Abstract Background: MRI measured functional tumor volume (FTV) can predict pathologic complete response (pCR) to neoadjuvant chemotherapy (NAC) as early as 3 weeks after treatment initiation (1), indicating the potential of using MRI to guide treatment de-escalation. We developed MRI based, subtype-specific models for predicting pCR, to be used as part of a de-escalation strategy combining MRI and inter-regimen core biopsy pathology in I-SPY 2. Methods: I-SPY 2 patients underwent MRI exams at pre-treatment, early treatment (3 wks), inter-regimen (12 wks), and pre-surgery. pCR was assessed at surgery. FTV was calculated semi-automatically for every MRI (2). Subtype-specific FTV-based MRI prediction models were trained using FTV measurements at baseline, early treatment and inter-regimen timepoints for patients enrolled in I-SPY 2 between May 2010 and November 2016. Breast cancer subtype was defined by hormone receptor (HR) and human epidermal growth factor receptor 2 (HER2) status. The MRI prediction model was then used to predict probability of achieving pCR for each patient at inter-regimen, based on their subtype. The therapy de-escalation strategy focuses on achieving high positive predictive value (PPV, correct identification of patients with pCR) while maximizing sensitivity (proportion of patients with pCR identified by the test). Predicted probability above a specified threshold was considered a positive test for pCR. This study investigated the tradeoff between PPV and sensitivity within subtype groups defined by HR and HER2, and held to constraints set by probability thresholds at the 1st, 2nd (median) and 3rd quartile. Results: A total of 814 patients were included in the analysis. Median age was 49 (range: 24 - 77) years. The pCR rate was 36% (289/814). Table 1 shows patient number and pCR rate by HR/HER2 subtype. The subtype-specific MRI models consist of the predictors: change of FTV (dFTV) at inter-regimen for HR+/HER2- and HR-/HER2+; dFTV at early treatment for HR+/HER2+; pre-treatment FTV and dFTV at inter-regimen for triple negatives. The highest probability varied by subtype: 0.24 for HR+/HER2-, 0.61 for HR+/HER2+, 0.73 for HR-/HER2+, 0.68 for triple negatives. The maximum PPV was 67% for HR+/HER2- and 100% for all other subtypes. Table 1 shows the tradeoff between PPV and sensitivity by subtype using the 1st, median, and 3rd quartile thresholds. Conclusions: Our data demonstrate that PPV and sensitivity vary by breast cancer subtype when the probability threshold generated by MRI model increases from low to high quartile. Results from this study suggest that the probability threshold for recommending treatment de-escalation should be selected carefully based on breast cancer subtype. Imaging results will be combined with core biopsy information obtained at the 12-week timepoint to further improve overall accuracy. Table 1 Tradeoff between positive predictive value (PPV) and sensitivity at different levels of PPVCohortNpCR rateProbability thresholdPPV (%)Sensitivity (%)HR+/HER2-32820% (64/328)1st quartile0.1824912nd quartile0.2128773rd quartile0.233552HR+/HER2+13239% (51/132)1st quartile0.2744862nd quartile0.4053693rd quartile0.515837HR-/HER2+7166% (47/71)1st quartile0.6775852nd quartile0.7181623rd quartile0.728934Triple negative(HR-/HER2-)28345% (127/283)1st quartile0.3454912nd quartile0.5262703rd quartile0.607341 Citation Format: Wen Li, David C Newitt, Jessica Gibbs, Lisa J Wilmes, Ella F Jones, Natsuko Onishi, Bonnie N Joe, Elissa Price, John Kornak, Christina Yau, Denise M Wolf, Barbara LeStage, Susan Samson, the I-SPY 2 Imaging Working Group, the I-SPY 2 Consortium, Laura J Esserman, Nola M Hylton. Subtype-specific MRI models to guide selection of candidates for de-escalation of neoadjuvant therapy [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PD6-05.