A simplified bedside model using age, NRS-2002, and ADL to predict postoperative delirium in elderly hip fracture patients: a retrospective cohort study
- 1Geriatric Diseases Institute of Chengdu, Department of Geriatrics, Chengdu Fifth People's Hospital (The Second Clinical Medical College, Affiliated Fifth People's Hospital of Chengdu University of Traditional Chinese Medicine)
Abstract
Background: Postoperative delirium (POD) is the most common neuropsychiatric complication after hip fracture surgery in elderly patients. Existing prediction models rely on cognitive assessments or intraoperative parameters that are difficult to obtain in emergency settings. This study aimed to develop and internally validate a simplified model using age, Nutritional Risk Screening 2002 (NRS-2002), and Activity of Daily Living (ADL) dependency to predict POD.
Methods: This single-center retrospective cohort study included patients aged >=60 years who underwent hip fracture surgery at Chengdu Fifth People’s Hospital between January 2019 and August 2023. Patients with pathological fractures or pre-existing delirium were excluded. Multiple imputation by chained equations (MICE, m=20) was used for missing data, with pooled estimates combined using Rubin’s rules. To prevent data leakage, all imputation models were fitted strictly within the training folds of cross-validation, and all internal validation metrics were computed on out-of-fold (OOF) predictions. Discrimination was assessed by the area under the receiver operating characteristic curve (AUC) with 5-fold stratified cross-validation repeated 20 times; calibration was assessed by the calibration slope, calibration-in-the-large, the Hosmer-Lemeshow test, and the Brier score. Complete-case analysis and delta-adjusted MNAR sensitivity analyses were performed.
Results: A total of 387 patients were included, with 43 (11.1%) developing POD. All three variables were independent predictors: age (OR = 1.096, 95% CI: 1.039-1.157, P<0.001), NRS-2002 (OR = 2.075, 95% CI: 1.370-3.142, P<0.001), and ADL (OR = 1.593, 95% CI: 1.094-2.319, P = 0.015). The cross-validated AUC was 0.815 (95% CI across repeats: 0.810-0.819; pooled OOF AUC 0.812), with a calibration slope of 1.068 (95% CI: 0.885-1.252), Brier score of 0.0848, and Hosmer-Lemeshow P = 0.086. At the optimal cutoff (0.115), sensitivity was 88.4% (38/43) and the negative predictive value reached 97.9%, with an observed POD rate of 2.1% among patients below the cutoff; the positive predictive value was 24.8%. The complete-case analysis (N = 200, 21 events) was underpowered, prone to selection bias, and is reported purely as a directional sensitivity check.
Conclusion: This simplified three-variable model demonstrates moderate discrimination and acceptable calibration. With an NPV of 97.9% but a modest PPV of 24.8%, it is best suited as a bedside rule-out screening tool rather than for identifying high-risk patients. The NRS-2002 estimate attenuated under MNAR sensitivity analysis and is interpreted with caution. Multicenter prospective external validation, incorporating systematic cognitive screening, is needed before clinical application.
Related
- Development and Validation of a Perioperative Risk Update Model for Postoperative Delirium in Patients Aged ≥ 80 Years Undergoing Hip Fracture Surgery: A Retrospective Cohort Study
- Commentary: Incidence and independent risk factors for postoperative delirium in ICU patients after cardiopulmonary bypass cardiac surgery: a retrospective cohort study
- Association of preoperative systemic immune-inflammation index and geriatric nutritional risk index with prolonged postoperative length of stay in older patients with hip fracture : A retrospective cohort study
- Stress hyperglycemia ratio is associated with delirium in elderly cardiac surgery patients in the postoperative intensive care unit: a multicenter retrospective cohort study.
- U-Shaped Association Between Preoperative Fasting Blood Glucose and Postoperative Delirium in Elderly Diabetic Patients: A Multicentre Retrospective Cohort Study
BibTeX
@article{chen2026simplified,
title = {{A simplified bedside model using age, NRS-2002, and ADL to predict postoperative delirium in elderly hip fracture patients: a retrospective cohort study}},
author = {Qinyuan Chen and Shanping Chen and Lijuan Guan and Yaoxuan Wu and Zuowei Pei},
journal = {Frontiers in Psychiatry},
year = {2026},
month = oct,
doi = {10.3389/fpsyt.2026.1963389},
url = {https://www.frontiersin.org/articles/10.3389/fpsyt.2026.1963389/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A simplified bedside model using age, NRS-2002, and ADL to predict postoperative delirium in elderly hip fracture patients: a retrospective cohort study
AU - Qinyuan Chen
AU - Shanping Chen
AU - Lijuan Guan
AU - Yaoxuan Wu
AU - Zuowei Pei
PY - 2026
DA - 2026/10/19
JO - Frontiers in Psychiatry
DO - 10.3389/fpsyt.2026.1963389
UR - https://www.frontiersin.org/articles/10.3389/fpsyt.2026.1963389/full
AB - Background: Postoperative delirium (POD) is the most common neuropsychiatric complication after hip fracture surgery in elderly patients. Existing prediction models rely on cognitive assessments or intraoperative parameters that are difficult to obtain in emergency settings. This study aimed to develop and internally validate a simplified model using age, Nutritional Risk Screening 2002 (NRS-2002), and Activity of Daily Living (ADL) dependency to predict POD. Methods: This single-center retrospective cohort study included patients aged >=60 years who underwent hip fracture surgery at Chengdu Fifth People’s Hospital between January 2019 and August 2023. Patients with pathological fractures or pre-existing delirium were excluded. Multiple imputation by chained equations (MICE, m=20) was used for missing data, with pooled estimates combined using Rubin’s rules. To prevent data leakage, all imputation models were fitted strictly within the training folds of cross-validation, and all internal validation metrics were computed on out-of-fold (OOF) predictions. Discrimination was assessed by the area under the receiver operating characteristic curve (AUC) with 5-fold stratified cross-validation repeated 20 times; calibration was assessed by the calibration slope, calibration-in-the-large, the Hosmer-Lemeshow test, and the Brier score. Complete-case analysis and delta-adjusted MNAR sensitivity analyses were performed. Results: A total of 387 patients were included, with 43 (11.1%) developing POD. All three variables were independent predictors: age (OR = 1.096, 95% CI: 1.039-1.157, P<0.001), NRS-2002 (OR = 2.075, 95% CI: 1.370-3.142, P<0.001), and ADL (OR = 1.593, 95% CI: 1.094-2.319, P = 0.015). The cross-validated AUC was 0.815 (95% CI across repeats: 0.810-0.819; pooled OOF AUC 0.812), with a calibration slope of 1.068 (95% CI: 0.885-1.252), Brier score of 0.0848, and Hosmer-Lemeshow P = 0.086. At the optimal cutoff (0.115), sensitivity was 88.4% (38/43) and the negative predictive value reached 97.9%, with an observed POD rate of 2.1% among patients below the cutoff; the positive predictive value was 24.8%. The complete-case analysis (N = 200, 21 events) was underpowered, prone to selection bias, and is reported purely as a directional sensitivity check. Conclusion: This simplified three-variable model demonstrates moderate discrimination and acceptable calibration. With an NPV of 97.9% but a modest PPV of 24.8%, it is best suited as a bedside rule-out screening tool rather than for identifying high-risk patients. The NRS-2002 estimate attenuated under MNAR sensitivity analysis and is interpreted with caution. Multicenter prospective external validation, incorporating systematic cognitive screening, is needed before clinical application.
SN - 1664-0640
ER -