Heart failure & cardiomyopathies
Artificial Intelligence Approaches for Phenotyping Heart Failure in U.S. Veterans Health Administration Electronic Health Record
Authors' conclusion
These findings suggest that NLP and ML models are efficient AI tools to phenotype HF in big EHR data to create contemporary HF registries for clinical studies of effectiveness, quality improvement, and hypothesis generation.
Study context
- Population
- US Veterans Affairs (VA) patients with International Classification of Diseases (ICD) codes for heart failure (HF) in the electronic health record (EHR)gold standard cohort
- Participants
- 200 patients
- Comparison
- Natural language processing (NLP), machine learning (ML) and NLP + ML models vs ICD code approaches for HF phenotyping
- Outcome
- HF classification vs gold standard HF (manual chart adjudication)
Methodological detailsmatched development cohort, gold standard cohort with manual chart review
- Matched development cohort
- Gold standard cohort with manual chart review
Performance
As reported in the abstract
HF classification in gold standard cohort
| Test or model | Sensitivity | PPV |
|---|---|---|
| NLP | 0.86 | 0.83 |
| ML | 0.88 | 0.77 |
| NLP + ML | 0.83 | 0.85 |
| 1 discharge code | 0.21 | 0.97 |
| 2 outpatient codes | 0.54 | 0.88 |
Sensitivity
NLP0.86
ML0.88
NLP + ML0.83
1 discharge code0.21
2 outpatient codes0.54
PPV
NLP0.83
ML0.77
NLP + ML0.85
1 discharge code0.97
2 outpatient codes0.88
Further findings
| Outcome | NLP | ML |
|---|---|---|
| Model development (testing) | ||
| PPV | 0.81 | 0.86 |
| Sensitivity | 0.77 | 0.86 |
- Model development (testing): development cohort n = 20 000 (10 000 HF, 10 000 matched non-HF)
Characteristics and prognosis of NLP + ML-classified HF
- Characteristically and prognostically similar to gold standard HF
Original publication
Artificial Intelligence Approaches for Phenotyping Heart Failure in U.S. Veterans Health Administration Electronic Health Record. ESC Heart Fail. Published 14 June 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →