Identification of correlative factors and development of a preliminary classification model for delirium superimposed...

This study aimed to identify the correlative factors of delirium in patients with dementia and to preliminarily develop a classification model for probable delirium superimposed on dementia (DSD).

A total of 143 dementia inpatients from a tertiary hospital were enrolled. Participants were assessed using the Mini-Mental State Examination (MMSE) to evaluate baseline cognitive status, and the Confusion Assessment Method-Chinese Revision (CAM-CR) was applied to categorize participants into the probable delirium group (CAM-CR > 19) and the non-probable delirium group (CAM-CR ≤ 19). Assessments of clinical characteristics included the Generalized Anxiety Disorder Seven-Item Scale (GAD-7), 30-Item Geriatric Depression Scale (GDS-30), Nurses' Global Assessment of Suicide Risk (NGASR), Social Support Rating Scale (SSRS), Barthel Index of Activities of Daily Living (ADL), and Brøset Violence Checklist (BVC). Independent-samples t-tests, chi-square tests, and binary logistic regression were used to identify correlative factors and to develop the classification model. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC).

The prevalence of CAM-CR-assessed probable delirium was 40.6%. Binary logistic regression identified three independent correlative factors of probable delirium in patients with dementia: the number of comorbid physical diseases [odds ratio (OR) = 1.458, 95% confidence interval (CI): 1.121-1.897], level of social support utilization (OR = 0.833, 95% CI: 0.703-0.987), and risk level of violence (medium risk: OR = 4.677, 95% CI: 1.373-15.929; high risk: OR = 13.073, 95% CI: 3.560-48.002). The final classification model, visualized via a nomogram, demonstrated an apparent AUC of 0.767 (95% CI: 0.689-0.844). Bootstrap validation (1,000 iterations) yielded an optimism-corrected AUC of 0.744 (bias = 0.023), and fivefold cross-validation produced a mean AUC of 0.735 (SD = 0.113). The calibration curve showed good agreement between predicted and observed probabilities (mean absolute error = 0.024). The Hosmer-Lemeshow test indicated good calibration (χ ² = 3.477, p = 0.901).

The number of comorbid physical diseases, utilization of social support, and violence risk were significant correlative factors of probable delirium in dementia inpatients. The nomogram represented a proof-of-concept classification tool for DSD screening. However, given the limitations of the study design, sample size, and the absence of external validation, this model should be regarded as a preliminary tool that requires external validation before clinical application.