IRAIS 2026 Proceedings
Abstract
Breast cancer risk prediction is not merely a classification task but a clinical decision-support problem. Risk assessment models estimate an individual's probability of developing breast cancer over a specified time horizon to guide screening, preventive counseling, genetic testing referrals, and risk-reducing interventions (American Cancer Society, 2023). Because breast cancer risk arises from the interaction of demographic, reproductive, genetic, lifestyle, and clinical factors (Dolatkhahi et al., 2021), effective decision support requires more than predictive accuracy. The resulting artifact must integrate clinical knowledge, provide transparent and interpretable explanations, and generate robust predictions across alternative evaluation settings (Lundberg & Lee, 2017; Lundberg et al., 2020). Accordingly, this study conceptualizes breast cancer risk prediction as the design and evaluation of an explainable clinical decision-support artifact rather than simply a machine learning model. The novelty of this study is not any single method alone, but the integration of expert validation, structural risk-factor modeling, and screening-oriented prediction into one reusable clinical decision-support design process. This perspective aligns with the Design Science Research Methodology (DSRM), which emphasizes the creation and rigorous evaluation of artifacts that solve important real-world problems while generating knowledge of prescriptive design (Peffers et al., 2007).
Research Question
This research addresses the following design question: “How can an explainable decision-support artifact be designed and evaluated to support breast cancer risk prediction by integrating expert-validated risk factors, structural modeling, and robust predictive analytics?” Following this question, the artifact was designed around four objectives: grounding in expert-validated clinical risk factors, organization of those factors into interpretable structural roles, screening-oriented predictive performance that prioritizes clinical sensitivity, and stable performance across alternative evaluation conditions.
Data and Methodology
This study adopts Design Science Research Methodology (DSRM) (Peffers et al., 2007) to guide the design, development, demonstration, and evaluation of an explainable breast cancer risk prediction artifact. The artifact is an integrated decision-support model artifact rather than a fully deployed clinical software system. Its inputs are patient-level breast cancer risk factors, and its primary output is a predicted probability of breast cancer risk. The intended users are clinicians who may use the artifact during patient intake or screening reviews to support follow-up prioritization. The artifact comprises a Knowledge Component grounded in expert consensus, a Structural Component that models the relationships among the validated risk factors, and a Predictive Component based on machine learning.
Contributions and Conclusion
This study makes three main contributions. First, it contributes to Decision Support Systems research by framing breast cancer risk prediction as an artifact design problem rather than only a machine learning classification task. This is important because clinical decision support requires not only predictive accuracy, but also interpretability, stability, and usefulness for decision making (Vasey et al., 2022). Second, the study contributes methodologically by offering a reusable design pattern for clinical prediction artifacts: validate the input factors with experts, model the structural roles of those factors before prediction, embed them into a machine learning evaluation pipeline, and assess the final artifact for sensitivity, interpretability, and stability. Combining expert validation, structural modeling, and predictive analytics therefore provides a transparent pathway for designing clinically meaningful prediction artifacts. Third, the study offers practical value by demonstrating how an explainable and robust artifact can be assembled to support breast cancer screening prioritization.
Overall, this research shows that combining expert-validated risk factors with explainable and robust machine learning can produce a decision-support artifact that is more useful than a standalone predictive model. In use, the artifact can support clinicians by converting patient-level risk factors into a predicted probability that can inform screening review and follow-up prioritization.
References
American Cancer Society. (2023). American Cancer Society recommendations for the early detection of breast cancer.
Dolatkhahi, K., Azar, A., Karimi, T., & Hadizadeh, M. (2021). Diagnosis of breast cancer using machine learning. Payavard Salamat, 15(4), 340–352.
Lundberg, S., & Lee, S.-I. (2017). A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems, 30.
Lundberg, S. M., Erion, G., Chen, H., DeGrave, A., Prutkin, J. M., Nair, B., Katz, R., Himmelfarb, J., Bansal, N., & Lee, S.-I. (2020). From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2(1), 56–67.
Peffers, K., Tuunanen, T., Rothenberger, M. A., & Chatterjee, S. (2007). A design science research methodology for information systems research. Journal of Management Information Systems, 24(3), 45–77.
Vasey, B., Nagendran, M., Campbell, B. et al. (2022). Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. Nature Medicine, 28, 924–933.
Recommended Citation
Dolatkhahi, Kasra and Mulgund, Pavankumar, "Designing Explainable Decision Support for Breast Cancer Risk Prediction" (2026). IRAIS 2026 Proceedings. 4.
https://aisel.aisnet.org/irais2026/4
Abstract Only