Validation and Comparison of 4 Fall Risk Assessment Tools for Older Adults in Chinese Nursing Homes: A Prospective Cohort Study.

Objectives: This study evaluates and compares the predictive performance of 4 widely used fall risk assessment tools-Morse Fall Scale (MFS), St. Thomas's Risk Assessment Tool in Falling Elderly Inpatients (STRATIFY), Hendrich II Fall Risk Model (Hendrich II), and Timed Up and Go Test (TUGT)-in Chinese nursing homes, with a focus on optimizing cutoff values for better applicability.

Design: A prospective cohort study.

Setting and participants: The study was conducted in 4 nursing homes in China, including 866 older adults capable of providing informed consent and completing verbal communication.

Methods: Participants were assessed using the 4 fall risk tools, and their fall events were recorded over 6 months. Predictive performance was evaluated using sensitivity, specificity, positive predictive value, negative predictive value, and area under the receiver operating characteristic curve. Calibration curves were generated to assess the agreement between predicted and observed fall probabilities.

Results: Using the original cutoff values, the MFS (≥45) and TUGT (≥12 seconds) both showed high sensitivity (0.889 and 0.933, respectively) but low specificity (0.284 and 0.261, respectively). In contrast, the STRATIFY (≥2) and Hendrich II (≥5) exhibited high specificity (0.964 and 0.827, respectively) but low sensitivity (0.117 and 0.328, respectively). After optimization, the MFS (≥65) improved specificity (0.592) with moderate sensitivity (0.689), the STRATIFY (≥1) increased sensitivity (0.856) while reducing specificity to 0.407, the Hendrich II (≥2) achieved specificity of 0.519 with sensitivity of 0.739, and the TUGT (≥26.6 seconds) maintained high sensitivity (0.739) but had a specificity of 0.622. The TUGT demonstrated the strongest overall predictive accuracy (area under the receiver operating characteristic curve, 0.722).

Conclusions and implications: All tools showed limitations in balancing sensitivity and specificity. Adjusting thresholds improved performance but did not yield optimal results. The findings highlight the importance of tailoring fall risk assessments to specific populations with thresholds adjusted to optimize performance. Future research should explore integrating clinical assessments with data-driven predictive models to enhance fall risk evaluation in long-term care settings.