Artificial Intelligence-Enabled Detection and Management of Physiological Instability in Heart Failure: Results of the Randomized LINK-HF2 Trial.

Authors

Konstantinos Sideris, Carsten Schmalfuss, Neil Lewis, Karim Sallam, Biykem Bozkurt, Thomas C Hanff, Michael T Young, Richard Schofield, Christos P Kyriakopoulos, Heather Hanson, Lina Brinker, Peter Wohlfahrt, Mandeep R Mehra, Tempa Curry, Cheri Knecht, Stephanie Tuckett, Richard E Nelson, Susan L Zickmund, Jorie M Butler, Jay M Edelberg, Gregory H Kurio, Stephan Wegerich, Stylianos Serghiou, Matt Pipke, Josef Stehlik

Published in

JACC. Heart failure. Pages 103422. Oct 09, 2026. Epub Oct 09, 2026.

Abstract

Artificial intelligence (AI)-enabled wearable monitoring can detect individualized physiological deviations preceding heart failure (HF) decompensation, but whether these signals can be translated into improved outcomes remains uncertain.
This study aims to evaluate the feasibility of integrating AI-generated notifications with a structured clinician response algorithm into postdischarge HF care and to explore their association with HF events following hospitalization.
LINK-HF2 (Detection of Physiological Anomaly Using Artificial Intelligence and Prevention of Heart Failure Hospitalization) was a prospective, randomized, open-label, blinded-endpoint study conducted at 5 Veterans Affairs medical centers. Adults hospitalized for acute HF were enrolled at discharge and assigned to AI-guided monitoring with clinician notification or to standard care. All participants wore a continuous multisensor patch for up to 90 days. A personalized similarity-based model detected deviations in cardiorespiratory and activity signals and generated notifications. Implementation outcomes included feasibility, acceptability, and usability. The main exploratory clinical endpoint was HF rehospitalization. Secondary exploratory clinical endpoints included the composite outcome of all-cause unplanned rehospitalizations, emergency department visits, and all-cause mortality. All clinical endpoints were compared through Cox proportional hazards models over the follow-up period.
Among 171 analyzed participants, 80 (46.8%) were assigned to intervention and 91 (53.2%) to control. Median monitoring adherence was 72.4% of follow-up days; clinicians responded to communicated notifications within a median of 1 day (range: 1-5 days), and qualitative analysis demonstrated high patient acceptance and clinician engagement, whereas identifying workflow integration and response uncertainty as the primary barriers to implementation. In exploratory analyses, HF rehospitalization occurred in 13 (16.3%) patients in the intervention group and 17 (18.7%) in the control group (HR: 0.90, 95% CI: 0.43-1.84; P = 0.76). The composite clinical endpoint also did not differ between groups (HR 1.03; 95% CI: 0.63-1.70; P = 0.89). The intervention group had a lower rate of AI-detected physiological instability than controls (HR: 0.76; 95% CI: 0.60-0.97; P = 0.025).
AI-enabled continuous physiological monitoring was feasible and reduced AI-detected physiological instability. Although this feasibility study did not demonstrate a reduction in exploratory clinical events, these findings highlight that accurate physiological risk detection alone may be insufficient to modify clinical risk without mechanistically aligned and trusted therapeutic responses. (Detection of Physiological Anomaly Using Artificial Intelligence and Prevention of Heart Failure Hospitalization [LINK-HF2]; NCT04502563).

PMID:
42853165
Bibliographic data and abstract were imported from PubMed on 09 Oct 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.