Mortality Risk Prediction and Survival Modeling in Heart Failure Using the UCI Heart Failure Clinical Records Dataset

Abstract

Heart failure is one of the single biggest causes of cardiovascular mortality in the world and thus there is a need for powerful statistical tools that allow early identification of the risk and the evaluation of prognosis. In this study, we propose an integrated modelling strategy to simultaneously learn a multivariable logistic regression for the multi-class mortality classification problem and Cox proportional hazards regression for the time-to-event survival problem, using the UCI Heart Failure Clinical Records Dataset (N = 299). A systematic process translating from data preprocessing to exploratory analysis, univariate screening, multivariate modelling and diagnostic validation was followed and eleven clinical covariates were analyzed. Logistic regression identified age (OR = 1.057, p < .001), ejection fraction (OR = 0.932, p < .001), serum creatinine (OR = 1.938, p < .001), and creatinine phosphokinase (OR = 1.000, p = .042) as significant independent predictors of mortality, achieving an ROC-AUC of 0.752. Cox regression confirmed age (HR = 1.048, p < .001), ejection fraction (HR = 0.952, p < .001), serum creatinine (HR = 1.379, p < .001), anaemia (HR = 1.584, p = .034), high blood pressure (HR = 1.609, p = .028), and creatinine phosphokinase (HR = 1.000, p = .026) as significant time-dependent hazard determinants, with a concordance index of 0.741. The dual-model framework illustrates complementary use for diagnosis and prognosis in the context of evidence-based management of heart failure.