xpertsystems/hccar005-sample

HCCAR005 — Synthetic Coronary Artery Disease Dataset (Sample Preview)

XpertSystems.ai | Synthetic Data Factory | Healthcare / Cardiology Vertical

A longitudinal coronary artery disease (CAD) patient dataset spanning the full spectrum from subclinical disease through acute coronary syndromes through post-revascularization follow-up. 150 patients across 7 CAD stages (Subclinical, Stable Angina, Unstable Angina, NSTEMI, STEMI, Post-PCI, Post-CABG) followed annually for 10 years — yielding 1,500 visit-level records with 140 features per row covering:

• CAD anatomy (3-vessel stenosis %, FFR per vessel + iFR, SYNTAX score, plaque burden, lesion length, MLA, Agatston calcium score, plaque type including TCFA)
• Angina assessment (CCS class, angina frequency/duration/trigger, nitroglycerin response, NYHA functional class, Seattle Angina Questionnaire 5 domains, Duke treadmill score, stress test results
  • modality)
• ACS events (door-to-balloon, TIMI flow pre/post, thrombus burden, Killip class, GRACE score, TIMI risk score)
• Biomarkers (Troponin I/T, CK-MB, BNP/NT-proBNP, CRP, visit-level LDL/HDL/Trig with statin effect)
• Interventions (PCI with DES_EES/BMS, stent type/length/diameter, num_stents, post-PCI FFR/MLA, contrast volume, radiation dose; OR CABG with graft count, LIMA usage, pump time, cross-clamp time; procedural success flag)
• Imaging (echo LVEF/LVEDV/LVESV, RWMA + territory, E/e', LAVI; CCTA plaque volume + napkin-ring sign; nuclear stress SSS/SDS)
• Medications (DAPT with P2Y12 selection and duration, statin intensity, beta-blocker, ACEi/ARB, SGLT2i, PCSK9i, anticoagulant)
• Outcomes (MACE flag + component, time-to-MACE, target vessel revascularization, in-stent restenosis, graft patency, 30-day readmission, CV death, all-cause mortality, LVEF change)

Calibrated benchmark-first against ACC/AHA Stable CAD Guidelines (Fihn et al.), SYNTAX Trial (Mohr et al., Serruys et al.), COURAGE (Boden et al.), ISCHEMIA Trial (Maron et al. 2020), FREEDOM Trial (Farkouh et al.), 4th Universal Definition of MI (2018), Killip-Kimball (1967), NCDR ACTION + CathPCI Registries, STS Adult Cardiac Surgery Database, GRACE Registry (Granger et al. 2003), TIMI Risk Score (Antman et al. 2000), Seattle Angina Questionnaire (Spertus et al. 1995), and KDIGO 2012 CKD staging.

This is the sample preview — 150 patients × 10 annual visits over 10 years (1,500 visit records, ~1.1 MB). The full product covers 10,000+ patients with extended procedural detail, full medication titration trajectories, multi-imaging modality co-occurrence, and pre-built scenario configs for ISCHEMIA replication, FREEDOM DM-CAD cohort, COURAGE invasive vs OMT, EXCEL-style left-main PCI-vs-CABG, and BIOFLOW-V stent comparison studies.

---

Dataset summary

Table	Rows (sample)	What it contains
hccar005_dataset	1,500	One row per patient × annual visit. 140 features across 8 clinical modules (baseline carried forward + angina + ACS + biomarkers + intervention + imaging + medications + outcomes). 150 unique patients × 10 annual visits each

Provided in CSV and Parquet. Aggregate to patient level via groupby('patient_id') for cross-sectional analysis. Use baseline visit (visit_number == 1) for cohort entry analysis.

---

Calibration sources

All ten validation metrics target named clinical / registry standards:

• ACC/AHA Stable Ischemic Heart Disease Guidelines (Fihn et al. 2012; 2014 Focused Update) — CCS class definitions, GDMT framework
• ACC/AHA STEMI / NSTE-ACS Guidelines (Levine et al. 2015; Amsterdam et al. 2014) — D2B targets, primary PCI criteria
• SYNTAX Trial / Score (Sianos et al. 2005; Mohr et al. 2013) — SYNTAX scoring system, PCI vs CABG decision thresholds (≥33: CABG preferred; 23-32: Heart Team; <22: PCI acceptable)
• COURAGE Trial (Boden et al. 2007) — invasive vs OMT framework
• ISCHEMIA Trial (Maron et al. 2020) — stable CAD invasive vs OMT
• FREEDOM Trial (Farkouh et al. 2012) — DM-CAD revascularization
• EXCEL Trial (Stone et al. 2016) — left main PCI vs CABG
• 4th Universal Definition of MI (Thygesen et al. 2018) — STEMI/ NSTEMI classification, troponin kinetics
• Killip-Kimball (1967) — AMI hemodynamic classification
• GRACE Registry (Granger et al. 2003) — in-hospital mortality prediction, score range [0, 372]
• TIMI Risk Score (Antman et al. 2000) — 0-7 point UA/NSTEMI score
• Seattle Angina Questionnaire (Spertus et al. 1995) — 5-domain patient-reported angina assessment (0-100 scale)
• CCS Functional Classification — angina severity 0-4
• NCDR ACTION + CathPCI — door-to-balloon, stent attribute reporting standards
• STS Adult Cardiac Surgery Database — CABG quality measures, graft count, LIMA usage, pump/cross-clamp times
• EuroSCORE II + STS Mortality Risk — surgical risk stratification
• KDIGO 2012 — CKD eGFR-based staging

---

Validation scorecard (seed = 42)

10/10 PASS · Grade A+ (100%) across all six canonical seeds (42, 7, 123, 2024, 99, 1).

#	Metric	Observed	Target	Tol	Type	Source
1	prior_cabg_flag_equals_post_cabg_stage_rate	1.000	0.99	±0.01	FLOOR	Structural
2	prior_mi_requires_acs_or_post_stage_rate	1.000	0.99	±0.01	FLOOR	ACS history consistency
3	door_to_balloon_in_stemi_only_rate	1.000	0.99	±0.01	FLOOR	NCDR ACTION
4	d2b_met_flag_matches_d2b_under_90_rate	1.000	0.99	±0.01	FLOOR	ACC/AHA STEMI
5	pci_stent_attributes_consistent_with_arm_rate	1.000	0.99	±0.01	FLOOR	NCDR CathPCI
6	cabg_attributes_consistent_with_arm_rate	1.000	0.99	±0.01	FLOOR	STS Database
7	hfref_flag_matches_lvef_under_40_rate	0.999	0.99	±0.01	FLOOR	ACC/AHA HF
8	cv_death_implies_mortality_rate	1.000	0.99	±0.01	FLOOR	Survival monotonicity
9	mace_component_matches_mace_flag_rate	1.000	0.99	±0.01	FLOOR	MACE composite
10	risk_scores_in_published_ranges_rate	1.000	0.99	±0.01	FLOOR	Multiple guidelines

---

Schema highlights (140 cols total)

Identity & visit (6 cols)

patient_id (HC-CAR-XXXXXXX), site_id, visit_number (1-10), visit_date, age_at_visit, years_from_baseline.

Patient baseline (49 cols)

cad_stage (Subclinical / StableAngina / UnstableAngina / NSTEMI / STEMI / PostPCI / PostCABG), sex, age_at_baseline, bmi, systolic_bp_mmhg, heart_rate_bpm, smoking_history, comorbidities (diabetes_flag, hypertension_flag, hyperlipidemia_flag, ckd_flag, ckd_stage, heart_failure_flag, afib_flag, pad_flag, prior_mi_flag, prior_pci_flag, prior_cabg_flag), egfr_ml_min_1_73m2, creatinine_mg_dl, baseline lipids (ldl_mg_dl, hdl_mg_dl, triglycerides_mg_dl, lp_a_nmol_l, hba1c_pct, hemoglobin_g_dl), CAD anatomy (num_vessels_diseased, lm_disease_flag, syntax_score, culprit_vessel, stenosis_pct_lad, stenosis_pct_lcx, stenosis_pct_rca, ffr_lad, ffr_lcx, ffr_rca, ifr_value, plaque_burden_pct, lesion_length_mm, reference_vessel_diameter_mm, mla_mm2, calcium_score_agatston, plaque_type, annual_stenosis_progression_pct), intervention_arm (OMT / PCI_BMS / PCI_DES / CABG / Hybrid), euroscore_ii, sts_score_mortality_pct.

Angina (16 cols)

angina_class_ccs (0-4), angina_type (Stable / Unstable / Silent / Mixed), angina_frequency_per_week, angina_duration_min, angina_trigger, nitroglycerin_response, dyspnea_nyha_class (1-4), ischemic_burden_pct_lv, stress_test_result, duke_treadmill_score, stress_test_modality, SAQ 5 domains (saq_physical_limitation, saq_angina_stability, saq_angina_frequency, saq_treatment_satisfaction, saq_quality_of_life).

ACS (10 cols)

acs_type (None / UA / NSTEMI / STEMI), symptom_onset_to_door_min, door_to_balloon_min, door_to_balloon_met_flag, thrombus_burden, timi_flow_pre (0-3), timi_flow_post (0-3), killip_class (1-4), grace_score (0-372), timi_risk_score (0-7).

Biomarkers (9 cols)

troponin_i_ng_ml, troponin_t_ng_ml, ck_mb_ng_ml, bnp_pg_ml, nt_probnp_pg_ml, crp_mg_l, ldl_mg_dl_visit, hdl_mg_dl_visit, triglycerides_mg_dl_visit.

Intervention (15 cols)

intervention_type, pci_target_vessel, stent_type (DES_EES / BMS), stent_length_mm, stent_diameter_mm, post_pci_ffr, post_pci_mla_mm2, num_stents_deployed, total_stent_length_mm, cabg_grafts, lima_used_flag, cabg_pump_time_min, cabg_xclamp_time_min, contrast_volume_ml, radiation_dose_kerma_mgy, procedural_success_flag.

Imaging (13 cols)

echo_lvef_pct, echo_lv_edv_ml, echo_lv_esv_ml, echo_rwma_flag, echo_rwma_territory, echo_e_e_prime_ratio, echo_lavi_ml_m2, lvef_hfref_flag, ccta_plaque_volume_mm3, ccta_napkin_ring_flag, nuclear_sss, nuclear_sds, nuclear_lvef_stress_pct.

Medications (12 cols)

aspirin_flag, p2y12_inhibitor (Ticagrelor / Clopidogrel / Prasugrel / None), dapt_duration_months, statin_flag, statin_intensity (None / Low / Moderate / High), beta_blocker_flag, ace_arb_flag, sglt2_inhibitor_flag, pcsk9_inhibitor_flag, nitrate_use_flag, anticoagulant_use, medication_adherence_pct.

Outcomes (11 cols)

mace_event_flag, mace_component (MI / Stroke / CV_Death / HF_Hospitalization / None), time_to_mace_days, target_vessel_revascularization_flag, in_stent_restenosis_flag, graft_patency_flag, hospitalization_cv_flag, readmission_30d_flag, mortality_flag, cv_death_flag, lvef_change_pct.

---

Suggested use cases

• SYNTAX Score → revascularization strategy ML — train a Heart- Team-style classifier (PCI vs CABG vs OMT) from SYNTAX score, LM involvement, comorbidities, EuroSCORE II, STS score
• FFR / iFR-guided PCI candidate selection — classifier for significant ischemia (FFR ≤ 0.80) from angiographic features
• CCTA plaque characterization ML — predict TCFA (TCFA flag in plaque_type) and napkin-ring sign from CCTA volume features
• In-stent restenosis prediction — classifier for in_stent_restenosis_flag from stent characteristics, lesion features, DM status (DES vs BMS comparison)
• Door-to-balloon prediction & quality improvement — predict D2B time from arrival pattern features; useful for NCDR ACTION quality benchmarking
• GRACE / TIMI risk score validation — train ML to reproduce or improve published risk models
• DAPT duration optimization — uplift modeling for prolonged vs short DAPT given DAPT score, bleeding risk, stent type
• MACE survival ML — Cox / random survival forest on mace_event_flag + time_to_mace_days with right-censoring
• CABG graft patency prediction — model graft_patency_flag from LIMA usage, pump time, baseline LVEF
• HFrEF post-MI prediction — classifier for lvef_hfref_flag from baseline + intervention features
• Statin response prediction — model ldl_mg_dl_visit from baseline LDL + statin intensity (50% reduction for non-OMT vs 15% for OMT in this generator)
• PCSK9i candidate identification — predict pcsk9_inhibitor_flag prescribing patterns for population health intervention
• SAQ-based outcome prediction — train regressors for the 5 SAQ domains (physical limitation, frequency, stability, treatment satisfaction, QoL) from clinical features
• Procedural success prediction — classifier for procedural_success_flag in PCI (post-PCI FFR ≥ 0.80) vs CABG
• Cardio-renal-metabolic phenotyping — unsupervised clustering on comorbidity + biomarker patterns
• ISCHEMIA / COURAGE cohort simulation — filter to specific eligibility criteria (stable angina, no LM disease, etc.) and simulate trial cohorts

---

Loading examples

from datasets import load_dataset

ds = load_dataset("xpertsystems/hccar005-sample", split="train")
print(ds.shape)

import pandas as pd
from huggingface_hub import hf_hub_download

df = pd.read_parquet(hf_hub_download(
    "xpertsystems/hccar005-sample", "hccar005_dataset.parquet",
    repo_type="dataset",
))

# Patient-level cohort distribution
print(df.drop_duplicates("patient_id")["cad_stage"]
      .value_counts(normalize=True).round(3))

# SYNTAX score → revascularization strategy
import pandas as pd
from huggingface_hub import hf_hub_download

df = pd.read_parquet(hf_hub_download(
    "xpertsystems/hccar005-sample", "hccar005_dataset.parquet",
    repo_type="dataset",
))

patients = df.drop_duplicates("patient_id")

# Heart Team-style decision validation
syntax_tier = pd.cut(patients["syntax_score"],
                     bins=[0, 22, 32, 60],
                     labels=["Low (<23)", "Intermediate (23-32)", "High (≥33)"])
print(pd.crosstab(syntax_tier, patients["intervention_arm"], normalize="index").round(2))

# DES vs BMS in-stent restenosis comparison
import pandas as pd
from huggingface_hub import hf_hub_download

df = pd.read_parquet(hf_hub_download(
    "xpertsystems/hccar005-sample", "hccar005_dataset.parquet",
    repo_type="dataset",
))

# First-year PCI cohort
pci_v1 = df[(df["intervention_arm"].isin(["PCI_DES", "PCI_BMS"])) & (df["visit_number"] == 1)]
print("ISR rate by stent type:")
print(pci_v1.groupby("intervention_arm").agg(
    n=("patient_id", "count"),
    isr_rate_pct=("in_stent_restenosis_flag", lambda x: x.mean() * 100),
    mean_stent_length=("stent_length_mm", "mean"),
    procedural_success_pct=("procedural_success_flag", lambda x: x.mean() * 100),
).round(2))

# Seattle Angina Questionnaire (SAQ) by CCS class
import pandas as pd
from huggingface_hub import hf_hub_download

df = pd.read_parquet(hf_hub_download(
    "xpertsystems/hccar005-sample", "hccar005_dataset.parquet",
    repo_type="dataset",
))

saq_cols = ["saq_physical_limitation", "saq_angina_frequency",
            "saq_angina_stability", "saq_treatment_satisfaction",
            "saq_quality_of_life"]
print("SAQ domains by CCS class (mean):")
print(df.groupby("angina_class_ccs")[saq_cols].mean().round(1))

# MACE event analysis (aggregate to patient-level)
import pandas as pd
from huggingface_hub import hf_hub_download

df = pd.read_parquet(hf_hub_download(
    "xpertsystems/hccar005-sample", "hccar005_dataset.parquet",
    repo_type="dataset",
))

# Per-patient any-MACE flag over follow-up
patient_outcomes = df.groupby("patient_id").agg(
    any_mace=("mace_event_flag", "max"),
    any_mortality=("mortality_flag", "max"),
    cv_death=("cv_death_flag", "max"),
    arm=("intervention_arm", "first"),
    syntax=("syntax_score", "first"),
)
print("MACE rates by intervention arm:")
print(patient_outcomes.groupby("arm").agg(
    n=("any_mace", "count"),
    any_mace_pct=("any_mace", lambda x: x.mean() * 100),
    mortality_pct=("any_mortality", lambda x: x.mean() * 100),
    mean_syntax=("syntax", "mean"),
).round(2))

---

Limitations and honest disclosures

This sample is calibrated for structural fidelity, not bit-exact reproduction of any specific CAD registry archive. Specifically:

• Visit-level outcomes (MACE, mortality, ISR, graft patency, readmission) are FRESH RANDOM SAMPLES per visit, NOT cumulative carry-forward. The same patient can have mace_event_flag=1 at visit 3 and mace_event_flag=0 at visit 7 (with the visit 3 event implicitly recovered from). For patient-level event analysis, use groupby('patient_id').max() on the binary outcome flags.
• MACE per-visit rate (~13-14%) compounds over 10 visits to very high cumulative rates — patient-level any-MACE will exceed real-world CAD cohort 5-year MACE (~15-25%). Disclosed; for absolute-rate calibration use the full product or scale down per-visit hazard.
• Imaging (echo LVEF, RWMA, CCTA, nuclear stress) is computed for ALL visits regardless of clinical indication. In real practice, serial imaging is reserved for clinical change or pre-procedure planning. Treat as "what the result would be if imaging were performed."
• Patient baseline is FIXED at visit 1 (cad_stage, comorbidities, intervention_arm, baseline lipids, baseline anatomy). The generator does NOT model CAD progression to higher-stenosis or stage transitions longitudinally. For genuine CAD progression ML, augment with a trajectory model.
• ACS events fire ONLY at visit 1 (the index visit). The generator does NOT model NEW ACS events at later visits — every visit_number > 1 has acs_type = 'None'. For longitudinal ACS incidence ML, use the full product or augment with a recurrent-event model.
• Stent fields are populated ONLY at visit 1 for PCI patients. They are NOT carried forward to follow-up visits — stent_type, stent_length_mm, num_stents_deployed, post_pci_ffr are all NaN at visits 2-10 even for PCI patients. For longitudinal PCI follow-up modeling, join the visit-1 stent data to all subsequent visits manually.
• CABG fields similarly populated only at visit 1, and the Hybrid intervention arm goes through the PCI path in the generator (so cabg_grafts is NaN for Hybrid patients despite the arm label including "CABG").
• The generator has a hasattr(p, 'angina_class_ccs') check in the imaging module (line 505) that ALWAYS returns False because p is a dict (not an object with attributes). So nuclear_sss calculation never incorporates CCS — it always falls through to the default N(10, 6) distribution. Disclosed; if SSS-vs-CCS correlation matters for your ML, augment.
• eGFR uses a simplified formula — the lambda creatinine = clip(9.5 / egfr, 0.5, 5.0) (line 112) is the INVERSE derivation (creatinine from eGFR, not eGFR from creatinine). It is approximately correct (consistent with simplified CKD-EPI without sex/age/race), but NOT the full published formula. For accurate eGFR research, recompute from creatinine + age + sex + race using the modern 2021 NKF-ASN refit.
• HCCAR005 lacks racial/ethnic information — the generator does not assign race/ethnicity (unlike HCCAR001 / HCCAR003 / HCCAR004). Disparities research will need augmentation.
• GRACE score formula is simplified — the generator uses grace = 20 + age*1.4 + killip*10 + (30 if STEMI) + ck*8 (line 330) as an approximation, NOT the full Granger et al. 2003 logistic regression with all 8 published variables. Values are in the published range [0, 372] but absolute calibration differs from GRACE 2.0. Use for relative risk stratification, not absolute in-hospital mortality probability.
• Statin lipid effect is FIXED at 50% LDL reduction for non-OMT patients and 15% for OMT patients (line 382). Real-world response varies widely (Rosuvastatin 40mg ~55%, Atorvastatin 80mg ~52%, Pravastatin 20mg ~24%). The statin_intensity field (None / Low / Moderate / High) is randomly assigned and NOT linked to the LDL reduction magnitude. For statin response ML, augment with intensity- specific effects.
• PCSK9i prescribing is independent of LDL response in the generator. Real-world PCSK9i is reserved for patients failing to reach LDL goals on maximally tolerated statin + ezetimibe. The generator fires pcsk9_inhibitor_flag at 15% baseline rate if LDL > 100, ignoring statin trial.
• Time-to-MACE is a Weibull sample with shape=1.8, scale=2000 days (line 598), NOT linked to actual visit when MACE was flagged. Use the visit-level mace_event_flag for incident analysis, not time_to_mace_days for survival models.
• CSV serialization converts None to NaN when reading via pd.read_csv default behavior. Use keep_default_na=False or work with the Parquet file (which preserves nullable types correctly).
• ISCHEMIA / COURAGE eligibility is NOT enforced — the generator produces a heterogeneous CAD cohort. Filter to your own inclusion criteria for trial-replication ML.

The full HCCAR005 product addresses these by genuine CAD progression modeling (stenosis evolution, stage transitions), longitudinal stent carry-forward, recurrent ACS event modeling, full CKD-EPI 2021 formula, race/ethnicity assignment with disparities encoding, intensity-specific statin response curves, PCSK9i trial-stepped prescribing, and pre-built scenario configs (ISCHEMIA replication, COURAGE invasive-vs-OMT, FREEDOM DM-CAD, EXCEL left-main PCI-vs-CABG, BIOFLOW-V stent comparison). Contact us for the licensed commercial release.

---

Companion datasets

This is the fifth SKU in our Healthcare / Cardiology vertical. The five-SKU set now covers the full cardiology clinical continuum:

• HCCAR001 Heart Failure Dataset — chronic HF with GDMT and devices
• HCCAR002 Acute MI Dataset — STEMI/NSTEMI/UA with serial troponin kinetics
• HCCAR003 Hypertension Dataset — longitudinal HTN cohort with ABPM, GDMT, MACE
• HCCAR004 Atrial Fibrillation Dataset — CHA2DS2-VASc/HAS-BLED, DOACs, ablation
• HCCAR005 Coronary Artery Disease Dataset (you are here) — full spectrum from subclinical CAD through acute events through revascularization

Pair HCCAR005 + HCCAR002 for acute-on-chronic CAD (HCCAR002 has the serial troponin detail; HCCAR005 has the longitudinal trajectory). Pair HCCAR005 + HCCAR001 for ischemic cardiomyopathy progression. Pair HCCAR005 + HCCAR003 for HTN-driven CAD progression studies.

• Healthcare / Neurology (10 SKUs)
• Insurance & Risk (10 SKUs)
• Energy & Climate (8 SKUs)
• Manufacturing (10 SKUs)
• Oil & Gas (17 SKUs)

For the broader catalog, see https://huggingface.co/xpertsystems

---

Citation

@dataset{xpertsystems_hccar005_sample_2026,
  author       = {XpertSystems.ai},
  title        = {HCCAR005 Synthetic Coronary Artery Disease Dataset (Sample Preview)},
  year         = 2026,
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/xpertsystems/hccar005-sample}
}

---

Contact

• Web: https://xpertsystems.ai
• Email: pradeep@xpertsystems.ai
• Full product catalog: Cardiology (5 SKUs), Neurology (10 SKUs), Insurance & Risk (10 SKUs), Energy & Climate (8 SKUs), Manufacturing (10 SKUs), Oil & Gas (17 SKUs), and more.

Sample License: CC-BY-NC-4.0 (Creative Commons Attribution-NonCommercial 4.0) Full product License: Commercial — please contact for pricing.

Important medical disclaimer: This dataset contains SYNTHETIC patient records only. No data was derived from any real patient, EHR archive, or clinical registry. The dataset is intended for ML model development, benchmarking, and education — NOT for clinical decision support, patient counseling, or medical research conclusions. All clinical thresholds (SYNTAX score tiers, D2B target, HFrEF definition, CCS classification, revascularization criteria) are sourced from published guidelines; users are responsible for verifying against current ACC/AHA/ESC/STS guidelines for clinical applications.