Abstract Homologous recombination deficiency (HRD) phenotype often observed in tumors with BRCA1/2 deficiency is considered to be associated with efficacy of PARP inhibitors and platinum-based therapies. However, analysis on the degree of HRD requires comprehensive “genomic scar” analyses like Myriad MyChoice, therefore, we sought to develop a simple HRD prediction model in breast cancer patients without compromising its predictive value. HRD score as a consequence of genomic scars was calculated counting loss of heterozygosity score (LOH), telomeric allelic imbalance score (TAI), and large-scale state transition score (LST) for three independent breast cancer cohorts; TCGA (n= 744), ICGA (International Cancer Genome Consortium, n= 477), and Japanese AYA cohorts (n= 46). The HRD-high phenotype was defined as HRD scores ≥ 42. Genetic and pathological factors in HRD-high cases were determined by analyzing the sequencing data and DNA methylation chip analysis data as well as clinico-pathological information of the cohorts. The sequencing data comprised somatic mutation status of 82 significantly mutated genes (SMGs), germline/somatic mutations of 25 cancer susceptibility genes and three other HR-related genes. First, we adopted an Elastic Net regularized regression approach that controls for co-varying features within high-dimensional data in a TCGA breast cancer (Area under the curve [AUC] = 0.876), then applied the model onto two validation datasets. Among 16 selected features in the model, features with top-10 coefficient values were hypermethylation status, germline mutations of BRCA1, BRCA2, and PALB2, somatic mutations of BRCA1, BRCA2, PALB2 and TP53, triple negative subtype, and higher nuclear grade. The AUC values in ICGC and Japanese AYA were 0.899 and 0.941, respectively. Second, a simple predictive model for the HRD-high phenotype was developed using a TCGA data set (Area under the curve [AUC] = 0.838) based on only 5 features which were statistically significant (P < 0.01) in logistic regression analysis; germline BRCA1, BRCA2 and somatic TP53 mutations, triple negative subtype and higher nuclear grades. Its prediction power was validated in ICGC (AUC = 0.873) and Japanese AYA cohorts (AUC = 0.936). After excluding most features in the comprehensive Elastic Net model, the simple prediction model derived from five features still showed high AUC values comparable to those from the comprehensive models in the validation sets. These five features can be assessed by gene panel tests and daily pathological analyses. Thus, this study clarifies genomic and pathological factors associated with HRD phenotype of breast cancer and provides a simple predictive model of HRD to identify breast cancer patients who would benefit from PARP and/or platinum therapies in the clinical setting. Citation Format: Maki Tanioka, Tomoko Watanabe, Takayuki Honda, Hirohiko Totsuka, Eri Arai, Yae Kanai, Kouya Shiraishi, Kenji Tamura, Takashi Kohno. Simple vs. comprehensive prediction models of homologous recombination deficiency based on mutational and clinical features in three independent breast cancer datasets [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 3551.