Background: The landmark TOAD trial (Lancet Oncol 2016) demonstrated an overall survival benefit for immediate compared to delayed androgen deprivation therapy (ADT) for patients with recurrent prostate cancer or unsuitable for curative treatment. However, the extent to which TOAD trial results can be replicated in real-world population-based observational data is unknown. For these reasons, we used TOAD trial data and criteria to define a similar retrospective cohort of men with rising PSA after definitive treatment of localized prostate cancer to compare immediate versus delayed ADT for recurrent non-metastatic disease. Methods: We used two data sources to conduct this study: 1) TOAD trial data, and 2) national electronic health record, cancer registry, administrative, pharmacy, and laboratory data for men treated with radical prostatectomy or radiotherapy for localized prostate cancer and laboratory confirmed biochemical recurrence (BCR) between 2005- 2015 in the Veterans Health Administration. We defined BCR according to guidelines and used National Death Index data through 2019 for survival outcomes. We designated randomization to the ‘immediate’ and ‘delayed’ arms for men receiving ADT within (immediate) or after (delayed) 2 years of BCR in the observational data. We used Cox proportional hazards regression models adjusting for age, PSA at BCR, comorbidities, and Gleason score, stratified by prior ADT exposure, to compare survival outcomes for immediate versus delayed ADT. Results: We identified 9,866 men with BCR after radical prostatectomy, 3,431 after radiotherapy alone, and 2,703 after radiotherapy and adjuvant ADT. The mean age at randomization for the TOAD cohort was older (74.8 vs. 75.9 years, delayed vs. immediate) compared to our observational cohort (68.1 vs. 67.0 years, delayed vs. immediate). In unadjusted survival analyses, we found both overall and prostate cancer-specific survival were higher among patients with delayed ADT (log-rank p-value < 0.001 for both). Our multivariable survival models among men with and without prior ADT exposure also demonstrated survival benefit associated with delayed ADT (prior ADT exposure, adjusted hazard ratio (aHR) 2.3, 95% confidence interval (CI) 1.7, 3.0, no prior ADT exposure 2.6, 95% CI 2.4, 2.8). Conclusions: The benefit of immediate ADT administration compared to delayed as reported within the TOAD study was not found in a comparable observational cohort of men with non-metastatic recurrent prostate cancer. Even after multivariable adjustment, we found clear association of delayed ADT administration with improved survival outcomes suggesting selection bias and unobserved confounding, or that delayed ADT may be warranted for selected patient populations necessitating further investigation. Source of Funding: NCI R01 CA242559.