Trajectories in physical performance and fall prediction in older adults: A longitudinal population-based study.

Abstract

A physical performance evaluation can inform fall risk in older people, however, the predictiveness of a one-time assessment is limited. The trajectory of physical performance over time has not been well characterized and might improve fall prediction. We aimed to characterize trajectories in physical performance and determine if fall prediction improves using trajectories of performance. This was a cohort design using data from the National Health and Aging Trends Study. Physical performance was measured by the short physical performance battery (SPPB) with scores ranging from 0 (worst) to 12 (best). The trajectory of SPPB was categorized using latent class modeling and slope-based multilevel linear regression. We used Cox proportional hazards models with an outcome of time to ≥2 falls from annual self-report to assess predictiveness after adding SPPB trajectories to models of baseline SPPB and established non-physical-performance-based variables. The sample was 5969 community-dwelling Medicare beneficiaries aged ≥65 years. The median number of annual SPPB evaluations was 4 (IQR, 3-7). Mean baseline SPPB was 9.2 (SD, 3.0). The latent class model defined SPPB trajectories over a range of two to nineteen categories. The mean slope from the slope-based model was -0.01 SPPB points/year (SD, 0.14). Discrimination of the baseline SPPB model to predict time to ≥2 falls was fair (Harrell's C, 0.65) and increased after adding the non-performance-based predictors (Harrell's C, 0.70). Discrimination slightly improved with the SPPB trajectory category variable that had the best fit (Harrell's C, 0.71) but did not improve with the SPPB linear slope. Calibration with and without the trajectory categories was similar. We found that the trajectory of physical performance did not meaningfully improve upon fall prediction from a baseline physical performance assessment and established non-performance-based information. These results do not support longitudinal SPPB assessments for fall prediction.