Abstract
Background: Electronic health records (EHRs) provide abundant routine clinical care data to support research, but often lack phenotypic information needed to discern relevant subphenotypes. Left ventricular ejection fraction (EF) is required to discriminate the different heart failure (HF) phenotypes [i.e. HF with preserved (HFpEF), mid-range (HFmrEF) and reduced (HFrEF) EF], but is not collected in EHRs. This may represent a major limitation when using EHRs in HF research. Aim of this analysis was to create an algorithm that identifies HF EF phenotypes in routine care data using available patient’s characteristics.
Methods: We included 42,061 HF patients from the Swedish Heart Failure Registry that collects data on EF and biomarkers. For the primary analysis we performed two logistic regression models to predict 1) HFrEF and HFmrEF vs. HFpEF; and 2) HFrEF vs. HFmrEF and HFpEF. In the secondary analysis we performed a multivariable multinomial logistic regression to create a prediction model for all 3 separate HF phenotypes: HFrEF vs. HFmrEF vs. HFpEF. The models included 22 predictors: age, sex, NT-proBNP, NYHA class, mean arterial pressure, heart rate, Body Mass Index (BMI), estimated Glomerular Filtration Rate (eGFR), history of ischaemic heart disease, atrial fibrillation, Chronic Obstructive Pulmonary Disease (COPD), diabetes, hypertension, anemia, cancer in previous 3 years, valvular disease, device therapy, RAS inhibitors, beta-blockers, diuretics, MRA, digoxin.
Results (Figure 1): The C-statistic, that measures the discriminative ability of a predictive model, was 0.774 (95% confidence interval 0.769 - 0.780) for HFrEF and HFmrEF vs. HFpEF, 0.756 (95% CI 0.752 – 0.761) for HFrEF vs. HFmrEF and HFpEF, 0.683 (95% CI 0.677 - 0.688) for HFpEF vs. HFmrEF vs. HFrEF. Strongest predictors for the HFpEF phenotype were older age, female sex, presence of hypertension, atrial fibrillation and anemia, while those for HFrEF were presence of a cardiac device (implantable cardioverter defibrillator or cardiac resynchronization therapy), increased levels of NT-proBNP, use of RAS-inhibitors and beta-blockers.
Conclusion: Baseline patient characteristics from EHR could be used to identify different EF phenotypes in HF. Both logistic models categorizing HFmrEF with either HFrEF or HFpEF performed better than the multinomial model. This may be explained by different phenotypes coexisting within HFmrEF and part of HFmrEF patients having transitioning EF.
Methods: We included 42,061 HF patients from the Swedish Heart Failure Registry that collects data on EF and biomarkers. For the primary analysis we performed two logistic regression models to predict 1) HFrEF and HFmrEF vs. HFpEF; and 2) HFrEF vs. HFmrEF and HFpEF. In the secondary analysis we performed a multivariable multinomial logistic regression to create a prediction model for all 3 separate HF phenotypes: HFrEF vs. HFmrEF vs. HFpEF. The models included 22 predictors: age, sex, NT-proBNP, NYHA class, mean arterial pressure, heart rate, Body Mass Index (BMI), estimated Glomerular Filtration Rate (eGFR), history of ischaemic heart disease, atrial fibrillation, Chronic Obstructive Pulmonary Disease (COPD), diabetes, hypertension, anemia, cancer in previous 3 years, valvular disease, device therapy, RAS inhibitors, beta-blockers, diuretics, MRA, digoxin.
Results (Figure 1): The C-statistic, that measures the discriminative ability of a predictive model, was 0.774 (95% confidence interval 0.769 - 0.780) for HFrEF and HFmrEF vs. HFpEF, 0.756 (95% CI 0.752 – 0.761) for HFrEF vs. HFmrEF and HFpEF, 0.683 (95% CI 0.677 - 0.688) for HFpEF vs. HFmrEF vs. HFrEF. Strongest predictors for the HFpEF phenotype were older age, female sex, presence of hypertension, atrial fibrillation and anemia, while those for HFrEF were presence of a cardiac device (implantable cardioverter defibrillator or cardiac resynchronization therapy), increased levels of NT-proBNP, use of RAS-inhibitors and beta-blockers.
Conclusion: Baseline patient characteristics from EHR could be used to identify different EF phenotypes in HF. Both logistic models categorizing HFmrEF with either HFrEF or HFpEF performed better than the multinomial model. This may be explained by different phenotypes coexisting within HFmrEF and part of HFmrEF patients having transitioning EF.
| Original language | English |
|---|---|
| Pages (from-to) | 215-215 |
| Journal | European Journal of Heart Failure |
| Volume | 21 |
| Publication status | Published - May 2019 |