Abstract
Objectives
To develop a multimodal pediatric critical care datamart supporting predictive modeling and decision support tool development, integrating high-resolution physiologic and clinical data and future clinical deployment.
Materials and Methods
We developed a continuously expanding datamart integrating electronic health record data, high-resolution telemetry, and extracorporeal membrane oxygenation (ECMO)-domain datasets. The platform links static and longitudinal time-series variables with expert-curated neurologic outcomes for both ECMO and non-ECMO patients, enabling trajectory-based analyses.
Results
The datamart currently includes 25 762 pediatric patients, of whom 395 received ECMO support. The datamart captures granular longitudinal physiologic, laboratory, medication, and telemetry data suitable for dynamic predictive modeling.
Discussion
Existing ECMO prognostication tools rely on static variables and lack appropriate control cohorts. This datamart enables trajectory-based multimodal modeling, that reflects evolving physiology and neurologic outcomes.
Conclusion
This platform provides a scalable foundation for predictive modeling across pediatric critical care, beyond ECMO, to support precision decision-making and outcomes research.
Keywords: extracorporeal membrane oxygenation, pediatric critical care, machine learning, clinical decision support, neurologic outcomes
Background and significance
Extracorporeal membrane oxygenation (ECMO) is a lifesaving therapy for children with severe, refractory, cardiac or respiratory failure, with over 20 000 children treated internationally in the past 5 years, including over 13 500 in North America alone.1 Despite technological advancements and expanding use, pediatric ECMO carries substantial risks with mortality rates up to 61% and neurologic complications, such as seizures, in up to 40% of patients.2–4 The complexity and heterogeneity of intensive care unit (ICU) data, coupled with the high-risk nature of ECMO, make its initiation one of the most consequential decisions in clinical care. However, decisions rely on multidisciplinary consensus and remain largely unsupported by objective decision support tools.
Existing ECMO decision aids are limited to ECMO prognostication scores with significant shortcomings.5 These scores were derived from sparse, registry-style data captured at or near ECMO initiation, failing to incorporate the dynamic, time-varying nature of ICU trajectories. They overlook evolving physiology and organ dysfunction critical for outcome predictions and were developed exclusively from ECMO patients, without non-ECMO control comparators. Additionally, while long-term neurologic and functional outcomes of ECMO survivors are increasingly recognized as critical,6–8 existing tools primarily focus on short-term mortality. Consequently, clinicians lack reliable, data-driven guidance for ECMO candidacy and timing in relation to meaningful neurologic outcomes.
These limitations have driven efforts to leverage high-resolution ICU data and machine learning to improve ECMO decision-making. Prior studies have demonstrated the feasibility of using granular physiologic time-series data to predict ECMO utilization and outcomes. Early institutional models employed Gradient Boosting Tree (GBT) techniques to identify patients at risk of ECMO hours in advance of initiation.9 Subsequent models incorporated Long Short Term Memory (LSTM) Recurrent Neural Network (RNN) to capture evolving physiologic trajectories utilizing large national datasets.10 Additionally, a Treatment Variational AutoEncoder (TVAE) model validated with multinational data showed promising performance at identifying patients who benefited from ECMO based on a survival outcome.11 These studies highlighted the promise of machine learning, but were limited by sparsely available data, heterogeneity across sites and a focus on a single underlying pathology for ECMO utilization. To address these gaps, we developed a comprehensive Pediatric Critical Care Datamart (PCCD) that integrates high-granularity, time-resolved ICU data with expert-curated neurologic outcomes, to enable scalable, multimodal predictive modeling reflecting the complexity of critical illness and ECMO care.
Objectives
The objectives of this manuscript are to describe the design principles, system architecture, and data pipelines of the multimodal Pediatric Critical Care Datamart (PCCD); to outline the integrated data sources, cohort structure and longitudinal data density across ECMO and non-ECMO cohorts; and to demonstrate the datamart’s capacity to support trajectory-based predictive modeling through its relational structure, temporal resolution, and outcome linkage. Specifically, we present the architectural framework used to harmonize electronic health records (EHR) data, high-resolution bedside telemetry and ECMO-domain-specific datasets, into a unified, relational, and queryable platform. We detail the construction of retrospective and prospective cohorts spanning general pediatric critical care and ECMO populations, and describe the key design principles, temporal alignment, cross-domain linkage, and scalability for future model development. Together, these objectives establish the datamart as enabling infrastructure for future machine learning analyses and decision support research.
Materials and methods
The PCCD sources data from June 2018 to present, at Washington University at St Louis/St Louis Children’s Hospital, a single-center quaternary pediatric ICU with a high-volume ECMO program. Development was led by a multidisciplinary team including domain experts within pediatric critical care medicine, pediatric neurocritical care, ECMO clinicians; clinical informaticians; data engineers; and collaborators in biostatistics and machine learning. The datamart was designed using standard clinical database principles, including a relational schema with normalized domain-specific tables, consistent primary and secondary keys, and time-aligned tables, to support scalable longitudinal analyses.12,13 The PCCD integrates data from 3 core sources (1) all extractable EHR fields, (2) high-resolution bedside telemetry data, and (3) ECMO-specific data not otherwise captured in discrete EHR fields, including expert-curated neurologic outcomes (Figure 1, Digital Content S1). Data are standardized using harmonized patient and encounter identifiers and timestamp conventions to enable longitudinal integration. Automated extract-transform-load (ETL) pipelines interface with the institutional EHR datalake and include validation steps for data completeness, timestamp synchronization, and range plausibility. Missingness and outlier reports are generated and reviewed to ensure data integrity. The datamart supports structured query language (SQL), Python analytics, and machine learning algorithms, with visualization tools (Power BI and Tableau) to streamline data exploration. The datamart is updated monthly, adding new patients and refreshing historical data. This project was approved by the Institutional Review Board (IRB 201812081) with a waiver of consent for retrospective data use. The datamart is housed within a HIPAA-compliant institutional Microsoft Azure Databricks cloud platform with role-based access controls. Analyses presented here are descriptive and intended to demonstrate datamart capabilities rather than draw clinical inference.
Figure 1.
The multi-source pediatric ICU datamart schematic. Figure illustrating the datamart structure, integrating 3 data pipelines: extractable EHR data, high-resolution telemetry, and ECMO-specific datasets. Representative tables are shown for each data source, including medications, imaging, laboratory results, disposition, and flowsheets from the EHR; BedMaster and Etiometry telemetry tables containing physiologic and ventilator parameters; and ECMO-specific tables encompassing ELSO registry tables as run details, survival, and complications and neuro-outcomes data including EEG, neuroimaging, and functional outcomes. The datamart is organized into retrospective and prospective cohorts, supporting longitudinal, high-resolution analyses of pediatric ICU and ECMO populations. ECMO, extracorporeal membrane oxygenation; EHR, electronic health record; ELSO, Extracorporeal Life Support Organization; ICU, intensive care unit; EEG, electroencephalography.
Results
To date, the datamart includes 25 762 patients, of whom 395 received ECMO support, organized into 2 cohorts: a retrospective cohort (June 2018-October 2023) and a prospective cohort updated monthly. All summaries reported here reflect the retrospective cohort unless otherwise specified. The retrospective cohort includes 10 348 patients across 11 592 distinct ICU encounters, including 222 patients (2.1%) supported on ECMO. The ECMO encounters accounted for approximately 3% of all ICU encounters, with a significantly longer ICU length of stay (9.5 [4.5-18.5] vs 4.5 [2.5-7.5] days, P < .001) and higher hospital mortality (41.9% vs 4.4%, P < .001) in the ECMO compared to the non-ECMO cohorts (Table 1).
Table 1.
Demographics of the ECMO and non-ECMO patients in the retrospective cohort.a
| Median [IQR] | Total Cohort | Non-ECMO | ECMO |
|---|---|---|---|
| 10 348 | 10 126 (98%) | 222 (2%) | |
| Age at first encounter, years | 4.45 [0.48, 12.38] | 4.45 [1.47, 12.38]b | 1.47 [0.48, 12.38]b |
| Height, inches | 43.9 [34.4-60.5] | [43.9 [34.4-60.5]b | 38.0 [28.5-56.9]b |
| Weight, kg | 21.43 [13.14-51.01] | 21.43 [13.14-51.01]b | 15.51 [9.59-46.28]b |
| Female gender, n (%) | 4525 (43.73) | 4422 (43.67) | 103 (46.4) |
| Race, n (%) | |||
| American Indian or Alaska Native | 44 (0.43) | 43 (0.42) | 1 (0.45) |
| Asian | 159 (1.54) | 154 (1.52) | 5 (2.25) |
| Black or African American | 3085 (29.81) | 3033 (29.95)d | 52 (23.42)d |
| Other Pacific Islander | 46 (0.44) | 45 (0.44) | 1 (0.45) |
| White | 6740 (65.13) | 6581 (64.99)d | 159 (71.62)d |
| Other | 16 (0.15) | 16 (0.16) | 0 |
| Unknown | 30 (0.29) | 28 (0.28) | 2 (0.9) |
| Ethnicity, N (%) | |||
| Hispanic | 423 (4.09) | 412 (4.07) | 11 (4.95) |
| Non-Hispanic | 9751 (94.23) | 9543 (94.24) | 208 (93.69) |
| ICU LOS, days | 4.6 [2.5, 8] | 4.5 [2.5, 7.5]b | 9.5 [4.5, 18.5]b |
| Mortality, n (%) | 534 (5.16%) | 441 (4.36%)c | 93 (41.89%)c |
ECMO, extracorporeal membrane oxygenation; ICU, intensive care unit; LOS, length of stay.
P < .001 Mann-Whitney test for continuous data.
P < .001 Fisher-exact test for categorical data.
P < .05 Fisher-exact test for categorical data.
Table 2 summarizes the PCCD data domains by source and attribute count. The EHR data pipeline captures all extractable discrete fields from the institutional transition to the current EHR system in mid-2018 and supports historical backfilling. These include admission, discharge, and transfer events; demographics; comorbidities; flowsheets; medication orders and administration; dialysis; laboratory; imaging; procedures; isolation status; ventilation; and disposition. All EHR data are linked longitudinally across encounters using harmonized patient and encounter identifiers and synchronized timestamps to enable reconstruction of complete ICU trajectories.
Table 2.
Datamart table structure.a
| EHR | Attributes | Telemetry | Attributes | ECMO-Domain | Attributes |
|---|---|---|---|---|---|
| ADT | 16 | Bedmaster | 15 | ELSO | |
| Comorbidities | 10 | Etiometry computed data | 37 | ECMO log | 36 |
| Demographics | 18 | Cannula | 23 | ||
| Dialysis | 8 | Cardiac diagnoses | 9 | ||
| Dialysis history | 9 | Etiometry labs | 13 | Cardiac procedures | 14 |
| ECG | 10 | ||||
| ECMO cohort report | 37 | Etiometry data | 624 | Complications | 11 |
| Encounters | 24 | Diagnoses | 12 | ||
| Flowsheets | 9 | – | – | Equipment | 19 |
| Home medications | 6 | – | – | Membrane lung | 15 |
| Imaging | 11 | – | – | Organisms | 15 |
| Imaging impressions | 7 | – | – | Patient run | 151 |
| Inclusion Encounter | 9 | – | – | Pre-ECMO support | 9 |
| Isolation | 9 | – | – | ||
| Lab results | 27 | – | – | Procedures | 12 |
| Lab results susceptibility | 16 | – | – | ECMO pump | 15 |
| – | – | Run detail | 14 | ||
| LDA | 28 | – | – | Outcomes | |
| Medication administration | 17 | – | – | Neuro-outcomes | 163 |
| – | – | – | – | ||
| Medication orders | 23 | – | – | – | – |
| Patient inclusion reason | 6 | – | – | – | – |
| – | – | – | – | ||
| PFT | 17 | – | – | – | – |
| MRN merge history | 2 | – | – | – | – |
| Procedure orders | 6 | – | – | – | – |
| Registry cohort report | 11 | – | – | – | – |
| Vaccinations | 9 | – | – | – | – |
| Vaping | 9 | – | – | – | – |
| Ventilation | 10 | – | – | – | – |
EHR, electronic health record; ECMO, extracorporeal membrane oxygenation; ADT, admission/discharge/transfer; ELSO, extracorporeal life support organization; ECG, electrocardiograph; LDA, lines, drains and airway; MRN, medical record number; PFT, pulmonary function test. Each domain represents distinct data pipelines integrated into the datamart. Attribute counts reflect variables per table. Totals: EHR = 27 tables; Telemetry = 4 tables; ECMO-Domain = 17 tables.
High-resolution telemetry data are sourced from BedMasterEx and Etiometry, which are deployed across all ICUs. Physiologic and ventilator parameters are initially captured at 5-s resolution then summarized as 1-minute medians across the ICU course. These data streams provide near-continuous representation of variables such as heart rate, blood pressure, respiratory rate, oxygen saturation, and ventilator settings in addition to derived physiologic indices, complementing the discrete EHR documentation to enable high-resolution temporal analyses.
The ECMO-specific data pipeline integrates institutional ECMO logs with data submitted to the Extracorporeal Life Support (ELSO) registry, including detailed ECMO run characteristics such as cannulation configuration, circuit components, membrane lung use, and complications. This domain is further linked to expert-curated neurologic outcomes. Neurologic data include the interpretation of continuous video electroencephalography (EEG) monitoring for at least the first 48 hours of ECMO support, serial neuroimaging (daily head ultrasound for infants, clinically indicated head computed tomography, CT, and post-ECMO brain magnetic resonance imaging, MRI, prior to discharge), and functional outcomes assessed using Functional Status Scale (FSS),7,8,14 at pre-illness baseline and hospital discharge. For non-ECMO patients, neurologic outcomes are limited to survival, neuroimaging and available EHR-derived indicators.
Overall, the PCCD integrates 27 EHR tables (346 attributes), 4 telemetry data tables with 674 attributes, and 17 ECMO-specific tables (355 attributes in the ELSO tables and 163 in the neuro-outcomes domain). Collectively, these encompass over 625 million discrete flowsheet entries, 9.5 million medication records, 13.2 million laboratory results, and more than 2.5 billion telemetry observations. These data provide synchronized, longitudinal, high-resolution representations of ICU care across ECMO and non-ECMO populations.
To validate that the PCCD meets the temporal and multimodal design goals, we examined data density and availability across ECMO and non-ECMO ICU encounters (Figures 2 and 3). Figure 2 shows that across ICU days 0-20, ECMO encounters exhibited higher median (Figure 2A) and higher end (95th percentile, Figure 2B) data density across physiologic monitoring, laboratory results, medication administration, imaging, and clinical documentation compared to non-ECMO encounters. These patterns reflect the expected increased monitoring and documentation intensity characteristic of ECMO care. Figure 3 evaluates the within-day temporal granularity and data availability on ICU day 1. Figure 3A shows higher median within-day data availability across modalities in ECMO encounters, especially for physiologic telemetry, flowsheet documentation, and medication administration. Figure 3B quantifies the differences in data availability between cohorts at a fine temporal scale, showing consistently greater 5-minute bin-level data availability for ECMO encounters across most modalities. These figures demonstrate consistent, high-frequency data capture within ICU days, supporting granular temporal analyses and multimodal predictive modeling.
Figure 2.
Data density by modality and ICU day in ECMO and non-ECMO encounters. (A) Heatmaps depicting median data density, defined as the median number of recorded observations per ICU encounter per day for each modality, stratified by ECMO status. Rows represent clinical and physiologic data modalities, columns represent ICU day relative to ICU admission (day 0 to 20), and color intensity corresponds to the median number of observations per encounter-day on a logarithmic scale. Non-ECMO encounters are shown in the top panel and ECMO encounters in the bottom panel. Across all modalities, ECMO encounters demonstrated consistently higher median data density than non-ECMO encounters throughout the ICU stay, reflecting more frequent physiologic monitoring, laboratory testing, imaging, medication administration, and clinical documentation. These patterns captured in the datamart reflect expected differences in care intensity and patient trajectory between cohorts. (B) Heatmaps depicting high-end data density, defined as the 95th percentile of recorded observations per ICU encounter per day for each modality, stratified by ECMO status. Rows represent clinical and physiologic data modalities, columns represent ICU day relative to ICU admission (day 0 to 20), and color intensity corresponds to the 95th percentile of observations per encounter-day on a logarithmic scale. Non-ECMO encounters are shown in the top panel and ECMO encounters in the bottom panel. At the higher end of ICU data generation, ECMO encounters exhibited greater data density across most modalities, highlighting periods of increased monitoring and documentation that further distinguish ECMO care from standard ICU management. ICU, intensive care unit; ECMO, extracorporeal membrane oxygenation.
Figure 3.
Within-day temporal granularity and modality-level data availability in ECMO and non-ECMO encounters. (A) Heatmaps depicting within-day data availability by modality on ICU day 1, summarized as the median percentage of hourly time bins containing at least 1 observation, stratified by ECMO status. Rows represent clinical and physiologic data modalities, columns represent hour within ICU day 1, and color intensity corresponds to the percent of encounter-hours with data present. Non-ECMO encounters are shown in the top panel and ECMO encounters in the bottom panel. Across modalities, ECMO encounters demonstrated consistently greater within-day data availability, especially for physiologic telemetry, flowsheet documentation, and medication administration, reflecting sustained high-frequency monitoring and management throughout the ICU day. (B) Within-day differences in modality-level data availability on ICU day 1, depicted as the percent difference in mean 5-min bin-level encounter availability between ECMO and non-ECMO encounters. Panels represent individual data modalities, and values indicate the difference in the percentage of encounters with at least 1 observation within each 5-min interval across ICU day 1. Positive values indicate greater availability in ECMO encounters. ECMO care was associated with consistently higher within-day data availability across most modalities, especially in Etiometry telemetry, laboratory results, and medication administration, highlighting increased monitoring intensity and complexity characteristic of ECMO management. In both panels, a modality was considered present if at least 1 observation occurred within the corresponding time bin. ICU day 1 was defined as the first full 24-h period following ICU admission. ICU, intensive care unit; ECMO, extracorporeal membrane oxygenation.
Discussion
This work describes the design, implementation, and data availability of a multimodal pediatric critical care datamart integrating structured EHR, high-resolution telemetry, and ECMO-specific data within a single architecture. To our knowledge, few pediatric ICU databases capture this level of granularity, temporal resolution, multimodal integration, and longitudinal linkage across the full ICU population while incorporating domain-specific data such as ECMO.
Our results demonstrate that the Pediatric Critical Care Datamart meets foundational requirements to support future predictive modeling and advanced analytics. The system enables longitudinal patient-level linkage across encounters, synchronized timestamps across data sources, and integration of high-frequency physiologic telemetry with structured clinical data. Figures 2 and 3 provide evidence that these design goals are achieved, showing dense, consistent, time-resolved data capture across modalities. ECMO encounters demonstrated higher data density and within-day availability, reflecting expected care complexity, and confirming faithful representation of real-world ICU workflows. Inclusion of the full ICU population enables identification of non-ECMO comparator cohorts, addressing a key limitation of prior ECMO registry-based analyses. While ECMO was selected as a proof-of-concept subdomain due to its clinical importance and data complexity, the PCCD was intentionally designed as a general ICU resource. The relational architecture, standardized identifiers, and time-resolved pipelines are extendible to other ICU therapies such as renal replacement therapy or plasma exchange.
Several limitations warrant consideration. High-resolution data are susceptible to outliers and implausible values. While current ETL pipelines include automated range checks and missingness reporting, future modeling efforts will require additional physiologic plausibility filtering and age-specific normalization. In addition, the datamart is currently derived from a single center and expert-curated neurologic outcomes are available primarily for ECMO patients. Planned multicenter expansion and alignment with common data models will be essential for broader generalizability.
Looking forward, the datamart is designed to support future analytic studies that leverage longitudinal, time-resolved ICU data, including machine learning approaches suitable for dynamic risk modeling such as GBT and RNN. While such models are not presented here, the system architecture, temporal alignment, and cohort structure described in this work establish the necessary prerequisites for their development and evaluation. In parallel, alignment with emerging data harmonization efforts, as the Extracorporeal Life Support (ECLS) Common Data Model (CDM),15 positions the datamart for future multicenter expansion and external. These features support the pathway toward scalable predictive models in ECMO and broader pediatric critical care, contingent on subsequent analytic and funding efforts.
Conclusion
We present a comprehensive, multimodal Pediatric Critical Care Datamart integrating EHR, high-resolution telemetry, and ECMO-specific data into a unified analytic platform. This system demonstrates the feasibility of capturing synchronized, time-resolved ICU data across ECMO and non-ECMO populations and provides a scalable foundation for future predictive modeling and decision support development in pediatric critical care, with ECMO as an initial proof-of-concept application.
Supplementary Material
Contributor Information
Miranda Edmunds, Division of Pediatric Critical Care, Department of Pediatrics, Washington University in St. Louis, St. Louis, MO 63110, United States.
Aditi Gupta, Institute for Informatics, Data Science and Biostatistics, Washington University in St. Louis, St. Louis, MO 63110, United States.
Inez Oh, Institute for Informatics, Data Science and Biostatistics, Washington University in St. Louis, St. Louis, MO 63110, United States.
Levi Kaster, Institute for Informatics, Data Science and Biostatistics, Washington University in St. Louis, St. Louis, MO 63110, United States.
Jonathan Sagel, Division of Pediatric Critical Care, Department of Pediatrics, Washington University in St. Louis, St. Louis, MO 63110, United States.
Andrew Michelson, Institute for Informatics, Data Science and Biostatistics, Washington University in St. Louis, St. Louis, MO 63110, United States; Department of Medicine, Washington University in St. Louis, St. Louis, MO 63110, United States.
Yuyao Zhuge, Department of Energy, Environmental & Chemical Engineering, Washington University in St. Louis, St. Louis, MO 63110, United States.
Philip Payne, Institute for Informatics, Data Science and Biostatistics, Washington University in St. Louis, St. Louis, MO 63110, United States; Department of Medicine, Washington University in St. Louis, St. Louis, MO 63110, United States.
Ahmed Said, Division of Pediatric Critical Care, Department of Pediatrics, Washington University in St. Louis, St. Louis, MO 63110, United States; Institute for Informatics, Data Science and Biostatistics, Washington University in St. Louis, St. Louis, MO 63110, United States.
Author contributions
Miranda Edmunds (Visualization, Writing—original draft, Writing—review & editing), Aditi Gupta (Methodology, Writing—review & editing), Inez Oh (Methodology, Writing—review & editing), Levi Kaster (Methodology, Visualization, Writing—review & editing), Jonathan Sagel (Methodology), Andrew Michelson (Methodology), Yuyao Zhuge (Methodology), Philip Payne (Methodology, Validation, Writing—review & editing), and Ahmed Said (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing—original draft, Writing—review & editing)
Supplementary material
Supplementary material is available at JAMIA Open online.
Funding
This project was funded, in whole or in part, by the Foundation for Barnes Jewish Hospital and their generous donors. This publication was supported by the National Institutes of Health (NIH)/National Center for Advancing Translational Sciences (NCATS), grant UL1TR002345. A.S. has received research support from the Children’s Discovery Institute Faculty Development Award at Washington University in St Louis. For the remaining authors, none were declared.
Conflicts of interest
None declared.
Data availability
The datamart contains protected health information and therefore cannot be publicly shared. However, the datamart’s data model, relational schema, entity-relationship diagrams, cohort definitions, and analytic codes used to generate the figures in this manuscript are publicly available via GitHub (https://github.com/Saidahmed75/Pediatric-Critical-Care-Datamart.git). This repository includes documentation describing the datamart structure and design principles to support transparency and reproducibility of the analytic framework. Access to patient-level data is available to investigators following Washington University in St Louis IRB approval and institutional data-use agreement.
References
- 1. Organization ELS. ECLS Registry Report & International Summary of Statistics. Secondary ECLS Registry Report & International Summary of Statistics. 2025. https://www.elso.org/registry/internationalsummaryandreports.aspx
- 2. Bauer Huang SL, Said AS, Smyser CD, Lin JC, Guilliams KP, Guerriero RM. Seizures are associated with brain injury in infants undergoing extracorporeal membrane oxygenation. J Child Neurol. 2021;36:230-236. 10.1177/0883073820966917 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Said AS, Guilliams KP, Bembea MM. Neurological monitoring and complications of pediatric extracorporeal membrane oxygenation support. Pediatr Neurol. 2020;108:31-39. 10.1016/j.pediatrneurol.2020.03.014 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Felling RJ, Kamerkar A, Friedman ML, et al. Neuromonitoring during ECMO support in children. Neurocrit Care. 2023;39:701-713. 10.1007/s12028-023-01675-8 [DOI] [PubMed] [Google Scholar]
- 5. Shah N, Said AS. Extracorporeal support prognostication-time to move the goal posts? Membranes (Basel). 2021;11:537. 10.3390/membranes11070537 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Turner AD, Streb MM, Ouyang A, et al. Long-term neurobehavioral and functional outcomes of pediatric extracorporeal membrane oxygenation survivors. ASAIO J. 2024;70:409-416. 10.1097/MAT.0000000000002135 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Hennigan C, Cheung E, Garey D, Zygmunt A, Morris M, Geneslaw A. 35: Understanding the ecmo, recovery, and outcome experience of pediatric respiratory ECMO survivors. Critic Care Med. 2022;50:18. 10.1097/01.ccm.0000806608.47241.0f [DOI] [Google Scholar]
- 8. Yuhang Y, Ni Y, Tiening Z, Lijie W, Wei X, Chunfeng L. Functional status of pediatric patients after extracorporeal membrane oxygenation: a five-year single-center study. Front Pediatr. 2022;10:917875. 10.3389/fped.2022.917875 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Xue B, Shah N, Yang H, et al. Multi-horizon predictive models for guiding extracorporeal resource allocation in critically ill COVID-19 patients. J Am Med Inform Assoc. 2023;30:656-667. 10.1093/jamia/ocac256 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Zhu D, Xue B, Shah N, Payne PRO, Lu C, Said AS. Multi-modal prediction of extracorporeal support-a resource intensive therapy, utilizing a large national database. JAMIA Open. 2025;8:ooae158. 10.1093/jamiaopen/ooae158 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Xue B, Said AS, Xu Z, et al. Assisting clinical decisions for scarcely available treatment via disentangled latent representation. In: Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, 2023:5360-5371. [Google Scholar]
- 12. Murphy SN, Weber G, Mendis M, et al. Serving the enterprise and beyond with informatics for integrating biology and the bedside (i2b2). J Am Med Inform Assoc. 2010;17:124-130. 10.1136/jamia.2009.000893 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Reps JM, Schuemie MJ, Suchard MA, Ryan PB, Rijnbeek PR. Design and implementation of a standardized framework to generate and evaluate patient-level prediction models using observational healthcare data. J Am Med Inform Assoc. 2018;25:969-975. 10.1093/jamia/ocy032 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Pollack MM, Holubkov R, Glass P, et al. ; Eunice Kennedy Shriver National Institute of Child Health and Human Development Collaborative Pediatric Critical Care Research Network. Functional status scale: new pediatric outcome measure. Pediatrics. 2009;124:e18-e28. 10.1542/peds.2008-1987 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Rieder C, Zhuk O, Said A, Alexander PMA, Hoechter DJ. The ECLS CDM Expansion Enables the OMOP CDM to Store Extracorporeal Life Support Data. OHDSI Europe Symposium. 2024. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datamart contains protected health information and therefore cannot be publicly shared. However, the datamart’s data model, relational schema, entity-relationship diagrams, cohort definitions, and analytic codes used to generate the figures in this manuscript are publicly available via GitHub (https://github.com/Saidahmed75/Pediatric-Critical-Care-Datamart.git). This repository includes documentation describing the datamart structure and design principles to support transparency and reproducibility of the analytic framework. Access to patient-level data is available to investigators following Washington University in St Louis IRB approval and institutional data-use agreement.