Late-life depression often extends to the marital partner, yet mechanisms of dyadic interdependence and their cross-cultural generalizability remain unclear.

We analyzed harmonized longitudinal data from couples in the China Health and Retirement Longitudinal Study (CHARLS; n = 3,532; 5 waves, 2011–2020), the Health and Retirement Study (HRS; n = 2,332; 8 waves, 2006–2020), and the English Longitudinal Study of Ageing (ELSA; n = 1,895; 9 waves, 2002–2018). Depressive symptoms (CES-D) and activities of daily living (ADL)/instrumental ADL (IADL) limitations were measured at every wave. Complementary actor–partner interdependence models (APIM) and random-intercept cross-lagged panel models (RI-CLPM) were applied, with multiplicity adjustment, full-information maximum likelihood (FIML), and 20-imputation triangulation.

Within-couple correlations of depressive symptoms were positive in every cohort (r = 0.34–0.43 in CHARLS, 0.14–0.20 in HRS, 0.23–0.29 in ELSA). APIM partner effects were robust in CHARLS but sparse elsewhere. After partialling out stable between-couple similarity, RI-CLPM within-person cross-spouse associations were attenuated, reaching consistent significance under FIML and multiple imputation only in ELSA. Four-variable RI-CLPMs showed bidirectional depression–disability coupling, strongest in HRS and ELSA, with cross-spouse function-to-depression effects in Western cohorts but not in CHARLS. Sex distinguishability was not supported, and half-longitudinal mediation through ADL/IADL did not survive multiplicity correction.

Concordance in late-life depressive symptoms replicates across three major aging cohorts; in CHARLS, it primarily reflects stable between-couple similarity, whereas modest within-person prospective coupling persists in ELSA and in the multiple-imputation HRS estimates. Bidirectional depression–disability coupling in Western cohorts is the most replicable within-person signal, supporting combined depression management and functional rehabilitation for older couples.