Early detection of heart failure using in-patient longitudinal electronic health records
2024

Early Detection of Heart Failure Using Electronic Health Records

Sample size: 183894 publication 10 minutes Evidence: high

Author Information

Author(s): Drozdov Ignat, Szubert Benjamin, Murphy Clare, Brooksbank Katriona, Lowe David J.

Primary Institution: Institute of Cardiovascular and Medical Sciences, BHF Glasgow Cardiovascular Research Centre, University of Glasgow

Hypothesis

A simple Transformer neural network can be used to prospectively identify patients at increased risk of first hospitalisation due to heart failure.

Conclusion

The study demonstrates that a Transformer neural network trained on electronic health records can effectively predict the risk of first hospitalisation for heart failure.

Supporting Evidence

  • The model achieved an AUROC of 0.86, indicating good discriminatory power.
  • Sensitivity was 36.4% and specificity was 98.26%.
  • At a probability threshold of 90%, the model achieved 100% positive predictive value.
  • Performance was consistent across different patient demographics.
  • Model probabilities were well-calibrated with a Pearson’s R2-value of 0.99.

Takeaway

Researchers used a computer model to look at health records and found a way to predict who might get heart failure before it happens.

Methodology

The study used a nested case-control design with a Transformer neural network trained on longitudinal electronic health records.

Potential Biases

The retrospective nature of the study may not represent real-world prevalence and could affect performance metrics.

Limitations

The study may overlook pre-hospitalization stages of heart failure and was not validated on an out-of-sample dataset.

Participant Demographics

The average age of patients with first hospitalisation due to heart failure was 75.7 years for males and 80.4 years for females.

Statistical Information

P-Value

0.00099

Confidence Interval

95% CI: 33.33%-39.56%

Statistical Significance

p<0.05

Digital Object Identifier (DOI)

10.1371/journal.pone.0314145

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication