An interpretable integer score for angiographic coronary artery disease in the Cleveland cohort: derivation and cross-validated comparison with machine-learning models

Background: Machine-learning classifiers are often proposed for the non-invasive diagnosis of coronary artery disease (CAD), but they are rarely compared with simple scores that clinicians can compute by hand. Objective: To compare a transparent integer points score with six alternative models for the diagnosis of angiographic CAD. Methods: The public Cleveland Clinic angiography cohort was used (303 patients, 138 with at least 50% stenosis in a major coronary artery). Nine clinically dichotomised items were defined (male sex, age ≥55 years, asymptomatic chest pain, exerciseinduced angina, ST depression ≥1 mm, flat or downsloping ST segment, maximum heart rate <140 bpm, at least one vessel visible on fluoroscopy, and a thallium perfusion defect), weighted by rounding twice their logistic coefficients, and recalibrated to absolute risk. The score was compared with a unit-weight item count, logistic regression with all 13 predictors, L1-penalised logistic regression, random forest, gradient boosting and a support vector machine. Evaluation used 20 repeats of stratified five-fold cross-validation, with all derivation steps confined to training folds. Results: The points score (range 0–19) achieved a mean area under the receiver operating characteristic curve (AUC) of 0.915, a Brier score of 0.113 and a calibration slope of 0.88. Full logistic regression performed similarly (AUC 0.912), whereas random forest (0.902), support vector machine (0.894) and gradient boosting (0.887) had lower AUC than the score in every one of the 20 repeats. Observed CAD prevalence rose from 0% (0/30) at 02 points to 96.8% (60/62) at ≥15 points. Decision curve analysis showed net benefit for the score comparable to or greater than that of the ensemble models across threshold probabilities of 0.10–0.70. The score was well calibrated in both women (observed-to-expected ratio 1.00) and men (1.00), whereas the random forest over-predicted risk in women. Journal Healthcare Treatment Development (JHTD) Journal homepage: https://journal.hmjournals.com/index.php/JHTD Journal Healthcare Treatment Development (JHTD) ISSN: 2799-1148  2 Conclusions: In this dataset, a transparent 19-point score performed at least as well as machine-learning ensembles while offering advantages in interpretability and bedside use. External validation in contemporary cohorts is required before clinical application.

Authors

Publication Details

Journal
Journal Healthcare Treatment Development
Published
2026-11-18
DOI
https://doi.org/10.55529/jhtd.62.1.14
Primary Topic
Artificial Intelligence in Healthcare
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

An interpretable integer score for angiographic coronary artery disease in the Cleveland cohort: derivation and cross-validated comparison with machine-learning models

Samidin Begaly uulu, Matsakov Kanybek, Inamzhanova Gulnara
Journal Healthcare Treatment Development
Artificial Intelligence in Healthcare
article

An interpretable integer score for angiographic coronary artery disease in the Cleveland cohort: derivation and cross-validated comparison with machine-learning models

Samidin Begaly uulu, Matsakov Kanybek, Inamzhanova Gulnara
article en

Abstract

Background: Machine-learning classifiers are often proposed for the non-invasive diagnosis of coronary artery disease (CAD), but they are rarely compared with simple scores that clinicians can compute by hand. Objective: To compare a transparent integer points score with six alternative models for the diagnosis of angiographic CAD. Methods: The public Cleveland Clinic angiography cohort was used (303 patients, 138 with at least 50% stenosis in a major coronary artery). Nine clinically dichotomised items were defined (male sex, age ≥55 years, asymptomatic chest pain, exerciseinduced angina, ST depression ≥1 mm, flat or downsloping ST segment, maximum heart rate <140 bpm, at least one vessel visible on fluoroscopy, and a thallium perfusion defect), weighted by rounding twice their logistic coefficients, and recalibrated to absolute risk. The score was compared with a unit-weight item count, logistic regression with all 13 predictors, L1-penalised logistic regression, random forest, gradient boosting and a support vector machine. Evaluation used 20 repeats of stratified five-fold cross-validation, with all derivation steps confined to training folds. Results: The points score (range 0–19) achieved a mean area under the receiver operating characteristic curve (AUC) of 0.915, a Brier score of 0.113 and a calibration slope of 0.88. Full logistic regression performed similarly (AUC 0.912), whereas random forest (0.902), support vector machine (0.894) and gradient boosting (0.887) had lower AUC than the score in every one of the 20 repeats. Observed CAD prevalence rose from 0% (0/30) at 02 points to 96.8% (60/62) at ≥15 points. Decision curve analysis showed net benefit for the score comparable to or greater than that of the ensemble models across threshold probabilities of 0.10–0.70. The score was well calibrated in both women (observed-to-expected ratio 1.00) and men (1.00), whereas the random forest over-predicted risk in women. Journal Healthcare Treatment Development (JHTD) Journal homepage: https://journal.hmjournals.com/index.php/JHTD Journal Healthcare Treatment Development (JHTD) ISSN: 2799-1148  2 Conclusions: In this dataset, a transparent 19-point score performed at least as well as machine-learning ensembles while offering advantages in interpretability and bedside use. External validation in contemporary cohorts is required before clinical application.

Journal Healthcare Treatment DevelopmentVol. 6(2)
Artificial Intelligence in Healthcare
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon