Using instrumented everyday gait to predict falls in older adults using the WHS cohort

点击翻译按钮获取中文摘要

Among community-living older adults, falls are a leading cause of injury, disability, injury-related death, and high medical costs. Despite decades of research, the proportion of older adults who fall has not declined. Identifying older adults at risk of falls remains a major public health priority. Exercise and other interventions can lower fall risk; however, new tools are needed to determine who is most likely to benefit from early interventions. Early research linking fall risk to gait measures obtained in the clinic (e.g., average speed, stride variability) contributed significantly to the understanding of the prediction of fall risk. Studies have also shown that older adults who are more active have reduced risks of falls and fall-related injury. However, critical gaps remain. Exciting advances in digital medicine and remote monitoring using wearable devices have afforded new and more widely accessible opportunities for evaluating the relationships between Daily Living Gait (DLG) and Daily Living Physical Activity (DLPA) to injurious falls in older adults. Measures of DLG (e.g., gait speed, cadence, variability, and how these vary throughout the week) and measures of DLPA (e.g., activity levels and activity fragmentation) can all be derived from a single accelerometer worn for 1 week. While growing evidence suggests that DLG and DLPA do a better job at predicting falls than conventional in-clinic measures, studies to date have been relatively small and have not focused on the prediction of injurious falls. Moreover, little is known about the utility of combining DLG and DLPA measures to predict injurious falls. To address these gaps, we will leverage: 1) an existing large dataset of older women enrolled in the Women’s Health Study (WHS) and 2) advances in wearable technology and machine learning. From 2011 to 2015, 17,466 WHS women wore a tri-axial accelerometer during waking hours for a week; they also regularly self-reported their physical activity levels and health history. We propose to evaluate, for the first time, if and how DLG and DLPA measures predict fall-related injuries in this aging cohort (average age=72 years at the time of accelerometer wear) using records of injurious falls from the Centers for Medicare & Medicaid Services (CMS). Primary Aims 1 and 2 will evaluate which specific measures of DLG and DLPA are associated with the risk of injurious falls in the subsequent year after assessment, using statistical and machine learning approaches that use time-to-event analyses (with and without adjustments for covariates). Primary Aim 3 will evaluate whether utilizing measures of both DLG and DLPA is more strongly associated with the risk of injurious falls than utilizing each of these measures alone. We will also determine if self-reported exercise history is associated with DLG and DLPA, and explore whether markers of DLG and DLPA are associated with risks of injurious falls over more extended periods of 5 and 10 years, as secondary and exploratory aims. By taking advantage of a unique, large dataset, our multi-disciplinary team will identify potential “signatures” to identify high-risk adults who may benefit from early fall prevention strategies and markedly accelerate the potential of using digital markers of fall risk.