Cardiovascular Risk & Heart Failure Cohort (10,000 Patients, 10-Year Follow-Up)

This upload contains the full technical documentation and a free evaluation sample (500 patients, 4,210 annual visit records) — not the full dataset. The complete dataset — 10,000 patients, 77,164 annual visit records, up to 10 years of follow-up — is available at: https://sentineldata.com.ua/dataset/cardiovascular-risk-heart-failure-cohort Sample contents: sample_patients.parquet500 rows sample_visits.parquet4,210 rows sample_outcomes.parquet500 rows sample_true_state.parquet500 rows CSV mirrors of every sample table are included, plus a short guide on opening Parquet files. The sample is the first 500 patients by ID (CARD-00001–CARD-00500), all from the primary-prevention stratum — it does not include secondary-prevention or heart-failure patients. Both of those strata, and the full 10-year outcome history, are only in the complete dataset linked above. Fully synthetic. Contains no real patient data. Every generative parameter is traceable to a published, verified source. 1. Overview Cardiovascular Risk & Heart Failure Cohort is a fully synthetic, longitudinal cohort of 10,000 patients followed for up to 10 years (77,164 annual visit records), built for cardiovascular risk-prediction modeling, heart-failure research, and synthetic-control / methods-development work. Baseline risk is driven by SCORE2 / SCORE2-OP (ESC 2021) and the ASCVD Pooled Cohort Equations (ACC/AHA 2013); annual outcomes are generated by a competing-risk simulation (MI, stroke, CV death, HF hospitalization, non-CV death, loss to follow-up), with heart-failure mortality anchored to MAGGIC and modified by guideline-directed medical therapy (GDMT) hazard ratios. 2. Cohort Design StratumNShareDescription Primary prevention5,50055%No established ASCVD, no heart failure Secondary prevention2,00020%Established ASCVD (prior MI / stroke) Heart failure2,50025%Prevalent HF at enrollment, all NYHA classes (HFpEF 48.4% / HFrEF 41.2% / HFmrEF 10.4%) Enrollment age 40–85 (HF stratum shifted 5–15 years older, per HF epidemiology). Sex: 50.6% women / 49.4% men. Race/ethnicity: 72.3% White NH, 11.5% Black NH, 9.4% Hispanic, 6.8% Other. Baseline risk-factor prevalence: smoking 11.9%, diabetes 19.0%, hypertension 61.9%, atrial fibrillation 3.1%, CKD 14.8%. 3. Risk & Outcome Model Baseline risk: SCORE2 (ages 40–69) and SCORE2-OP (ages ≥70), ESC 2021; ASCVD Pooled Cohort Equations (ACC/AHA 2013, White/Black coefficients) computed alongside for every patient. Annual competing-risk simulation: MI, stroke, CV death, HF hospitalization, non-CV death (national life-table anchored), and loss to follow-up (1.5%/yr, +0.5pp if age ≥75). Secondary prevention: hazard multiplier 2.0 (Antithrombotic Trialists’ Collaboration, Lancet 2009 anchor). Heart failure stratum: MAGGIC-anchored annual mortality (12.1% HFpEF, 14.1% HFrEF) and 18.5%/yr HF hospitalization, modified by GDMT hazard ratios — ARNI 0.84, beta-blocker 0.66, MRA 0.70, SGLT2i 0.87. Medication effects: statin HR 0.78, aspirin HR 0.88 on MI. Risk-factor drift: SBP, lipids, eGFR, HbA1c, BMI, EF trajectories, and NT-proBNP within-person variability all drift annually by age band. 4. Files & Schema FileRowsColsGrain patients.parquet10,00035one row per patient (baseline) visits.parquet77,16417one row per attended annual visit outcomes.parquet10,00012one row per patient (10-year follow-up) true_state.parquet10,0004latent generative truth per patient CSV mirrors of all four tables are included; Parquet is the canonical format. HF-specific fields (subtype, EF, NYHA, NT-proBNP, HF duration, prior HF hospitalization) are structural nulls for the 7,500 non-HF patients — every other field is complete. 5. Headline Outcome Rates (10 Years) MetricPrimarySecondaryHFOverall Any CV event (incl. CV death)17.8%26.4%79.5%34.9% All-cause mortality15.7%16.2%65.9%28.3% First events across the cohort: HF hospitalization 1,488, CV death 1,188, MI 598, stroke 220. Loss to follow-up (censored): 12.8%. Mean follow-up: 7.57 years. 6. Quality Assurance 69 of 69 validation gates pass — null audit, shape, distributions vs. verified targets, ranges, clinical coherence (LDL ≤ TC−HDL; HF subtype ↔ EF bands; NYHA ↔ HF stratum; diabetes ↔ HbA1c; eGFR ↔ CKD; GDMT only in HF; risk ordering HF > secondary > primary), types, competing-risk partition, and outcome consistency. The generator is deterministic: re-running it with master seed 20260909 reproduces bit-identical parquet files (md5-verified). 7. Provenance 58 generative parameters: 37 VERIFIED (live-checked DOI/PMID), 17 ESTIMATE_SOURCED (explicit anchors), 3 MIXED, 0 unsourced — backed by an 84-entry literature registry. Full audit trail (search queries, verification dates) ships in constants_audit.csv. 8. Known Limitations PCE coefficients are validated to age 79; ages 80–85 are extrapolated. Hispanic/Other race groups use White PCE coefficients (standard practice). Annual visit granularity — no within-year timing, no imaging, no free text. Visit count (~77k) is below a nominal 100k because mortality (28.3%) and censoring (12.8%) truncate follow-up and per-visit attendance is 90%/yr — realistic longitudinal behavior, not a defect. Medication effects are applied as constant hazard ratios; no adherence dynamics are modeled. 9. Use Cases Suitable for: cardiovascular risk-prediction modeling, heart-failure outcomes research, competing-risk and survival-analysis methods development, synthetic-control benchmarking, and ML training/evaluation on richly annotated longitudinal data. NOT suitable for: clinical decision-making — this dataset is fully synthetic and contains no real patient data. 10. License & Distribution Format: Apache Parquet + CSV mirrors. Delivery: One-time purchase; download link provided after payment. A free stratified sample is available for evaluation before purchase.