Purpose: Factors associated with invasive recurrence (REC) of ductal carcinoma in situ (DCIS) are less known. This study was aimed at identifying better biomarkers to predict the prognosis of DCIS. Methods: RNA extracted from formalin-fixed paraffin-embedded blocks of twenty-four pure DCIS cases was subjected to differential gene expression analysis. The DCIS cases were selected by matching age and estrogen receptor status. Sixteen REC-free and 8 invasive-REC cases with disease-free interval of > 5 years were analyzed. Immunohistochemistry (IHC) staining was used to validate sixty-one independent pure DCIS cases, including invasive-REC (n = 16) and REC-free (n = 45) cases. Results: Eight differentially expressed genes (DEGs) were statistically significant (log 2-fold change [FC] < –1 or > 1 and p < 0.001). Less than ½ fold expression of CUL1, androgen receptor (AR), RPS27A, CTNNB1, MAP3K1, PRKACA, GNG12, MGMT genes was observed in the REC group compared to the no evidence of disease group. AR and histone deacetylase 1 (HDAC1) genes were selected for external validation (AR: log 2-FC − 1.35, p < 0.001, and HDAC1: log 2-FC − 0.774, p < 0.001). External validation showed that the absence of AR and high HDAC1 expression were independent risk factors for invasive REC (hazard ratio [HR], 5.04; 95% confidence interval [CI], 1.24–20.4; p = 0.023 and HR, 3.07; 95% CI, 1.04–9.04; p = 0.042). High nuclear grade 3 was also associated with long-term invasive REC. Conclusion Comparative gene expression analysis of pure DCIS revealed 8 DEGs among recurring cases. External validation with IHC suggested that the absence of AR and overexpression of HDAC1 are associated with a greater risk of long-term invasive REC of pure DCIS.