Chronic Heart Disease, Chronic Heart Failure, Heart Failure
Conditions
Brief summary
PRE-DETECT-HF is a prospective, single-arm observational study evaluating a voice-based machine learning algorithm for early detection of heart failure decompensation. 123 patients hospitalized for acute decompensated or de-novo heart failure will be enrolled across three sites in the Netherlands and Spain. Patients make daily voice recordings via a smartphone app and answer symptom questions for 6 months. The algorithm analyzes voice patterns compared to a baseline recording at discharge. Treatment decisions are based on symptom data only; voice-based predictions are analyzed retrospectively after study completion. The primary endpoint is sensitivity of the voice-based software in detecting heart failure deterioration, defined as heart failure hospitalization, or intensification of heart failure therapy. Secondary endpoints include app adherence, usability, and associations between voice data and blood biomarkers.
Detailed description
Heart failure decompensation is often detected too late by conventional symptom and weight monitoring, leaving insufficient time to intervene. Invasive alternatives such as implantable pulmonary artery pressure monitors are effective but require surgical implantation. Voice-based digital biomarkers offer a promising non-invasive approach, as fluid overload may produce detectable changes in vocal features. Patients begin voice recordings during hospitalization while still volume overloaded. At home, patients record daily using standardized and variable text content. The voice-based algorithm extracts biomechanical vocal features and calculates a risk score. Healthcare providers access a dashboard showing symptom-based notifications and may adjust therapy at their discretion. Voice-derived risk scores are withheld during the study and analyzed retrospectively. Study visits occur at months 3 and 6 (in-clinic) and month 1 (telephone). Blood samples are collected at baseline, month 3, and month 6 for analysis of traditional (NT-proBNP, creatinine) and novel biomarkers. Usability and quality of life are assessed via questionnaires distributed throughout the study period.
Interventions
OTHERDaily Voice Recording and Symptom Monitoring
Patients use the mobile app daily to record voice samples and answer symptom-related questions. Voice recordings are analyzed by a algorithm, which extracts vocal biomechanical features. Healthcare providers receive notifications based on symptom data only and may adjust therapy at their discretion. Voice-derived risk scores are not shared with clinicians during the study and are analyzed retrospectively after study completion.
Sponsors
Noah Labs
Hospital Clinic of Barcelona
Maastricht University
Zuyderland Medical Centre
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
* Informed consent provided * Currently hospitalized for acutely decompensated HF or de-novo HF * Age: 18 years and above
Exclusion criteria
* Inability to provide consent * Pregnancy * Life-expectancy lower than 1 year due to a condition other than HF * Planned cardiac intervention within the next 6 months (e.g. valve replacement, bypass surgery) * Disabling mental diseases (e.g., Alzheimer's disease) * Symptoms mainly caused by chronic disease other than HF such as chronic obstructive pulmonary disease * Inability to use a smartphone or a tablet computer despite support by informal caregiver if required * Insufficient knowledge of the local language * Previous operations on organs involved in generation of voice (vocal tract, vocal folds, etc.) * Participation in another interventional study within 30 days of inclusion
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Sensitivity of Voice-Based Software in Detecting Heart Failure Deterioration | 6 month | Sensitivity of the voice-based prediction in detecting heart failure deterioration, defined as heart failure-related hospitalization, or intensification of heart failure therapy due to worsening heart failure. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Unexplained Alert Rate per Patient-Year | 6 month | Number of voice-based alerts not associated with clinical deterioration, reported as a single rate per patient-year of follow-up. |
| Alert Lead Time in Days | 6 month | Median number of days prior to a heart failure deterioration event that the voice-based algorithm generates an alert, reported in days. |
| Adherence to voice-based monitoring | 6 month | Adherence to voice-based monitoring in number and percentage of days with at least one transmitted voice recording. |
| App Usability via In-App Questionnaires | 6 month | User experiences, expectations, and acceptance assessed via standardized in-app questionnaires on a 7-point Likert scale. |
| Quality of Life using the Kansas City Cardiomyopathy Questionnaire | 6 months | Kansas City Cardiomyopathy Questionnaire (KCCQ) overall summary score (range 0-100, higher scores indicate better health status) at baseline, month 3, and month 6. |
Countries
Netherlands, Spain
Outcome results
None listed