Other Affiliation: MGH Institute of Health Professions
Other Affiliation: Duke University
Affiliation: School of Medicine, Department of Allied Health Sciences
Other Affiliation: Boston University
Introduction: Community-dwelling, ambulatory stroke survivors fall at very high rates in the first 3-6 months. Current inpatient clinical assessments for fall risk have inadequate predictive accuracy. We found that a pre-discharge obstacle-crossing test has excellent specificity (83%) but lacks acceptable sensitivity (67%) for identifying would-be fallers and non-fallers post discharge. Hypothesis: We assessed the hypothesis that combining the obstacle-crossing test with other highly discriminatory fall risk factors would compensate for the obstacle test’s fair sensitivity and yield an instrument with superior prediction accuracy. Methods: 45 ambulatory stroke survivors (60±11 years old, 15±11 days post stroke) being discharged home completed a battery of physical performance-based and self-reported measures 1-5 days prior to discharge. After discharge, participants were prospectively followed and classified as fallers (≥1 fall) or non-fallers at 3 months. Pre-discharge measures with the largest effect sizes for differentiating fallers and non-fallers were combined into a composite index. Several variations of the composite index were examined to optimize accuracy. Results: A 4-item discharge composite index significantly predicted fall status at 3-months. The goodness of fit of the regression model was significantly better than the obstacle-crossing test alone, χ2(1) = 6.036, p=0.014. Furthermore, whereas the obstacle-crossing test had acceptable overall accuracy (AUC 0.78, 95% CI 0.60-0.90), the composite index had excellent accuracy (AUC 0.85, 95% CI 0.74-0.96). Combining the obstacle-crossing test with only the step test produced a model of equivalent accuracy (AUC 0.85, 95% CI 0.73-0.96) and with better symmetry between sensitivity and specificity (0.71, 0.83) than the 4-item composite index (0.86, 0.67). This 2-item index was validated in an independent sample of n=30 and with bootstrapping 1000 samples from the pooled cohorts. The 4-item index was internally validated with bootstrapping 1000 samples from the derivation cohort plus n=9 additional participants. Conclusion: This study provides convincing proof-of-concept that strategic aggregation of performance-based and self-reported mobility measures, including a novel and demanding obstacle-crossing test, can predict post-discharge fallers with excellent accuracy. Further instrument development is warranted to construct a brief aggregate tool that will be pragmatic for inpatient use and improve identification of future post-stroke fallers before the first fall.