Nonsurgical treatment options are increasingly needed for endometrial atypical hyperplasia (AH) and endometrioid endometrial cancer (EEC). Despite promising initial response rates, prospective long-term data and determinants for relapse are limited. Follow-up data from patients in our prospective phase II trial of LIUD for AH/G1EEC were collected from medical records. Spatial transcriptomics (Nanostring GeoMX digital spatial profiling) with in silico cell type deconvolution and pathway analyses were employed on longitudinal biopsy samples from five patients across pre-treatment, on-treatment, and relapse. Of 43 participants exhibiting initial response to LIUD, 41 had follow-up data. Sixteen (39%) experienced relapse. Clinical factors associated with shorter response duration included younger age, initial diagnosis of G1EEC, lack of response at six months, premenopausal status, and Hispanic ethnicity (p<0.05), but only six-month response status remained a significant predictor in a multivariate model (p=0.023). LIUD increased abundance of NK cells (DMCP-counter score=46.13, FDR=0.004) and cytotoxic lymphocytes (DMCP-counter score=277.67, FDR=0.004), as well as lymphocyte cytotoxicity markers PRF1 (log2FC=1.62, FDR=0.025) and GZMA (log2FC=2.47, FDR=0.008). NK cells were reduced at relapse (DMCP-counter score=-55.96, FDR=0.02). Immune-related pathways (IFNα-response and TGFβ-signaling) were enriched at relapse (FDR<0.05). IDO1 expression, reflecting immune exhaustion, was upregulated at relapse (FDR<0.05). Upfront resistance and relapse after initial response to LIUD for AH/G1EEC impacts nearly half of patients, remaining a major hurdle for non-surgical treatment of AH/G1EEC. Molecular studies evaluating longitudinal biopsies from a small cohort implicate immune mechanisms at relapse, including reversal of progestin-related immunomodulation and increased immune exhaustion.