This research is Trending!
Artificial Intelligence–Enabled Detection and Management of Physiological Instability in Heart Failure
Konstantinos SiderisCarsten SchmalfussNeil LewisKarim SallamBiykem BozkurtThomas C. HanffMichael T. YoungRichard SchofieldChristos P. KyriakopoulosHeather HansonLina BrinkerPeter WohlfahrtMandeep R. MehraTempa CurryCheri KnechtStephanie TuckettRichard E. NelsonSusan ZickmundJorie M. ButlerJay M. EdelbergGregory H. KurioStephan WegerichStylianos SerghiouMatt PipkeJosef Stehlik
Why the study?
Artificial intelligence-enabled wearable monitoring can detect individualized physiological deviations preceding heart failure decompensation, but whether these signals can be translated into improved outcomes remains uncertain.
Does AI-guided continuous physiological monitoring with clinician notifications improve implementation outcomes and reduce heart failure rehospitalization in adults recently discharged for acute heart failure?
Key Points
- To assess the feasibility of combining wearable AI-driven physiological alerts with clinician response protocols, and to explore their impact on postdischarge heart failure rehospitalizations.
- The LINK-HF2 prospective, randomized, open-label, blinded-endpoint trial enrolled 171 adults across 5 Veterans Affairs medical centers upon hospital discharge for acute heart failure (NCT04502563).
- Participants wore a continuous multisensor patch for up to 90 days and were assigned to AI-guided monitoring with clinician notification (n=80) or standard care (n=91).
- Cox proportional hazards models evaluated exploratory clinical endpoints including heart failure rehospitalization, a composite of all-cause unplanned rehospitalizations, emergency visits, and death.
- Heart failure rehospitalization occurred in 13 (16.3%) intervention patients versus 17 (18.7%) controls (HR: 0.90; 95% CI: 0.43–1.84; P = 0.76).
- The composite clinical endpoint did not differ between the intervention and control arms (HR: 1.03; 95% CI: 0.63–1.70; P = 0.89).
Study Design
Type
RCT (n=171)
Blinding
open-label, blinded-endpoint
Randomization
randomized
Multicenter
Yes
Structured PICO
Does AI-guided continuous physiological monitoring with clinician notifications improve implementation outcomes and reduce heart failure rehospitalization in adults recently discharged for acute heart failure?
P
Population
171 adults hospitalized for acute heart failure, randomized at discharge to AI-guided monitoring or standard care and monitored for up to 90 days.
I
Intervention
AI-guided continuous multisensor patch monitoring for up to 90 days, with personalized similarity-based notifications of cardiorespiratory and activity deviations sent to clinicians for a structured response.
C
Comparator
Standard care (participants also wore the continuous multisensor patch for up to 90 days without clinician notification).
O
Outcome
Implementation outcomes (feasibility, acceptability, usability) and exploratory clinical endpoint of heart failure rehospitalization over up to 90 days follow-up.hard clinical
Main Result
Hazard Ratio: 0.9 (95% CI 0.43–1.84)
Absolute Event Rate: 16.3% vs 18.7%
p-value: p=0.76
Limitations
- Workflow integration and response uncertainty were primary barriers to implementation
- Study was exploratory and not powered to demonstrate a reduction in clinical events
Key expert perspectives
Captured external expert commentary on this paper, strongest first. Original sources are linked where available.
“The central insight from LINK-HF2 is that in AI-enabled medicine, accurate physiological risk detection alone does not modify disease trajectory unless it is coupled with consistent mechanistically aligned and trusted therapeutic responses.”
HVHarriette Van SpallCardiologist, McMaster University
“This very small feasibility trial was not powered to show a reduction in clinical endpoints. The bigger issue is translating EHR alerts to action, integrating care across organ-based siloes, and if proven successful (mixed evidence to date), sustaining & scaling.”
GFGregg FonarowCardiologist, UCLA
“The authors argue this is the key point: Clinicians responded quickly, but alerts rarely led to treatment changes.”
Implication
AI-enabled continuous physiological monitoring post-discharge in heart failure patients is feasible but did not significantly reduce clinical events, suggesting accurate detection requires more effective therapeutic responses to modify risk.
Abstract
BACKGROUND: 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. OBJECTIVES: 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. METHODS: 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. RESULTS: 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). CONCLUSIONS: 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).
Discussion
No takes yet. Share an insight, caveat, or question.
Cite This Study
Sideris et al. (2026) conducted an RCT in Heart Failure (n=171). AI-guided monitoring with clinician notification vs. standard care was evaluated on HF rehospitalization (HR 0.90, 95% CI 0.43-1.84, p=0.76).