A Systematic Review on Prediction Models for Postoperative Delirium in Non-cardiac Surgery Patients

• Published in: Asian Journal of Advanced Research and Reports Vol. 19; no. 6; pp. 112 - 126
• Main Authors: Iddi, Asha Khatib, Bah, Chernor Sulaiman, Mbambara, Bongani, Si, Yanna, Huang, Kaizong, Zou, Jianjun
• Format: Journal Article
• Language: English
• Published: 14.06.2025
• Subjects: Computer Science; Medical Imaging
• ISSN: 2582-3248; 2582-3248
• DOI: 10.9734/ajarr/2025/v19i61046

Introduction: Numerous risk prediction models (RPMs) for postoperative delirium (POD) following non-cardiac surgery have been developed and validated recently. However, the robustness and applicability of these models require further investigation. Methods: Using PRISMA-2020 guidelines and PROBAST checklist, studies on POD RPMs in non-cardiac surgery patients were searched from PubMed and Google Scholar from January 2021 to December 2023. Inclusion criteria were: (a) adults (aged ≥18 years), (b) non-cardiac surgery patients, (c) development and/or validation of delirium RPMs, and (d) full papers in English. Exclusion criteria were studies not meeting these inclusion parameters. Results: Twelve studies included non-cardiac surgery patients with varying rates of POD (3.22% to 38.30%). The Confusion Assessment Method (CAM) was commonly used for assessing POD risk, with logistic regression being the most employed prediction model. Predictors often found were age, intraoperative blood loss, albumin levels, anesthesia duration, and ICU stays. Internal validation was done in 75% of all the models included. The area under the curve (AUC) ranged from 0.68 to 0.94 for internal validation and from 0.630 to 0.880 for external validation sets. Additionally, most of the models showed a minimal risk of bias (83.3%) and were considered to have a low concern regarding their applicability (75%). Conclusion: Based on this review, current RPMs for POD among non-cardiac surgery patients exhibit high accuracy, low risk of bias, and minimal concerns regarding their applicability. We recommend that future research prioritize the external validation of existing models to improve their clinical utility.