Handgrip strength for predicting incident ADL disability in older adults with arthritis: a multi-cohort external validation study

Handgrip strength is an established marker of functional health, but a prognostic association does not guarantee that its incremental predictive value or derived absolute risks transfer across populations. We evaluated the cross-cohort transportability of handgrip-based prediction of incident disability in basic activities of daily living (ADL) and whether alternative handgrip representations improved external performance. Models were developed in the Survey of Health, Ageing and Retirement in Europe (SHARE; 2011–2013) and applied without refitting to the Health and Retirement Study (HRS; 2010–2012), the English Longitudinal Study of Ageing (ELSA; 2012–2015), and the China Health and Retirement Longitudinal Study (CHARLS; 2015–2018). Participants were aged ≥ 65 years, reported physician-diagnosed arthritis or rheumatism and current pain, and had no baseline difficulty in five basic ADLs. A no-handgrip benchmark was compared with models adding raw handgrip strength, handgrip strength divided by height squared, locally age- and sex-standardized handgrip strength, or European Working Group on Sarcopenia in Older People 2 (EWGSOP2) weakness. Discrimination, calibration, Brier score, and net benefit were assessed, with optimism correction and sensitivity analyses of prediction horizon, handgrip protocol, Asian Working Group for Sarcopenia (AWGS) thresholds, and mortality. Analytic samples comprised 3658 participants (585 events) in SHARE, 696 (93) in HRS, 639 (132) in ELSA, and 328 (121) in CHARLS. Raw handgrip strength increased the SHARE area under the receiver operating characteristic curve (AUC) from 0.681 to 0.692 (optimism-corrected, 0.675 to 0.685), but this increment did not transport: AUCs with versus without handgrip strength were 0.706 versus 0.711 in HRS, 0.661 versus 0.665 in ELSA, and 0.620 versus 0.628 in CHARLS. For the raw-handgrip model, observed-to-expected ratios were 0.70, 1.19, and 1.67, respectively; accounting for the 3-year CHARLS follow-up reduced its ratio to 1.22 but did not correct its calibration slope. No alternative representation, including AWGS weakness, consistently improved performance. Among older adults who reported physician-diagnosed arthritis or rheumatism and current pain, handgrip strength was prognostic but added little reproducible predictive value across cohorts. Rescaling handgrip strength did not resolve differences in calibration, which partly reflected differing follow-up intervals. Local validation and recalibration are needed before handgrip-based absolute risks are used elsewhere. Not applicable.