Explainable ECG-derived cardiac aging pattern assessment using PTB-XL with a lightweight web prototype

Q: What is the clinical evidence from this study?

Study design: Observational. Population: Cardiac aging (n=21373). Intervention: 12-lead ECG-derived statistical features. Primary outcome: Older-age group classification (age ≥ 65 years) (AUROC 0.805, 95% CI 0.786-0.823).

Ye-Ji Park; Jae-Hyun Jo; Ye-seul Park; Jin-Hyoung Jeong

doi:10.1038/s41598-026-69354-0

September 19, 2026Scientific ReportsOpen Access

Explainable ECG-derived cardiac aging pattern assessment using PTB-XL with a lightweight web prototype

View Full Paper

Ask AI

Bookmark

Share

Key result

Explainable ECG-based machine learning identifies adults aged ≥65 years with ~0.81 AUROC.

AUROC 0.805
95% CI 0.786-0.823
n=21,373

Why the study?

The potential of explainable features from standard 12-lead ECG to identify older adults and assess cardiac aging patterns was not established.

Can an XGBoost model utilizing explainable 12-lead ECG statistical features accurately identify older adults and estimate cardiac aging patterns?

Key Points

To determine whether explainable features extracted from standard 12-lead electrocardiography can identify older adults and assess cardiac aging patterns using machine learning.
Analyzed 21,373 12-lead ECG records from individuals aged 18–89 years from the PTB-XL dataset, defining older age as 65 years or older.
Extracted 176 lead-specific statistical features reflecting amplitude, variability, energy, waveform changes, and zero-crossing patterns from 10-s, 100-Hz recordings.
Compared logistic regression, random forest, and XGBoost models for age classification and regression, applying five-fold patient-level grouped validation and regression-based age-bias correction.
XGBoost achieved an AUROC of 0.805 (95% CI 0.786–0.823), an accuracy of 0.723 (95% CI 0.703–0.742), and a Brier score of 0.181 (95% CI 0.173–0.189) for older-age classification, with a mean AUROC of 0.797 ± 0.007 in patient-level grouped validation.
In chronological age regression, the XGBoost model achieved a mean absolute error of 10.31 years (95% CI 9.97–10.64) and an R-squared of 0.431 (95% CI 0.398–0.460).
The accelerated cardiac aging group showed a lower prevalence of normal ECG findings (30.3% vs. 62.7%, p=1.35e-50) and a higher prevalence of abnormalities (68.8% vs. 38.0%, p=6.93e-46) than the younger cardiac pattern group.

Study Design

Type

Observational (n=21,373)

Structured PICO

Can an XGBoost model utilizing explainable 12-lead ECG statistical features accurately identify older adults and estimate cardiac aging patterns?

P

Population

21,373 ECG records from adults aged 18 to 89 years from the PTB-XL dataset were analyzed to evaluate explainable ECG-derived cardiac aging patterns.

E

Exposure

XGBoost machine learning model utilizing 176 explainable statistical features extracted from 10-s, 100-Hz, standard 12-lead ECG signals.

C

Comparator

Low-information baseline models using biological sex alone.

O

Outcome

Older-age group classification (chronological age >= 65 years) and continuous chronological age prediction (ECG-predicted age).surrogate

Main Result

Effect estimate: AUROC 0.805 (95% CI 0.786-0.823)

Explainable statistical features derived from standard 12-lead ECGs can estimate chronological age with moderate accuracy, and an age-bias-corrected cardiac age gap is associated with a higher prevalence of ECG abnormalities.

Limitations

Only a single public dataset (PTB-XL) was used, meaning no claim of generalizability can be made without independent external validation
Basic statistical features were used instead of clinically established ECG morphology features, limiting direct clinical interpretation
The corrected cardiac age gap was not evaluated against clinical outcomes such as mortality or major adverse cardiovascular events
Important clinical confounders like hypertension, diabetes, and medication use were not available in a harmonized structured format
Diagnostic association analyses may be partly circular because the corrected cardiac age gap and diagnostic classes were derived from the same ECG recordings
Explicit signal-quality assessment, artifact exclusion, and pacemaker- or rhythm-specific exclusion criteria were not implemented
The web prototype requires pre-extracted ECG features and has not undergone usability testing or clinical workflow evaluation
Broader applicability of the framework remains speculative
Prototype should not be considered ready for clinical deployment, population screening, or general use without independent external validation
Remaining age-regression error limits practical individual-level age estimation
SHAP feature importance results explain the model's prediction basis but should not be interpreted as evidence of specific biological aging mechanisms

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

May flag accelerated cardiac aging on routine ECGs; leaves open prospective validation before informing clinical decisions.

Abstract

This study aimed to investigate whether explainable features extracted from standard 12-lead electrocardiography (ECG) can be used to identify older adults and assess ECG-derived cardiac aging patterns, and to implement the resulting model as a lightweight web-based interpretation prototype. A total of 21,373 ECG records from individuals aged 18–89 years were included from the publicly available PTB-XL dataset. Lead-specific statistical features related to amplitude, variability, energy, waveform change, and zero-crossing patterns were extracted from 10-s, 100-Hz, 12-lead ECG signals in records100. Adults aged 65 years or older were defined as the older-age group. Logistic regression, random forest, and XGBoost models were compared. XGBoost achieved moderate discrimination for older-age classification, with an AUROC of 0.805 in the PTB-XL strat_fold-based test set and a Brier score of 0.181. In five-fold patient-level grouped validation, the model achieved a mean AUROC of 0.797 ± 0.007 with no patient overlap between training and test folds. In age regression, XGBoost achieved a mean absolute error of 10.31 years. The age-dependent bias of the raw cardiac age gap was removed using regression-based correction. The accelerated cardiac aging group, defined as the upper quartile of the corrected cardiac age gap, showed a lower prevalence of normal ECG findings and higher prevalence of ECG abnormalities, MI-related patterns, ST-T changes, conduction disturbances, and hypertrophy than the younger cardiac pattern group. These findings suggest that explainable ECG-derived features may support exploratory cardiac aging pattern assessment and provide a basis for research-oriented web-based interpretation tools. However, any broader applicability of this framework remains speculative, and the prototype should not be considered ready for clinical deployment, population screening, or general use without independent external validation.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Park et al. (2026) conducted an observational in Cardiac aging (n=21,373). 12-lead ECG-derived statistical features was evaluated on Older-age group classification (age ≥ 65 years) (AUROC 0.805, 95% CI 0.786-0.823). An XGBoost model using explainable ECG-derived statistical features identified adults aged 65 years or older with an AUROC of 0.805, and the derived corrected cardiac age gap was associated with ECG abnormalities.

synapsesocial.com/papers/6ad0fd1915427f6a84040483 https://doi.org/10.1038/s41598-026-69354-0

View Full Paper

Ask AI

Bookmark

Share

Explainable ECG-derived cardiac aging pattern assessment using PTB-XL with a lightweight web prototype

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Explainable ECG-derived cardiac aging pattern assessment using PTB-XL with a lightweight web prototype

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading