Movement Intensity in the Minutes Before a Fall: A Lag-Controlled, Activity-Matched Time-Stratified Case-Crossover Analysis of Home Monitoring in Multiple Sclerosis

Background: Fall-risk research in multiple sclerosis usually asks who falls and does not address what changes in the minutes before a fall. Methods: We analyzed continuous home inertial monitoring of 26 people with multiple sclerosis using a time-stratified case-crossover design. Each fall was compared with referent windows from the same participant on other days at the same clock time. Results: After one participant whose accelerometer was not on a 1 g scale was excluded, 48 falls and 285 referents remained in 12 informative strata. Per standard deviation, gravity-removed dynamic acceleration had an odds ratio of 3.05 (95% CI 1.66–5.63). The estimate persisted when the window ended 10, 30, or 60 s before the annotated fall, although these designs overlap the primary one and are not independent tests. Referents from pre-fall days only gave 3.46 (1.67–7.16). When activity was held comparable, the estimate fell to 2.11 (1.17–3.79) within ambulatory windows and to 1.38 (0.86–2.22, p = 0.183) against referents from walking periods. Conclusions: These results suggest that much of the association reflects the fact that people fall while they are moving. No released fall lies in unrecorded time, which indicates that the public dataset retains sensor-captured falls.