Academic Editor: Michael H. Dahan
Background: To design a software-applied predictive model relating
patients clinical and pathological traits associated with sentinel lymph-node
total tumor load to individually establish the need to perform an axillary
lymph-node dissection. Methods: Retrospective observational study
including 127 patients with breast cancer in which a sentinel lymph-node biopsy
was performed with the one step nucleic acid amplification method and a
subsequent axillary lymph-node dissection. We created various binary multivariate
logistic regression models using non-automated methods to predict the presence of
metastasis in non-sentinel lymph-nodes, including Log total tumor load,
immunohistochemistry, multicentricity and progesterone receptors. These
parameters were progressively added according to the simplicity of their
evaluation and their predictive value to detect metastasis in non-sentinel
lymph-nodes. Results: The final model was selected for having maximum
discriminatory capability, good calibration, along with parsimony and
interpretability. The binary logistic regression model chosen was the one which
identified the variables Log total tumor load, immunohistochemistry,
multicentricity and progesterone receptors as predictors of metastasis in
non-sentinel lymph-nodes. Harrell’s C-index obtained from the area under the
curve of the predicted probabilities by Model 4 was 0.77 (95% CI, 0.689–0.85;
p
Breast cancer is a frequent entity whose management has evolved in recent years. Nowadays there is a tendency towards more conservative techniques. In the last decade of the 20th century, sentinel lymph node (SLN) biopsy replaced systematic axillary lymph node dissection (ALND), becoming the standard procedure for staging the axilla in breast cancer patients with clinically node-negative axilla [1, 2, 3, 4]. This caused a decrease of ALND rates [5] along with its associated comorbidity [6]. In aims to assess semiquantitatively the state of the SLN, the “One Step Nucleic Acid Amplification” (OSNA) method was proposed [7, 8, 9, 10, 11]. It is based on the quantification of Cytokeratin 19 (CK19) mRNA, which is expressed in more than 95% of breast cancer cases [12], and it is associated with total tumor load (TTL). The OSNA method shows a rentability comparable to conventional histologic techniques, and greatly benefit patients with clinically node-negative axilla [10, 11].
Determination of the different values of SLN TTL allows a different surgical approach to the axilla. In cases with an undetectable or low TTL, the ALND may be safely avoided [13, 14, 15]. Patients with detected micrometastasis in SLN have a disease-free and overall survival comparable to those who received an ALND [16]. Moreover, patients with a T1-2 tumor who had macrometastasis in two or less SLN and received conservative surgery, radiotherapy and adjuvant systemic therapy, showed similar results in terms of survival [13].
On the other hand, TTL of CK19 mARN correlates with the presence of metastasis in non-sentinel lymph nodes (NSLN), thus it is considered the most important predictive factor for the presence of metastasis in NSLN [17]. This is the reason why several cut-off points have been published for SLN TTL to determine when to perform an ALND [18, 19, 20]. Said cut-off points vary between 2150 copies, established by Terretano et al. [20], and 15,000 copies of CK19 mARN, defined by Peg et al. [19]. However, SNL TTL is not the only predictive factor for metastasis in NSLN. Previous studies tried to identify predictive factors for metastasis in NSLN in aims avoid ALND [21, 22, 23, 24]. Clinical and pathological factors have been described in aims to improve the predictive capability of SNL TTL. In this regard, we consider that a unique cut-off point of CK19 mARN copies would not be enough to predict the probabilities of metastasis in NSLN [25], given that there are other factors to be considered. Thus, our objective is to design a software-applied predictive model relating patients clinical and pathological traits associated with SNL TTL to individually establish the need to perform an ALND.
An observational retrospective study was carried out, including 127 patients with breast cancer in which a SLN biopsy was performed with the OSNA method and a subsequent ALND. Patients were consecutively recruited between October 1st 2010 and April 31st 2018.
Inclusion criteria were as follow: patients who had surgery for an invasive T1-3 breast carcinoma, which expressed CK19, with clinically node-negative axilla, and a normal preoperative axillar ultrasound or a lymph node biopsy with no evidence of metastasis. Patients with neoadjuvant chemotherapy treatment, previous ipsilateral axillary surgery, cancer recurrence or extensive in situ ductal carcinoma, were excluded from the study.
Variables studied were: age, menopausal status, menopause age, parity, number of births, ALND, tumor size, tumor histology type (ductal, lobular or others), multicentricity (presence of 2 or more tumor foci in different quadrants of the same breast or foci more than 5 centimeters from the primary focus), multifocality, lymphovascular invasion, tumor histological grade according to Modified Bloom-Richardson (tubules, nuclei and mitosis), estrogen receptors (ER), progesterone receptors (PR), y cHer-2 protein (HER2), Ki-67, SLN (macro or micrometastasis), NSLN (presence or absence of metastasis) and SLN TTL. Immunohistochemistry (IHC) classification was based in previously stablished criteria [26]: “Luminal A-like” (all of: ER and PR positive, HER2 negative and Ki-67 low); “Luminal B-like HER2 negative” (ER positive, HER2 negative and at least one of: Ki-67 high, PR negative or low); “Luminal B-like HER2 positive” (ER positive, HER2 over- expressed or amplified, any Ki-67, any PR); “HER2 positive non-luminal” (HER2 over- expressed or amplified, ER and PR absent); “Triple negative ductal” (ER and PR absent, HER2 negative).
Prior to surgery, and after the breast cancer diagnosis, all patients were assessed with an axillar ultrasound. Axillary lymph nodes suspicious for metastasis were those who shown cortical thickening of 2–3 mm, focal bulge, round shape, partial or complete absence of the fatty hilium, non-hiliar cortical blood flow, or a complete or partial replacement by tumoral tissue. A core needle biopsy was performed in patients with suspicious lymph-nodes. If the presence of metastasis was confirmed patients were excluded from the study.
During surgery, SLN biopsy was performed according to the established protocol
in our unit, marking the node with a radiopharmaceutical and blue dye. The
radiopharmaceutical used was 99mTc albumin nanocolloid, which was injected in the
intradermic periareolar area a day prior to surgery. The dye used was methylene
blue. Once the patient was under anesthesia, 2 mL of methylene blue was injected
in the four quadrants of the periareolar area (0.5 mL per quadrant). Once the SLN
was located and extirped, it was sent to the Anatomy Pathology unit for
application of the OSNA method according to the existing literature [9]. The
amplification rate was assessed by specthophotometry, and the number of CK19 mARN
copies was calculated in relation to a standard curve. Macrometastasis were
defined as the existence of more than 5000 copies/
Afterwards a Level II ALND was performed and NSLN were histologically assessed after being processed with a hematoxylin and eosin stain. Tissue blocks of NSLN were selected with a width of 3 microns, in a 200 microns interval, to determine the presence or absence of metastasis.
Statistical analysis was performed with the statistics software IBM SPSS version 22 (IBM, Armonk, NY, USA). Mean and standard deviations were determined for numeric variables, while percentages were used for qualitative variables. Comparisons of numeric variables was evaluated using Student’s t-test, while the Chi-square test was used for comparisons of qualitative variables. Individual predictive values were evaluated using a receiver operating characteristic curve and the area under the curve (AUC). All statistic comparisons were performed with a two-tailed test, and statistical significance was set at 0.05.
We created various binary multivariate logistic regression models using non-automated methods to predict the presence of metastasis in NSLD, including Log TTL, immunohistochemical (IHC), multicentricity and progesterone receptors (PR). These parameters were progressively added according to the simplicity of their evaluation and their predictive value to detect metastasis in NSLN.
We implemented and compared four binary logistic regression models (Table 1). A goodness-of-fit test was applied (logarithmic probability of –2) as well as the Hosmer and Lemeshow test for each model. Then we determined the Harrell’s C-index (a statistics index to measure the goodness of fit for regression models, which analyzes its capability to discriminate the presence or absence or the event) for those models with an adequate fit to evaluate their discriminatory capability (obtained as the AUC of the predicted probabilities predicted by the model). The slope and calibration graph were also obtained.
Model | Parameters included in the predictive model |
Model 1 | Log TTL |
Model 2 | Log TTL and IHC |
Model 3 | Log TTL, IHC and multicentricity |
Model 4 | Log TTL, IHQ, multicentricity and PR |
TTL, total tumor load; IHC, immunohistochemical; PR, progesterone receptors. |
The final model was selected for its maximum discriminative capability and calibration graph, according to the principles of parsimony and interpretability. Models were calibrated by the slope and calibration graph. Once the definitive binary multivariate regression model was identified, we developed a software to predict the presence of metastasis in NSLN with the objective to make this model applicable for clinical practice.
A total of 127 patients who required an ALND were recruited. Of the population
studied, 37% (47/127) had metastasis in NSLN, while 63% (80/127) did not.
Characteristics of both groups are shown in Tables 2,3,4. Multicentricity is more
frequent in ALND with metastatic NSLN (23.4% vs 10.0%; p: 0.069). IHC
classification showed that Luminal A-like tumors were more frequent in patients
with no metastatic NSLN (57.4% vs 48.1%; p: 0.359). When comparing
axillary surgery characteristics between both groups, we observed that patients
with metastatic NSLN has higher rates of macrometastasis (93.6% vs 63.7%;
p: 0.005) and TTL (917772.13
Mean |
|||||
ALND with metastatic NSLN (n: 47) | ALND without metastatic NSLN (n: 80) | p | OR (95% CI) | p | |
Age | 54.21 |
57.18 |
0.325 | 0.98 (0.95; 1.00) | 0.208 |
Menopausal status | 27/47 (57.4%) | 43/80 (53.8%) | 0.715 | 1.16 (0.56; 2.40) | 0.686 |
Age of menopause | 49.33 |
47.84 |
0.158 | 1.07 (0.96; 1.18) | 0.217 |
Parity | 43/47 (91.5%) | 71/80 (88.8%) | 0.766 | 1.36 (0.39; 4.70) | 0.624 |
Number of births | 2.56 |
2.45 |
0.381 | 1.06 (0.80; 1.41) | 0.673 |
ALND, axillary lymph node dissection; NSLN, nonsentinel lymph nodes. |
Mean |
||||||
ALND with metastatic NSLN (n: 47) | ALND without metastatic NSLN (n: 80) | p | OR (95% CI) | p | ||
Tumor size | 20.29 |
20.67 |
0.641 | 0.99 (0.96; 1.03) | 0.838 | |
Histological type | ||||||
Ductal | 32/47 (68.1%) | 64/80 (80.0%) | 0.074 | 2.50 (0.28; 22.31) | 0.412 | |
Lobular | 14/47 (29.8%) | 11/80 (13.8%) | 6.36 (0.65; 62.69) | 0.113 | ||
Others | 1/47 (2.1%) | 5/80 (6.2%) | ||||
Multicentricity | 11/47 (23.4%) | 8/80 (10.0%) | 0.069 | 2.75 (1.02; 7.44) | 0.046 | |
Multifocality | 5/47 (10.6%) | 8/80 (10.0%) | 1 | 1.07 (0.33; 3.49) | 0.909 | |
Lymphovascular invasion | 15/47 (31.9%) | 27/80 (33.8%) | 1 | 0.92 (0.43; 1.98) | 0.832 | |
Tumor histological grade | ||||||
1 | 8/47 (17.0%) | 9/80 (11.4%) | 0.664 | |||
2 | 24/47 (51.1%) | 44/80 (55.7%) | ||||
3 | 15/47 (31.9%) | 26/80 (32.9%) | ||||
Tubules | ||||||
1 | 3/47 (6.4%) | 2/80 (2.5%) | 0.279 | |||
2 | 10/47 (21.3%) | 26/80 (32.9%) | ||||
3 | 34/47 (72.3%) | 51/80 (64.6%) | ||||
Nuclei | ||||||
1 | 2/47 (4.3%) | 1/80 (1.3%) | 0.419 | |||
2 | 21/47 (44.7%) | 30/80 (38.0%) | ||||
3 | 24/47 (51.1%) | 48/80 (60.7%) | ||||
Mitosis | ||||||
1 | 27/47 (57.4%) | 40/80 (50.6%) | 0.383 | |||
2 | 14/47 (29.8%) | 21/80 (26.6%) | ||||
3 | 6/47 (12.8%) | 18/80 (22.8%) | ||||
ALND, axillary lymph node dissection; NSLN, nonsentinel lymph nodes. |
Mean |
||||||
ALND with metastatic NSLN (n: 47) | ALND without metastatic NSLN (n: 47) | p | OR (95% CI) | p | ||
IHC | ||||||
Luminal A-like | 27/47 (57.4%) | 38/80 (48.1%) | 0.811 | |||
Luminal B-like HER2 negative | 17/47 (36.2%) | 31/80 (39.2%) | ||||
Luminal B-like HER2 positive | 2/47 (4.3%) | 6/80 (7.6%) | ||||
HER2 positive nonluminal | 0/47 (0%) | 1/80 (1.3%) | ||||
Triple-negative | 1/47 (2.1%) | 3/80 (3.8%) | ||||
IHC (Grouped) | ||||||
Luminal A-like | 27/47 (57.4%) | 38/80 (48.1%) | 0.359 | 0.69 (0.33; 1.42) | 0.311 | |
No Luminal A-like | 20/47 (42.6%) | 41/80 (51.9%) | ||||
ER | 46/47 (97.9%) | 76/80 (95.0%) | 0.651 | 2.42 (0.26; 22.33) | 0.435 | |
PR | 37/47 (78.7%) | 69/80 (86.3%) | 0.325 | 1.69 (0.66; 4.36) | 0.274 | |
HER2 positive | 2/47 (4.3%) | 7/80 (8.9%) | 0.482 | 0.46 (0.10; 2.30) | 0.342 | |
Ki 67 (%) (Grouped) | ||||||
33/47 (70.2%) | 47/80 (59.5%) | 0.255 | 0.62 (0.29; 1.34) | 0.228 | ||
14/47 (29.8 %) | 32/80 (40.5%) | |||||
SLN | ||||||
Micrometastasis | 3/47 (6.4%) | 29/80 (63.7%) | 8.34 (2.38; 29.26) | 0.001 | ||
Macrometastasis | 44/47 (93.6%) | 51/80 (63.7%) | ||||
TTL | 917772.13 |
335574.25 |
0.005 | 1 (1.00; 1.00) | 0.120 | |
Log TTL | 5.11 |
4.28 |
0.005 | 1.31 (1.31; 2.61) | ||
ALND, axillary lymph node dissection; NSLN, nonsentinel lymph nodes; SLN, sentinel lymph node; IHC, immunohistochemical; ER, estrogen receptors; PR, progesterone receptors; TTL, total tumor load. |
We used several binary logistic regression models to predict the presence of metastasis in NSLN. Harrell’s C-index values of models oscillated between 0.68 and 0.77, determined as the AUC of the probability predicted (Table 5). Bivariate logistic regression models were performed linking metastasis in NSLN (positive/negative) and every single one of the identified variables as prognostics of positivity or negativity of metastasis in NSLN. These models led us to selecting the variable Log TTL in Model 1. The addition of variables in the subsequent models was made attending to the increase of the predictive capability of the models, their calibration (Hosmer-Lemeshow) and their discrimination capability (Harrell’s C-index). The final model was selected for having maximum discriminatory capability, good calibration, along with parsimony and interpretability. The binary logistic regression model chosen was the one which identified the variables Log TTL, IHC, multicentricity and PR as predictors of metastasis in NSLN. Thus, these were the variables included in the final multivariate analysis, as can be seen in Table 5.
Models | Variables | OR 95% CI | Calibration (Homer-Lemeshow) p | Discrimination (Harrel’s C-index 95% CI) |
1 | Log TTL | 1.85 (1.31; 2.61) | 0.555 | 0.68 (0.59; 0.78) |
2 | Log TTL | 1.89 (1.34; 2.68) | 0.708 | 0.70 (0.61; 0.79) |
IHC | 0.56 (0.26; 1.23) | |||
3 | Log TTL | 1.87 (1.31; 2.66) | 0.347 | 0.73 (0.64; 0.82) |
IHC | 0.58 (0.26; 1.28) | |||
MC | 0.43 (0.15; 1.23) | |||
4 | Log TTL | 2.14 (1.45; 3.17) | 0.151 | 0.77 (0.69; 0.85) |
IHC | 0.30 (0.11; 0.78) | |||
MC | 0.34 (0.11; 1.07) | |||
PR | 6.68 (1.79; 24.87) |
Harrell’s C-index obtained from the AUC of the predicted probabilities by Model
4 was 0.77 (95% CI, 0.689–0.85; p
ROC curve for logistic regression model obtained from the
association between Log TTL, IHQ, multicentricity and PR. Area under ROC
curve 0.770 (95% CI, 0.688–0.852; p
Calibration graph of original logistic regression model obtained from the association between Log TTL, IHQ, multicentricity and PR.
The main finding of our study is that a model based in the TTL, IHC, multicentricity and PR can predict 77% of patients with metastasis in NSLN. Given the simplicity of this model, which includes only 4 parameters (TTL, IHC, multicentricity an PR), it is easy to apply intraoperatively in any breast cancer unit. When applying this proposed predictive model, any breast surgeon can easily predict the probability of metastasis in NSLN and then decide whether to perform an ALND during the surgery act (Fig. 3).
Example of using the binary model based on total tumor load (TTL), IHC, MC and PR as a predictor for metastasis in non-sentinel lymph-nodes. PR (progesterone receptors): 0 = Presence of progesterone receptors; 1 = No presence of progesterone receptors. IHC (Immunohistochemistry): 0 = Luminal A-like; 1 = No Luminal A-like. MC (multicentricity): 0 = multicentric tumor; 1 = Non-multicentric tumor.
Up until recently, nomograms have been based in SLN TTL, establishing different
AUC ranging between 0.66 and 0.86 [20, 27, 28, 29, 30, 31, 32, 33]. In our work, we described an AUC
of 0.77 (95% CI, 0.688–0.852; p
There have been publications of several cut-offs points for the copies of CK19
mARN in SLN TTL to decide on the ALND. Heilmann et al. [34] defined 7900
copies/
The association between clinical and pathological factors and the risk of metastasis in NSLN have been established in several studies, like the HER-2 status or the histological and nuclear grade described by Meretoja et al. [35]. Tumor size is also considered an important factor for metastasis in NSLN [36, 37, 38]. Furthermore, some factors have even been included in nomogram like the association of the TTL with tumor size, lymphovascular invasion, HER-2 status and number of metastatic SLN [28].
In contrast, Shimazu et al. [29] stated that nomograms using several pathological parameters are not practical for intraoperative decision-making. However, we consider that a simplify model like the one presented may be quite useful given the feasibility of use during surgery with a high predictive capacity.
We consider our study to have many strengths, one of them being the considerable sample size, with the inclusion of a representative sample of breast cancer patients undergoing ALND. Another notorious strength of our study is the fact that this proposed model allows for individual assessment of the need to perform ALND in the operating room, that can be made in a quick, effective and simple way without extending surgery time. Nonetheless, our study also has some limitations. We consider that a prospective design, along with the need for external validation might be considered for future studies. In addition, it would require an external validation to be able to apply it in routine practice.
In conclusion, the combination of TTL, IHC, multicentricity and PR can predict 77% of patients with metastasis in NSLN and said prediction may be made intraoperatively in a feasible manner.
SLN, Sentinel lymph node; ALND, Axillary lymph node dissection; OSNA, One step nucleic acid amplification; CK19, Cytokeratin 19; TTL, Total tumor load; NSLN, Non-sentinel lymph nodes; IHC, Immunohistochemical; ER, Estrogen receptors; PR, Progesterone receptors; MC, Multicentricity.
JAGM and JAS designed the research study. RGJ performed the research. AFP analyzed the data. JAGM, MSS, RGJ, and JAS wrote the manuscript. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript.
All subjects gave their informed consent for inclusion before they participated in the study. The study was conducted in accordance with the Declaration of Helsinki , and the protocol was approved (on January 29th 2019) by Andalucia’s Board of Biomedicine Ethics Committee (registration number: 1004-N-18).
Thanks to all the peer reviewers for their opinions and suggestions.
This research received no external funding.
The authors declare no conflict of interest.