303 patients referred for cardiac evaluation at the Cleveland Clinic Foundation in the late 1980s, with 13 clinical measurements and an angiographic diagnosis.
Single institution, single era (Cleveland Clinic Foundation, late 1980s). A referral population with 45.9% disease prevalence, far above a general population — probabilities are conditioned on having been referred. Predominantly male. No socioeconomic, comorbidity or medication data.
No feature correlates above 0.90 with the target (strongest: thal at +0.538). No duplicate rows, no constant columns, no entity identifier, no temporal ordering. One finding: ca and thal are results of invasive follow-up tests ordered only once disease is suspected. They are not leakage in the strict sense, but a model requiring them is not a screening model — so both are excluded from the served variant.
Version 1.0.0 · a74b7efa387b…
| Feature | Type | Role | Range / values | Missing | Description |
|---|---|---|---|---|---|
| age | Numeric | Feature | 29.0–77.0 years | — | Age at examination. |
| sex | Categorical | Feature | — | — | Recorded sex: 0 female, 1 male. |
| cp | Categorical | Feature | — | — | 1 typical angina, 2 atypical angina, 3 non-anginal pain, 4 asymptomatic. |
| trestbps | Numeric | Feature | 94.0–200.0 mm Hg | — | Resting blood pressure on admission. |
| chol | Numeric | Feature | 126.0–564.0 mg/dl | — | Serum cholesterol. |
| fbs | Categorical | Feature | — | — | 1 if fasting blood sugar exceeds 120 mg/dl. |
| restecg | Categorical | Feature | — | — | 0 normal, 1 ST-T abnormality, 2 probable left ventricular hypertrophy. |
| thalach | Numeric | Feature | 71.0–202.0 bpm | — | Maximum heart rate achieved during exercise testing. |
| exang | Categorical | Feature | — | — | 1 if exercise induced angina. |
| oldpeak | Numeric | Feature | 0.0–7.0 mm | — | ST depression induced by exercise relative to rest. |
| slope | Categorical | Feature | — | — | Slope of the peak exercise ST segment: 1 upsloping, 2 flat, 3 downsloping. |
| ca | Numeric | Excluded | 0.0–3.0 | 4 | Invasive. Excluded from the served model — unavailable in a screening setting. |
| thal | Categorical | Excluded | — | 2 | Invasive. Excluded from the served model — unavailable in a screening setting. |
| num | Integer | Target | 0.0–4.0 | — | Angiographic disease: 0 absent, 1-4 increasing severity. Binarised to presence/absence. |
This dataset is not redistributed here — its licence does not clearly permit it. Obtain it from the original source.
A classification study on 303 referred cardiac patients — and why the best-scoring model is not the one served.
Thirteen clinical measurements, 303 patients, and one question: how much of the diagnosis is visible before any invasive test is ordered? The study compares candidate models against a …
11 min read · Updated 15 Sep 2026