eCARTv5 Clinical Deterioration Suite (eCART)

The eCARTv5 Clinical Deterioration Suite is an adjunctive predictive indicator that employs a Gradient Boosted Machine learning algorithm to analyze real-time Electronic Health Record data, including vital signs and lab results, for early identification of adult ward patients at risk of death or ICU transfer within 24 hours. Its purpose is to provide automated risk stratification and trend analysis to support clinical teams in making informed decisions, thereby aiming to improve patient monitoring and allow for timely intervention.

The eCARTv5 Clinical Deterioration Suite ("eCART") is a cloud-based software device designed to aid clinical teams in identifying hospitalized adult ward patients (18 years or older) at risk of clinical deterioration, specifically death or ICU transfer. Integrated with the hospital's Electronic Health Record (EHR), eCART utilizes a Gradient Boosted Machine (GBM) learning algorithm that processes validated EHR data—including routine vital signs, laboratory results, and patient demographics—to generate a risk score scaled from 0 to 100. The system presents this score, along with trend analysis and the observed odds of deterioration in the next 24 hours, to trained medical professionals as an adjunctive clinical decision support tool, intended to supplement, not replace, standard care or clinical judgment; scoring is initiated by the documentation of any vital sign, using only confirmed data. Performance of eCART was assessed through extensive retrospective (2009-2023) and prospective (2023-2024) validation studies across three distinct health systems, involving over 1.7 million and 200,000 patient encounters, respectively. For predicting clinical deterioration (death or ICU transfer within 24 hours), the Area Under the Receiver Operating Characteristic Curve (AUROC) was 0.835 in the retrospective cohort and 0.828 in the prospective cohort. For mortality prediction within 24 hours, AUROCs were 0.923 (retrospective) and 0.913 (prospective). At a moderate-risk threshold (eCART score ≥93), retrospective sensitivity for deterioration was 51.8% with 93.1% specificity, and for mortality, sensitivity was 75.8% with 92.6% specificity; prospective results were comparable. Subgroup analyses for various comorbidities (congestive heart failure, COVID-19, chronic pulmonary disease, sepsis) and racial groups showed comparable performance, generally meeting predefined thresholds for AUROC and sensitivity.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the eCART algorithm's performance vary when predicting deterioration beyond the 24-hour window, for instance, at 48 or 72 hours?; What specific vital signs, laboratory data, and demographic features carry the most weight in the Gradient Boosted Machine model's risk score calculation for eCART?; Given the defined thresholds for moderate (eCART score ≥93) and high (eCART score ≥97) risk, what is the typical lead time provided by eCART before an actual clinical deterioration event (death or ICU transfer) occurs?; How does the eCART system's performance compare in patient populations with multiple co-existing comorbidities versus those with a single listed comorbidity from the subgroup analyses?; What is the impact on clinician workflow and actual patient outcomes (e.g., reduction in ICU transfers or mortality rates) in real-world settings where eCART is implemented alongside institution-specific standardized clinical workflows for elevated-risk patients?