Logistic Regression Analysis of Heart Disease Risk Factors in Erbil, Iraq

Abstract

Heart disease remains a major cause of illness and death. We analyzed hospital records from 297 patients in Erbil (Kurdistan Region) to identify risk determinants and to benchmark maximum-likelihood (MLE) against Bayesian logistic regression for prediction. Candidate predictors were age, systolic and diastolic blood pressure, body mass index (BMI), total cholesterol, smoking, and history of hypertension. The same logistic specification was estimated by MLE and by a Bayesian model with weakly informative priors. Performance was evaluated using AIC/BIC versus WAIC/LOOIC, ROC AUC, Brier score, and calibration; robustness was probed with observation-level bootstrap subsamples at 25%, 50%, and 93% (B = 1,000/1,500/3,500). Per-observation information criteria differed only slightly across methods, indicating comparable expected predictive fit. In the full cohort, the Bayesian model yielded well-calibrated probabilities and coherent uncertainty estimates, with modest discrimination. Age, smoking, cholesterol, BMI, and hypertension history were the most influential variables. Given the near-equivalence in fit and the Bayesian model’s superior handling of uncertainty and diagnostics, we adopt the Bayesian specification as the primary model. These results support pragmatic risk stratification for patients in Erbil and provide a reproducible template for side-by-side evaluation of Bayesian and MLE approaches in similar hospital settings.