运用临床和影像学方法预测乳腺癌患者乳头—乳晕复合体有无侵犯

Clinical and radiological predictors of nipple-areola complex involvement in breast cancer patients

• Marta D’Alonzo^e, 
• Laura Martincich^b, 
• Nicoletta Biglia^e, 
• Alberto Pisacane^c, 
• Furio Maggiorotto^a,
• Giovanni De Rosa^f, 
• Filippo Montemurro^d, 
• Franziska Kubatzki^a, 
• Piero Sismondi^e, 
• Riccardo Ponzone^a, , 

• ^a Division of Gynecological Oncology, Institute for Cancer Research and Treatment (IRCC) of Candiolo, Fondazione del Piemonte per l’Oncologia, Strada Provinciale 142, Km 3.95 – 10060 Candiolo, Turin, Italy
• ^b Division of Radiology, Institute for Cancer Research and Treatment (IRCC) of Candiolo, Fondazione del Piemonte per l’Oncologia, Strada Provinciale 142, Km 3.95 – 10060 Candiolo, Turin, Italy
• ^c Division of Pathology, Institute for Cancer Research and Treatment (IRCC) of Candiolo, Fondazione del Piemonte per l’Oncologia, Strada Provinciale 142, Km 3.95 – 10060 Candiolo, Turin, Italy
• ^d Academic Division of Medical Oncology, University of Turin, Institute for Cancer Research and Treatment (IRCC) of Candiolo, Fondazione del Piemonte per l’Oncologia, Strada Provinciale 142, Km 3.95 – 10060 Candiolo, Turin, Italy
• ^e Academic Division of Gynecological Oncology, University of Turin, A.O. Ordine Mauriziano, Largo Turati 62, 110128 Turin, Italy
• ^f Division of Pathology, A.O. Ordine Mauriziano, Largo Turati 62, 110128 Turin, Italy

• http://dx.doi.org/10.1016/j.ejca.2012.04.017, How to Cite or Link Using DOI

• Permissions & Reprints

---

Keywords

• Breast cancer; 
• Nipple-areola complex sparing mastectomy; 
• Magnetic resonance; 
• Mammography

---

1. Introduction

Nipple-areola sparing mastectomy (NSM) is currently considered as a possible alternative to skin sparing mastectomy (SSM) in selected breast cancer patients undergoing immediate breast reconstruction.^1 and 2The oncological safety of NSM, although not formally proven, is supported by a convincing rationale and also by many retrospective^{3, 4, 5, 6, 7 and 8} and prospective clinical data.⁹ In the absence of randomised clinical trials, whose implementation is often difficult in surgery,¹⁰ the introduction of NSM must be regulated and carried out gradually. This is essential to minimise the possible negative effects in terms of surgical and oncological risks.

A crucial issue is the possibility to predict preoperatively tumour involvement of the nipple-areola complex (NAC) in order to select which patients may be candidates to a NSM.¹¹ The risk of NAC involvement has been assessed in several retrospective analyses of large datasets, but the issue is of considerable interest because literature data on the best clinical and radiological predictors are inconsistent.¹

In the current study, we analysed a mono-institutional series of total mastectomies without NAC preservation with the following aims: (A) to estimate the actual rate of NAC involvement; (B) to assess the value of mammography (MX) and magnetic resonance imaging (MRI) in predicting NAC involvement; (C) to validate the performance of existing models for the prediction of NAC involvement.

2. Patients and methods

2.1. Patient selection

All patients submitted to total mastectomy from January 2000 to July 2010 were searched in the institutional database. Out of 1359 patients retrieved, we selected those who had a scheduled appointment at the follow up clinic within the next 4 weeks. This selection was decided in order to acquire the radiograms in the shorter and most efficient way. In fact, although all patients underwent MX before surgery, we had no copy available at the Institute. Overall, 78 out of 147 patients contacted brought their original MX (MX group). The second group included all patients submitted to mastectomy from January 2009 to March 2010 who underwent preoperative MRI of the breast (54 out of 179) at the Institute (MRI group). This limited time-frame was chosen to uniform the technique of image acquisition and processing. Indications for performing preoperative MRI were: age below 40 years of age, lobular histology, suspicion of multifocal disease, or equivocal MX findings. No overlap existed in the patients’ composition of the two groups.

Patients with inflammatory breast cancer, clinically evident tumour involvement of the NAC, Paget’s disease and phylloides tumour were excluded, as well as all patients with a negative and/or low quality preoperative MX or MRI or those who underwent preoperative chemotherapy. Patients with skin and/or nipple retraction were not excluded because this does not necessarily imply tumour infiltration. All clinical and pathological information such as age, menopausal status, tumour histology, location (central versus peripheral), multifocality and/or skin involvement were obtained through a prospectively maintained institutional database.

2.2. Radiological evaluation

All examinations were blindly reviewed by a radiologist (L.M.) with 15 years of experience in breast imaging. In the MX group and MRI groups, 61/78 (78.2%) and 39/54 (72.2%) cases were enrolled respectively. The remaining cases were excluded because either the quality of the images was too low (six MX and three MRI), or both (cranio-caudal and oblique) views were not available (four MX), or the lesion was not visible (six MX and one MRI), or the patient underwent preoperative chemotherapy (nine MRI) or the final diagnosis was phylloides tumour (one MX) or Paget’s disease (two MRI).

MX was performed by two standard views, cranio-caudal and mediolateral oblique, either with analogic or digital technique. The diameter of the lesion was defined as the maximum extension of suspicious opacity, architectural distortion and/or suspicious microcalcifications. In the case of bifocal, multifocal or multicentric lesions, these were considered as a single mass and the reference measure reflected the whole area occupied in the breast (Fig. 1). The distance between tumour and NAC (tumour–NAC distance) was defined as the minimum distance between the base of NAC and the nearest margin of the lesion. For both views, two measures were taken parallel to the nipple and to the ducts respectively and the shorter was chosen as the reference distance (Fig. 2). MRI examinations were acquired with a 1.5 T equipment (HDx Signa Excite, GE HealthCare, Milwaukee) and dedicated phased-array 8-channel coil, using a dynamic technique (DCE-MRI) following the recommended requirements.¹² Multiplanar reconstructions (MPR) from subtracted images (postcontrast series–precontrast acquisition) were used to assess the diameter of the lesion, which was defined as the maximum extent of suspicious enhancement. As for Mx, in the case of bifocal, multifocal or multicentric lesions, these were considered as a single mass and the reference measure reflected the whole area occupied in the breast. A conventional measure of the larger tumour foci was also recorded in this group. The tumour NAC-distance was measured by electronic calipers, on both axial and sagittal Maximum Intensity Projection (MIP) images. The distance between lesion and NAC was defined as the minimum distance between the base of the NAC and the nearest margin of the lesion.

2.3. Histopathological evaluation

In the original pathological report, gross and histologic assessment of the mastectomy specimens were performed according to an institutional protocol with conventional haematoxylin-eosin stained sections. The NAC was routinely inspected in all mastectomies, including a single sagittal section through the nipple. The subareolar margin was dissected immediately on the fresh specimen to reproduce a surgical margin at 4–5 mm from the base of the nipple. The NAC was considered involved in case of invasive ductal/lobular carcinoma and/or ductal carcinoma in situ within the subareolar margin, but not in the case of presence of lobular carcinoma in situ/lobular intraepithelial neoplasia. The pathological report was retrieved for all patients from the MX and MRI groups to examine whether NAC involvement was explicitly mentioned or not. For 13 dubious cases, the original pathological slides were blindly reviewed by two pathologists with specific experience on breast pathology (A.P. and G.D.R.) unaware of imaging data.

2.4. Prediction of NAC involvement

The likelihood of NAC involvement was calculated by applying the algorithms proposed by Schecter et al.¹³and Loewen et al.¹⁴ (Fig. 3). The sensitivity (Se), specificity (Sp), positive predictive value (PPV), negative predictive value (NPV) and accuracy (Acc) in predicting NAC involvement was calculated for all algorithms. Algorithms were tested using the MX and MR data separately. Receiving operating curves (ROC) were constructed to detect the best cut-off point to allow the optimal balance between Se and Sp either by MX or MRI.

2.5. Statistical analysis

Statistical significance was set at p values < 0.05 with two-tailed test. The quantitative variables were compared with the Pearson chi-square test or Fisher’s exact test. Qualitative variables were compared using the analysis of variance. The normality of variables was tested by the Kolmogorov–Smirnov procedure. Variables not normally distributed were analysed using the non-parametric Mann–Whitney U test. Multivariate analysis was performed using the logistic regression model. Statistical analysis was performed with SPSS for Windows.

3. Results

3.1. Correlation between clinical pathologic factors and NAC involvement

At histopathological analysis, 14 out of 100 cases showed tumour cells in the NAC. At univariate analysis, menopausal status, tumour histology, tumour grade, immunohistochemical characteristics and multifocality did not influence NAC status. Although positive axillary lymph node status and larger tumour size were more common in the NAC positive group, the difference did not reach statistical significance (p value: 0.09 and 0.08, respectively). Conversely, tumour stage was positively associated with NAC involvement (p value: 0.03) (Table 1). Tumour–NAC distance showed the highest association with NAC involvement (p value: 0.006), as well as two closely related parameters such as central tumour location (p value: 0.004) and presence of NAC retraction (p value: 0.001).

Since both the measurements of tumour size and tumour–NAC distance may vary with different imaging techniques, the analysis was performed separately for the MX and MRI groups. In the MX group, all significant parameters at the combined MX + MRI analysis lost their statistical significance. In contrast, in the MRI group both lesion size (p value: 0.008) and tumour–NAC distance (0.009) retained their statistical significance (Table 2). Tumour–NAC distance remained a significant predictor of NAC involvement by setting the cut off at 5 mm (p value: 0.003), 10 mm (p value: 0.002), 20 mm (p value: 0.029). We also performed a multivariate analysis which showed that a tumour–NAC distance ⩽10 mm was the only factor independently associated with NAC involvement (relative risk: 9.55; p value 0.05) (data not shown).

3.2. Prediction of NAC involvement by tumour–NAC distance

Tumour–NAC distance resulted in a key predictor of NAC involvement. Overall, the area under the curve (AUC) of the ROC curve was 0.782 (Fig. 4), and was larger for MRI than for Mx (0.888 and 0.677 respectively). The 10 mm cut-off value allowed the best balance between Se and Sp, being associated with a NPV for NAC involvement of 94% and 100% with MX and MRI respectively (Table 3).

3.3. Validation of the algorithms

The results from our series were used to validate the algorithms by Schecter et al.¹³ and Loewen et al.¹⁴Measurements obtained with MRI and MX were analysed separately (Table 4). Overall, the performance of Schecter’s algorithm¹³ was inferior as compared to the original series due to its very low Sp (35% and 22% for MX and MRI respectively), despite its good Se (86% and 100% for MX and MRI respectively). Conversely, we were able to reproduce Loewn’s original results (Se 57%, Sp 72% for MX) and even improve their overall performance when MRI was used for measuring tumour–NAC distance (Se 100%; Sp 69%).¹⁴

4. Discussion

Occult involvement of the NAC is reported in 0–58% of mastectomy specimens.¹ Larger series including at least 100 patients show lower variability of occult NAC involvement (5.6–26.4%),¹ and are consistent with our finding (14%).

The pathological protocol for NAC evaluation, as well as pathological data collection, have a major role in determining the rate of NAC involvement. In many series, the NAC was assessed exclusively through the consultation of pathological reports. According to Schecter et al.¹³ this is not appropriate since only 4 cases of NAC involvement were identified by simply checking medical records in a series of 48 mastectomies, with further 9 cases identified after a pathological review of all cases. At our institution the pathological protocol requires the inspection of the NAC in all mastectomy specimens. Therefore, whenever the report contained a description of the NAC status (free of or involved with cancer), we considered the data as reliable. Only if the NAC status was not clearly stated, we asked the pathologists to review the slides. In our series, this procedure identified 2 additional positive NAC (from 12 to 14) out of 13 cases re-evaluated.

In our study, NAC involvement was defined by the presence of invasive carcinoma and/or ductal carcinoma in situ at the subareolar margin. This boundary has clinical relevance as it represents the surgical plane of dissection of a NAC-sparing mastectomy. We used a 4–5 mm cut-off for our analysis since this is the mean thickness of the skin flaps of a mastectomy performed at the level of the superficial fascia dividing the subcutaneous fat from the breast glandular tissue.¹⁵ This was a conservative decision, since recent data¹and our own experience¹⁶ suggest that when performing a NAC sparing mastectomy the dissection plane can be conducted even closer to the base of the nipple, including the whole duct bundle, with a reasonably low risk of necrosis.

Our low rate of NAC positivity may depend by the pathological protocol which did not require serial coronal sections of the nipple and by the exclusion of lobular carcinoma in situ (LCIS) from our criteria of NAC involvement. Actually, in studies^17 and 18 that include LCIS, the incidence of malignant NAC involvement is significantly higher than the average (30–58%). We believe LCIS should not be regarded as evidence of NAC involvement since it cannot be considered a true cancer lesion and does not mandate further surgical excision when detected at the surgical margins.

Tumour size and tumour distance from the NAC show respectively a positive and negative correlation with NAC involvement in most studies.^{11, 18 and 19} We showed that the risk of NAC involvement is strongly related to tumour–NAC distance and also to tumour size when measured by MRI. In order to allow external comparison, tumour size for multifocal tumours was measured either considering the separated tumour foci as single mass or as the diameter of the largest lesion, without any significant change of the results (Table 1 and Table 2). Conversely, tumour size measured by Mx showed only a non-significant correlation with NAC involvement, probably because it mainly reflects the invasive component of the tumour, while MRI often includes also the surrounding non-invasive component of the disease. According to some authors,¹¹ ductal carcinoma in situ is the most frequent type of cancer detected in the NAC and thus MRI measurement may better reflect the true risk of NAC involvement. Furthermore, in contrast with Schecter et al.,¹³ but in agreement with others,^18 and 20 we found that NAC involvement was significantly related to central location of the lesion, skin and/or nipple retraction and stage of disease.

We validated two dedicated algorithms in order to identify those patients whose risk is so low that a NAC sparing mastectomy may be justified. To this aim we tested separately the performance of MX and MRI in the models. In the algorithm by Schecter et al.,¹³ size, distance and tumour stage independently predicted NAC involvement, while Loewen et al. used only distance in their model. By entering our data in the Schecter’s algorithm, we found significantly lower Sp, PPV and Acc.¹³ This was not influenced by our decision to measure multifocal lesions as a single mass; in fact the ‘size factor’ of the Schecter’s algorithm is set at 1 cm and all of our multifocal tumours were larger both considering the diameter of the larger lesion or the diameter of the area occupied by the cancer foci. Conversely, we obtained values of Se, Sp, PPV and NPV comparable to that achieved by Loewen et al. (Table 3).¹⁴ By analysing the data separately according to the type of imaging used, we found that MRI consistently outperformed Mx as far as Se, NPV and PPV were concerned in both models,^13 and 14 while Sp was significantly better for Mx as compared to MRI only in the Schechter’s model.

In a recent comprehensive review of the literature,² tumour size ⩽5 cm and tumour–nipple distance ⩾2 cm are proposed as inclusion criteria for performing a NAC sparing mastectomy, provided that retroareolar ducts do no show tumour involvement at pathological examination. Since our data suggest that tumour–NAC distance is the most important predictive factor, we tried to define the best cut-off to predict a low risk of NAC involvement. We found that the best compromise between specificity and sensitivity was obtained by setting the cut-off at 10 mm. Indeed, this cut-off at MRI was the only factor independently associated with NAC involvement at multivariate analysis and would have allowed to exclude NAC involvement in all cases of our series. Literature data on the best cut-off values of tumour–NAC distance are variable; only the study of Sacchini et al. support our 1 cm cut-off value,²¹ while most authors suggest that it should be set at ⩾2 cm.^{3,11 and 22}

Limitations of our study are mainly derived from its retrospective design. In order to limit selection biases, we prospectively included those patients with a scheduled appointment at the clinic, but we could not exclude image based- selection and study examination biases. Reader-order biases were avoided since the radiologist and pathologists were unaware of the pathological and radiological report respectively. Unfortunately, a proper comparison between MX and MRI was not possible in our study since only a few patients had both exams performed.

In conclusion, occult tumour involvement of the NAC is detected in a minority of breast cancer patients submitted to mastectomy. Although, several parameters may predict the likelihood of NAC involvement, our data suggest that multiparametric models do not increase the overall accuracy obtained by the measurement of tumour–NAC distance. A distance of at least 10 mm, especially if measured by MRI, allows optimal discriminative power between cases at high versus low likelihood of NAC involvement and can virtually exclude NAC involvement. Caution is required in interpreting our data, since numbers are small and the study may be underpowered to detect significant differences. Furthermore, only prospective studies of NAC-sparing mastectomy in breast cancer patients will be able to assess the best parameters to predict NAC involvement and long term follow up data will be required to assess the oncological safety of the procedure. In the meantime, it appears that the decision to preserve the NAC in breast cancer patients undergoing mastectomy may be guided by MRI measurement of tumour–NAC distance in addition to the intraoperative pathological evaluation of the retroareolar ducts.

Conflict of interest statement

None declared.