A Comparison of Local Recurrence Risk Estimates After Breast-Conserving Surgery for DCIS: DCIS Nomogram Vs Refined Oncotype DX Breast DCIS Score™

Abstract

Background

A DCIS Nomogram, integrating 10 clinicopathologic/treatment factors, and a Refined DCIS Score (RDS), incorporating a genomic assay and 3 clinicopathologic factors (Oncotype DX DCIS Score™), are available to estimate DCIS 10-year local recurrence risk (LRR). We compared these estimates.

Methods

Patients age ≥50 with DCIS size ≤2.5cm and a genomic assay available were identified. RDS within 1-2% of the range of Nomogram LRR estimates obtained by assuming use and non-use of endocrine therapy (Nomogram+/− ET), were defined as concordant. Assuming a 10-year risk threshold for recommending radiation of 10%, Nomogram+/−ET and RDS estimates were compared; threshold concordance was determined.

Results

In 54/59 (92%), the RDS and Nomogram+/−ET LRR estimates were concordant. In the remaining 5/59 (8%), the RDS LRR estimates were lower than the Nomogram+ET with an absolute difference of 3-8% and thus were discordant. For these 5, the RDS estimates of 10-year LRR were ≤10% (range 5-8%) and the Nomogram+ET estimates were ≥10% (range 11-14%). These 5 patients with both discordant and threshold-discordant estimates all had close margins (≤2mm).

Conclusions

Among 92% women ≥50 with DCIS ≤2.5cm, free-of-charge online Nomogram 10-year LRR estimates were concordant with those obtained with the commercially available RDS (≥$4600). Among the 8% with discordant risk estimates, the RDS appears to underestimate the LRR and may lead to inappropriate omission of RT. Unless other data show a clinically significant advantage for the RDS (Oncotype DX DCIS Score™), our data suggest that, for women ≥50 with DCIS ≤2.5cm, its use is not warranted.

INTRODUCTION

Ductal carcinoma in situ (DCIS) accounts for 20% of all newly diagnosed breast cancer.¹ Most patients with DCIS will be treated with breast-conserving surgery (BCS) and radiation,² due to the proven benefit of radiation in reducing the local recurrence risk (LRR) by about half.³ The meta-analysis of the 4 mature randomized trials of radiation after BCS for DCIS, accrued in the 1980s-90s, showed that the 10-year local recurrence rate of 28.1% after BCS alone was reduced to 12.9% with the addition of radiation, but it did not lessen 10-year breast cancer mortality.³

Efforts have been made to identify a subset of patients with low-risk DCIS who derive no benefit from radiation, especially since local recurrence rates have decreased in the decades since the 4 randomized trials.⁴ However, to date, such a subset has not been established. In a randomized trial of patients with low-risk DCIS, defined as low- or intermediate-grade DCIS ≤2.5cm with margins >3 mm, the addition of radiation resulted in a significant reduction in 12-year local failure rates (2.8% with radiation vs 11.4% without radiation).^5,6 Yet, given the lack of improvement in survival and the rare but potentially serious morbidity of radiation,⁷ there are continuing efforts to develop tools to accurately estimate recurrence risk so that patients and clinicians can weigh the advantages and disadvantages of various treatment options.

The Memorial Sloan Kettering DCIS Nomogram (Nomogram), available online at www.nomograms.org. and the Oncotype DX Breast DCIS Score™ (Genomic Health, Redwood City, CA), which is currently reported as a “Refined” DCIS Score (RDS) incorporating a genomic assay and 3 clinicopathologic factors, are 2 clinically available tools designed to estimate LRR in patients with DCIS treated with BCS.^8,9 Here we compared LRR estimates for women with lower-risk DCIS (age ≥50, DCIS size ≤2.5cm) treated with BCS without radiation, obtained by using the Nomogram and the RDS, as reported by Genomic Health since late 2017 (adjusted for age, size, and diagnosis year). Furthermore, assuming that radiation would be recommended if 10-year LRR met a threshold of ≥10%, we sought to determine how often these 2 assays would yield similar or discordant radiation recommendation results.

METHODS

After institutional review board approval, the pathology database at Montefiore Medical Center was searched to identify all patients age ≥50 with DCIS ≤2.5cm in size, without positive margins, treated with BCS and for which a DCIS Score was obtained.

Clinicopathological information was obtained by chart review: age, family history of breast cancer in a first- or second-degree relative, screen-detected or clinical presentation of DCIS, nuclear grade, necrosis, estrogen receptor status, number of surgical excisions, DCIS size, surgical margin status (negative if >2mm; close if >0mm and ≤2mm), and year of surgery. Nomogram 10-year LRR estimates were calculated using the online tool available at http://nomograms.mskcc.org/breast/DuctalCarcinomaInSituRecurrencePage.aspx. These estimates were calculated without radiation or endocrine therapy (Nomogram–ET), and without radiation but with endocrine therapy (Nomogram+ET); these 2 values defined the Nomogram+/− ET estimate range.

Oncotype DX Breast DCIS Score™ reports were reviewed, and the numerical score was recorded. Until late 2017, the Genomic Health report showed a 10-year LRR estimate corresponding to the DCIS Score that was defined by a single curve for all patients. Since late 2017 there have been 4 different curves that are adjusted for diagnosis year (≥2000), and that depend on age category (<50 vs ≥50 years old) and tumor size (≤1.0cm vs >1.0-≤2.5cm). For this analysis, we use the term RDS to designate the 10-year LRR estimate that is currently reported and calculated according to diagnosis year, age and tumor size.¹⁰ Tumor size was determined in accordance with College of American Pathologists (CAP) guidelines.¹¹

For DCIS Scores reported before 11/2017, we calculated RDS risk by using the Genomic Health curves (reported after 11/2017) for the appropriate age/size stratum. We used DigitizeIt (Braunschweig, Germany) software to apply individual patient DCIS Scores to the appropriate current curves to obtain the RDS estimate.¹² There were 2 patients with scores >70; for them, we used the maximum DCIS Score of 70, which is the approach used for such patients on Genomic Health RDS reports.

Ten-year LRR estimates obtained from Nomogram and RDS were compared. RDS is reported with a 95% confidence interval that is ≥5% for tumor size ≤1.0cm, and ≥8% for tumor size >1.0 and ≤2.5cm. RDS was designed to be unaltered by use of ET. Therefore, RDS LRR estimates within 1-2% of the Nomogram+/−ET estimate range were defined to be concordant.

We assessed the number of women with risk estimates ≥10% and ≥15%. Assuming a threshold for recommending radiation of 10% 10-year LRR, Nomogram+/−ET and RDS estimates were compared and defined to be threshold concordant if the estimates were concordant, or if the discordant estimates were on the same side of the 10% threshold (i.e., either both did or both did not estimate risk ≥10%).

All statistical analyses were conducted in R version 3.4.4 (R Core Development Team, Vienna, Austria).

RESULTS

A total of 59 women ≥50 who underwent BCS without positive margins for DCIS ≤2.5cm, and for whom a DCIS Score was available, were identified. All patients underwent BCS between 11/2011-12/2017. Table 1 presents clinicopathologic characteristics.

TABLE 1.

Clinicopathologic characteristics of 59 patients with DCIS

DCIS ductal carcinoma in situ

Characteristic		Median (range)
Age at surgery, years		67 (50-81)
DCIS size, cm		0.6 (0.2-2.5)
N (%)
Presentation	Radiologic	55 (93%)
	Clinical	4 (6.8%)
Family history of breast cancer	Yes	12 (20%)
	No	47 (80%)
Nuclear grade	1	10 (17%)
	2	34 (58%)
	3	15 (25%)
Necrosis present	Yes	35 (59%)
	No	24 (41%)
Size category	≤ 1 cm	42 (71%)
	> 1 cm and ≤ 2.5 cm	17 (29%)
Number of excisions	1	55 (93%)
	2	4 (6.8%)
Margin width	> 0 mm and ≤ 2 mm	12 (20%)
	> 2 mm	47 (80%)
Estrogen receptor	Positive	58 (98%)
	Negative	1 (1.7%)

Median Nomogram–ET 10-year LRR was 14% (range 9-27%), with 56 (95%) having a risk ≥10%, and 24 (41%) having a risk ≥15%. Median Nomogram+ET 10-year LRR was 7% (range 4-14%), with 12 (20%) having a risk ≥10% and none ≥15% (Table 2).

TABLE 2.

Number and proportion of women (n = 59) with local recurrence risk estimates in each risk category. Risk estimated by Nomogram, with endocrine therapy and without, and by Refined Oncotype DX DCIS Score™ (genomic assay adjusted for year of diagnosis, age, and size)

DCIS ductal carcinoma in situ

	10-year local recurrence risk estimate
Method of risk estimation	<10%	≥ 10%	≥15%
	N (%)	N (%)	N (%)
Nomogram, with endocrine therapy	47 (80%)	12 (20%)	0 (0%)
Nomogram, without endocrine therapy	3 (5%)	56 (95%)	24 (41%)
Refined DCIS Score	35 (59%)	24 (41%)	4 (7%)

Comparison of the reported LRR estimates from DCIS Scores prior to late 2017 (based solely on the DCIS Score) to the RDS LRR estimates reported since late 2017 (adjusted for diagnosis year, age, and size) showed that for every patient in our cohort (age ≥50, size ≤2.5cm), the RDS estimate was lower. Median RDS 10-year LRR estimate was 8% (range 5-17%), with 24 patients (41%) having a risk ≥10%, and with 4 (7%) ≥15% (Table 2).

Comparison of Nomogram and RDS local recurrence risk estimates

Overall, the Nomogram+/−ET and RDS risk estimates were concordant in 54/59 (92%) cases and discordant in 5/59 (8%)(Lig. 1, 2). Only 2 women (ages 76 and 78) had a higher RDS than Nomogram–ET LRR estimate; the estimates differed by 1% and 2%, and were therefore concordant [13% vs 12%, and 13% vs 11% (RDS vs Nomogram–ET)](Lig. 1). Both women had risk estimates meeting the 10% radiation threshold by both methods, and therefore were threshold-concordant.

Compared with Nomogram+ET, 16/59 patients had lower RDS LRR estimates. In 11, the absolute difference in risk estimates was only 1-2%, and the estimates were therefore concordant. In 5/59 (8%), RDS risk estimates were >1-2% lower than the Nomogram+ET (absolute difference 3-8%), and therefore discordant. For these 5, the RDS 10-year LRR estimates were <10% (range 5-8%) as compared to the Nomogram+ET estimates of ≥10% (range 11-14%), and therefore deemed threshold-discordant. These 5/59 (8%) patients with both discordant and threshold-discordant estimates all had close margins (≤2mm)(Fig. 1, 2).

Fig. 1.

Refined Oncotype DX Breast DCIS Score™ (RDS) and Nomogram with/without endocrine therapy risk estimates for each patient. Estimates were deemed concordant if the RDS was within 1-2% of the Nomogram with/without endocrine therapy range (open circles, concordant estimates; black solid squares, discordant estimates).

DCIS ductal carcinoma in situ

Fig 2.

Concordance of 10-year local recurrence risk estimates, as estimated by the Refined Oncotype DX Breast DCIS Score™ as compared to the Nomogram with or without endocrine therapy.

*A11 5 discordant estimates were in patients with close (> 0 mm, ≤ 2 mm) margins.

DCIS, ductal carcinoma in situ

Overall, 12 (20%) patients had margins ≤2mm. RDS LRR estimate was below the 10% radiation recommendation threshold in 7/12 (58%). In contrast, 11/12 (92%) met the ≥10% radiation threshold based on the Nomogram+ET, and all 12 (100%) reached it based on the Nomogram–ET LRR estimate (Fig. 3).

Fig 3.

Among 12 DCIS patients with close margins (≤ 2 mm), the proportion of patients with risk estimates that reached ≥ 10% 10-year local recurrence risk threshold using the Nomogram without endocrine therapy, Nomogram with endocrine therapy, and the Refined Oncotype DX Breast DCIS Score™.

DCIS ductal carcinoma in situ

DISCUSSION

Here we have found a remarkable concordance between the Nomogram and RDS 10-year LRR estimates among 59 women age ≥50 with DCIS ≤2.5cm.

The Nomogram was developed in a population of 1681 women with DCIS treated from 1991-2006 with BCS.⁸ Age, family history, clinical presentation, margin width, nuclear grade, necrosis, number of excisions necessary (included as a surrogate for size), year of surgery, and use of RT and ET were included; each is associated with LRR.

Age has been shown to heavily influence LRR in women who both do or do not receive radiation.^3,13 Cronin performed a multivariable LRR analysis by decade of age, stratifying by receipt of radiation and adjusting for 7 other clinicopathologic variables. In women who underwent BCS without radiation, the youngest cohort (age <40) had a 4.2-fold higher 10-year LRR than the oldest group (≥80).¹³

The meta-analysis of the 4 prospective randomized studies of radiation for women with DCIS treated with BCS showed that among women with DCIS treated with BCS alone, larger size of DCIS is associated with greater 10-year risk of LR.¹ Notably, however, pathologic size was missing in most women in these trials due to difficulties in determining microscopic extent of disease.³ Although the CAP has issued guidelines on determining DCIS size,¹¹, any DCIS size measurement must be considered an estimate¹⁴ because it is highly dependent on specimen sampling completeness, with optimal sampling requiring the entire specimen to be serially submitted. In a 2018 survey of over 800 pathology departments, only 29% reported routinely submitting the entire DCIS specimen.¹⁵ Furthermore, several methods exist for reporting DCIS size, and although CAP recommends use of the method which yields the largest size, only 32% of institutions do so. Recipients of the RDS report may not realize that the RDS estimate is highly dependent on the size provided. Recognizing the difficulty of determining DCIS size, the Nomogram does not use pathologic size, but instead includes number of excisions, which is a correlate of extent of disease, and, especially, radiographically occult disease.

Treatment time period has also been shown to have a large impact on LRR. Subhedar examined almost 3000 women with DCIS treated with BCS from 1978-2010, and examined outcomes based on treatment year.⁴ They found that among women who underwent BCS alone 10-year LRR fell over time, and even after adjustment for 9 clinicopathologic/treatment factors, there was a 38% reduction in risk for women treated in later vs earlier years. Surgery year is included in the Nomogram, and it is one of the factors that now refines the DCIS Score estimates.

The original DCIS score was a poor predictor of LRR, as shown in a multivariable analysis of 571 women from the Ontario cohort.¹⁶ Although the DCIS Score was associated with LRR (p=0.02) after adjusting for age, size, grade, necrosis, multifocality, and architectural growth pattern,¹⁶ each of these clinicopathologic variables had an effect on LRR of larger magnitude than did the DCIS Score. It was also applied to a subset (49%) of 327 women from ECOG5194 and found to be associated with LRR (p=0.02) after adjusting for menopausal status and size.⁹ However in both populations, the LRR associated with “intermediate” DCIS score was higher than that for “high” DCIS score. Importantly, no measure of predictive accuracy (either discrimination or calibration) of the DCIS score has been provided in any publication of which we are aware, for either the Ontario or the original ECOG population subset in which it was “validated”.^{9, 16–18}

In contrast, the Nomogram’s accuracy has been validated in at least 5 independent populations, constituting a total of 2594 patients.^19–23 Measures of discrimination (area under the receiver operator curve or c-index) range from 0.65-0.92, consistent with a good-to-excellent model, and calibration (correlation between observed and predicted values) has been very good to excellent in these populations.^19–23

In keeping with their previous finding that several clinicopathologic factors had a greater predictive value than did the DCIS Score,¹⁶ Rakovitch and colleagues recently created predictive models with age, size, multifocality, grade, necrosis, margin status, margin width, and radiation, with and without the DCIS Score. They found discrimination only increased from 0.68 to 0.70 with the addition of the DCIS Score, proving that most of the predictive ability of the model was from the clinicopathologic and treatment variables rather than the DCIS Score.¹⁷

Presumably because of these findings, “refined estimates” of LRR by DCIS Score, adjusted for age, size, and diagnosis year, were derived from a combination of subsets of the ECOG 5194 and Ontario populations (n=773).¹⁸ They are now provided when the DCIS Score is performed and reported (which we have referred to in the current study as the RDS). Examination of these LRR estimates shows that the estimates vary more due to patient age and DCIS size than from a change in DCIS Score from low to high, again demonstrating age and size contribute more to LRR estimation than does DCIS Score.¹⁸

Given the evidence provided by Rakovitch^16–18 showing that the contribution from clinicopathologic features to risk estimation is greater than the DCIS Score, and that the currently reported RDS incorporates 3 important clinicopathologic factors (age, size, diagnosis year), it should not be surprising that we found no clinically significant discordance between the Nomogram and RDS risk estimates in 92% of cases. RDS estimates were substantially lower than Nomogram+ET estimates in only 5 women; all had close margins. Margins ≤2mm have been clearly associated with a higher LRR for women receiving radiation^24,25 and appear to be even more important for those not receiving radiation.²⁶ While the Nomogram incorporates margin width, the RDS does not, suggesting the Nomogram provides a more realistic LRR estimate while the RDS underestimates the risk for those with close margins. Such underestimation could lead to inappropriate omission of radiation.

The current work is the first of which we are aware that directly compares the Nomogram and RDS; it highlights another important difference between them. The DCIS Score was designed to be unaltered by ET use,⁹ even though ET has been shown in randomized trials to reduce LRR substantially,^27–29 and even though of patients diagnosed ≥2000 in which the RDS was derived, about half (48.9%) received ET.¹⁸ This design renders the RDS insensitive to a treatment that clearly reduces LRR and thereby lessens the precision of the RDS (i.e., the RDS is not altered by ET use and thus LRR should be lower than the RDS estimate if the patient takes ET, and should be higher if she does not). Because the Nomogram incorporates ET use, the appropriate comparison is between the RDS and the range of estimates defined by Nomogram+/− ET. In the current series, the risk estimates derived from the Nomogram identify more women (without ET) at the highest risk ≥15% (41% vs 7% with RDS) and more women (with ET) at low risk <10% (80% vs 59% with RDS), which suggests better risk separation.¹⁸

An important limitation of our work is that actual local recurrence outcomes are not available for these recently treated patients. The best assessment of any prediction tool is obtained by comparing predicted to observed outcomes and assessing measures of discrimination and calibration. While discrimination and calibration in independent populations have been published for Nomogram predictions,^19–23 they have not for either the original DCIS Score nor the RDS estimates. Another limitation of our work is the number of patients; the DCIS Score was available only if ordered at the discretion of clinicians and patients after discussion and shared decision-making.

Conclusions

We have shown that in 92% of patients ≥50 with DCIS ≤2.5cm, a readily available, free-of-charge online Nomogram provided 10-year LRR estimates concordant with an RDS that incorporates a genomic assay and 3 clinicopathologic variables and costs $4620. The 8% of women who had discordant estimates all had close margins, which is incorporated into the Nomogram but not the RDS LRR estimates, thereby suggesting that the RDS likely underestimates their risk. Unless and until further data demonstrate a clinically significant advantage of the costly genomic assay, our data suggest that for women ≥50 with DCIS ≤2.5cm, use of the RDS for LRR estimation is not warranted.

Synopsis:

Most women age ≥50 with DCIS ≤2.5cm have risk estimates concordant between the free-of-charge Nomogram and Refined DCIS Score (RDS)(>$4600); among discordant estimates, RDS appears to underestimate risk. These findings suggest use of RDS is not warranted for these patients.