Predictors of invasion in patients with core-needle biopsy-diagnosed ductal carcinoma in situ and recommendations for a selective approach to sentinel lymph node biopsy in ductal carcinoma in situ

Abstract

BACKGROUND

Among patients with core-needle biopsy (CNB)-diagnosed ductal carcinoma in situ (DCIS), the guidelines for the selection of patients to undergo sentinel lymph node (SLN) biopsy are not well defined, and many patients with no invasion undergo an unnecessary SLN biopsy. The objective of this study was to identify the predictors of invasion in patients with CNB-diagnosed DCIS and, thus, to help determine the most appropriate candidates for SLN biopsy.

METHODS

The authors retrospectively evaluated 200 consecutive patients with CNB-diagnosed DCIS who underwent final excision at their institution between May 1, 2002 and June 30, 2005. Demographic data, the size and type of lesion on imaging studies, histologic features of DCIS on CNB, the number of cores taken, and the number of cores involved by DCIS were correlated with invasion on excision and SLN biopsy outcome.

RESULTS

Forty-one of 200 patients (21%) had invasive carcinoma diagnosed on final excision. Parameters that correlated with invasion were a mass lesion, lesion size >1.5 cm, high nuclear grade, and the presence of lobular cancerization on CNB. A multivariate logistic regression model was developed to predict invasion. At the time of breast surgery, 103 of 200 patients (52%) underwent SLN biopsy, and 34 had invasion diagnosed on final excision. Three patients had 1 positive SLN each.

CONCLUSIONS

A mass lesion, lesion size >1.5 cm, and the presence of lobular cancerization on CNB were significant, independent predictors of invasion in patients with DCIS. The authors recommended the inclusion of these predictors in the guidelines for selecting patients for SLN biopsy to reduce the number of unnecessary procedures. Cancer 2006. © 2006 American Cancer Society.

The diagnosis of noninvasive breast carcinoma has increased in recent years because of the increasing use of mammography. Ductal carcinoma in situ (DCIS) represents >20% of all newly diagnosed breast cancers, with >50,000 women diagnosed in the U.S. annually.1, 2 The cellular features of DCIS are similar to those of invasive ductal carcinoma; however, because the neoplastic cells in DCIS do not invade the basement membrane of the mammary ducts, DCIS does not have the ability to metastasize. The diagnosis of DCIS usually is made preoperatively by percutaneous core-needle biopsy (CNB). Ultrasound guidance can be used for mass lesions, whereas stereotactic automated biopsy or vacuum-assisted stereotactic mammotome devices are used for suspicious calcifications. The mammotome device extracts more contiguous and larger tissue fragments, so that more tissue can be obtained. In addition, the authors of 1 study suggested that this device reduces the rate of under-diagnosis of invasive disease on CNB.3 Reportedly, from 13% to 29% of patients who are diagnosed with DCIS on CNB demonstrate invasive carcinoma at final excision.4–12

Axillary lymph node status is the best prognostic indicator of survival for women with invasive breast cancer, therefore, axillary lymph node staging at the time of initial surgical excision for invasive disease is standard. The use of SLN surgery rapidly is replacing axillary lymph node dissection for axillary staging, because it has been shown that SLN biopsy is an accurate predictor of axillary lymph node status and because SLN surgery has lower associated morbidity than standard axillary surgery. Some investigators have strongly advocated the use of SLN surgery for patients with CNB-diagnosed DCIS. Advocates note that SLN surgery, at least for a selected group of patients with “high-risk” DCIS, avoids the need for a second operation for patients who prove to have invasive disease on final excision.13 However, SLN biopsy in patients with high-risk DCIS, the majority of whom do not actually have invasive disease, would mean subjecting many patients to unnecessary treatment. Therefore, axillary surgery in patients who have a diagnosis of pure DCIS remains controversial.

Currently, no definitive selection criteria exist for the use of SLN biopsy in patients with pure DCIS diagnosed on CNB. SLN biopsy typically is offered on a case-by-case basis. The surgeon has to estimate the probability of invasive carcinoma preoperatively according to the clinical, radiologic, and pathologic findings and must make a decision regarding axillary surgery based on these estimates. If axillary staging is not performed at the time of initial excision and invasive carcinoma is identified on final pathology, then the patient may have to undergo a second surgical procedure. Patients who undergo segmental mastectomy still are candidates for a SLN biopsy, but patients who undergo a mastectomy need to have a complete axillary lymph node dissection if axillary staging is required. Conversely, the routine use of SLN biopsy in all patients with DCIS is not justified, because the reported rate of positive SLNs (1–13%) and the extent of lymph node metastases are low even when invasive carcinoma is identified on excision.6, 14–16

In an effort to identify clinical, radiologic, and histologic predictors of invasion in patients with DCIS diagnosed on CNB and, thus, to help determine which patients with CNB-diagnosed DCIS are the most appropriate candidates for SLN biopsy, we retrospectively reviewed the records of a large series of patients with CNB-diagnosed DCIS who were treated recently at our institution. We evaluated patients with DCIS diagnosed by CNB to determine the incidence of invasive carcinoma on excision and the frequency and extent of SLN metastases to identify radiologic and histologic predictors of invasion. The results may help improve the selection process of patients for SLN surgery and, thus, reduce the number of patients who undergo unnecessary SLN procedures.

MATERIALS AND METHODS

Patients

After approval was granted by the Institutional Review Board of The University of Texas M. D. Anderson Cancer Center, we used a computer data base from the Department of Pathology to identify 209 consecutive patients who were diagnosed with pure DCIS on CNB from a total of 2720 patients who underwent breast CNB from May 1, 2002 to June 30, 2005. All women who had a diagnosis of DCIS with a suspicion of microinvasion and/or concurrent invasive carcinoma on CNB were excluded from the study. The 200 women who had slides of CNB specimens available for review constituted the population for this study. In 161 patients, the target lesion was suspicious calcifications, and the biopsy was performed with stereotactic guidance using a directional, vacuum-assisted mammotome device using a 9-gauge to 11-gauge needle. In the other 39 patients, the target lesion was a mass, and the biopsy was performed with ultrasound guidance using a 14-gauge to 18-gauge needle. The demographics, lesion size, and number of cores obtained at the time of the biopsy were recorded for each patient.

Histologic Evaluation

The histologic features of DCIS were retrospectively rereviewed by 2 pathologists (L.H. and E.R.). Two hematoxylin and eosin-stained levels of each core section were evaluated. The number of cores obtained, the number of cores involved by DCIS, and the extent of ductal involvement by DCIS (focal, ≤3 ducts; extensive, >3 ducts) were recorded. Histologic variables, including nuclear grade, architectural pattern of DCIS, and the presence or absence of lobular cancerization, necrosis, periductal fibrosis, and periductal lymphocytic infiltrate also were documented. For each of these variables, each tumor was assigned to 1 of 2 subgroups, which were defined as follows: nuclear grade, high (Grade 2 or 3) versus low (Grade 1); architectural pattern, solid and cribriform versus others, including micropapillary and papillary; necrosis, comedo versus noncomedo or no necrosis; lobular cancerization, present versus absent; periductal fibrosis, high (severe or moderate) versus low (mild or none); and lymphocytic infiltrate, high (severe or moderate) versus low (mild or none). From the pathology report of the final surgical excision, the type of surgical excision (mastectomy vs. segmental mastectomy), final diagnosis (presence or absence of invasive disease), and use of axillary surgery also were documented.

Statistical Analysis

Statistical analysis was performed by using SAS software (version 9.1; SAS Institute Inc., Cary, NC). P values <.05 were considered statistically significant. All continuous variables (age, lesion size, number of cores obtained, number of DCIS-involved cores, and extent of ductal involvement) were categorized into 2 groups, and the median observed value between the groups was considered the cut-off point. The chi-square test or the Fisher exact test was used to assess the association between the presence of invasive disease on final pathology review or SLN surgery and all analyzed variables. Logistic regression models were fitted to find the relation between the final diagnosis and the variables in univariate and multivariate analyses. For the multivariate logistic regression model, we used the stepwise procedure to perform the backward selection to obtain the most significant variables in the model. Specificity and sensitivity analyses were performed according to the multivariate logistic model using the variables that were associated with diagnosis as predictors.17

RESULTS

Factors Associated with Invasion on Excision in Patients with CNB-Diagnosed DCIS

Of 200 patients who had CNB-diagnosed DCIS, 41 patients (21%) had invasive carcinoma on final surgical excision, with the size of the invasive tumor ranging from 0.05 cm to 1.60 cm in greatest dimension. The factors we identified that were associated with invasive disease at final surgical excision were the presence of any mass lesion, lesion size >1.5 cm, high nuclear grade of DCIS, and the presence of lobular cancerization on CNB (Table 1). Univariate analysis revealed that invasion was almost 4 times as likely to occur in patients who had a lesion size >1.5 cm (odds ratio [OR], 3.56; 95% confidence interval [95% CI], 1.67–7.60) and almost 3 times as likely to occur in patients who had mass lesions compared with patients who had calcifications (OR, 2.78; 95% CI, 1.28–6.03) and in patients who had lobular cancerization compared with patients who were without lobular cancerization (OR, 2.70; 95% CI, 1.27–5.74) on CNB. Although we observed that high-nuclear-grade DCIS was almost 7 times more likely to be associated with invasion than low-nuclear-grade DCIS (OR, 6.76; 95% CI, 0.89–51.64), this finding did not reach statistical significance according to our univariate analysis (P < .0653) (Table 2, univariate model). The multivariate logistic regression model using a backward selection revealed that significant predictive factors of invasion on CNB were a lesion size >1.5 cm, a mass lesion rather than calcifications, and the presence of lobular cancerization (Table 2, multivariate model). None of the other recorded histologic variables, including architectural pattern and the presence or absence of necrosis, periductal fibrosis, or periductal lymphocytic infiltrate, correlated significantly with invasion. Moreover, the effect of sampling evaluated by recording the number of obtained cores, the number of cores involved by DCIS, and the extent of DCIS involvement in the cores did not correlate significantly with invasion.

Table 1. Association of Clinical, Radiologic, and Histologic Variablesby Chi-Square or Fisher Exact Tests after an Initial Diagnosisof Ductal Carcinoma In Situ on Core-Needle Biopsy

Variable	Diagnosis on excision: No. of patients (%)		P
	Invasive (n = 41)	Noninvasive (n = 159)
DCIS indicates ductal carcinoma in situ; CNB, core-needle biopsy.
Age
>57.5 y	21 (51.2)	79 (49.7)	.8610
≤57.5 y	20 (48.8)	80 (50.3)
Type of lesion on imaging
Calcifications	27 (65.9)	134 (84.3)	.0079
Mass	14 (34.2)	25 (15.7)
Size of lesion
>1.5 cm	30 (73.2)	69 (43.4)	.0007
≤1.5 cm	11 (26.8)	90 (56.6)
No. of cores obtained
>9	19 (46.3)	69 (43.4)	.7348
≤9	22 (53.7)	90 (56.6)
No. of cores involved
>5	18 (43.9)	56 (35.2)	.3046
≤5	23 (56.1)	103 (64.8)
Extent of DCIS on CNB
Extensive	36 (87.8)	123 (77.4)	.1396
Focal	5 (12.2)	36 (22.6)
Nuclear grade
High	40 (97.6)	136 (85.5)	.0325
Low	1 (2.4)	23 (14.5)
Pattern
Solid/cribriform	30 (73.2)	134 (84.3)	.0989
Other	11 (26.8)	25 (15.7)
Necrosis
Comedo	19 (46.3)	72 (45.3)	.9034
None/noncomedo	22 (53.7)	87 (54.7)
Lobular cancerization
Present	15 (36.6)	28 (17.6)	.0084
Absent	26 (63.4)	131 (82.4)
Periductal fibrosis
High	26 (63.4)	94 (59.1)	.6167
Low	15 (36.6)	65 (40.9)
Lymphocytic infiltrate
High	19 (46.3)	83 (52.2)	.5033
Low	22 (53.7)	76 (47.8)

Table 2. Odds Ratios and 95% Confidence Intervals for the Univariate Logistic Regression and Multivariate Logistic Regression Models for Ductal Carcinoma in Situ Invasion on Core-Needle Biopsy

Variable	Univariate model		Multivariate model
	OR (Wald 95% CI)	P	OR (Wald 95% CI)	P
OR indicates odds ratio; 95% CI, 95% confidence interval; DCIS, ductal carcinoma in situ. * Reference: calcifications. † Reference: focal. ‡ Reference: none or noncomedo necrosis.
Age >57.5 y	1.06 (0.54–2.11)	.8610
Mass lesion on imaging*	2.78 (1.28–6.03)	.0096	2.48 (1.1–5.62)	.0294
Lesion >1.5 cm	3.56 (1.67–7.6)	.0010	3.15 (1.44–6.88)	.0041
>9 cores obtained	1.13 (0.57–2.24)	.7349
>5 cores involved	1.44 (0.72–2.89)	.3059
Extensive DCIS†	2.11 (0.77–5.77)	.1466
High nuclear grade	6.76 (0.895–1.64)	.0653
Patterns other than solid	1.97 (0.87–4.43)	.1030
Comedo necrosis‡	1.04 (0.52–2.08)	.9033
Presence of lobular cancerization	2.70 (1.27–5.74)	.0100	2.78 (1.25–6.16)	.0120
Presence of periductal fibrosis	1.20 (0.59–2.44)	.6170
Absence of lymphocytic infiltrate	1.27 (0.64–2.52)	.5039

Factors Associated with Performance of SLN Biopsy

Of 200 patients with CNB-diagnosed DCIS, 103 patients underwent SLN surgery at the time of surgical excision. The variables we identified that were associated significantly with the use of SLN surgery included the presence of a mass lesion, lesion size >1.5 cm, and the use of mastectomy as the definitive surgical procedure (Table 3). The histologic variables reported in the pathology reports, including nuclear grade, the pattern of DCIS, and the presence or absence of necrosis in the CNB specimens, as well as the number of cores obtained were assessed by chi-square or Fisher exact tests to determine whether any of those variables had an association with the SLN procedures. None of those reported histologic variables on CNB were associated significantly with the performance of SLN biopsy.

Table 3. Association of Clinical, Radiologic, and Histologic Variables with Sentinel Lymph Node Biopsy for Ductal Carcinoma in Situ

Variable	SLN sampling performed: No. of patients (%)		P
	No (n = 97)	Yes (n = 103)
SLN indicates sentinel lymph node biopsy.
Age
>57.5 y	50 (51.6)	50 (48.5)	.6712
≤57.5 y	47 (48.5)	53 (51.5)
Procedure
Segmental resection	84 (86.6)	25 (24.3)	<.0001
Mastectomy	13 (13.4)	78 (75.7)
Size of lesion
>1.5 cm	36 (37.1)	63 (61.2)	.0007
≤1.5 cm	61 (62.9)	40 (38.8)
Type of lesion on imaging
Calcifications	86 (88.7)	75 (72.8)	.0047
Mass	11 (11.3)	28 (27.2)
No. of cores obtained
>9	44 (45.4)	44 (42.7)	.7067
≤9	53 (54.6)	59 (57.3)
Nuclear grade
High	84 (86.6)	92 (89.3)	.5538
Low	13 (13.4)	11 (10.7)
Pattern
Solid/cribriform	84 (86.6)	80 (77.7)	.1005
Other	13 (13.4)	23 (22.3)
Necrosis
Comedo	43 (44.3)	48 (46.6)	.7471
None/noncomedo	54 (55.7)	55 (53.4)

Factors Associated with the Prevalence of Metastases in SLNs

Of 103 patients who underwent SLN surgery, 34 patients had invasion on final excision (33%). Metastases in the SLN specimens were identified in 3 patients (1 involved lymph node in each patient) and measured 0.05 cm, 0.10 cm, and 0.60 cm. All 3 patients had invasive carcinoma on final excision with sizes ranging from 0.7 cm to 1.2 cm. All three patients underwent a completion axillary lymph node dissection to yield 14 lymph nodes, 20 lymph nodes, and 24 lymph nodes, respectively, and no additional positive lymph nodes were identified. In 1 patient who had a final diagnosis of DCIS on excision, we identified isolated cytokeratin-positive tumor cells in 1 SLN. No additional axillary surgery was performed for this patient. Because this was a small number of patients, a statistical analysis was not performed.

Three additional patients underwent axillary lymph node dissection with no prior SLN biopsy, because they had radiologic or clinical findings that were suggestive of lymph node metastases. One patient had a final diagnosis of DCIS and 11 negative axillary lymph nodes. Two patients had a final diagnosis of invasive carcinoma: One of those patients had metastatic carcinoma in 2 of 23 axillary lymph nodes, and the other patient had metastatic carcinoma in 25 of 29 lymph nodes.

DISCUSSION

On the basis of our current findings, we identified the presence of a mass lesion, lesion size >1.5 cm, lobular cancerization, and a high nuclear grade of DCIS as variables that were associated significantly with invasion. Furthermore, each of these variables, except nuclear grade, was an independent predictor of invasion according to our univariate and multivariate analyses.

We evaluated several histologic variables on CNB and correlated their association with invasion. High nuclear grade and the presence of comedo-type necrosis in DCIS have been indicated as significant predictors of microinvasion and progression of disease,18–22 and it has been demonstrated that those histologic variables are significant predictive factors of invasion in patients with DCIS in some studies4, 7, 19, 23 but not in others.5, 6, 8 According to our current study, comedo-type necrosis was not a significant factor for predicting invasion. High nuclear grade was associated significantly with invasion in our study, although it did not prove to be a significant independent predictor of invasion in univariate analysis or in the multivariate logistic regression model.

Moreover, other investigators have reported a significant correlation between periductal inflammation and invasion,7, 8, 24, 25 although our findings did not support this observation. We categorized our results as either a low degree or a high degree of periductal fibrosis and periductal inflammation. Although 2 pathologists agreed on the categorization, the separation largely was arbitrary. This lack of standard quantification of periductal fibrosis and inflammation may account in part for the inconsistent results from different studies.

It is noteworthy that, in our study, lobular cancerization by DCIS was a significant independent predictor of invasion in both univariate and multivariate analyses. To our knowledge, lobular cancerization has not been reported previously as an independent predictor of invasion. Renshaw12 reported that patients with DCIS that measured >4 mm in greatest dimension, together with lobular extension on CNB, had an increased risk of invasion on final excision on univariate analysis. In addition, lobular cancerization near the margin of initial excisional biopsy for DCIS reportedly was associated with an increased risk of residual DCIS on reexcision.26 Considering the results from other investigators and our own observations, we believe that routinely reporting the presence of lobular cancerization by DCIS on CNB specimens may prove useful to clinicians.

A summary of the results from recent studies that analyzed the rate of underestimation of invasion and potential predictive factors for invasion in patients with CNB-diagnosed DCIS is provided in Table 4. Underestimation of invasion in DCIS diagnosed on CNB most often is related to inadequate sampling of the lesional area. Although some authors have noted a similar underestimation rate of invasion with different biopsy approaches,6 others have reported significant differences based on the degree of sampling. In the studies by Yen et al.11 and Lee et al.8, excisional biopsy was associated with a significantly lower incidence of invasion than CNB. Although, in their study, Lee et al.8 did not observe a statistically significant difference between 14-gauge and 11-gauge core biopsies, Jackman et al.10 did report a statistically significant difference between 14-gauge large-core biopsies and vacuum-assisted 11-gauge or 14-gauge biopsies, with invasion rates of 20% and 11%, respectively. In addition, Jackman et al.10 observed that, if ≤10 cores per lesion were obtained, then it was more likely that invasion would be identified on excision. We examined the effect of sampling by recording the number of cores obtained on CNB, the number of cores involved by DCIS, and the extent of ductal involvement in the cores, and we observed no statistically significant difference in any of those parameters. Consistent with prior reports,5, 7, 10 we also demonstrated that the presence of a mass lesion and large size of the lesion (>1.5 cm) were associated significantly with invasion on final excision.

Table 4. Summary of Recent Studies that Evaluated Core Biopsy of Ductal Carcinoma in Situ for Predictive Factors of Invasive Disease on Surgical Excision

Year	Type of biopsy (Needle size)	Targeted lesion	No. with invasion (%)	Significant predicting factors of invasion	Insignificant factors	Reference
S indicates stereotactic biopsy; G, gauge; US, ultrasound guided biopsy.
2000	S (11 G), S (14 G)	Calcifications	17/59 (29)	Inflammatory infiltrate	Nuclear grade, necrosis, desmoplasia, no. of cores obtained, size of lesion, distribution and morphology of calcifications, associated density, level of suspicion, 11-G vs. 14-G core biopsy	Lee et al., 20008
2000	S (14 G) US (14 G)	Nonpalpable lesions	11/71 (15)	Not available	Not available	Verkooijen et al., 20009
2001	Large core (14 G), vacuum assisted (11 G and 14 G)	Mass or calcifications	183/1326 (14)	Large-core biopsy, mass, ≥10 cores per lesion, greatest dimension of lesion	Not available	Jackman et al., 200110
2001	S (14 G)	Calcifications	14/68 (21)	High nuclear grade, increasing number of calcifications (>40 flecks)	Morphology of calcifications, size of calcifications	Bagnall et al., 20014
2001	S (14 G), US (14 G)	Mass or calcifications	36/140 (26)	Mass	Nuclear grade, comedo necrosis, periductal fibrosis and periductal inflammation	King et al., 20015
2001	Core biopsy, excisional biopsy	Not available	29/224 (13)	None	Comedo necrosis, histologic pattern, nuclear grade, core biopsy vs. excisional biopsy	Cox et al., 20016
2002	S (11 G and 14 G), US (14 G)	Mass, asymmetry, or calcifications	17/74 (19)	Comedo necrosis histology with papillary/cribriform features, size >4 mm with lobular extension	Nuclear grade, comedo necrosishistology, necrosis, size of thelargest focus, aggregate size	Renshaw, 200212
2003	S (14 G)	Nonpalpable lesions	41/255 (16)	High nuclear grade,periductal inflammation,large area of calcifications	Periductal fibrosis	Hoorntje et al.,20037
2005	Core biopsy,excisional biopsy	Not available	80/398 (20)	Core needle biopsy,mammographic size, hightumor grade, comedo necrosis	Palpable lesion, pathologic size	Yen et al., 200511

Early studies of the routine examination of Level I and level II axillary dissection specimens in patients with DCIS demonstrated that the rate of metastasis was low (<1%).27 Considering the morbidity associated with axillary dissection, its use generally is not warranted in this group of patients.6 With the advent of SLN surgery in recent years, several institutions use SLN biopsy, either routinely or selectively, in patients with DCIS. In a study by Cox et al.,6 240 patients with a biopsy diagnosis of DCIS or DCIS with microinvasion underwent SLN biopsy as part of their surgical treatment. In that study, 37 patients (15%) were identified with positive SLNs, including 26 of 195 patients (13%) who had a final diagnosis of DCIS, 3 of 15 patients (20%) who had a final diagnosis of DCIS with microinvasion, and 8 of 30 patients (27%) who had a final diagnosis of invasive carcinoma. Those authors did not identify any significant histologic predictors of invasion and, thus, recommended SLN biopsy for all patients with DCIS. In contrast, Klauber-DeMore et al.,14 limited their study population in an effort to define better which patients had a high risk for invasion to avoid recommending SLN biopsy for all patients with DCIS. Those authors prospectively evaluated 76 patients with DCIS who underwent SLN biopsy who had at least 1 of the following characteristics: palpable or mammographic mass, multicentric disease that required mastectomy, histology suspicious but not diagnostic for microinvasion, and histologically high nuclear grade or necrosis. In that study, 9 patients (12%) with those parameters had positive SLNs identified.

The rate of positive SLNs in our study was low (3%), and all 3 patients who had SLN metastases also had invasive carcinoma on excision. In the 69 patients who had DCIS without invasion and underwent SLN surgery at excision, only 1 patient had a cluster of isolated tumor cells (<0.2 mm) in 1 SLN; currently, those cells are considered to have uncertain biologic significance. This is in contrast to several reports, which indicated that between 3% and 13% of patients who had a final diagnosis of DCIS had positive SLNs.6, 15, 16 It remains controversial whether micrometastasis identified in lymph nodes in patients with DCIS implies that there is invasive disease in the primary lesion. Lara et al.28 reported on 102 patients with DCIS who had a 10-year follow-up and noted a 13% rate of lymph node micrometastasis. Those authors did not report any association between disease recurrence and micrometastasis, suggesting that lymph node micrometastasis in patients with DCIS has no clinical significance. However, other investigators consider micrometastasis as a sign of occult, invasive disease and treat it as lymph node-positive, invasive carcinoma.11

Although SLN surgery has a lower rate of complications than complete axillary dissection, several adverse effects have been reported, including allergy to the blue dye and to the radiocolloid, lymphedema, and damage to the intercostal brachial nerves.13, 29, 30 In addition, it is an expensive procedure and may not be justified in the treatment of all patients with DCIS.31 Among the 103 patients who underwent SLN biopsy in our study, 69 patients (67%) did not have invasive disease on final excision. If some of these patients could be identified preoperatively as having a low risk for invasion, then they could be spared an unnecessary SLN procedure. For this purpose, using the 4 significant variables that we identified as associated with invasion, we performed specificity and sen sitivity analyses based on a logistic regression model to predict invasion on the basis of our data (Fig. 1; Tables 5, 6). In our analysis, the greatest sensitivity and specificity (73% and 60%, respectively) could be achieved when the probability cut-off point was set between 0.20 and 0.22. If it is applied clinically, then this model may help reduce the number of patients who undergo unnecessary SLN surgery.

Figure 1. This plot illustrates sensitivity and specificity versus the probability cut-off points.

Table 5. Logistic Regression Model for Predicting Invasion

Variable	OR	Wald 95% CI	P
OR indicates odds ratio; 95% CI, confidence interval. * Reference: calcifications.
Mass lesion on imaging*	2.416	1.061–5.503	.0357
Presence of lobular cancerization	2.466	1.104–5.511	.0278
Lesion >1.5 cm	3.055	1.392–6.707	.0054
High nuclear grade	4.750	0.598–37.716	.1405

Table 6. Probabilities of Different Combinations of the 4 Variables Based on the Logistic Model in Table 5

Type of lesion	Lobular CA	Size, Cm	Nuclear grade	Probability	Sensitivity	Specificity
Lobular CA indicates lobular cancerization.
Low probability
Calcifications	No	≤1.5	Low	0.01945	1	0
Mass	No	≤1.5	Low	0.04572	1	0.08176
Calcifications	Yes	≤1.5	Low	0.04663	1	0.08805
Calcifications	No	>1.5	Low	0.05713	1	0.09434
Calcifications	No	≤1.5	High	0.0861	0.97561	0.13208
Mass	Yes	≤1.5	Low	0.106	0.854	0.472
Mass	No	>1.5	Low	0.1277	0.85366	0.4717
Calcifications	Yes	>1.5	Low	0.130	0.854	0.472
Mass	No	≤1.5	High	0.1854	0.85366	0.48428
Calcifications	Yes	≤1.5	High	0.18854	0.85366	0.54088
High probability
Calcifications	No	>1.5	High	0.22351	0.73171	0.60377
Mass	Yes	>1.5	Low	0.265	0.512	0.843
Mass	Yes	≤1.5	High	0.35952	0.5122	0.84277
Mass	No	>1.5	High	0.41018	0.4878	0.85535
Calcifications	Yes	>1.5	High	0.41517	0.21951	0.90566
Mass	Yes	>1.5	High	0.63169	0.04878	0.98113

Another important question that arose from our study was whether axillary staging is necessary in all patients with disease that is upstaged from DCIS to invasive disease on final pathology. It is noteworthy that, among the 41 patients with invasive disease in our series, 22 patients (54%) had invasive carcinomas that were categorized as T1a. Other investigators also observed the small size of the invasive component on final excision in a large proportion of patients with an original diagnosis of DCIS on CNB.6, 32 It has been proposed previously that axillary treatment may not be necessary in patients with Stage T1a invasive tumors because of the low rate of positive lymph nodes in this group of patients.27 Indeed, in our series, all 3 patients with positive SLN had invasive tumors >0.5 cm. A larger study with longer clinical follow-up in this group of patients is needed. Nevertheless, if a patient with CNB-diagnosed DCIS decides to forgo SLN biopsy at the time of surgical excision because of a lack of preoperatively defined “high-risk” factors for invasion and an invasive carcinoma is identified on final pathology, then the patient still could undergo an SLN biopsy if it is indicated clinically.

In conclusion, in this retrospective review of 200 patients with a CNB-diagnosed DCIS, we observed an association between the presence of invasive carcinoma on final surgical excision and clinical, radiologic, and histologic variables. The presence of a mass lesion, lesion size >1.5 cm, lobular cancerization, and a high nuclear grade of DCIS were associated significantly with invasion, and all except nuclear grade were independent predictors of invasion in multivariate analysis. Based on our data, we created a logistic regression model to predict invasion on CNB and, hence, to allow a more selective approach to SLN surgery.

Acknowledgements

Supported by an institutional developmental grant to E.R