Abstract P5-07-04: Pathology data predicts MammaPrint result- The Magee MammaPrint equation

Abstract Introduction Prognosis for breast cancer patients may be determined using clinical-pathologic data (CPD) or gene expression profiling (GEP). We have previously reported a combined morphologic and immunohistochemical (IHC) method (Magee Equations) that can be used to estimate the Oncotype DX (ODX) recurrence score in clinical practice (http://path.upmc.edu/onlineTools/mageeequations.html ). MammaPrint (MP) is a 70 gene GEP assay that is used to assess prognosis. Our goal was to develop similar equations to estimate the MP result based on CPD. Methods The study included 344 patients who had MP and compete CPD in an IRB approved research setting. Using the available CPD, a logistic regression model was constructed to predict MP risk group classification. Odds ratios and corresponding 95% confidence intervals were calculated. Model estimates were used to create an equation to predict MP risk category. The risk group cut-off for predicted scores was chosen to give 0.90 sensitivity with resulting specificity of 0.51 based on the observed MP scores. This corresponds to 10% of subjects in the observed MP high risk group being predicted as low risk. The sensitivity was chosen to minimize the misclassification rate of the high risk group. Internal validation of the prediction model was determined using the bootstrap method. For each of the 1,000 iterations, the bootstrap sample was used to generate the prediction equation and the cutoff value as described above. Subsequently, this equation was used to predict the risk group for the observations not contained in the bootstrap sample (out-of-bag observations) and the resulting sensitivity and specificity were recorded. Results The logistic regression model from the observed data is: y = 0.7495(Age< 50) - 1.4135(Nottingham Grade 1) + 0.4644 (Nottingham Grade 2) + 1.7437 (Nottingham Grade 3) -0.0038*(ER H-Score)– 0.0044* (PR H-score) + 1.9220 (HER2 Positive) The probability of being classified as high risk is then given by: p=exp⁡(y)/(1+exp⁡(y)) The corresponding predicted risk group is defined as: Low Risk: p < 0.18 High Risk: p≥0.18 The 0.18 cut-off was chosen to yield 90% sensitivity. For subjects with a known MP result of high risk, the probability that the model will predict high risk is 0.90. As a result of this sensitivity, the specificity is 51%. For subjects with a known MP result of low risk, the probability that the model will predict low risk is 0.51. The sensitivity was chosen to minimize the misclassification rate of the high risk group. The median with 1st and 3rd quartiles for the cut-off value from the bootstrap prediction equations is 0.18 (0.16, 0.20). The median with 1st and 3rd quartiles for the sensitivity and specificity for the predicted scores of out-of-bag observations are 0.89 (0.85, 0.93) and 0.50 (0.45, 0.54), respectively. Conclusion The pathology data can predict for high risk MP result with a high sensitivity and moderate specificity. Citation Format: Dabbs DJ, Cooper KL, Brufsky A, Rosenzweig M, Bhargava R. Pathology data predicts MammaPrint result- The Magee MammaPrint equation. [abstract]. In: Proceedings of the Thirty-Eighth Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2015 Dec 8-12; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2016;76(4 Suppl):Abstract nr P5-07-04.