ECG-based deep learning for chronic kidney disease detection and cardiovascular risk prediction

Area of Science:

• Cardiology
• Nephrology
• Artificial Intelligence

Background:

• Chronic kidney disease (CKD) presents a global health challenge with low awareness.
• Deep learning models (DLMs) show potential in ECG interpretation for disease detection.
• ECG-based DLMs may offer novel avenues for early CKD identification.

Purpose of the Study:

• To develop and validate a DLM for detecting CKD using ECGs.
• To assess the DLM's ability to identify individuals at risk for CKD and its complications.
• To compare the DLM's predictive performance against traditional eGFR classification.

Main Methods:

• Utilized ECGs from 49,632 outpatients (66,587 total enrolled) with eGFR data (Jan 2010-Oct 2020).
• Developed DLMs using 72,618 ECGs, with internal validation on 16,955 patients and external validation on 10,476 patients.
• Primary outcome: CKD detection (eGFR < 60 mL/min/1.73 m²); secondary outcomes: all-cause mortality and major cardiovascular events.

Main Results:

• The DLM achieved AUCs of 0.885 (internal) and 0.861 (external) for CKD detection.
• DLM-identified CKD patients exhibited higher risks for CKD progression and cardiovascular disease.
• Positive DLM screens in patients without baseline CKD significantly increased incident CKD risk (HRs 2.14 and 1.38).
• DLM stratification outperformed eGFR in predicting adverse outcomes like stroke, heart failure, and atrial fibrillation.

Conclusions:

• An ECG-based DLM can identify individuals at risk for CKD and associated complications.
• This AI approach facilitates early detection and risk stratification in clinical settings.
• The model shows promise for identifying subclinical CKD before laboratory markers are apparent.