Background: Identifying modifiable risk factors for late-life depression is hindered by confounding bias and etiological complexity. This study estimated independent causal effects a Target Trial Emulation (TTE) framework.Methods: We applied a TTE framework to two complementary cohorts: CHARLS (2011–2020,) and CLHLS (2008– 2018). Following LASSO selection, we emulated parallel trials using Inverse Probability of Treatment Weighting to approximate randomization. Causal effects were estimated via weighted XGBoost T-learners for cumulative risk and Cox regression for temporal dynamics, supplemented by SHAP mechanistic analysis.Findings: In the younger-old CHARLS cohort, somatic conditions significantly accelerated disease onset: Physical Pain (RR=1.28; RD=11.52%; HR=1.41) and Arthritis (RR=1.26; RD=10.11%; HR=1.38) were potent drivers. Structural disadvantages, including Low Education (RR=1.25) and Rural Residence (RR=1.23), were robust factors. In the oldest-old CLHLS cohort, Poor Self-rated Health emerged as a prominent risk signal (RR=1.32; HR=1.33). Mechanistic analysis revealed sleep disturbance as a critical amplifierr in arthritis patients, whereas intergenerational economic support was a predictive feature for resilience in the oldest-old.Interpretation: Arthritis, pain, and structural disadvantage are robust causal drivers. The findings suggest an etiological transition where psychiatric risk shifts from individual accumulated deficits in the younger-old to reliance on extrinsic intergenerational support in the oldest-old. Furthermore, the consistency of poor self-rated health across cohorts validates its utility as a universal sentinel metric for early risk stratification in primary care. Prevention strategies should integrate clinical pain-sleep management with efforts to address socioeconomic inequities.