Aims: Evaluate the prevalence and predictive ability of a multidomain frailty index derived from routinely collected health data (electronic frailty index-acute hospital: eFI-AH) for in-hospital mortality and long hospital stays in patients with and without dementia.
Methods: Retrospective study of 30492 patients aged ≥65 years on their first admission to six public hospitals in NSW (22/6/2022-/7/2023). Dementia diagnosis determined from ICD-10-AM codes. The eFI-AH, derived from 54 items across multiple domains, was calculated for patients with ≥30 items assessed. Association between eFI-AH and dementia, adjusted for age and gender, was estimated using gamma regression. Predictive ability of the eFI-AH for mortality and long hospital stays (>10 days) was evaluated using ROC-AUC, PR-AUC and calibration plots.
Results: The eFI-AH was calculated for 23060/30492 patients (75.6%), of whom 2549 (10.1%) had dementia. Median eFI-AH score was higher in those with dementia (0.34, IQR 0.26-0.42) than without (0.17, IQR 0.11-0.27), although the gap narrowed with increasing age (p<0.05). Dementia was strongly associated with eFI-AH (adjusted OR 1.81, 95%CI 1.65-1.98). Patients with dementia had higher rates of mortality (dementia 9.5%, non-dementia 3.4%) and long hospital stay (dementia 42.7%, non-dementia 22.7%). The ROC-AUC for eFI-AH was lower in those with dementia than without for mortality (0.77 vs 0.88) and for long hospital stays (0.58 vs 0.79). Calibration plots revealed over-estimation in dementia for mortality (slope 1.36).
Conclusions: In this cohort of older inpatients, dementia diagnosis was associated with frailty index. The disparity in eFI-AH between patients with and without dementia decreased as age increased.