Predicting Nonsentinel Lymph Node Metastasis Using Lymphoscintigraphy in Patients with Breast Cancer

DISCUSSION

The most unique feature of the NCC model is its use of SLS imaging findings to predict NSLN metastasis. To the best of our knowledge, no model for predicting NSLN metastasis has included a covariate other than histopathologic parameters. We found that the ^99mTc-HSA SLS findings can be graded according to the flow and retention pattern and that this grading system has implications for NSLN status.

Because SLS is a procedure for identifying SLNs rather than for detecting SLN metastasis, there have been a limited number of studies of SLS as a tool for assessing axillary LN status. However, indirect effects of tumor infiltration can manifest as SLS findings. A few previous studies reported a correlation between decreased SLN uptake and increased risk of axillary involvement (13–15). A reasonable explanation for this phenomenon would be that the lymphatics become progressively infiltrated with tumor cells, blocking the passage of radiotracers (13). Despite this finding, SLS generally possesses limited utility for assessing axillary involvement because small metastases would scarcely affect lymphatic flow and SLN uptake. In this study, we demonstrated that SLS findings help discriminate between NSLN-positive patients and NSLN-negative patients rather than between axillary–LN-positive patients and axillary–LN-negative patients. SLS findings in SLN-positive–NSLN-negative patients were more similar to those in SLN-negative patients than to those in SLN-positive–NSLN-positive patients (Table 4).

This phenomenon could be explained in several ways. One explanation would be that the upstream lymphatic retention pattern arises as the axillary tumor burden increases. The grading system (Table 2) reflects the degree of impairment of lymphatic transport. As the grade increases, the radioactivity of the upstream lymphatic vessels increases while that of the SLNs decreases. This might be because the lymphatic flow becomes more sluggish as the tumor obstructs the lymphatic channel, increasing mean transit time through upstream lymphatic vessels and decreasing tracer delivery to SLNs. A grade 1 (focal) SLS pattern indicates fast lymphatic transport and good SLN uptake, suggesting minimal hindrance of lymphatic flow by infiltrating tumor cells in the SLNs (unless normal lymphatic drainage is disturbed by inadequate injection or previous excisional biopsy (16)). On the other hand, a grade 3 (serpentine) pattern suggests that lymphatic flow is significantly hindered by LN metastasis. As the disease progresses, the grade 3 pattern changes to a grade 4 (no uptake), indicating complete blockage of the lymphatic channels by tumor cells. A higher-grade pattern is thus related to a greater lymphatic tumor burden, and this may be related to an increased likelihood of NSLN metastasis.

Another explanation would be related to the fact that SLNs are defined by radiotracers. The theoretic definition of an SLN is the axillary LN or LNs to which breast cancer would first metastasize. However, this definition is not practical. In clinical practice, the SLN is operationally defined by radiocolloids or blue dyes. The operational definition of an SLN is the LN or LNs that accumulate radiotracers (or blue dyes) injected into the tumor basin. In the early phase of LN metastasis in which the metastatic tumor does not distort the lymphatic flow, these 2 definitions would generate the same result. However, several studies have suggested that cancerous involvement of the lymphatic system may influence the drainage pattern (13,17,18), raising the possibility that the operationally defined SLNs might change over time. An axillary LN that harbors tumor emboli in the early phase of metastasis would be an SLN both theoretically and operationally, but as the tumor grows and impedes afferent lymphatic flow, the radiotracer delivery to this LN decreases, and the LN might show no tracer uptake in the later stages of the disease. If that were the case, the LN would be operationally defined as an NSLN, despite the fact that it was initially an SLN. The SLS pattern of upstream retention and decreased LN uptake would imply some hindrance of lymphatic drainage. This hindrance could be associated with the presence of a metastasis-positive NSLN that was initially an SLN.

When survival analysis was performed according to the SLS findings, patients with focal grade showed a slight trend toward better survival than did those with nonfocal grades, although statistically not significant. Because of the excellent prognosis of the patients included in the study (only 12 recurrences of 295 cases), more patients or a longer follow-up period may be required to derive statistically valid conclusions.

A noteworthy point of the SLS procedure is that we used ^99mTc-HSA, the particle size of which (2–3 nm) is among the smallest of those of the radiocolloids available for clinical practice. The biokinetics of radiocolloids depend strongly on their particle sizes (19). Radiocolloids with large particle sizes, such as ^99mTc-tin colloid (50–600 nm) and ^99mTc-sulfur colloid (100–400 nm), migrate through the lymphatic channels much more slowly than does ^99mTc-HSA (20). Small or large particle size per se is neither an advantage nor a disadvantage, because small particle size may cause mapping beyond true SLNs while large particle size may cause nonvisualization of SLNs if the wait time is insufficient (16). However, the grading system developed in this study is applicable primarily to SLS using ^99mTc-HSA because, in our experience, SLS using radiocolloids with large particle sizes invariably shows a focal uptake pattern on delayed images, invalidating the grading system. The median transit time of ^99mTc-HSA to SLN is 5 min. Therefore, in our hospital, the SLS imaging protocol for ^99mTc-HSA is the acquisition of anterior images for 5 min immediately after injection followed by the acquisition of lateral images for another 5 min. Because of the rapid transport of the radiopharmaceutical, the flow through the lymphatic vessels was visualized by SLS and could be used for grading.

Using this visual grading of the SLS pattern as a covariate, we developed a model for predicting NSLN metastasis and compared its performance with that of the MSKCC model. Because this was a retrospective study, we could not prepare a prospective group as an independent dataset for the validation of the NCC model. Instead, we performed an internal validation procedure using a bootstrap method. The same bootstrap samples were used to evaluate the performance of the MSKCC model for model comparison. The bootstrap resampling method is reported to produce stable and unbiased estimates of true model performance in logistic regression analysis (21). The mean AUC of the NCC model was 0.812, which was significantly larger than the AUC of 0.728 (P < 0.001) of the MSKCC model.

We modified the MSKCC model in 2 ways. As described elsewhere (3,4,6,12), the method of detection variable of the MSKCC model is a shortcoming that limits interhospital applicability. When applying the MSKCC model to our patient dataset, we substituted the LN size category variable for the method of detection. This substitution has been reported to enhance the performance of the MSKCC model (12). The other modification was the use of histologic grade instead of nuclear grade. This substitution has been reported not to affect model performance (11).

Calibration is a measure of how accurately the predicted probabilities reflect the observed frequencies. The Hosmer and Lemeshow test suggested that the NCC model is well calibrated. FNR in the low-predicted-probability subgroup is another important measure (10). The basic intent of the model for predicting NSLN metastasis is to avoid complete ALND in SLN-positive patients with a low risk of additional NSLN metastasis. Therefore, the proportion of patients classified as at low risk for NSLN metastasis and the accuracy of the classification are important measures of the model performance. For the predicted probability of 0.10 or less, 8% of patients were classified as low risk by the NCC model. In contrast, 1% of patients were so classified by the MSKCC model (Table 7). The FNR was 0% for both models. For the predicted probability of 0.15 or less, the patient proportions of the NCC and the MSKCC models were 15.8% and 5.2%, respectively. The FNRs of the NCC and the MSKCC models were both 2.3%. This result is somewhat different from that of a previous study in which the patient proportion in the 0.10 or less probability group predicted by the MSKCC model was up to 35.8% (10). This difference may be because the proportion of patients included in the low-predicted-probability subgroup is influenced by the patient population and by the model characteristics.

There are a few limitations of this study. The size of LN metastasis variable used in the NCC model does not distinguish between SLN metastasis and NSLN metastasis. Considering the purpose of the model (i.e., the prediction of NSLN status), it would be appropriate for the size of LN metastasis variable to mean the size of the SLN metastasis. However, although the pathology reports for most patients did not specify whether the reported size of LN metastasis came from SLN or NSLN, the sizes of the SLNs were reported in most cases. SLNs are the first LNs in which the metastatic tumor starts to grow. Therefore, the largest LN metastasis can be expected to be found in the SLN in most cases. This consideration might mitigate the limitation.

Another limitation is that no independent dataset could be provided for validation of the established model in this retrospective study. The validity of the NCC model should be externally tested on many independent datasets to obtain sound proof of its accuracy.