AuricGrid-Laboratory's picture
Rename README cardiovascular hf dataset.md to README.md
6c2698b verified
|
Raw
History Blame Contribute Delete
11.4 kB

README — Cardiovascular Heart Failure Dataset

File: cardiovascular_hf_dataset-3.csv
Rows: 50,000 patients
Columns: 68 features
Domain: Cardiovascular medicine — Heart Failure risk prediction


Overview

This dataset contains multimodal patient-level data for heart failure (HF) risk modelling. Each row represents one patient encounter with demographic, clinical, laboratory, echocardiographic, symptomatic, pharmacological, and longitudinal variables. Three outcome targets are provided: a binary HF risk label, a future hospitalisation probability, and a 12-month survival probability.


Data Sources

Source N %
Preventive_Cardiology 20,094 40.2%
Hospital_Inpatient 15,052 30.1%
Annual_Checkup 14,854 29.7%

Column Dictionary

1. Patient Identifiers

Column Type Description
Patient_ID str Unique patient identifier (PT000001 – PT050000)
Source str Clinical setting of data collection

2. Demographics

Column Type Range / Categories Description
Age int 18 – 90 Age in years
Gender str Female, Male, Other Self-reported gender
Ethnicity str White, Black, Hispanic, Asian, Native_American, Other Self-reported ethnicity

3. Lifestyle & Socioeconomic

Column Type Range Description
BMI float 14.0 – 58.0 Body mass index (kg/m²)
Activity_Level str Sedentary / Low / Moderate / High / Very_High Self-reported physical activity category
SES_Score float 1.0 – 10.0 Socioeconomic status composite score
SES_Category str Low / Middle / High Derived SES category
Smoking_Status str Never / Former / Current Tobacco smoking history
Alcohol_Units_Week float 0.2 – 50.0 Alcohol consumption (units/week)
Physical_Activity_Min_Week float 0 – 448 Weekly moderate-intensity exercise (minutes); ~5% missing
Sleep_Hours float 3.0 – 11.9 Average nightly sleep duration; ~4.9% missing
Diet_Score float 1.0 – 10.0 Dietary quality composite score; ~5% missing
Stress_Score float 1.0 – 10.0 Perceived stress score; ~4.9% missing

4. Comorbidities (binary flags: 0 = absent, 1 = present)

Column Prevalence
Hypertension 44.8%
Diabetes 28.4%
CAD (Coronary Artery Disease) 23.9%
Prior_MI (Myocardial Infarction) 4.9%
Arrhythmia 19.8%
CKD (Chronic Kidney Disease) 20.5%
Family_HF_History 24.0%

5. Vital Signs

Column Type Range Description
Heart_Rate_bpm int 38 – 135 Resting heart rate (bpm)
SBP_mmHg int 72 – 210 Systolic blood pressure (mmHg)
DBP_mmHg int 43 – 132 Diastolic blood pressure (mmHg)
Respiratory_Rate int 8 – 33 Respiratory rate (breaths/min)
SpO2_pct float 87.7 – 100.0 Peripheral oxygen saturation (%)

6. Biomarkers & Laboratory Values (~9–10% missing)

Column Unit Description
BNP_pgmL pg/mL B-type natriuretic peptide — primary HF biomarker
NT_proBNP_pgmL pg/mL N-terminal pro-BNP — HF severity marker
Troponin_I_ngmL ng/mL Cardiac troponin I — myocardial injury marker
HbA1c_pct % Glycated haemoglobin — glycaemic control
LDL_mgdL mg/dL Low-density lipoprotein cholesterol
HDL_mgdL mg/dL High-density lipoprotein cholesterol
Triglycerides_mgdL mg/dL Serum triglycerides
Creatinine_mgdL mg/dL Serum creatinine — renal function
eGFR_mLmin mL/min Estimated glomerular filtration rate
Sodium_mEqL mEq/L Serum sodium — electrolyte
Potassium_mEqL mEq/L Serum potassium — electrolyte
CRP_mgL mg/L C-reactive protein — systemic inflammation

7. Electrocardiographic (ECG) Variables

Column Type Description
QTc_Interval_ms int Corrected QT interval (ms); range 340–574
QRS_Duration_ms int QRS complex duration (ms); range 60–171
ST_Abnormality binary ST-segment abnormality present (22.0%)
LVH_Flag binary Left ventricular hypertrophy on ECG (29.4%)
AF_Presence binary Atrial fibrillation detected (26.5%)

8. Echocardiographic Variables (~10% missing)

Column Unit Description
Ejection_Fraction_pct % Left ventricular ejection fraction (15–80%)
LV_Wall_Thickness_mm mm Left ventricular wall thickness (6–18.4 mm)
Cardiomegaly binary Cardiomegaly present (25.6%)
Echo_Abnormalities_Score 0–5 Composite echocardiographic abnormality score

9. Symptoms (binary flags)

Column Prevalence Description
Dyspnea 35.4% Shortness of breath
Fatigue 36.2% Fatigue or generalised weakness
Chest_Discomfort 29.1% Chest pain or pressure
Peripheral_Edema 31.9% Ankle/leg swelling
Exercise_Intolerance 36.3% Reduced exercise capacity
Dizziness 21.1% Lightheadedness or dizziness
Symptom_Burden_Count int (0–6) Total number of symptoms present (mean 1.9)

10. Medications (binary flags: 0 = not prescribed, 1 = prescribed)

Column Prevalence
ACE_Inhibitor 30.5%
Beta_Blocker 33.9%
Diuretic 28.6%
Statin 37.0%
Anticoagulant 28.1%

11. Healthcare Utilisation

Column Type Range Description
Prior_Hospitalizations int 0 – 9 Number of prior hospital admissions
ED_Visits_Per_Year int 0 – 9 Emergency department visits in the past year

12. Longitudinal / Trend Variables (~17–18% missing)

Column Unit Description
Weight_Fluctuation_3mo_kg kg Weight change over 3 months (− = loss, + = gain)
BP_Trend_mmHg_month mmHg/month Blood pressure trend over recent months
HRV_ms ms Heart rate variability (5–88 ms; higher = better autonomic tone)

13. Outcome Variables

Column Type Description
HF_Risk_Binary 0 / 1 Primary label — heart failure risk (26.1% positive)
Future_Hospitalization_Risk float [0–1] Predicted probability of future hospitalisation
Survival_Probability_12mo float [0–1] Predicted probability of survival at 12 months

Class imbalance note: HF positive cases represent 26.1% of the dataset. Consider SMOTE, class weighting, or stratified sampling before model training.


Missing Data Summary

Missingness tier Columns Approx. missing
Low (~5%) Physical activity, sleep, diet, stress scores 4.9–5.0%
Moderate (~9–10%) All biomarkers, echocardiographic variables 9.2–10.4%
High (~17–18%) Weight fluctuation, BP trend, HRV 17.6–18.1%

Recommended imputation strategies: median imputation or MICE for continuous variables; mode imputation for binary flags. Missing echo variables may reflect patients who did not undergo echocardiography (MCAR/MAR assumption should be validated).


Suggested Use Cases

  • Binary classification: HF risk prediction (HF_Risk_Binary)
  • Regression: 12-month survival probability or future hospitalisation risk
  • Survival analysis: time-to-event modelling using Survival_Probability_12mo
  • Missing data benchmarking: graduated missingness across feature tiers
  • Multi-source generalisation studies: three distinct clinical settings
  • Feature importance analysis: identifying key biomarker and symptom drivers of HF

Notes

  • All Patient_IDs are unique (no duplicate rows).
  • The dataset is synthetic or de-identified; do not attempt patient re-identification.
  • Echocardiographic variables (Ejection_Fraction_pct, LV_Wall_Thickness_mm, Cardiomegaly, Echo_Abnormalities_Score) share a correlated missingness pattern, suggesting they are collected together as part of an echo workup.
  • BNP and NT-proBNP are highly correlated; consider using only one in models to avoid multicollinearity.
  • Age range spans 18–90 years; consider age-stratified analyses for paediatric-excluded adult cohorts.

README generated for cardiovascular_hf_dataset-3.csv — 50,000 rows × 68 columns.