ECG for high-throughput screening of multiple diseases: Proof-of-concept using multi-diagnosis deep learning from population-based datasets

Q: What is the clinical evidence from this study?

Study design: Observational. Population: Multiple diseases (ICD-wide range) (n=239852). Intervention: Deep learning model based on first in-hospital ECG vs. ICD-10 diagnostic codes (ground truth). Primary outcome: Discriminative performance (AUROC) for predicting ICD codes and categories.

Weijie Sun; Sunil V. Kalmady; Amir Salimi; Nariman Sepehrvand; Eric Ly; Abram Hindle; Russell Greiner; Padma Kaul

doi:10.48550/arxiv.2210.06291

October 6, 2022Open Access

ECG for high-throughput screening of multiple diseases: Proof-of-concept using multi-diagnosis deep learning from population-based datasets

View Full Paper

Ask AI

Bookmark

Share

Key result

A deep learning model applied to a patient's first in-hospital ECG successfully predicted 128 specific diseases and 68 disease categories with an AUROC greater than 80% in external validation.

Why the study?

Most deep learning diagnostic predictions using ECGs have been limited to a small set of cardiac diseases in selected cohorts, despite ECG abnormalities occurring across non-cardiovascular conditions.

Does a deep learning model applied to a patient's first in-hospital ECG accurately predict a wide range of diseases in a population-based cohort?

Study Design

Type

Observational (n=239,852)

Multicenter

Yes

Structured PICO

Does a deep learning model applied to a patient's first in-hospital ECG accurately predict a wide range of diseases in a population-based cohort?

P

Population

239,852 adult patients with emergency department visits or hospitalizations who had at least one 12-lead ECG, used to train and validate deep learning models for predicting a wide range of ICD-10 diagnoses.

E

Exposure

Deep learning (ResNet-based) model applied to the patient's first in-hospital 12-lead ECG to predict multiple ICD-10 diagnoses.

O

Outcome

Discriminative performance (AUROC and AUPRC) for predicting ICD-10 codes and categories from the first in-hospital ECG.

A deep learning model applied to routine 12-lead ECGs can identify a wide range of cardiovascular and non-cardiovascular diseases with high discriminative performance, providing a proof-of-concept for high-throughput screening.

Limitations

Low precision (PPV) for many diseases due to low prevalence rates, making it more suitable for 'rule out' screening.
Diagnostic labels from administrative health records may not be gold-standard ground truth without proper adjudication.
Latent ECG features used by the model may be related to comorbidities, medications, or lifestyle factors rather than underlying pathology.
Potential selection bias as the cohort is restricted to patients who had at least one ECG.
Generalizability across hospitals and robustness against gender or ethnic biases need further investigation.
Low precision (PPV) for many diseases due to low prevalence rates
Diagnostic labels from administrative records lack gold-standard adjudication
Latent ECG features may reflect comorbidities, medications, or lifestyle rather than direct pathology
Selection bias as cohort is restricted to patients who had at least one ECG

Implication

Provides proof-of-concept for using routine ECGs in high-throughput disease.

Abstract

Electrocardiogram (ECG) abnormalities are linked to cardiovascular diseases, but may also occur in other non-cardiovascular conditions such as mental, neurological, metabolic and infectious conditions. However, most of the recent success of deep learning (DL) based diagnostic predictions in selected patient cohorts have been limited to a small set of cardiac diseases. In this study, we use a population-based dataset of >250,000 patients with >1000 medical conditions and >2 million ECGs to identify a wide range of diseases that could be accurately diagnosed from the patient's first in-hospital ECG. Our DL models uncovered 128 diseases and 68 disease categories with strong discriminative performance.

Discussion

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

Cite This Study

Sun et al. (2022) conducted an observational in Multiple diseases (ICD-wide range) (n=239,852). Deep learning model based on first in-hospital ECG vs. ICD-10 diagnostic codes (ground truth) was evaluated on Discriminative performance (AUROC) for predicting ICD codes and categories. A deep learning model applied to a patient's first in-hospital ECG successfully predicted 128 specific diseases and 68 disease categories with an AUROC greater than 80% in external validation.

synapsesocial.com/papers/6a8ef8d096767b15a7fffe63 https://doi.org/10.48550/arxiv.2210.06291

View Full Paper

Ask AI

Bookmark

Share

ECG for high-throughput screening of multiple diseases: Proof-of-concept using multi-diagnosis deep learning from population-based datasets

Study Design

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

ECG for high-throughput screening of multiple diseases: Proof-of-concept using multi-diagnosis deep learning from population-based datasets

Study Design

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading