Rather than relying on a single abnormal vital sign, the study will examine how multiple physiologic signals change together over time and whether changes in the relationships among these signals provide useful information about impending clinical deterioration. Individual participant physiologic patterns may be characterized longitudinally to account for differences between individuals and changes within the same individual over time. The primary objective is to evaluate the predictive accuracy of multimodal digital biomarkers for protocol-defined acute clinical deterioration occurring within 72 hours. The study is observational and does not assign participants to an investigational treatment.
FUSION-72 (Multimodal Physiologic Signal Fusion for 72-Hour Prediction of Acute Clinical Deterioration) is a prospective, longitudinal, observational study designed to investigate multimodal digital biomarkers associated with acute clinical deterioration. The central scientific premise is that clinically meaningful deterioration may be preceded not only by an abnormal value in an individual physiologic parameter, but also by coordinated or discordant changes occurring across multiple physiologic, behavioral, and symptom domains. The study will collect longitudinal data from multiple digital and physiologic sources, which may include wearable electrocardiography (ECG), photoplethysmography (PPG), peripheral oxygen saturation (SpO2), temperature, gait and activity characteristics, sleep measures, voice characteristics, cough characteristics, and participant-reported symptoms. These multimodal measurements will be temporally aligned and analyzed to characterize changes within individual participants and relationships among physiologic, behavioral, acoustic, and symptom-derived signals. Analyses will evaluate whether multimodal signal fusion provides additional predictive information beyond individual digital biomarker streams or conventional single-parameter measurements. The primary objective is to evaluate the predictive accuracy of a prespecified multimodal digital biomarker model for protocol-defined acute clinical deterioration occurring within a subsequent 72-hour prediction window. Secondary objectives include evaluation of time-dependent predictive performance, lead time to detection, incremental predictive value of multimodal signal relationships, model calibration, and robustness of predictive performance across prespecified participant and clinical subgroups. The study is observational. No investigational treatment is assigned, and participation does not direct, replace, or require modification of usual clinical care or medical management.
Study Type
OBSERVATIONAL
Enrollment
10,000
Prospective collection and analysis of multimodal digital biomarkers derived from wearable and remote-monitoring technologies, including ECG, PPG, oxygen saturation, temperature, gait, sleep, voice, cough, and participant-reported symptom data. The monitoring is observational and does not direct or replace clinical care, does not assign treatment, and does not require modification of participants' usual medical management. Multimodal signals will be temporally aligned and evaluated for changes in inter-signal relationships associated with protocol-defined clinical deterioration occurring within the subsequent 72-hour prediction window.
Truway Health, Inc.
New York, New York, United States
Predictive Accuracy of Multimodal Digital Biomarker Fusion for Acute Clinical Deterioration Within 72 Hours
Performance of the prespecified multimodal digital biomarker model in predicting protocol-defined acute clinical deterioration during the subsequent 72-hour prediction window. Model performance will be evaluated using the area under the receiver operating characteristic curve (AUROC), with additional assessment of the area under the precision-recall curve (AUPRC), sensitivity, specificity, positive predictive value, negative predictive value, and calibration.
Time frame: At initiation of each prespecified prediction episode and 72 hours after initiation of each prespecified prediction episode.
Time-Dependent Predictive Performance of Individual and Combined Digital Biomarkers
Comparison of predictive performance associated with individual and multimodal combinations of electrocardiography (ECG), photoplethysmography (PPG), peripheral oxygen saturation (SpO2), temperature, gait, sleep, voice, cough, and symptom-derived features for prediction of protocol-defined acute clinical deterioration.
Time frame: At initiation of each prespecified prediction episode and 72 hours after initiation of each prespecified prediction episode.
Lead Time to Detection of Acute Clinical Deterioration
Time interval, measured in hours, between the earliest prespecified model-generated deterioration signal meeting protocol-defined prediction criteria and occurrence of the corresponding protocol-defined acute clinical deterioration event.
Time frame: For each protocol-defined acute clinical deterioration event, during the 72 hours preceding event occurrence and ending at the time of event occurrence.
Incremental Predictive Value of Multimodal Signal Relationships
Assessment of whether incorporating prespecified temporal relationships and interactions among multiple physiologic, behavioral, acoustic, and symptom-derived signals improves prediction of protocol-defined acute clinical deterioration compared with models based on individual signal domains or conventional single-parameter measurements.
Time frame: At initiation of each prespecified prediction episode and 72 hours after initiation of each prespecified prediction episode.
Model Calibration for 72-Hour Clinical Deterioration Risk
Agreement between predicted probabilities of protocol-defined acute clinical deterioration and observed event frequencies within the subsequent 72-hour prediction horizon, assessed using prespecified calibration metrics.
Time frame: At 72 hours after initiation of each prespecified prediction episode.
Robustness of Multimodal Prediction Across Participant and Clinical Subgroups
Evaluation of predictive performance across prespecified participant and clinical subgroups, including age category, sex, baseline health status, comorbidity burden, care setting, and availability or completeness of individual digital biomarker streams.
Time frame: At 72 hours after initiation of each prespecified prediction episode.
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.