A nomogram constructed using intraoperative ex vivo shear-wave elastography precisely predicts metastasis of sentinel lymph nodes in breast cancer

In our previous prospective study, a total of 228 axillary LNs, including sentinel LNs obtained from fifty-five patients who underwent breast cancer surgery in Gangnam Severance Hospital from May 2014 to April 2015, were investigated [20]; these patients were utilized as the development cohort. Another 80 patients diagnosed with breast cancer underwent breast cancer surgery at the same hospital from August 2015 to March 2016 were prospectively enrolled. Patients with stage IV cancer or those who received neoadjuvant chemotherapy were excluded. Unlike in the development cohort, 217 sentinel LNs obtained from these 80 patients examined using intraoperative ex vivo SWE were included in the validation cohort.

Our study was approved by the Institutional Review Board of Gangnam Severance Hospital and conducted in accordance with the Good Clinical Practice guidelines and the Declaration of Helsinki principles.

SLNB was performed using 99mTc-labeled tin colloid. Intradermal injection of 0.4 mL 30 MBq (0.8 mCi) 99mTc tin colloid diluted in normal saline solution was performed at four peri-areolar sites. Sentinel LNs were determined by employing a gamma detector (Gamma Detection System, Neoprobe Corporation). The LNs showing a high radioactivity were dissected, after which the gamma detector was used again to confirm the correct sentinel LNs. All radioactive LNs with an account equal to or greater than 10% of the highest radioactive LN were removed. When suspicious LNs not detected by the gamma probe were found after SLNB, they were removed and examined as non-sentinel LNs. If the result of frozen pathology was positive for malignancy, ALND was performed.

Excised LNs were delivered to the radiology part. Then, B-mode ultrasound with ex vivo SWE was performed as mentioned in our previous study [20]. In summary, the nodal size and elasticity values, such as mean stiffness and elasticity ratio of the sentinel LNs to the adjacent fat tissues, were examined during surgery via B-mode ultrasound and SWE using the Aixplorer ultrasound system (Super Sonic Imagine) and ShearWave^TM elastography mode [21, 22] with a 4–15-MHz linear transducer. During the examination, the radiologists need to handle the specimen carefully so as not to exert pressure on the excised LNs by the ultrasound transducer and sufficient contact jelly was used to avoid artificial stiffness. The longest diameter of the LN which was hypoechoic compared with the adjacent fat tissue was measured in B-mode ultrasound (Fig. 1). After few seconds of immobilization to allow the SWE image to stabilize, mean stiffness was measured by placing a fixed 2-mm circular region-of-interest (ROI) box (Q-BOX^TM; SuperSonic Imagine) on the stiffest region of the excised LNs. The elasticity ratio was calculated automatically when the first ROI was placed on the stiffest region of excised LNs, and the other ROI was immediately placed on the adjacent fat tissue (Fig. 1). Four radiologists with a clinical experience of more than 2 years performed B-mode ultrasound with SWE, and one radiologist with more than 10 years of clinical experience reviewed the results, i.e., nodal size and elasticity values, such as mean stiffness and elasticity ratio.

After the radiologic examination was finished, the pathologic evaluation of sentinel LNs was performed. The excised sentinel LNs were cut into 2–3-mm sizes, and all slices were assessed. The size of the metastatic lesions within the excised nodes was reported in accordance with the staging system of the American Joint Committee on Cancer as follows: > 2 mm, macro-metastasis; 0.2–2.0 mm, micro-metastasis; and < 0.2 mm, isolated tumor cells [23]. In this study, sentinel LN metastasis was defined as macro-metastasis.

A nomogram was established on the basis of the results of the multivariate analysis in the development cohort. Thereafter, 200 bootstrap samples were used for internal validation of the development cohort, and external validation was performed with an independent cohort. The performance of the constructed nomograms was quantified with respect to discrimination and calibration [24]. The discriminative power of whether the constructed nomogram can correctly predict the probability of LN metastasis was quantified using the area under the receiver operating characteristic curve. Calibration was performed to identify the agreement between the observed outcomes and predicted probabilities of metastasis among the excised LNs. To verify the suitability of the constructed nomograms, the Hosmer-Lemeshow goodness-of-fit test was performed [25, 26]. The software used to perform these analyses was the SAS program (ver 9.2, SAS Institute Inc.) or the R Statistical Package (ver 3.3.3, Institute for Statistics and Mathematics).

The characteristics of the patients were analyzed using the chi-square test and independent two-sample t test. The mean elasticity values were compared between the non-metastatic and metastatic sentinel LNs using the independent two-sample t test. Analysis was performed using the SPSS software ver 23 (SPSS).

We previously reported the characteristics and elasticity values of the harvested axillary LNs measured using B-mode ultrasound and intraoperative ex vivo SWE in the development cohort [20]. Briefly, the nodal size and elasticity values, such as mean stiffness and elasticity ratio, were associated with axillary LN metastasis in the multivariable analysis. The nomogram was constructed on the basis of significant factors from the multivariable analysis in the development cohort to predict the probability of axillary LN metastasis (Fig. 2a). From binary logistic regression model for multivariable analysis, we obtained intercept and regression coefficient of each significant factor. The probability of axillary LN metastasis was calculated as follows: 1/(1 + exp(−A)), where A =  − 4.9898 + 0.1498 × (nodal size) + 0.0345 × (mean stiffness) + 0.5321 × (ratio)

In the nomogram, the estimated probability of LN metastasis could be obtained by summing the scores of each variable and locating such on the total score scale. For instance, a patient with a 16-mm nodal size (11 points), 70-kPa mean stiffness (13 points), and 5 ratio (14 points) would score 38 points, which can be converted into 92.29% probability of LN metastasis (Fig. 2b).

The area under the curve (AUC) was 0.856 (95% confidence interval (CI), 0.783–0.929) in the development cohort (Fig. 3a). The p value obtained using the Hosmer-Lemeshow goodness-of-fit test was 0.354, indicating a good fit of the model. When internal validation was performed using the 200 bootstrap samples, the mean absolute error was as low as 0.029. Moreover, the calibration plots of the nomogram showed good agreement between the observed and predicted outcomes (Fig. 3b).

In the validation cohort, 10 of the 80 patients had sentinel LN metastases, all of which were in the N1 stage. Tumor size of the patients with sentinel LN metastasis was larger than that of the patients without sentinel LN metastasis (2.35 cm vs 1.61 cm, p = 0.045). The other clinicopathologic factors were no significantly different between the patients with and without sentinel LN metastasis (Table 1). The median age of all patients was 55 (range, 22–76) years. The findings for estrogen receptor (ER) and human epidermal growth factor receptor 2 (HER2) were positive in 67 (83.8%) and 15 (18.8%) patients, respectively. T1 stage and grade 2 were the most common classifications in 52 (65.0%) and 43 (53.8%) patients, respectively. The number of elderly patients and patients with HER2-negative findings and low N stage was higher in the validation cohort than in the development cohort (Table 2).

Of the 217 sentinel LNs, 15 (6.9%) were metastatic, and 202 (93.1%) were non-metastatic, including seven (3.2%) micro-metastatic LNs. The metastatic LNs tended to be larger than non-metastatic LNs, but it was not significant (8.70 mm vs 7.20 mm, p = 0.123) (Table 3). The mean stiffness of the metastatic sentinel LNs was significantly greater than that of the non-metastatic sentinel LNs (23.54 vs 10.41 kPa, p = 0.005) (Fig. 4). Furthermore, the elasticity ratio of the metastatic sentinel LNs was significantly higher than that of the non-metastatic sentinel LNs (3.24 vs 1.49, p = 0.028). Meanwhile, there was no difference in nodal size and elasticity values according to histological types of breast cancer (Table 4).

The nomogram was validated with the independent set including only 217 sentinel LNs. The discriminative power was good, with an AUC of 0.791 (95% CI, 0.668–0.915) (Fig. 3c). The calibration plots of the nomogram showed good agreement between the observed and predicted outcomes (Fig. 3d).