SHM Converge Conference Coverage

External Validation of Fall-Injury Risk Model Reveals Need for Simplified, Universal Predictive Tool

Kate Young

Key Highlights

  • External validation of the Saga Fall-Related Injury Risk Model (SFIRM) across eight hospitals showed reduced predictive accuracy compared with the original internal setting.
  • Five common predictors were consistently associated with fall-related injuries: age, male sex, history of falls, diabetes, and bedriddenness ranks.
  • Fall-related injuries occurred in 1.5% of patients, with injured patients having notably longer hospital stays.
  • Area Under the Curve (AUC) for predictive performance varied widely across facilities, from 0.300 to 0.719.

In a large, multicenter retrospective study presented at the Society of Hospital Medicine's SHM Converge 2025 in Las Vegas, NV, researchers externally validated the SFIRM, a previously developed prediction model for fall-related injuries, and identified common predictors across eight Japanese hospitals. The study found that although the model demonstrated good internal discrimination (AUC 0.772), its performance diminished in external hospital settings, indicating variability in predictive utility across institutions.

Falls and associated injuries remain a serious challenge in hospital settings due to their potential to cause residual disabilities, decreased activities of daily living (ADL), and increased medical costs. The importance of early identification of patients at high risk for fall-related injuries is underscored by the need to mitigate these adverse outcomes and improve patient safety, particularly in the growing aging population.

Researchers conducted a retrospective observational study using data from adult inpatients (n = 144,777) admitted to eight hospitals between April 2018 and March 2021. The study included acute, chronic, and tertiary acute care facilities. Patients under 20 years of age and those with missing data regarding fall severity or history were excluded. A total of 51 patients experienced falls, with 35 sustaining fall-related injuries. The model’s discrimination was assessed using AUC analysis, and facility-specific analyses were performed to identify consistent risk factors.

data from study

Among the 2376 randomly sampled patients, the median age was 70 years, and 54% of the patients were men. Fall-related injuries occurred in 1.5% of the population. Injured patients were older (median age 76 years) and had significantly longer hospital stays (median 26 days vs. 9 days). The model’s external validation across facilities showed varied AUC values, ranging from 0.300 to 0.719, highlighting inconsistent predictive performance. Despite this, five predictors—age, male sex, history of falls, diabetes, and bedriddenness ranks—were identified as consistently associated with fall-related injuries in multivariate analyses, suggesting their utility in simplified risk stratification tools.

“The discrimination of the SFIRM were lower in external validation across diverse backgrounds,” the study authors concluded. “These five items can be effectively utilized in a simple prediction model for fall-related injuries, enhancing patient safety across various hospitals.”

Reference:
Katsuki NE, Hirata R, Yaita S, et al. External validation of a prediction model for fall-related injuries and common predictors across multiple hospitals: a retrospective observational study. Presented at: Society of Hospital Medicine Converge; 2025; Las Vegas, NV. Accessed April 22, 2025. https://shmconverge.hospitalmedicine.org/