Abstract
<title>Abstract</title> <p>Background: Falls are a leading cause of hospitalization, disability, and mortality among older adults. In Medicare Annual Wellness Visits (AWVs), fall-related screening often relies on self-reported fall history, yet falls are frequently underreported—creating gaps in identifying functional vulnerability. Multi-domain, behaviorally anchored functional indicators (e.g., strength, balance, gait stability) can reveal fall-risk indicator burden independent of fall recall, but are difficult to administer at scale within AWV workflows. This study evaluated whether a conversational AI health assessment system, deployed as part of AWV health risk assessments, can deliver multi-domain functional screening at scale and identify high fall-risk indicator burden among older adults who report no falls in the past year (“hidden indicator burden”). Methods: This cross-sectional observational study analyzed data from an AI-powered, telephone-based health risk assessment deployed within OhioHealth’s Medicare AWV program. Community-dwelling older adults (N = 511) with valid responses on fall history and seven functional fall-risk indicators were included. High fall-risk indicator burden was defined as ≥2 indicators. Chi-square testing, unadjusted and adjusted logistic regression, and sensitivity analyses with alternative thresholds were conducted. Results: Mean age was 75.6 years (SD = 6.8); 59.5% were female. Among participants denying falls (n = 378), 32.3% met criteria for high indicator burden (≥2). Among those reporting falls (n = 133), 81.2% met criteria. Fall history was strongly associated with AI-derived high indicator burden (χ² = 95.16, p < .001; OR = 9.07, 95% CI: 5.58–14.73) and remained robust after adjustment for age and gender (adjusted OR = 8.65, 95% CI: 5.26–14.21). With a ≥3 indicator threshold, 18.3% of those denying falls still demonstrated substantial multi-domain burden. Conclusions: When integrated into Medicare AWV health risk assessments, conversational AI can scale standardized capture of functional fall-risk indicators and reveal a substantial hidden indicator burden among older adults who deny falling. This approach may help prioritize follow-up assessment and prevention services; prospective validation is needed to establish prediction of incident falls. Trial registration Not applicable. This study analyzed observational data from a deployed health risk assessment program and did not evaluate a healthcare intervention on human participants.</p>