Group-based trajectory modeling (GBTM) can be applied using either time-based or age-based approaches to identify frailty trajectories; however, the comparative effectiveness of the methods for mortality prediction remains unclear. This study aimed to compare time-based and age-based trajectory modeling in identifying frailty trajectories and their predictive value for mortality among Japanese community-dwelling older adults. Of 1085 community-dwelling older adults aged ≥ 65 years, 512 participants with at least two frailty assessments who remained independent during 2011–2017 were included. Frailty was assessed using the Kihon Checklist. Both time-based and age-based GBTM used data from 2011, 2014, and 2017 to identify trajectories. Agreement between methods was assessed using Cohen’s Kappa coefficient. Kaplan–Meier survival analysis and Cox proportional hazards models were used to evaluate mortality prediction during follow-up (May 2017–March 2021). Among the 512 participants (mean age 72 ± 6 years; 54.7% female), both models identified two frailty trajectories: low increasing (88.3% vs. 83.8%) and high increasing (11.7% vs. 16.2%) groups for the time-based and age-based models, respectively. The methods showed substantial agreement (Kappa = 0.63, P < 0.001). During follow-up, 48 participants (9.4%) died, with a median survival time of 27 (IQR 18–33) months. The high increasing group showed a higher mortality risk than the low increasing group in the time-based (adjusted HR = 3.0, 95% CI = 1.4–6.5, P = 0.006) and age-based models (adjusted HR = 2.5, 95% CI = 1.2–5.1, P = 0.01). Both models showed comparable discriminative ability (C-index: 0.77 vs. 0.77) and model fit (AIC: 451 vs. 453). In this three-wave Japanese community-based older adult study, both time-based and age-based GBTM identified similar two-group frailty trajectories with substantial agreement and effectively identified high-risk populations for mortality. These findings suggest that both modeling methods are suitable for frailty trajectory analysis and mortality risk prediction in this population; however, their generalizability to other populations requires further validation.
Yang et al. (Fri,) studied this question.