TABLE 3.
Mechanisms of Implementation for the Use of Data Science in Critical Care
| Mechanisms | Relevant Examples in Critical Care |
|---|---|
| Big Data Analytics Resources (per Groves et al [52], Waller and Fawcett [53], Chen and Zhang [54] as cited in Galetsi et al [49, 50]) | |
| Data types | Clinical data in the electronic medical record for hospitalization or pre-/post-ICU care; clinical device data (e.g., waveforms, IV pumps, ventilators, dialysis machines); questionnaires (e.g., family satisfaction, patient-reported outcomes); diagnostic and procedure codes; drug utilization and wastage data |
| Analytical resources | Society for Simulation in Healthcare; PhysioNet (e.g., Medical Information Mart for Intensive Care IV, eICU); National Clinical Cohort Collaborative |
| Phillips eICU Research Collaborative Database; Amsterdam University Medical Centers Database | |
| Big Data Analytics Capabilities, per Groves et al (52) as cited in Galetsi et al (49, 50) | |
| Monitoring, prediction/simulation, data mining, evaluation, and reporting | Monitoring of high-fidelity waveforms and processed data from bedside monitors |
| Prediction models: Disease-based (e.g., sepsis) (55); intervention-based (e.g., hemodynamic monitoring); prediction of changes in stability and acuity (56, 57) | |
| Mining of ICU data lakes (i.e., centralized repositories for storage of large amounts of raw data) | |
| Integrated evaluation and reporting of complex data streams and thresholds | |
| Created values | |
| Better diagnosis for provision of more personalized health care | Rapid diagnosis, predictive and prognostic enrichment (e.g., Covid-19) (58, 59) |
| Supporting/replacing professionals’ decision-making with automated algorithms | Machine learning-enabled clinical decision support (60) |
| New business models, products, and services | Prediction models |
| Enabling experimentation, expose variability, and improve performance | Synthetic datasets modeling and experimentation where ethical considerations would not allow randomized trials |
| Healthcare information sharing and coordination | Mapping of clinical data to common data models (25) |
| Creating data transparency | Data cleaning and post-processing systems |
| Identifying patient care risk | ICU scoring systems (e.g., SOFA, pediatric SOFA) (61–63) |
| Early warning systems (64) | |
| Prediction of ICU readmissions (65) | |
| Offering customized actions by segmenting populations | Population enrichment for interventional trials |
| Reducing expenditure while maintaining quality | Improved efficiency in treatment of sepsis |
| Protecting privacy | Cyber security and privacy systems in the ICU |
eICU = electronic ICU, SOFA = Sequential Organ Failure Assessment.