CLEWICU System

The CLEWICU System is an adjunctive predictive analytics tool employing machine learning to forecast the likelihood of hemodynamic instability in adult critical care patients, using EHR and monitoring data. Its methodology involves validated models, CLEWHI and CLEWLR, which provide physiological insights predicting events up to 8 hours in advance to clinicians for enhanced patient assessment, with provisions for ongoing model adaptation to diverse clinical environments.

The CLEWICU System is an analytical software product designed to assist clinicians in hospital critical care settings by predicting a patient's likelihood of future hemodynamic instability. It utilizes machine learning models operating on patient data imported from Electronic Health Records (EHR) and patient monitoring devices. The system comprises two main components: the ClewICUServer, a backend platform that processes data and computes the CLEWHI index (likelihood of hemodynamic instability requiring vasopressor/inotrope support) and CLEWLR index (indication of low risk for deterioration), and the ClewICUnitor, a web-based user interface that displays these predictions and associated notifications. The system is intended for adult patients (18 years and over) across various critical care units including ICUs, ED critical care areas, and PACUs, providing adjunctive information to support clinical assessment; predictions are for reference only and should not solely guide therapeutic decisions. The scientific basis of the CLEWICU System lies in its machine learning algorithms, which have been trained and validated to identify patterns indicative of impending hemodynamic changes, predicting events up to 8 hours in advance. Recent validation, using a reduced set of 50 input features, was conducted through a retrospective cohort study involving two large, independent healthcare datasets (UMass eICU and MIMIC-III). The CLEWHI model demonstrated sensitivities of 63-69% and positive predictive values (PPVs) of 10-12%, while the CLEWLR model achieved specificities around 90% and sensitivities of 35.5-47%, meeting all predefined performance targets. These results confirm that the system maintains clinically acceptable predictive performance with a streamlined input dataset. The system also incorporates a Predetermined Change Control Plan, allowing for model retraining with new or modified datasets to adapt to diverse hospital EMR systems and monitoring protocols and potentially improve sensitivity, while adhering to strict validation protocols and maintaining established performance standards for predicting hemodynamic events.

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 reduction in model features from 80 to 50 impact the system's ability to generalize across patient populations with varying data availability and completeness in real-world critical care settings?; What specific physiological parameters or EHR data points constitute the 50 features used by the current CLEWHI and CLEWLR models, and what is the relative importance of these features in predicting hemodynamic instability?; Given the PPV for the CLEWHI model is around 10-12%, what is the anticipated clinical workflow integration and how are clinicians guided to interpret and act upon these predictions to minimize alert fatigue and optimize patient benefit?; The PCCP allows for model retraining with new datasets; what governance and ongoing monitoring processes are in place to ensure that model performance does not degrade over time or with the introduction of data from new, diverse hospital systems?; How does the system define 'hemodynamic instability requiring vasopressor/inotrope support,' and what is the typical validated lead time of prediction provided by the CLEWHI index for an impending event?