CLEWICU System (ClewICUServer and ClewICUnitor)

The CLEWICU System is an adjunctive clinical decision support tool that employs machine learning algorithms to analyze comprehensive ICU patient data, aiming to predict hemodynamic instability and identify low-risk patients. Its science relies on models trained and validated on retrospective clinical data to provide early, actionable insights for clinicians, thereby potentially improving patient management in critical care settings.

The CLEWICU System, comprising the ClewICUServer and ClewICUnitor, is a stand-alone analytical software product designed to provide clinicians with physiological insights into the likelihood of future hemodynamic instability in adult intensive care unit (ICU) patients, aged 18 years and over. It also aims to identify patients at low risk for clinical deterioration. The system utilizes machine learning models to process patient data imported from Electronic Health Records (EHR) and other medical devices. The ClewICUServer computes two key indicators: the CLEWHI index, representing the likelihood of hemodynamic instability requiring vasopressor or inotrope support, and the CLEWLR indication, signifying a patient is at low risk. The ClewICUnitor, a web-based interface, displays these indices, associated notifications, and overall unit status. It is intended as an adjunctive tool, meaning its predictions are for reference only and should not be the sole basis for therapeutic decisions. The system's technological characteristics involve receiving near-real-time patient health data (including vital signs, nursing assessments, flowsheet data, medications, and lab data), analyzing this data to generate predictions of future clinically significant cardiac events, and alerting healthcare providers when a patient’s risk reaches a pre-defined threshold. Performance was assessed through software verification and validation, risk analysis, cybersecurity review, and human factors testing with 15 intended users. A retrospective clinical validation was conducted using a dataset from WakeMed Health System, for which a tagging system was developed and validated against human physician readers to establish clinical truth. The Hemodynamic Instability (HI) model demonstrated a sensitivity of 60.6%, a Positive Predictive Value (PPV) of 22.3%, and a median lead-time for true positive alerts of 3.0 hours. The Low Risk (LR) model showed a specificity of 95.7% and a sensitivity of 21.4%.

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 reported Positive Predictive Value of 22.3% for the Hemodynamic Instability model impact its clinical utility and the potential for alarm fatigue in a busy ICU setting?; What specific patient subgroups within the ICU (e.g., surgical, medical, trauma) were included in the WakeMed dataset, and was any variability in model performance observed across these subgroups?; Could the CLEWLR (Low Risk) model, with a sensitivity of 21.4%, be reliably used to de-escalate monitoring or intervention intensity, and what are the safety implications?; What methodology was used to validate the 'tagging system' against human physician readers, and what was the level of agreement achieved (e.g., Kappa statistic)?; Beyond the listed inputs (vital signs, nursing assessments, flowsheet data, medications, lab data), are there plans to incorporate other data sources, such as imaging or free-text notes, to enhance predictive accuracy?