Disease activity of rheumatoid arthritis (RA), evaluated as Disease Activity Score (DAS), is associated with joint destruction. Since joint destruction reflects the history of disease activities, we hypothesized that time-averaged disease activity would better correlate with joint destruction than one-time disease activity. We recruited RA patients in IORRA (n = 557) and KURAMA (n = 204) cohorts, and calculated time-averaged DAS28 to model a modified Sharp/van der Heijde score (SHS). We evaluated the fitting of the model using time-averaged DAS28 among 1000 models in which we randomly picked up one-time DAS28. We also used clinical disease activity index (CDAI) or data in the BeSt study (European population). After conditioning on autoantibody and disease duration, time-averaged DAS28 showed significant improvement of model fitting compared with one-time DAS28 in both cohorts (p = 0.001 and 0.034, respectively). Time-averaged CDAI also showed a better fit. Integration of multiple DAS fit SHS better in the BeSt study. A good fit of time-averaged DAS could be observed using five to six time points of DAS. In conclusion, time-averaged disease activity fits the joint destruction model better than one-time disease activity. Usage of time-averaged disease activity as a covariate would increase the power of studies to identify novel correlates of joint destruction.