Baihui Wu, Zhaoyan Chen, Fangyuan Tian
Background: Potentially inappropriate medication (PIM) use among older people is a serious public health problem associated with increased adverse drug events. PIM refers to medications where the adverse risks outweigh the potential benefits. Identifying the risk factors for PIM is essential for optimizing prescription practices and improving patient safety. Objectives: To establish a risk prediction model for potentially inappropriate medications (PIMs) in older patients with depression, providing guidance to optimize medication plans, reduce adverse drug reactions, and improve treatment outcomes and quality of life. Methods: Prescriptions for depression patients among all hospitals in the Chengdu area were taken as an example. A significant factor influencing PIM risk was identified through univariate and multivariate logistic regression analyses, and a nomogram was constructed. The discrimination and calibration of the model were evaluated via receiver operating characteristic (ROC) curves. Results: According to the analysis of the nomogram drawn from the prescriptions of patients with depression, it can be found that the department name, reimbursement, hospital grade, age, number of diseases, sleep disorders, hypertension, cerebrovascular disease and so on each have p < 0.05 for PIM. Data from the Chengdu area (n = 4629) were divided into a training set (n = 3548) and an internal validation set (n = 1081), with Zhengzhou data (n = 1620) used as the external validation set. ROC curve analysis revealed that the area under the curve (AUC) for the training set was 0.721, that for the internal validation set was 0.668, and that for the external validation set was 0.663. Conclusions: The prediction model based on these factors has predictive value for PIM use in older patients with depression, It shows some discriminatory ability, but its external performance is moderate and its calibration is insufficient. Nevertheless, it can be used for preliminary judgment of patients' medication and as an auxiliary screening tool. Further optimization and prospective validation are still needed.