Breast Cancer
Conditions
Keywords
Artificial Intelligence, Digital Pathology, Neoadjuvant Chemotherapy, Pathological Complete Response, Foundation Model, Treatment Response Prediction, Breast Neoplasms
Brief summary
This study evaluates the real-world clinical workflow integration of a previously developed artificial intelligence (AI) prognostic test in breast cancer patients receiving neoadjuvant chemotherapy, and validates its accuracy in predicting treatment response. The Ataraxis AI test analyzes digitized images of tumor biopsy slides combined with basic clinical information (age, tumor stage, hormone receptor status) to generate a risk score. Prior studies showed the AI test can predict cancer recurrence with accuracy comparable to or better than existing genomic tests. The study has two stages: * Stage 1 (30 patients): Assess whether the AI test can be practically integrated into routine clinical workflow, including ease of use, report clarity, and time requirements. * Stage 2 (70-120 additional patients): Validate the accuracy of AI-predicted pathological complete response (pCR) rates against actual surgical outcomes. This study uses a blinded design where treating physicians remain blinded to AI results until post-surgical pCR assessment. AI analysis is performed by the research coordinator in collaboration with Ataraxis. After pCR evaluation, AI results are disclosed and physicians complete surveys assessing hypothetical treatment changes. This design eliminates AI influence on treatment decisions and ensures independent validation. Participants are adults with Stage I-III breast cancer planned for neoadjuvant chemotherapy. The study involves no additional procedures beyond standard care except for completing surveys about the AI test experience.
Interventions
DIAGNOSTIC_TESTmulti-modal foundation AI test
Multi-modal AI test combining digital pathology features from H\&E-stained core needle biopsy slides with clinical information (age, molecular biomarkers, TNM stage) to generate a continuous risk score (0-1) predicting pathological complete response. Results provided as reference information only; does not influence treatment decisions.
Sponsors
Young-Joon Kang
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
FEMALE
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
* Histologically confirmed Stage I-III invasive breast cancer * Planned for neoadjuvant chemotherapy * H\&E-stained slides available from core needle biopsy * Age 18 years or older * Able to provide written informed consent
Exclusion criteria
* Metastatic breast cancer (Stage IV) * Not a candidate for neoadjuvant chemotherapy * H\&E slides not obtainable from core needle biopsy * Unable to provide informed consent
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Stage 1 - Feasibility: Clinical Workflow Compatibility Score | Within 4 weeks after surgery following NAC completion (approximately 5-7 months per participant) | Mean score on 5-point Likert scale assessing AI system integration into existing clinical workflow, including ease of use, report comprehensibility, credibility, and time burden. Higher scores indicate better compatibility. |
| pCR Prediction: pCR Prediction Accuracy (AUC-ROC) | Within 4 weeks after surgery following NAC completion (approximately 5-7 months per participant) | Area under the receiver operating characteristic curve for AI-predicted pCR probability versus actual pathological complete response status (defined as ypT0/is ypN0). |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Subtype-specific pCR Prediction Accuracy | Within 4 weeks after surgery following NAC completion (approximately 5-7 months per participant) | AUC-ROC for pCR prediction analyzed separately for each molecular subtype: TNBC, HER2+, and HR+/HER2-. Descriptive statistics only. |
| Sensitivity and Specificity of pCR Prediction | Within 4 weeks after surgery following NAC completion (approximately 5-7 months per participant) | Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for AI-predicted pCR using predefined risk thresholds. |
| AI Test Processing Time | Within 2 weeks after enrollment | Time in days from data upload to Ataraxis platform to AI result receipt. |
| Hypothetical Treatment Change Rate | After AI result disclosure following surgery (approximately 5-7 months per participant) | Proportion of cases where physicians indicate they would have modified treatment, assessed retrospectively after AI result disclosure following surgery and pCR evaluation. Includes regimen change, pembrolizumab addition/removal, cycle adjustment, or NAC omission. |
| Correlation Between AI Score and Established Prognostic Factors | Within 4 weeks after surgery following NAC completion (approximately 5-7 months per participant) | Spearman correlation coefficients between AI risk score and Ki67, tumor grade, clinical T stage, and clinical N stage. |
Countries
South Korea
Contacts
CONTACTYoung Joon Kang, Ph.D.
Outcome results
None listed