Validation of a Fall Predictive Model for Inpatients in Japanese Long Term Care Hospitals

TL;DR

This study validated a fall prediction model for older patients in Japanese long-term care hospitals, showing it works well in identifying those at risk.

Contribution

The study validated the Saga Falls Risk Model 2 (SFRM2) for use in long-term care hospitals, where fall risk is particularly high.

Findings

SFRM2 had an AUC of 0.889, indicating strong predictive accuracy for falls in long-term care hospital patients.

The model showed a sensitivity of 77.9% and a negative predictive value of 96.6% at the optimal cutoff score.

The fall incidence rate was 4.4 per 1000 patient-days among the studied population.

Abstract

Background:
The Saga Falls Risk Model 2 (SFRM2) is a simplified fall prediction model that we recently developed. It uses eight items that are easy to assess at the time of admission to an acute care hospital. However, patients in long-term care hospitals have poor activities of daily living and a high risk of falls compared to those in acute care hospitals. Although effective fall predictive models exist for long-term care hospitals, their accuracy remains suboptimal. This study aimed to validate the SFRM2 for predicting falls in long-term care hospital patients.

Methods:
This multicenter retrospective observational study was conducted in three long-term care hospitals in Japan from April 2018 to March 2021. All inpatients aged ≥20 years were included. The eight items of the SFRM2 (age, sex, emergency admission, department of admission, hypnotic medication use, history of falls, eating independence, and Bedriddenness rank) and in-hospital falls were collected from medical records. The accuracy of SFRM2 was assessed by calculating the area under the curve (AUC) and shrinkage coefficient, as well as the sensitivity, specificity, positive predictive value, and negative predictive value.

Results:
Among the 1182 patients (median age: 86 years, 538 males) included in the analysis, 140 (11.8%) experienced in-hospital falls. The fall incidence rate was 4.4 per 1000 patient-days. SFRM2 exhibited an AUC of 0.889 (95% confidence interval: 0.861-0.916), consistent with the actual incidence of falls, with a shrinkage coefficient of 0.975. The cutoff score for SFRM2 on the Youden index was -2.14, with a sensitivity of 77.9%, specificity of 84.7%, positive predictive value of 40.6%, and negative predictive value of 96.6%.

Conclusion:
SFRM2 showed good discriminative ability in external validation at long-term care hospitals. Its applicability in this setting may be advantageous due to the relatively stable condition of older inpatients compared to those in acute care hospitals.

Figures