Development and validation of a model predictive of occult nipple involvement in women undergoing mastectomy

Background: This prospective study aimed to build a predictive model using preoperative information to aid selection for nipple-sparing mastectomy. Methods: Two hundred consecutive skin-sparing mastectomy specimens without overt nipple involvement were evaluated. Demographic, preoperative pathology and imaging information was collected. Nipple specimens (2 x 2 x 2 cm) were sectioned at 3-mm intervals. Haematoxylin and eosin-stained slides were examined by a breast pathologist for involvement by turnout. Logistic regression analyses of 65 therapeutic procedures identified factors associated with occult involvement and created a predictive model. This was tested on specimens from a further 65 therapeutic procedures. Results: Occult nipple involvement was noted in 32 (24.6 per cent) of 130 mastectomy specimens. In the training set, imaging diameter of the lesion and its distance from the nipple predicted nipple involvement on univariable analysis (P = 0.011 and P = 0.014 respectively). The multivariable logistic regression model was validated in the test set. The areas under the receiver-operating characteristic curve were 0.824 and 0.709 for the training and test sets respectively. Conclusion: Three-quarters of women undergoing mastectomy did not have occult nipple involvement. A clinical tool including turnout size and distance from the nipple has been developed to improve patient selection for nipple-sparing mastectomy.