Auric-Grid/E1.M1-Atrial-Fibrillation-Risk-Prediction-Older-Adults

AFib Synthetic Dataset — README

Overview

This is a synthetic clinical dataset of 10,000 patient records designed to support research and machine learning work on Atrial Fibrillation (AFib) risk prediction. Each row represents one patient and includes demographic, clinical, lifestyle, lab, and wearable-derived features, along with a binary AFib diagnosis label.

Important: All data is synthetically generated and does not represent real patients.

---

File

Property	Value
Filename	afib_synthetic_dataset_10k.csv
Format	CSV (comma-separated)
Rows	10,000 patients
Columns	34 features
Target variable	Atrial_Fibrillation_Diagnosis

---

Target Variable

Column	Type	Description
Atrial_Fibrillation_Diagnosis	Binary (0/1)	1 = AFib diagnosed; 0 = no AFib

Class distribution: 17.5% positive (1,749 AFib cases), 82.5% negative — a moderately imbalanced classification task.

---

Feature Reference

Identifiers

Column	Type	Description
Patient_ID	String	Unique anonymized patient identifier (e.g., PID-768BEE683C)

Demographics

Column	Type	Values / Range
Age	Integer	55–90 years (mean ≈ 73)
Sex	Categorical	Female (52.5%), Male (47.5%)
Ethnicity	Categorical	White, Hispanic, Black, Asian, Other
BMI	Float	17.0–45.0 kg/m² (mean ≈ 28.5)

Lifestyle Factors

Column	Type	Values
Smoking_Status	Categorical	Never, Former, Current
Alcohol_Consumption	Categorical	Moderate, Heavy; 39.8% missing
Physical_Activity_Level	Categorical	Sedentary, Low, Moderate, Active
Stress_Level	Float	1–10 scale (mean ≈ 5.5); 4.2% missing

Comorbidities (Binary 0/1)

Column	Prevalence	Description
Sleep_Apnea	7.8%	Diagnosed sleep apnea
Hypertension	7.8%	Diagnosed hypertension
Diabetes	3.1%	Diagnosed diabetes
Coronary_Artery_Disease	3.3%	Diagnosed CAD
Prior_Stroke	1.1%	History of stroke
Family_History_CVD	38.1%	Family history of cardiovascular disease

Medications

Column	Type	Values
Medication_Use	Categorical	Antihypertensive, Statin, Anticoagulant, Multiple; 48.5% missing

Vital Signs & Labs

Column	Type	Range	Missing
Resting_Heart_Rate	Integer (bpm)	45–108	None
Systolic_BP	Integer (mmHg)	90–182	None
Diastolic_BP	Integer (mmHg)	55–115	None
Cholesterol_Total	Float (mg/dL)	120–309	7.7%
HDL	Float (mg/dL)	25–90	7.1%
LDL	Float (mg/dL)	50–244	9.3%
Triglycerides	Float (mg/dL)	50–328	7.0%
Blood_Glucose	Float (mg/dL)	65–209	9.1%

Cardiac Assessments

Column	Type	Description	Missing
ECG_Abnormalities	Binary (0/1)	Abnormal ECG findings (12.9% positive)	None
Heart_Rate_Variability	Float (ms)	HRV from clinical measurement, 10–106 ms	5.2%
AFib_Risk_Score	Float	Composite risk score, 0–7.9 (mean ≈ 2.1)	None

Symptoms

Column	Type	Values
Episodes_of_Palpitations	Categorical	Never, Rare, Occasional, Frequent
Dizziness_History	Binary (0/1)	7.9% positive
Chest_Discomfort	Binary (0/1)	5.1% positive

Wearable Device Data

Column	Type	Description	Missing
Wearable_Device_Data_Available	Binary (0/1)	57.5% of patients have wearable data	None
Average_Daily_Steps	Float	Steps/day, 500–12,488 (mean ≈ 5,301)	45.7% (patients without wearables)
HRV_Wearable_Trend	Categorical	Stable, Improving, Declining	42.5% (patients without wearables)

---

Missing Data Summary

Several columns have substantial missing rates, often by design (e.g., wearable columns are only populated when Wearable_Device_Data_Available = 1).

Column	Missing (%)	Likely Reason
Average_Daily_Steps	54.3%	No wearable device
HRV_Wearable_Trend	42.5%	No wearable device
Medication_Use	48.5%	Not on tracked medication
Alcohol_Consumption	39.8%	Not reported
Cholesterol_Total / LDL / etc.	7–9%	Lab not recorded
Stress_Level	4.2%	Self-report not collected
Heart_Rate_Variability	5.2%	Clinical HRV not measured

---

Suggested Use Cases

• Binary classification: Predict Atrial_Fibrillation_Diagnosis from patient features.
• Risk score modeling: Predict or validate AFib_Risk_Score as a continuous outcome.
• Missing data handling: Benchmark imputation strategies on realistic clinical missingness patterns.
• Wearable data integration: Explore the added predictive value of Average_Daily_Steps and HRV_Wearable_Trend as a subgroup analysis.
• Fairness/bias analysis: Examine model performance across Sex, Ethnicity, and age groups.

---

Notes

• All values are synthetically generated. Distributions are intended to be clinically plausible but do not reflect any real population.
• The dataset is not suitable for clinical decision-making.
• The AFib positive rate (17.5%) is higher than general population estimates; this may reflect a deliberate enrichment of at-risk patients.