Improving rare event prediction in imbalanced clinical data using cost-sensitive logistic regression

Rare outcomes are common in clinical research and often result in severe class imbalance. Under such conditions, traditional logistic regression tends to favor the majority class and shows limited sensitivity to clinically important rare events. Although complex machine learning methods may improve prediction, their limited interpretability can restrict clinical use. Therefore, an effective and interpretable modeling strategy for highly imbalanced clinical data remains needed. We developed a cost-sensitive logistic regression (CS-LR) model by incorporating different misclassification costs for different classification outcomes into the logistic loss function. The proposed framework was designed to prioritize rare but clinically important outcomes while retaining odds ratio interpretation. Simulation studies were conducted across four covariate scenarios, imbalance ratios from 1% to 30%, and sample sizes of 500 and 2000. CS-LR was compared with traditional logistic regression, cost-sensitive support vector machines, and AdaBoost. External validation was performed using a liver fibrosis dataset and a post-ESD bleeding dataset. Across simulated scenarios, CS-LR showed stable convergence and improved minority-class detection, particularly under small sample ( N = 500) and highly imbalanced (positive class ratio < 5%) settings. The advantages were most evident in recall, F1-score, and G-mean, while AUC remained broadly comparable. In the liver fibrosis validation, CS-LR consistently achieved higher recall and G-mean than traditional logistic regression across different sample sizes and positive class ratios. In the post-ESD bleeding dataset, CS-LR showed the most favorable overall performance among the compared models while retaining interpretable coefficients and odds ratio estimates. CS-LR provides an interpretable and clinically practical framework for rare event prediction in highly imbalanced medical data. It may be particularly useful in clinical prediction tasks where missed positive cases have important consequences, such as cancer screening, postoperative complication prediction, and detection of uncommon adverse clinical events.