H-HQNN: Hybrid quantum-classical neural network with feature selection by heuristics to predict heart failure.

Cardiovascular diseases are the most dominant cause of mortality across the entire world, claiming 17.9 million lives every year. According to recent trends, the mortality rate as a result of heart failure continues to increase particularly in low- and middle-income areas which means that early diagnostic tools desperately need to be discovered. Conventional biomarkers like NT-proBNP, troponins and creatinine levels are useful, although their sensitivity and specificity are low, particularly in identifying heart failure onset before it is actually clinically apparent. We suggest addressing these issues by proposing the following Hybrid Quantum-Classical Neural Network with Heuristics-Based Feature Selection that can be used to predict the onset of heart failure: H-HQNN: A Hybrid Quantum-Classical Neural Network with Heuristics-Based Feature Selection. Given this, SMOTE is used to balance the clinical data and reduce class imbalance, whereas the most informative cardiac biomarkers and clinical features are selected using a two-stage process in which Recursive Feature Elimination (RFE) first prunes redundant attributes, followed by a dual-phase optimization where PSO refines the feature subset and Cuckoo Search (CS) enhances global exploration to avoid local minima, resulting in a compact and highly relevant feature set. These features are then represented in a variational quantum circuit (4 qubits) and are implemented as ZZFeatureMap, Pauli-Z gates, EfficientSU2, and RealAmplitudes, which are combined in a hybrid quantum-classical neural network. Performance was estimated using stratified 5-fold cross-validation on a single public dataset; external validation on independent cohorts has not yet been performed. The average accuracy of our model is 96.62% ± 3.45 with precision of 96.11, recall of 96.10, F1-score of 96.54 and AUC = 0.9604. The most efficient fold has 98.30% accuracy and AUC = 0.9872 with a confusion matrix of almost perfect balance in true positive and true negative rates. H-HQNN is more efficient and interpretable as it achieves better discriminative power, with fewer features, compared to classical baselines (e.g., Random Forests, deep neural networks). H-HQNN presents an opportunity toward novel methodology of identifying early heart failure by utilizing quantum-classical hybrid modeling and heuristic feature selection to ensure progress in preventative cardiology and therapeutic intervention in particular where advanced imaging studies or invasive measurements of biomarkers are not feasible.