Development of a multi-institutional dataset to validate a novel inflammatory breast cancer diagnostic score.

Background: Susan G. Komen, the Inflammatory Breast Cancer (IBC) Research Foundation, and the Milburn Foundation convened patient advocates, clinicians, and researchers to propose novel quantitative scoring rubrics for IBC diagnosis. In this study, we developed a multi-institutional clinical dataset to test and validate the proposed scoring system.

Methods: IBC (N = 988) and non-IBC (N = 332) cases were identified at 2 institutions with dedicated multidisciplinary IBC programs. The non-IBC cohort included consecutive cT4b and cT4c patients. Standard operating procedures (SOPs) were developed for all ambiguous findings and languages. Three different methods were used for the imputation of missing data, resulting in 3 separate datasets. The sensitivity, specificity, and area under the receiver operator characteristic curve (AUC-ROC) were used to assess the discrimination of the proposed scoring rubric.

Results: The distribution of "true IBC" cases was 19.7% very likely IBC, 49.1% strong possibility of IBC, 0.4% weak possibility of IBC, 0.1% very unlikely IBC, and 30.7% unknown; corresponding groupings for true non-IBC cases were 0.6% very likely IBC, 51.8% strong possibility of IBC, 9.9% weak possibility of IBC, 2.1% very unlikely IBC, and 35.5% unknown. The AUC-ROC values for missing data imputation methods were similar (0.83-0.84); exploratory score refinement improved the AUC-ROC to 0.88-0.89.

Conclusion: Using the largest multi-institutional IBC clinical database to date, the score has been validated and is available for clinical use at https://www.komen.org/ibc-calc to assist health-care providers and their patients in IBC diagnosis. Exploratory score refinement demonstrates the potential to increase specificity; however, any change requires separate validation.