Work

Wearable activity-sequence measures and subsequent handgrip strength in adults aged 65 years and older: measurement characteristics and longitudinal associations

Year: 2026
Type: article
Abstract: Background Temporal organization of everyday activity may provide information complementary to conventional movement-volume summaries. We examined the measurement properties of activity-state sequence descriptors and their longitudinal associations with handgrip strength. Methods We analyzed SHARE Wave 8 three-day thigh accelerometry and Wave 9 handgrip data. There were 490 participants aged at least 65 years with eligible sequences; follow-up grip was available for 410 and the negative-code-corrected, fully adjusted model included 391. Five activity states were coded in 5-second epochs. The original research workflow prioritized run-collapsed LZ76, while the H1–H3 conditional-entropy analysis was an exploratory evaluation whose final screening took place after initial outcome analyses. Daily-average ICC and correlation-based redundancy were distinguished from a new post hoc two-window evaluation of the identical nonlinear 72-hour scores among 407 participants with six valid days in the same wear episode; inference was limited to within-wear agreement. Participant-level analyses were replayed locally from the three licensed SHARE source archives. Results In corrected complete cases, the adjusted 72-hour H1–H3 coefficient was β = 0.535 kg/SD (95% CI -0.335 to 1.404; P = 0.228), with ΔR²=0.00067. The original historically prioritized run-collapsed LZ76 measure gave β = 0.906 kg/SD (95% CI 0.171 to 1.641; P = 0.016), ΔR²=0.00301; this age-restricted contrast is not an independent preregistered confirmatory test. Fifty repeated, strictly foldwise 10-fold validation analyses yielded mean ΔR²=-0.000341 and mean ΔRMSE = 0.0038 kg on adding H1–H3. Twenty-draw predictive-mean-matching sensitivity gave β = 0.471 kg/SD (95% CI -0.380 to 1.322; P = 0.278). In a post hoc within-wear paired-window analysis ( n = 407), the complete-72-hour H1–H3 had ICC(A,1) = 0.775 (participant-bootstrap 95% interval 0.723–0.822), and run-collapsed LZ76 had ICC(A,1) = 0.876 (0.848–0.898). Conclusions Activity-sequence descriptors differed in their relationships with conventional movement measures and later handgrip strength. Among participants with two complete 72-hour windows within one wear episode, H1–H3 and run-collapsed LZ76 showed absolute-agreement ICCs of 0.775 and 0.876, respectively. H1–H3 showed an imprecise handgrip association and no mean gain in repeated cross-validated prediction. These findings highlight the value of assessing temporal activity organization and its measurement properties alongside conventional activity summaries. Clinical trial registration Not applicable.
Author: Yifu Zhu
Language: English
Field: Medicine
Open Access status: gold
Estimated APC paid: $3,190
OpenAlex ID: W7234427601
FWCI:
Cites:
Cited by:
Related to: 10