Impact of an artificial intelligence-driven telemonitoring programme on hospital admissions in patients with heart failure

Background

Artificial Intelligence (AI) provides a broad spectrum of resources for clinical practice, including telemedical monitoring tools for symptom management. Heart failure (HF) remains one of the most prevalent conditions among the elderly people, characterised by high readmission rates and significant healthcare resource consumption. To mitigate these admissions, various follow-up strategies have been developed within Heart Failure Units. The advent of AI offers tools that allow for the management of larger patient cohorts with reduced time and effort, aimed at optimising clinical control.

Purpose

To evaluate the impact of an AI-driven telephonic monitoring tool on hospital admission rates in patients with heart failure following hospital discharge.

Methods

We implemented an AI application featuring a human-like voice interface that conducts fortnightly telephone interviews to detect clinical changes and address HF decompensation early. The tool assesses parameters including general well-being, clinical symptoms and signs, blood pressure, heart rate, and medication adherence. Alerts are categorised as:

Mild: Increased fatigue, blood pressure fluctuations, heart rate abnormalities, worsening oedema, or decreased diuresis.

Moderate: General malaise, chest pain, increased dyspnoea, or emergency department (ED) contact.

Severe: Paroxysmal nocturnal dyspnoea, orthopnoea, haematuria, dysuria, weight gain, hospitalisation, or death.

A single moderate/severe alert or ≥2 mild alerts triggers a notification to the attending physician for medical intervention (telephonic review, pharmacological adjustment, or in-person assessment). Statistical analyses were performed to compare the number of hospital admissions in the six months prior to inclusion against the six months following implementation.

Results

The study included 107 patients from an Internal Medicine department. Inclusion criteria required at least one HF-related admission or emergency department (ED) visit in the preceding six months, preserved functional status (Barthel Index >60), and no cognitive impairment. 39 patients were excluded due to mortality or loss to follow-up. Admissions and ED visits for HF or comorbid decompensation were analysed, excluding unrelated surgical or traumatic causes. The mean number of admissions prior to monitoring was 1.87, decreasing to 0.63 post-implementation, representing a 66% reduction (p <0.001).

Conclusion

The impact of the AI tool was favourable, significantly reducing hospital admissions and ED attendings. While confounding factors such as inpatient education and increased patient self-awareness may contribute to these results, the tool appears to positively influence self-care and treatment adherence. AI-driven tools facilitate rigorous monitoring of larger patient populations, potentially reducing resource consumption by covering more patients with diminished clinician burden.Stratification of alertsFor image description, please refer to the figure legend and surrounding text.Reduction in admissionsFor image description, please refer to the figure legend and surrounding text.