-bed PICU. PATIENTS: All patients admitted to the PICU between 2015 and 2021 with at least 24 hours of ICU stay. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We calculated daily pSOFA scores up to 30 days, or until death or discharge from the PICU, if earlier. We used the joint longitudinal and time-to-event data model for the dynamic prediction of 30-day in-hospital mortality. The dataset, which included 9146 patients with a 30-day in-hospital mortality of 2.6%, was divided randomly into training (75%) and validation (25%) subsets, and subjected to 40 repeated stratified cross-validations. We used dynamic area under the curve (AUC) to evaluate the discriminative performance of the model. Compared with the admission-day pSOFA score, AUC for predicting mortality between days 5 and 30 was improved on average by 6.4% (95% CI, 6.3-6.6%) using longitudinal pSOFA scores from the first 3 days and 9.2% (95% CI, 9.0-9.5%) using scores from the first 5 days. CONCLUSIONS: Compared with admission-day pSOFA score, longitudinal pSOFA scores improved the accuracy of mortality prediction in PICU patients at a single center. The pSOFA score has the potential to be used dynamically for the evaluation of patient conditions.
Asunto(s)
Enfermedad Crítica , Mortalidad Hospitalaria , Unidades de Cuidado Intensivo Pediátrico , Puntuaciones en la Disfunción de Órganos , Humanos , Unidades de Cuidado Intensivo Pediátrico/estadística & datos numéricos , Estudios Retrospectivos , Masculino , Femenino , Niño , Preescolar , Lactante , Enfermedad Crítica/mortalidad , Adolescente , Estudios Longitudinales , Curva ROC , PronósticoRESUMEN
BACKGROUND: Real-time surveillance of emerging infectious diseases necessitates a dynamically evolving, computable case definition, which frequently incorporates symptom-related criteria. For symptom detection, both population health monitoring platforms and research initiatives primarily depend on structured data extracted from electronic health records. OBJECTIVE: This study sought to validate and test an artificial intelligence (AI)-based natural language processing (NLP) pipeline for detecting COVID-19 symptoms from physician notes in pediatric patients. We specifically study patients presenting to the emergency department (ED) who can be sentinel cases in an outbreak. METHODS: Subjects in this retrospective cohort study are patients who are 21 years of age and younger, who presented to a pediatric ED at a large academic children's hospital between March 1, 2020, and May 31, 2022. The ED notes for all patients were processed with an NLP pipeline tuned to detect the mention of 11 COVID-19 symptoms based on Centers for Disease Control and Prevention (CDC) criteria. For a gold standard, 3 subject matter experts labeled 226 ED notes and had strong agreement (F1-score=0.986; positive predictive value [PPV]=0.972; and sensitivity=1.0). F1-score, PPV, and sensitivity were used to compare the performance of both NLP and the International Classification of Diseases, 10th Revision (ICD-10) coding to the gold standard chart review. As a formative use case, variations in symptom patterns were measured across SARS-CoV-2 variant eras. RESULTS: There were 85,678 ED encounters during the study period, including 4% (n=3420) with patients with COVID-19. NLP was more accurate at identifying encounters with patients that had any of the COVID-19 symptoms (F1-score=0.796) than ICD-10 codes (F1-score =0.451). NLP accuracy was higher for positive symptoms (sensitivity=0.930) than ICD-10 (sensitivity=0.300). However, ICD-10 accuracy was higher for negative symptoms (specificity=0.994) than NLP (specificity=0.917). Congestion or runny nose showed the highest accuracy difference (NLP: F1-score=0.828 and ICD-10: F1-score=0.042). For encounters with patients with COVID-19, prevalence estimates of each NLP symptom differed across variant eras. Patients with COVID-19 were more likely to have each NLP symptom detected than patients without this disease. Effect sizes (odds ratios) varied across pandemic eras. CONCLUSIONS: This study establishes the value of AI-based NLP as a highly effective tool for real-time COVID-19 symptom detection in pediatric patients, outperforming traditional ICD-10 methods. It also reveals the evolving nature of symptom prevalence across different virus variants, underscoring the need for dynamic, technology-driven approaches in infectious disease surveillance.
Asunto(s)
Biovigilancia , COVID-19 , Médicos , SARS-CoV-2 , Estados Unidos , Humanos , Niño , Inteligencia Artificial , Estudios Retrospectivos , COVID-19/diagnóstico , COVID-19/epidemiologíaRESUMEN
OBJECTIVES: Identification of children with sepsis-associated multiple organ dysfunction syndrome (MODS) at risk for poor outcomes remains a challenge. We sought to the determine reproducibility of the data-driven "persistent hypoxemia, encephalopathy, and shock" (PHES) phenotype and determine its association with inflammatory and endothelial biomarkers, as well as biomarker-based pediatric risk strata. DESIGN: We retrained and validated a random forest classifier using organ dysfunction subscores in the 2012-2018 electronic health record (EHR) dataset used to derive the PHES phenotype. We used this classifier to assign phenotype membership in a test set consisting of prospectively (2003-2023) enrolled pediatric septic shock patients. We compared profiles of the PERSEVERE family of biomarkers among those with and without the PHES phenotype and determined the association with established biomarker-based mortality and MODS risk strata. SETTING: Twenty-five PICUs across the United States. PATIENTS: EHR data from 15,246 critically ill patients with sepsis-associated MODS split into derivation and validation sets and 1,270 pediatric septic shock patients in the test set of whom 615 had complete biomarker data. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: The area under the receiver operator characteristic curve of the modified classifier to predict PHES phenotype membership was 0.91 (95% CI, 0.90-0.92) in the EHR validation set. In the test set, PHES phenotype membership was associated with both increased adjusted odds of complicated course (adjusted odds ratio [aOR] 4.1; 95% CI, 3.2-5.4) and 28-day mortality (aOR of 4.8; 95% CI, 3.11-7.25) after controlling for age, severity of illness, and immunocompromised status. Patients belonging to the PHES phenotype were characterized by greater degree of systemic inflammation and endothelial activation, and were more likely to be stratified as high risk based on PERSEVERE biomarkers predictive of death and persistent MODS. CONCLUSIONS: The PHES trajectory-based phenotype is reproducible, independently associated with poor clinical outcomes, and overlapped with higher risk strata based on prospectively validated biomarker approaches.
Asunto(s)
Biomarcadores , Hipoxia , Fenotipo , Choque Séptico , Humanos , Biomarcadores/sangre , Femenino , Masculino , Niño , Preescolar , Lactante , Choque Séptico/sangre , Choque Séptico/mortalidad , Choque Séptico/diagnóstico , Hipoxia/diagnóstico , Hipoxia/sangre , Unidades de Cuidado Intensivo Pediátrico , Insuficiencia Multiorgánica/diagnóstico , Insuficiencia Multiorgánica/mortalidad , Insuficiencia Multiorgánica/sangre , Adolescente , Sepsis/diagnóstico , Sepsis/complicaciones , Sepsis/sangre , Sepsis/mortalidad , Reproducibilidad de los Resultados , Medición de Riesgo/métodos , Estudios Prospectivos , Encefalopatía Asociada a la Sepsis/sangre , Encefalopatía Asociada a la Sepsis/diagnóstico , Curva ROC , Puntuaciones en la Disfunción de ÓrganosRESUMEN
OBJECTIVES: Generative language models (LMs) are being evaluated in a variety of tasks in healthcare, but pediatric critical care studies are scant. Our objective was to evaluate the utility of generative LMs in the pediatric critical care setting and to determine whether domain-adapted LMs can outperform much larger general-domain LMs in generating a differential diagnosis from the admission notes of PICU patients. DESIGN: Single-center retrospective cohort study. SETTING: Quaternary 40-bed PICU. PATIENTS: Notes from all patients admitted to the PICU between January 2012 and April 2023 were used for model development. One hundred thirty randomly selected admission notes were used for evaluation. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Five experts in critical care used a 5-point Likert scale to independently evaluate the overall quality of differential diagnoses: 1) written by the clinician in the original notes, 2) generated by two general LMs (BioGPT-Large and LLaMa-65B), and 3) generated by two fine-tuned models (fine-tuned BioGPT-Large and fine-tuned LLaMa-7B). Differences among differential diagnoses were compared using mixed methods regression models. We used 1,916,538 notes from 32,454 unique patients for model development and validation. The mean quality scores of the differential diagnoses generated by the clinicians and fine-tuned LLaMa-7B, the best-performing LM, were 3.43 and 2.88, respectively (absolute difference 0.54 units [95% CI, 0.37-0.72], p < 0.001). Fine-tuned LLaMa-7B performed better than LLaMa-65B (absolute difference 0.23 unit [95% CI, 0.06-0.41], p = 0.009) and BioGPT-Large (absolute difference 0.86 unit [95% CI, 0.69-1.0], p < 0.001). The differential diagnosis generated by clinicians and fine-tuned LLaMa-7B were ranked as the highest quality in 144 (55%) and 74 cases (29%), respectively. CONCLUSIONS: A smaller LM fine-tuned using notes of PICU patients outperformed much larger models trained on general-domain data. Currently, LMs remain inferior but may serve as an adjunct to human clinicians in real-world tasks using real-world data.
Asunto(s)
Inteligencia Artificial , Unidades de Cuidado Intensivo Pediátrico , Humanos , Estudios Retrospectivos , Diagnóstico Diferencial , Niño , Masculino , Femenino , Preescolar , Lactante , Cuidados Críticos/métodos , AdolescenteRESUMEN
OBJECTIVE: To quantify the increase in pediatric patients presenting to the emergency department with suicidality before and during the COVID-19 pandemic, and the subsequent impact on emergency department length of stay and boarding. METHODS: This retrospective cohort study from June 1, 2016, to October 31, 2022, identified patients ages 6 to 21 presenting to the emergency department at a pediatric academic medical center with suicidality using ICD-10 codes. Number of emergency department encounters for suicidality, demographic characteristics of patients with suicidality, and emergency department length of stay were compared before and during the COVID-19 pandemic. Unobserved components models were used to describe monthly counts of emergency department encounters for suicidality. RESULTS: There were 179,736 patient encounters to the emergency department during the study period, 6,215 (3.5%) for suicidality. There were, on average, more encounters for suicidality each month during the COVID-19 pandemic than before the COVID-19 pandemic. A time series unobserved components model demonstrated a temporary drop of 32.7 encounters for suicidality in April and May of 2020 (p<0.001), followed by a sustained increase of 31.2 encounters starting in July 2020 (p = 0.003). The average length of stay for patients that boarded in the emergency department with a diagnosis of suicidality was 37.4 hours longer during the COVID-19 pandemic compared to before the COVID-19 pandemic (p<0.001). CONCLUSIONS: The number of encounters for suicidality among pediatric patients and the emergency department length of stay for psychiatry boarders has increased during the COVID-19 pandemic. There is a need for acute care mental health services and solutions to emergency department capacity issues.
Asunto(s)
COVID-19 , Suicidio , Humanos , Niño , Estudios Retrospectivos , Pandemias , COVID-19/epidemiología , Servicio de Urgencia en HospitalRESUMEN
Background: Identifying phenotypes in sepsis patients may enable precision medicine approaches. However, the generalisability of these phenotypes to specific patient populations is unclear. Given that paediatric cancer patients with sepsis have different host response and pathogen profiles and higher mortality rates when compared to non-cancer patients, we determined whether unique, reproducible, and clinically-relevant sepsis phenotypes exist in this specific patient population. Methods: We studied patients with underlying malignancies admitted with sepsis to one of 25 paediatric intensive care units (PICUs) participating in two large, multi-centre, observational cohorts from the European SCOTER study (n = 383 patients; study period between January 1, 2018 and January 1, 2020) and the U.S. Novel Data-Driven Sepsis Phenotypes in Children study (n = 1898 patients; study period between January 1, 2012 and January 1, 2018). We independently used latent class analysis (LCA) in both cohorts to identify phenotypes using demographic, clinical, and laboratory data from the first 24 h of PICU admission. We then tested the association of the phenotypes with clinical outcomes in both cohorts. Findings: LCA identified two distinct phenotypes that were comparable across both cohorts. Phenotype 1 was characterised by lower serum bicarbonate and albumin, markedly increased lactate and hepatic, renal, and coagulation abnormalities when compared to phenotype 2. Patients with phenotype 1 had a higher 90-day mortality (European cohort 29.2% versus 13.4%, U.S. cohort 27.3% versus 11.4%, p < 0.001) and received more vasopressor and renal replacement therapy than patients with phenotype 2. After adjusting for severity of organ dysfunction, haematological cancer, prior stem cell transplantation and age, phenotype 1 was associated with an adjusted OR of death at 90-day of 1.9 (1.04-3.34) in the European cohort and 1.6 (1.2-2.2) in the U.S. cohort. Interpretation: We identified two clinically-relevant sepsis phenotypes in paediatric cancer patients that are reproducible across two international, multicentre cohorts with prognostic implications. These results may guide further research regarding therapeutic approaches for these specific phenotypes. Funding: Part of this study is funded by the Eunice Kennedy Shriver National Institute of Child Health and Human Development.
RESUMEN
Though electronic health record (EHR) systems are a rich repository of clinical information with large potential, the use of EHR-based phenotyping algorithms is often hindered by inaccurate diagnostic records, the presence of many irrelevant features, and the requirement for a human-labeled training set. In this paper, we describe a knowledge-driven online multimodal automated phenotyping (KOMAP) system that i) generates a list of informative features by an online narrative and codified feature search engine (ONCE) and ii) enables the training of a multimodal phenotyping algorithm based on summary data. Powered by composite knowledge from multiple EHR sources, online article corpora, and a large language model, features selected by ONCE show high concordance with the state-of-the-art AI models (GPT4 and ChatGPT) and encourage large-scale phenotyping by providing a smaller but highly relevant feature set. Validation of the KOMAP system across four healthcare centers suggests that it can generate efficient phenotyping algorithms with robust performance. Compared to other methods requiring patient-level inputs and gold-standard labels, the fully online KOMAP provides a significant opportunity to enable multi-center collaboration.
RESUMEN
OBJECTIVES: "Cumulative excess oxygen exposure" (CEOE)-previously defined as the mean hourly administered Fio2 above 0.21 when the corresponding hourly Spo2 was 95% or above-was previously shown to be associated with mortality. The objective of this study was to examine the relationship among Fio2, Spo2, and mortality in an independent cohort of mechanically ventilated children. DESIGN: Retrospective cross-sectional study. SETTING: Quaternary-care PICU. PATIENTS: All patients admitted to the PICU between 2012 and 2021 and mechanically ventilated via endotracheal tube for at least 24 hours. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Among 3,354 patients, 260 (8%) died. Higher CEOE quartile was associated with increased mortality (p = 0.001). The highest CEOE quartile had an 87% increased risk of mortality (95% CI, 7-236) compared with the first CEOE quartile. The hazard ratio for extended CEOE exposure, which included mechanical ventilation data from throughout the patients' mechanical ventilation time rather than only from the first 24 hours of mechanical ventilation, was 1.03 (95% CI, 1.02-1.03). CONCLUSIONS: Potentially excess oxygen exposure in patients whose oxygen saturation was at least 95% was associated with increased mortality.
Asunto(s)
Hospitalización , Respiración Artificial , Humanos , Niño , Estudios Transversales , Respiración Artificial/efectos adversos , Estudios Retrospectivos , OxígenoRESUMEN
Objective: Identification of children with sepsis-associated multiple organ dysfunction syndrome (MODS) at risk for poor outcomes remains a challenge. Data-driven phenotyping approaches that leverage electronic health record (EHR) data hold promise given the widespread availability of EHRs. We sought to externally validate the data-driven 'persistent hypoxemia, encephalopathy, and shock' (PHES) phenotype and determine its association with inflammatory and endothelial biomarkers, as well as biomarker-based pediatric risk-strata. Design: We trained and validated a random forest classifier using organ dysfunction subscores in the EHR dataset used to derive the PHES phenotype. We used the classifier to assign phenotype membership in a test set consisting of prospectively enrolled pediatric septic shock patients. We compared biomarker profiles of those with and without the PHES phenotype and determined the association with established biomarker-based mortality and MODS risk-strata. Setting: 25 pediatric intensive care units (PICU) across the U.S. Patients: EHR data from 15,246 critically ill patients sepsis-associated MODS and 1,270 pediatric septic shock patients in the test cohort of whom 615 had biomarker data. Interventions: None. Measurements and Main Results: The area under the receiver operator characteristic curve (AUROC) of the new classifier to predict PHES phenotype membership was 0.91(95%CI, 0.90-0.92) in the EHR validation set. In the test set, patients with the PHES phenotype were independently associated with both increased odds of complicated course (adjusted odds ratio [aOR] of 4.1, 95%CI: 3.2-5.4) and 28-day mortality (aOR of 4.8, 95%CI: 3.11-7.25) after controlling for age, severity of illness, and immuno-compromised status. Patients belonging to the PHES phenotype were characterized by greater degree of systemic inflammation and endothelial activation, and overlapped with high risk-strata based on PERSEVERE biomarkers predictive of death and persistent MODS. Conclusions: The PHES trajectory-based phenotype is reproducible, independently associated with poor clinical outcomes, and overlap with higher risk-strata based on validated biomarker approaches.
RESUMEN
Objective: To identify a cohort of COVID-19 cases, including when evidence of virus positivity was only mentioned in the clinical text, not in structured laboratory data in the electronic health record (EHR). Materials and Methods: Statistical classifiers were trained on feature representations derived from unstructured text in patient EHRs. We used a proxy dataset of patients with COVID-19 polymerase chain reaction (PCR) tests for training. We selected a model based on performance on our proxy dataset and applied it to instances without COVID-19 PCR tests. A physician reviewed a sample of these instances to validate the classifier. Results: On the test split of the proxy dataset, our best classifier obtained 0.56 F1, 0.6 precision, and 0.52 recall scores for SARS-CoV2 positive cases. In an expert validation, the classifier correctly identified 97.6% (81/84) as COVID-19 positive and 97.8% (91/93) as not SARS-CoV2 positive. The classifier labeled an additional 960 cases as not having SARS-CoV2 lab tests in hospital, and only 177 of those cases had the ICD-10 code for COVID-19. Discussion: Proxy dataset performance may be worse because these instances sometimes include discussion of pending lab tests. The most predictive features are meaningful and interpretable. The type of external test that was performed is rarely mentioned. Conclusion: COVID-19 cases that had testing done outside of the hospital can be reliably detected from the text in EHRs. Training on a proxy dataset was a suitable method for developing a highly performant classifier without labor-intensive labeling efforts.
RESUMEN
OBJECTIVES: Untangling the heterogeneity of sepsis in children and identifying clinically relevant phenotypes could lead to the development of targeted therapies. Our aim was to analyze the organ dysfunction trajectories of children with sepsis-associated multiple organ dysfunction syndrome (MODS) to identify reproducible and clinically relevant sepsis phenotypes and determine if they are associated with heterogeneity of treatment effect (HTE) to common therapies. DESIGN: Multicenter observational cohort study. SETTING: Thirteen PICUs in the United States. PATIENTS: Patients admitted with suspected infections to the PICU between 2012 and 2018. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We used subgraph-augmented nonnegative matrix factorization to identify candidate trajectory-based phenotypes based on the type, severity, and progression of organ dysfunction in the first 72 hours. We analyzed the candidate phenotypes to determine reproducibility as well as prognostic, therapeutic, and biological relevance. Overall, 38,732 children had suspected infection, of which 15,246 (39.4%) had sepsis-associated MODS with an in-hospital mortality of 10.1%. We identified an organ dysfunction trajectory-based phenotype (which we termed persistent hypoxemia, encephalopathy, and shock) that was highly reproducible, had features of systemic inflammation and coagulopathy, and was independently associated with higher mortality. In a propensity score-matched analysis, patients with persistent hypoxemia, encephalopathy, and shock phenotype appeared to have HTE and benefit from adjuvant therapy with hydrocortisone and albumin. When compared with other high-risk clinical syndromes, the persistent hypoxemia, encephalopathy, and shock phenotype only overlapped with 50%-60% of patients with septic shock, moderate-to-severe pediatric acute respiratory distress syndrome, or those in the top tier of organ dysfunction burden, suggesting that it represents a nonsynonymous clinical phenotype of sepsis-associated MODS. CONCLUSIONS: We derived and validated the persistent hypoxemia, encephalopathy, and shock phenotype, which is highly reproducible, clinically relevant, and associated with HTE to common adjuvant therapies in children with sepsis.
Asunto(s)
Encefalopatías , Sepsis , Choque Séptico , Niño , Humanos , Insuficiencia Multiorgánica/etiología , Relevancia Clínica , Reproducibilidad de los Resultados , Fenotipo , Encefalopatías/complicaciones , Hipoxia/etiologíaRESUMEN
Bedside measurement of heart rate (HR) change (HRC) may provide an objective physiologic marker for when brain death (BD) may have occurred, and BD testing is indicated in children. OBJECTIVES: To determine whether HRC, calculated using numeric HR measurements sampled every 5 seconds, can identify patients with BD among patients with catastrophic brain injury (CBI). DESIGN SETTING AND PARTICIPANTS: Single-center, retrospective study (2008-2020) of critically ill children with acute CBI. Patients with CBI had a neurocritical care consultation, were admitted to an ICU, had acute neurologic injury on presentation or during hospitalization based on clinical and/or imaging findings, and died or survived with Glasgow Coma Scale (GCS) less than 13 at hospital discharge. Patients meeting BD criteria (BD group) were compared with those with cardiopulmonary death (CD group) or those who survived to discharge. MAIN OUTCOMES AND MEASURES: HRC was calculated as the interquartile range of HR divided by median HR using 5-minute windows with 50% overlap for up to 5 days before death or end of recording. HRC was compared among the BD, CD, and survivor groups. RESULTS: Of 96 patients with CBI (69% male, median age 4 years), 28 died (8 BD, 20 CD) and 20 survived (median GCS 9 at discharge). Within 24 hours before death, HRC was lower in BD compared with CD patients or survivors (0.01 vs 0.03 vs 0.04, p = 0.001). In BD patients, HRC decreased at least 1 day before death. HRC discriminated BD from CD patients and survivors with 90% sensitivity, 70% specificity, 44% positive predictive value, 96% negative predictive value (area under the receiver operating characteristic curve 0.88, 95% CI, 0.80-0.93). CONCLUSIONS AND RELEVANCE: HRC is a novel digital biomarker that, with further validation, may be useful as a classifier for BD in the overall course of patients with CBI.
RESUMEN
OBJECTIVES: To examine whether escalating antimicrobial treatment in pediatric oncology and hematopoietic cell transplantation (HSCT) patients admitted to the PICU is supported by culture data or affects patient outcomes. DESIGN: Retrospective cross-sectional study. SETTING: Quaternary care PICU. PATIENTS: Patients younger than 18 years old who were admitted to the PICU at Boston Children's Hospital from 2012 to 2017 with a diagnosis of cancer or who had received HSCT and who had suspected sepsis at the time of PICU admission. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Of 791 PICU admissions for 544 patients that met inclusion criteria, 71 (9%) had escalation of antimicrobial therapy. Median Pediatric Logistic Organ Dysfunction (PELOD) score was higher in the escalation group (4 vs 3; p = 0.01). There were 14 admissions (20%) with a positive culture in the escalation group and 110 (15%) in the no escalation group ( p = 0.31). In the escalation group, there were only 2 (3%) cultures with organisms resistant to the initial antimicrobial regimen, compared with 28 (4%) cultures with resistant organisms in the no escalation group ( p = 1). Mortality in the escalation group was higher (17%) compared with the nonescalation group (5%; p < 0.001). The escalation group had more acute kidney injury (AKI) (25%) during treatment compared with the no escalation group (15%; p = 0.04), although this difference was not statistically significant when controlling for age, neutropenia, and PELOD-2 score (odds ratio, 1.75; 95% CI, 0.95-3.08; p = 0.06). CONCLUSIONS: Few patients who had escalation of antimicrobials proved on culture data to have an organism resistant to the initial antimicrobials, and more patients developed AKI during escalated treatment. While the escalation group likely represents a sicker population, whether some of these patients would be safer without escalation of antimicrobial therapy warrants further study.
Asunto(s)
Lesión Renal Aguda , Antiinfecciosos , Trasplante de Células Madre Hematopoyéticas , Neoplasias , Niño , Humanos , Lactante , Adolescente , Estudios Retrospectivos , Estudios Transversales , Neoplasias/tratamiento farmacológico , Antiinfecciosos/uso terapéutico , Trasplante de Células Madre Hematopoyéticas/efectos adversos , Unidades de Cuidado Intensivo PediátricoRESUMEN
Objective: To identify a cohort of COVID-19 cases, including when evidence of virus positivity was only mentioned in the clinical text, not in structured laboratory data in the electronic health record (EHR). Materials and Methods: Statistical classifiers were trained on feature representations derived from unstructured text in patient electronic health records (EHRs). We used a proxy dataset of patients with COVID-19 polymerase chain reaction (PCR) tests for training. We selected a model based on performance on our proxy dataset and applied it to instances without COVID-19 PCR tests. A physician reviewed a sample of these instances to validate the classifier. Results: On the test split of the proxy dataset, our best classifier obtained 0.56 F1, 0.6 precision, and 0.52 recall scores for SARS-CoV2 positive cases. In an expert validation, the classifier correctly identified 90.8% (79/87) as COVID-19 positive and 97.8% (91/93) as not SARS-CoV2 positive. The classifier identified an additional 960 positive cases that did not have SARS-CoV2 lab tests in hospital, and only 177 of those cases had the ICD-10 code for COVID-19. Discussion: Proxy dataset performance may be worse because these instances sometimes include discussion of pending lab tests. The most predictive features are meaningful and interpretable. The type of external test that was performed is rarely mentioned. Conclusion: COVID-19 cases that had testing done outside of the hospital can be reliably detected from the text in EHRs. Training on a proxy dataset was a suitable method for developing a highly performant classifier without labor intensive labeling efforts.
RESUMEN
OBJECTIVE: Electronic Health Record (EHR) based phenotyping is a crucial yet challenging problem in the biomedical field. Though clinicians typically determine patient-level diagnoses via manual chart review, the sheer volume and heterogeneity of EHR data renders such tasks challenging, time-consuming, and prohibitively expensive, thus leading to a scarcity of clinical annotations in EHRs. Weakly supervised learning algorithms have been successfully applied to various EHR phenotyping problems, due to their ability to leverage information from large quantities of unlabeled samples to better inform predictions based on a far smaller number of patients. However, most weakly supervised methods are subject to the challenge to choose the right cutoff value to generate an optimal classifier. Furthermore, since they only utilize the most informative features (i.e., main ICD and NLP counts) they may fail for episodic phenotypes that cannot be consistently detected via ICD and NLP data. In this paper, we propose a label-efficient, weakly semi-supervised deep learning algorithm for EHR phenotyping (WSS-DL), which overcomes the limitations above. MATERIALS AND METHODS: WSS-DL classifies patient-level disease status through a series of learning stages: 1) generating silver standard labels, 2) deriving enhanced-silver-standard labels by fitting a weakly supervised deep learning model to data with silver standard labels as outcomes and high dimensional EHR features as input, and 3) obtaining the final prediction score and classifier by fitting a supervised learning model to data with a minimal number of gold standard labels as the outcome, and the enhanced-silver-standard labels and a minimal set of most informative EHR features as input. To assess the generalizability of WSS-DL across different phenotypes and medical institutions, we apply WSS-DL to classify a total of 17 diseases, including both acute and chronic conditions, using EHR data from three healthcare systems. Additionally, we determine the minimum quantity of training labels required by WSS-DL to outperform existing supervised and semi-supervised phenotyping methods. RESULTS: The proposed method, in combining the strengths of deep learning and weakly semi-supervised learning, successfully leverages the crucial phenotyping information contained in EHR features from unlabeled samples. Indeed, the deep learning model's ability to handle high-dimensional EHR features allows it to generate strong phenotype status predictions from silver standard labels. These predictions, in turn, provide highly effective features in the final logistic regression stage, leading to high phenotyping accuracy in notably small subsets of labeled data (e.g. n = 40 labeled samples). CONCLUSION: Our method's high performance in EHR datasets with very small numbers of labels indicates its potential value in aiding doctors to diagnose rare diseases as well as conditions susceptible to misdiagnosis.
Asunto(s)
Registros Electrónicos de Salud , Aprendizaje Automático Supervisado , Algoritmos , Modelos Logísticos , FenotipoRESUMEN
Sedation and analgesia (SA) management is essential practice in the pediatric intensive care unit (PICU). Over the past decade, there has been significant interest in optimal SA management strategy, due to reports of the adverse effects of SA medications and their relationship to ICU delirium. We reviewed 13 studies examining SA practices in the PICU over the past decade for the purposes of reporting the study design, outcomes of interest, SA protocols used, strategies for implementation, and the patient-centered outcomes. We highlighted the paucity of evidence-base for these practices and also described the existing gaps in the intersection of implementation science (IS) and SA protocols in the PICU. Future studies would benefit from a focus on effective implementation strategies to introduce and sustain evidence-based SA protocols, as well as novel quasi-experimental study designs that will help determine their impact on relevant clinical outcomes, such as the occurrence of ICU delirium. Adoption of the available evidence-based practices into routine care in the PICU remains challenging. Using SA practice as an example, we illustrated the need for a structured approach to the implementation science in pediatric critical care. Key components of the successful adoption of evidence-based best practice include the assessment of the local context, both resources and barriers, followed by a context-specific strategy for implementation and a focus on sustainability and integration of the practice into the permanent workflow.
RESUMEN
OBJECTIVE: Accurately assigning phenotype information to individual patients via computational phenotyping using Electronic Health Records (EHRs) has been seen as the first step towards enabling EHRs for precision medicine research. Chart review labels annotated by clinical experts, also known as "gold standard" labels, are essential for the development and validation of computational phenotyping algorithms. However, given the complexity of EHR systems, the process of chart review is both labor intensive and time consuming. We propose a fully automated algorithm, referred to as pGUESS, to rank EHR notes according to their relevance to a given phenotype. By identifying the most relevant notes, pGUESS can greatly improve the efficiency and accuracy of chart reviews. METHOD: pGUESS uses prior guided semantic similarity to measure the informativeness of a clinical note to a given phenotype. We first select candidate clinical concepts from a pool of comprehensive medical concepts using public knowledge sources and then derive the semantic embedding vector (SEV) for a reference article (SEVref) and each note (SEVnote). The algorithm scores the relevance of a note as the cosine similarity between SEVnote and SEVref. RESULTS: The algorithm was validated against four sets of 200 notes that were manually annotated by clinical experts to assess their informativeness to one of three disease phenotypes. pGUESS algorithm substantially outperforms existing unsupervised approaches for classifying the relevance status with respect to both accuracy and scalability across phenotypes. Averaging over the three phenotypes, the rank correlation between the algorithm ranking and gold standard label was 0.64 for pGUESS, but only 0.47 and 0.35 for the next two best performing algorithms. pGUESS is also much more computationally scalable compared to existing algorithms. CONCLUSION: pGUESS algorithm can substantially reduce the burden of chart review and holds potential in improving the efficiency and accuracy of human annotation.