Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
November 21, 2025Frontiers in Cardiovascular MedicineOpen Access

The CatBoost machine learning model integrating electrocardiogram, phonocardiogram, and clinical features accurately detected heart failure with an area under the curve of 0.998.

View Full Paper
Ask AI
Bookmark
Share

Why the study?

Heart failure is a major health concern with poor prognosis, creating an urgent clinical need for an easy and accurate screening method.

Does a machine learning model incorporating integrated ECG and PCG features accurately screen for heart failure in adult patients?

Population

1,017 individuals in China

Comparison

Five machine learning algorithms incorporating clinical, PCG, and ECG parameters

Design

Multicenter diagnostic validation study

Key result

The CatBoost machine learning model integrating electrocardiogram, phonocardiogram, and clinical features accurately detected heart failure with an area under the curve of 0.998.

Authors

KCKok Han CheeCLChengyu LiuHSHongwei Sun

Discussion

Loading...

Member takes

Overview

May aid noninvasive HF screening; hypothesis-generating and should not yet change practice.

Study Design

Type

Cohort (n=1,017)

Blinding

Independent reviewers blinded to feature analyses and model predictions

Multicenter

Yes

Structured PICO

Does a machine learning model incorporating integrated ECG and PCG features accurately screen for heart failure in adult patients?

P
Population
1,017 adults aged 18 and older without a prior history of heart failure or severe heart disease underwent AI-PECG screening to develop and validate a machine learning model for early heart failure detection.
E
Exposure
Machine learning-based screening model (CatBoost algorithm) incorporating 17 features including conventional risk factors (age, CRP, HR, HB), phonocardiogram (PCG) parameters, and electrocardiogram (ECG) parameters.
O
Outcome
Diagnostic performance for heart failure detection, measured by area under the curve (AUC), sensitivity, and specificity.surrogate

Main Result

Effect estimate: AUC 0.998 (95% CI 0.996-1.000)

A machine learning model integrating ECG and PCG features with clinical variables demonstrated excellent diagnostic performance for early heart failure screening.

Limitations

  • Development of the HF detection model depended on features of ECGs and PCGs extracted from manufacturer-specific software, which may require retraining for other software.
  • The data-driven feature selection strategy needs further assessment compared to mixed strategies.
  • The sample size of 1,017 patients may limit generalizability to larger datasets with varying patient characteristics across different institutions.
  • Development of the model depended on features extracted from manufacturer-specific software, requiring retraining for other software.
  • A mixed strategy for feature selection needs to be assessed in the future.
  • The study encompassed only 1,017 patients from three hospitals, and performance could differ in larger datasets with varying patient characteristics across different institutions.

Cite This Study

Chee et al. (2025) conducted a cohort in Heart failure (n=1,017). CatBoost machine learning model integrating ECG, PCG, and clinical features was evaluated on Area under the curve (AUC) for heart failure detection (CatBoost model, training set) (AUC 0.998, 95% CI 0.996-1.000). The CatBoost machine learning model integrating electrocardiogram, phonocardiogram, and clinical features accurately detected heart failure with an area under the curve of 0.998.

synapsesocial.com/papers/6a8c1c125ef9ff12d7895f4fhttps://doi.org/10.3389/fcvm.2025.1613577
View Full Paper
Ask AI
Bookmark
Share