Background <p>Delirium is a frequent complication in hospitalised patients, associated with long-term cognitive impairment, prolonged hospital stays, and increased mortality. Early risk assessment is essential for implementing preventive strategies. The Mayo Delirium Prediction (MDP) tool, developed using data from a large academic hospital in the United States, includes separate models for medical and surgical patients. The MDP is available in its original and recalibrated versions. However, its performance in non-US population remains unknown. The objective of this study is to externally validate the MDP tool in a cohort of hospitalised patients from a Swiss private hospital.</p> Methods <p>This retrospective validation study used routinely collected clinical data from adult medical and surgical inpatients admitted to a Swiss hospital in May and June 2023. Delirium diagnosis was based on the Delirium Observation Screening Scale (DOSS). Predictive performance was assessed using the Area Under the Receiver Operating Characteristic curve (AUROC) and calibration plots. Additional sensitivity analyses were performed to assess the model’s robustness and mitigate possible biases in the evaluation.</p> Results <p>The original medical MDP tool achieved the best predictive performance in the external validation cohort of 947 patients, with an AUROC of 0·87 (95% CI: 0·85–0·90). The original surgical MDP tool showed relatively lower performance in the external validation cohort of 1212 patients, with an AUROC of 0·73 (95% CI: 0·69–0·78). Performance remained acceptable across all sensitivity analyses, with only a modest reduction in AUROC. The original MDP tool consistently outperformed recalibrated versions across all scenarios.</p> Conclusions <p>The MDP tool, particularly the medical model, maintained strong predictive performance in an independent setting. These findings support its generalisability and potential integration into routine clinical workflows for delirium prevention.</p>