Skip to main content

Table 4 All variables used in the final model

From: Diagnostic signature for heart failure with preserved ejection fraction (HFpEF): a machine learning approach using multi-modality electronic health record data

Variable

HFpEF versus HFrEF

HFpEF versus non-HF

AGE

1

1

AO V2 MAX

1*

0

Aortic Valve Insufficiency

1

0

Atrial Fibrillation

1

1

BMI

1

1

Bisoprolol

0

1

Ca

0

1

Diabetes Mellitus, Non-Insulin-Dependent

1

0

Dilated

0

1*

Disease

1

1

Diuretics

1

1

Dyspnea

0

1*

EF(CUBED)

1*

1

EF(MOD-BP)

1*

0

EF(MOD-SP2)

1

0

EF(MOD-SP4)

1*

0

Edema

0

1

Fasting

1

1*

Full blood count normal

1*

1

Furosemide

1

1*

Furosemide 40 MG

1

1

Furosemide Oral Tablet

0

1

HEIGHT

1

0

Hypertensive disease

1

1

Hypokinesia

1

0

Increase dose

0

1

Kidney Failure, Acute

0

1

LA VOLUME (2D BIPLANE)

1

0

Little LOS

0

1

MV A MAX VEL

0

1

Mobility as a finding

0

1

Never smoked tobacco

0

1

Nitroglycerin

0

1

O/E—blood pressure reading

1

0

Patient address

1

1

Pharmacologic Substance

0

1*

Pitting edema

1

0

Pulmonary Edema

0

1

RR

1

0

SBP

1

1

Severe (severity modifier)

1

0

Swelling of lower limb

1

1

Tachycardia, Ventricular

1

0

WEIGHT

1

0

Risk factors

0

1

  1. The columns “HFpEF versus HFrEF” and “HFpEF versus non-HF” indicate use of a variable in either model where 1 = used and 0 = unused. Asterisks indicate the top 5 most important variables in each model according to SHAP analysis. All variables in upper case are structured features, all other features are derived from NLP