Heart failure & cardiomyopathies
Unsupervised Machine Learning to Identify Subphenotypes Among Cardiac Intensive Care Unit Patients with Heart Failure
Authors' conclusion
Unsupervised machine learning clustering can identify multiple distinct clinical HF phenotypes within the CICU population that display differing mortality profiles both in-hospital and at 1 year. Mortality was lowest for the uncomplicated HF phenotype and highest for the hypoperfused phenotype.
Study context
- Population
- Mayo Clinic cardiac ICU patients with an admission diagnosis of heart failure (HF), 2007–2018, with complete admission laboratory data6,008 screened
- Participants
- 4,877 patients
- Exposure
- HF phenotypes identified by clustering of admission laboratory values
- Outcome
- In-hospital mortality
Methodological detailsunsupervised machine learning (consensus k-means clustering), single-centre
- Unsupervised machine learning (consensus k-means clustering)
- Single-centre
Main finding
As reported in the abstract
Each group vs uncomplicatedAdjusted odds ratio (CI)P value
Iron-deficient
2.18(1.38–3.48)
P < 0.001Cardiorenal
2.11(1.37–3.32)
P < 0.001Inflamed
1.79(1.18–2.76)
P = 0.007Hypoperfused
4.32(2.89–6.62)
P < 0.001- Overall in-hospital mortality
- 10.7%
Observational data: associations, not effects.
Further findings
| Outcome | Values | Adjusted HR (CI) | P value | |
|---|---|---|---|---|
| 1-year mortalityInflamedvs UncomplicatedMultivariable-adjusted Cox models | Uncomplicated 11.9% · Inflamed 26.8% · Iron-deficient 33.8% · Cardiorenal 41.2% · Hypoperfused 52.3% | 1.56 (1.27–1.92) | P < 0.001 | |
| 1-year mortalityIron-deficientvs UncomplicatedMultivariable-adjusted Cox models | 2.47 (2.00–3.04) | P < 0.001 | ||
| 1-year mortalityCardiorenalvs UncomplicatedMultivariable-adjusted Cox models | 2.41 (1.97–2.95) | P < 0.001 | ||
| 1-year mortalityHypoperfusedvs UncomplicatedMultivariable-adjusted Cox models | 3.43 (2.82–4.18) | P < 0.001 | ||
| 0.20.5125Adjusted hazard ratio (log scale) |
- 1-year mortality — Overall 1-year mortality
- 31.5%
1-year mortality
- Similar findings for post-discharge 1-year mortality
Original publication
Unsupervised Machine Learning to Identify Subphenotypes Among Cardiac Intensive Care Unit Patients with Heart Failure. ESC Heart Fail. Published 19 August 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →