PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of jnciLink to Publisher's site
 
J Natl Cancer Inst. 2016 May; 108(5): djv372.
Published online 2015 December 16. doi:  10.1093/jnci/djv372
PMCID: PMC4849804

Outcomes of Active Surveillance for Ductal Carcinoma in Situ: A Computational Risk Analysis

Abstract

Background:

Ductal carcinoma in situ (DCIS) is a noninvasive breast lesion with uncertain risk for invasive progression. Usual care (UC) for DCIS consists of treatment upon diagnosis, thus potentially overtreating patients with low propensity for progression. One strategy to reduce overtreatment is active surveillance (AS), whereby DCIS is treated only upon detection of invasive disease. Our goal was to perform a quantitative evaluation of outcomes following an AS strategy for DCIS.

Methods:

Age-stratified, 10-year disease-specific cumulative mortality (DSCM) for AS was calculated using a computational risk projection model based upon published estimates for natural history parameters, and Surveillance, Epidemiology, and End Results data for outcomes. AS projections were compared with the DSCM for patients who received UC. To quantify the propagation of parameter uncertainty, a 95% projection range (PR) was computed, and sensitivity analyses were performed.

Results:

Under the assumption that AS cannot outperform UC, the projected median differences in 10-year DSCM between AS and UC when diagnosed at ages 40, 55, and 70 years were 2.6% (PR = 1.4%-5.1%), 1.5% (PR = 0.5%-3.5%), and 0.6% (PR = 0.0%-2.4), respectively. Corresponding median numbers of patients needed to treat to avert one breast cancer death were 38.3 (PR = 19.7–69.9), 67.3 (PR = 28.7–211.4), and 157.2 (PR = 41.1–3872.8), respectively. Sensitivity analyses showed that the parameter with greatest impact on DSCM was the probability of understaging invasive cancer at diagnosis.

Conclusion:

AS could be a viable management strategy for carefully selected DCIS patients, particularly among older age groups and those with substantial competing mortality risks. The effectiveness of AS could be markedly improved by reducing the rate of understaging.

Almost one in 1300 mammograms will detect ductal carcinoma in situ (DCIS) (1), which is considered the earliest detectable form of breast cancer. Over 50 000 women in the United States will be diagnosed with DCIS this year alone (2), almost exclusively in asymptomatic individuals. Without treatment, it is estimated that only 14% to 53% of DCIS will progress to invasive cancer (3,4). However, we currently lack markers to identify the likelihood and rate of DCIS progression. Thus, current National Comprehensive Cancer Network guidelines (5) recommend that DCIS be treated with a combination of surgery, radiation, and hormonal therapy upon diagnosis—recommendations similar to those for patients with invasive cancer except for the omission of chemotherapy.

DCIS was rarely encountered prior to introduction of mammography (4). There is growing concern that for some patients, particularly those with DCIS, breast cancer screening may unintentionally cause harm by introducing additional procedures, promoting anxiety (6,7), and detecting cancers that may never cause illness or death (8). In fact, advances in epidemiology and biology have shown that cancer is a conglomerate of many conditions and that screening uncovers lesions that may never impact a person’s health if left undetected (9–13). By definition, treatment of these entities can only result in harm, without improving either survival or quality of life. Overdiagnosis and overtreatment resulting from mammographic screening has been estimated to be as high as one in four patients diagnosed with breast cancer (14), although the absence of a standard definition of overdiagnosis leads to uncertainty around this estimate (15).

As a noninvasive lesion with poorly quantified progression risk, DCIS is a disease in which a watchful waiting or active surveillance (AS) approach may be rational. Clinical trials of AS for DCIS have been considered, but there are important challenges regarding the ethics, feasibility, and cost of a prospective randomized trial. Moreover, there has been reluctance to randomize patients to AS in the absence of outcome data, even in view of evidence that if DCIS does progress to invasive cancer and is detected at a stage confined to the breast, the five-year disease-specific survival is extremely high, at 98.5% (16), compared with 99% for treated DCIS (17). Because only 3% of patients diagnosed with DCIS elect no surgery or radiation therapy, which might best represent AS (17), there is virtually no evidence regarding outcomes for women who choose AS.

In a proposed AS scenario—similar to the approach many have adopted for early-stage prostate cancer—DCIS patients would enter a surveillance program consisting of regular follow-up screens with treatment only if invasive disease is diagnosed, instead of receiving usual care (UC) consisting of treatment at the time of diagnosis. The purpose of our study was to estimate the disease-specific cumulative mortality (DSCM) of AS and to compare the projected outcome with the DSCM of the UC scenario.

Methods

We developed a computational risk projection model that combines the natural history of breast cancer with AS and directly integrates outcome data from the National Cancer Institute’s Surveillance, Epidemiology, and End Results (SEER) database of the United States (1999–2011) (18). The model parameters are summarized in Table 1, and details about model development and parametrization are provided in the Supplementary Methods (available online). All simulations were performed using R (19).

Table 1.
Literature-based model parameters*

Computational Model

The natural history of DCIS and breast cancer was modeled using a Markov chain model (Supplementary Figure 1 and Supplementary Methods, available online) (20). DCIS either progresses to invasive ductal carcinoma (IDC) with probability p0 (progressive DCIS) or regresses/remains indolent (nonprogressive DCIS). Progressive DCIS then advances to localized IDC at rate α, upon which it spreads to the regional lymph nodes at rate β and eventually metastasizes to distant organs at rate γ. Based on the theory outlined in the Supplementary Methods (available online), we then derived formulas for the time-dependent probabilities pl(t), pr(t), and pd(t) of the lesion being localized, regional, or distant, respectively (here, t > 0 measures the time elapsed since progression of DCIS).

The AS strategy consists of three phases (Figure 1). Phase 1: the initial diagnosis is performed by vacuum-assisted core biopsy. Since roughly one in five DCIS diagnoses with this technique contain an undetected invasive portion (21), we introduce the probability pinv of understaged IDC at initial diagnosis; otherwise the lesion is purely in situ (true DCIS). Phase 2: the patient enters the AS program and is screened with a combination of mammography and magnetic resonance imaging (MRI) for malignancy every Δt months. For the model, we assume the highest performance of breast imaging surveillance for detection of invasive cancer as reported in studies of surveillance for women at increased risk of breast cancer (see also Supplementary Table 3, available online). In contrast to patients with understaged IDC (phase 2a), patients with true DCIS (phase 2b) may not develop a screen-detectable malignancy in their remaining lifespan. The true-positive rate, or screening sensitivity, is denoted by psen. Phase 3: once diagnosed with localized, regional, or distant IDC according to the temporal stage distribution pl(t), pr(t), and pd(t), the patient is treated according to UC with the attendant outcomes in SEER. Finally, we accounted for the possibility of death from an unrelated cause by incorporating competing risks for patients diagnosed with DCIS in SEER.

Figure 1.
Active surveillance strategy. 1) Ductal carcinoma in situ (DCIS) is diagnosed at time t=0 and is either understaged (with probability pinv) or classified as true DCIS otherwise. 2a) If understaged, the lesion is already localized at t=0 and starts progressing ...

Model Inputs

The model inputs are divided into three groups: 1) natural history parameters (Table 1), 2) control parameters that specify the AS strategy (Table 1), and 3) disease-specific and competing mortalities. 1) The natural history parameters describe the disease dynamics in absence of treatment. The probability to progress from DCIS to IDC (p0) and the progression rate (α) are estimated based on natural history studies (Supplementary Table 1 and Supplementary Methods, available online).

Estimates for the progression rates from localized to regional (β) and regional to distant (γ) disease were calculated by combining the stage distribution of invasive cancer in SEER prior to the use of screening mammography together with estimates for the average time between onset of subclinical invasive disease and clinical detection of the lesion. The latter time span between onset and detection is referred to as the mean sojourn time and has been estimated by multiple breast cancer screening studies (Supplementary Table 2 and Supplementary Methods, available online). As anticipated, the uncertainty ranges for these rates are large as they accommodate all different biological subtypes of invasive breast cancer. 2) Control parameters specify the surveillance program: the screening interval [increment]t, the screening sensitivity psen, and the understaging probability pinv. These parameters are adjustable within technological constraints. 3) SEER-based, nonparametric competing risk estimates (22) were used to characterize outcome measures: DSCM for treated IDC patients and risk of dying from unrelated causes.

Patient Characteristics in SEER

To evaluate the AS strategy, we considered three virtual patient groups by age at DCIS diagnosis: 40 (range = 39–41), 55 (range = 54–56), and 70 (range = 69–71) years. For the nonparametric, SEER-based outcome estimates, we included female breast cancer patients age 18 years and older diagnosed between 1991 and 2011. We excluded patients with a first cancer diagnosis other than breast cancer, those without microscopic disease confirmation, those identified at autopsy/death certificate only, those with no or unknown surgical treatment, and those with incomplete survival times or 0 days of survival. In addition, the following DCIS- and IDC-specific selection criteria were applied.

DCIS Cohort

Patients were included if diagnosed with ICD-O-3 behavior code in situ. The following ICD-O-3 histology codes were excluded: 8520/2 (lobular carcinoma in situ), 8522/2 (intraductal and lobular in situ carcinoma), 8524/2 (lobular CIS with other CIS), and 8720/2 (melanoma in situ). The final cohort comprised 120 862 patients.

IDC Cohort

Patients were included if diagnosed with ICD-O-3 behavior code malignant. The following ICD-O-3 histology codes were excluded: 8520/3 (lobular carcinoma, not otherwise specified), 8522/3 (infiltrating duct and lobular carcinoma), 8524/3 (infiltrating lobular mixed with other types of cancer), 8720/3 (malignant melanoma, not otherwise specified), and 8743/3 (superficial spreading melanoma). Patients with unknown SEER historic stage were also excluded from the analysis. The final cohort comprised 505 848 patients.

Main Outcome Measures

The main outcome measure was the 10-year DSCM for AS, defined as the probability of dying of breast cancer within 10 years of original DCIS diagnosis. In addition to the baseline control parameters (pinv = 18.9%, psen = 80%, [increment]t = 6 months), we also considered a scenario of improved control parameters (pinv = 10%, psen = 90%, [increment]t = 6 months) to model improved diagnostic performance of imaging consisting of a false-negative rate of invasive cancer of 10%, compared with 18.9% as published in a large meta-analysis (21). To quantify outcome uncertainty because of uncertainty in the model inputs, Monte Carlo simulations were performed as follows: Control parameters were fixed (baseline or improved), natural history parameters were sampled uniformly at random from their uncertainty ranges (Table 1), and SEER-based mortality estimates were sampled based on asymptotic normality. By discarding the 2.5% most extreme values of the model projection on either side of the median, a 95% projection range (PR) was obtained. The DSCM for UC was calculated using SEER-based, nonparametric competing risk estimates (22), and standard 95% confidence intervals (CIs) were computed.

To compare each of the two AS strategies (baseline and improved control parameters) with the UC scenario, the difference between the 10-year DSCM of AS and UC strategies was computed. Because this difference yields negative values for certain parameter sets in the model, we also calculated a conditional difference to reflect the plausible assumption that delayed treatment (AS) cannot yield a better outcome than immediate treatment (UC). Finally, taking AS as the reference strategy, the corresponding number of patients needed to treat (NNT) immediately (with UC) to save one life after 10 years was calculated. Again, Monte Carlo runs with a better outcome for AS compared with UC were excluded to ensure positivity of the conditional NNT estimates.

To rank the control parameters according to their mortality reduction potential, we performed a global sensitivity analysis (Supplementary Methods, available online). In brief, we discretized the control parameter space and computed the local sensitivity ratio (SR) for each of the three control parameters and across the entire grid. To obtain a global sensitivity measure, the local SR distributions were compared between control parameters.

Results

Breast Cancer Mortality of Active Surveillance vs Usual Care

Ten years after DCIS diagnosis, the projected median DSCM for the AS strategy with the baseline control parameters was found to be 3.5% (95% PR = 2.5–6.0) at age 40 years, 2.8% (95% PR = 1.9–4.8) at age 55 years, and 2.8% (95% PR = 1.9–4.8) at age 70 years (Figure 2A and Table 2). The SEER-based estimate of the 10-year DSCM for UC increased from 0.9% (95% CI = 0.6 to 1.3) at age 40 years, to 1.3% (95% CI = 1.0 to 1.7) at age 55 years, and to 2.7% (95% CI = 2.1 to 3.2) at age 70 years. Under the assumption that AS cannot outperform UC, the projected median differences in 10-year DSCM between AS and UC when diagnosed at ages 40, 55, and 70 years were 2.6% (PR = 1.4–5.1%), 1.5% (PR = 0.5–3.5%), and 0.6% (PR = 0.0–2.4), respectively. Finally, the age-stratified 10-year mortalities because of other causes of death were estimated to be 1.3% (95% CI = 0.8 to 1.7), 4.0% (95% CI = 3.5 to 4.6), and 16.5% (95% CI = 15.3 to 17.8), respectively.

Figure 2.
Cumulative mortality—active surveillance (AS) disease-specific projections vs usual care (UC) disease-specific and competing risks. The model-based, projected disease-specific cumulative mortality (DSCM) for the AS strategy (median: black line ...
Table 2.
Ten-year cumulative mortality—active surveillance disease-specific projections vs usual care disease-specific and competing risks

Under improved control parameters with a lower understaging risk (pinv = 10%) and a higher screening sensitivity (psen = 90%), lower mortality projections were obtained (Figure 2B and Table 2). In particular, the median conditional difference in 10-year DSCM between AS and UC strategies was found to be 1.7% (95% PR = 0.5–4.4) at age 40 years, 0.9% (95% PR = 0.0–3.1) at age 55 years, and 0.6% (95% PR = 0.0–2.2) at age 70 years. Projections for additional age groups (40, 50, 60, 70, and 80 years) and alternate screening intervals (12 and 24 months) are found in Supplementary Tables 5–7 (available online).

Numbers Needed to Treat

The conditional NNT were computed by treating AS as the reference strategy (Table 2). For the baseline control parameters, the NNT were 38.3 (95% PR = 19.7–69.9), 67.3 (95% PR = 28.7–211.4), and 157.2 (95% PR = 41.1–3872.8) at ages 40, 55, and 70 years, respectively. Based on improved control parameters, the corresponding NNT were substantially higher at 57.9 (95% PR = 22.7–211.2), 109.8 (95% PR = 32.3–2113.5), and 178.4 (95% PR = 45.7–3941.4), respectively. See Supplementary Tables 5–7 (available online) for additional age groups and screening intervals.

The Role of the Control Parameters

Sweeping across the entire control parameter space, the 10-year DSCM for AS was found to be most sensitive to the understaging probability pinv, an inherent limitation of diagnosis. For the age group 55 years, the mean local SR for pinv was found to be 7.1 (SD = 0.7). In contrast, the mean local SR for screening sensitivity psen and screening interval [increment]t were one order of magnitude smaller, at -0.5 (SD = 0.6) and 0.58 (SD = 0.3), respectively. In consequence, the 10-year DSCM for AS is most effectively reduced by decreasing pinv. This is illustrated in Figure 3, where the changes in 10-year DSCM are visualized for varying pinv and psen ([increment]t = 6 months, fixed). Similar results were found for age groups 40 and 70 years (data not shown).

Figure 3.
Control parameter sensitivity for age at diagnosis 55 years. Sensitivity of the 10-year disease-specific cumulative mortality (DSCM) to the control parameters [increment]t (screening interval) and pinv (probability of understaging invasive ductal carcinoma ...

Discussion

Although it is estimated that mammography has reduced breast cancer mortality by over 20% (23), there is growing concern that for some patients, particularly those with DCIS, breast cancer screening may unintentionally cause harm by intervening in patients who may never progress to disseminated cancer. Over the past few years, 28.3% of DCIS patients received a mastectomy and 69.5% a lumpectomy (17). These interventions can lead to a range of physical and psychological consequences during and after treatment that might outweigh a limited survival improvement (6,7,24). Motivated by the absence of robust biomarkers for stratification by progression risk, we considered an alternative active surveillance management strategy for DCIS patients. In this scenario, DCIS patients were assumed to undergo follow-up screening and to receive treatment only upon diagnosis with IDC. Based on a computational risk projection model, we estimated the disease-specific mortality for AS after diagnosis with DCIS and compared our projections to the SEER-based mortality estimates of UC. To our knowledge, this is the first study to model and quantify the mortality impact of AS in DCIS.

As expected, the projected 10-year DSCM for patients with AS was found to be higher than for patients with UC for most parameter sets, but the projected difference in outcome between AS and UC decreased with age and depended on the risk of understaged IDC at presentation. Assuming a six-month screening interval and baseline screening sensitivity for invasive cancer, the 10-year conditional difference in DSCM decreased from 2.6% (1.4–5.1) at age 40 years to 0.6% (0.0–2.4) at age 70 years. The corresponding median NNT to avert one breast cancer death at 10 years was approximately 160 for patients age 70 years at diagnosis.

Another effectiveness measure for AS is obtained by comparing the DSCM for AS with the cumulative mortality because of non–breast cancer related causes of death. As expected, the latter was found to be highly age-dependent. Ten-year mortality because of competing risks was found to be smaller than AS mortality at age 40 years, roughly equal to the latter at age 55 years, and about six-fold higher at age 70 years. In other words, a patient age 70 years of average health diagnosed with DCIS electing an AS strategy has a six-fold higher likelihood of dying from other causes than of breast cancer within 10 years. Based on these results, we believe it reasonable to discuss active surveillance for select patients, especially those with substantial comorbidities and advanced age where the difference in mortality risk between AS and UC may be overshadowed by the competing mortality risk. Defining patient characteristics for which AS should be recommended over UC will require further research.

The risk of a subsequent invasive event provides an alternative measure of comparison between AS and UC (Supplementary Methods, available online). For AS, this corresponds to the risk of being diagnosed with IDC during a follow-up screen. Comparing the corresponding AS projections to empirical UC estimates, we found that the risk of a subsequent invasive event was about two- to five-fold higher for AS compared with UC (Supplementary Figure 5 and Supplementary Methods, available online). However, it is important to emphasize that the increased risk of subsequent invasive disease is much larger than the corresponding increase in disease-related mortality, and in most cases the treatment for an early-stage breast cancer would be the same as that for DCIS. In addition, for DCIS cases without progression no intervention would be necessary.

Previously, several models of breast cancer screening and treatment have been developed (25–32). Most of these studies used cancer registry data for model calibration and validation. However, because we focused on a non-evidence-based alternative scenario, we directly incorporated SEER-based outcome estimates into the model. In addition to the global sensitivity analysis, we also performed a variance decomposition analysis to evaluate the respective contributions of the different natural history parameters to the projection uncertainty (Supplementary Methods, available online). In brief, we found that DCIS-specific parameters (propensity of and time to progression) played the most important roles, followed by the transition rate from localized to regional IDC.

Our study has several limitations. First, inherent limitations of the SEER database (33,34) with respect to outcome ascertainment and cause of mortality attribution may impact the validity of our projections. Second, local recurrence rates in treated DCIS patients have decreased over the past years (35), suggesting that the employed SEER-based estimates from 1991–2011 may not be representative of the current DSCM in UC patients. However, similar trends were observed in patients with invasive cancer (36). Therefore, the projected difference in DSCM between UC and AS from this model is expected to provide a rational estimate, even in light of current trends in improved treatment outcomes. Third, in patients with estrogen-sensitive DCIS lesions, AS could be combined with administration of antihormonal therapy. Indeed, adjuvant tamoxifen therapy has been shown to reduce the risk of subsequent breast cancer in DCIS patients undergoing usual care (37–39), and it is conceivable that antihormonal therapy may be beneficial for AS patients as well. Because SEER rarely recorded estrogen status of DCIS lesions in earlier years, we were not able to model the impact of antihormonal therapy during AS. Finally, because we modeled a hypothetical scenario, it was not possible to systematically validate our projections against data. However, the projection ranges were in good agreement with a SEER-based analysis of low-grade DCIS patients who did not receive surgery (40). In addition, extensive sensitivity analyses were performed to compensate for this shortcoming.

Importantly, the DSCM projections of the delayed treatment strategy depend on screening frequency, diagnostic accuracy, and screening sensitivity. The ability to rule out concurrent IDC at initial presentation (ie, low risk of understaged IDC) is most critical for reducing AS mortality. Future reduction of the understaging risk through more accurate biopsy techniques and improved imaging modalities will increase the effectiveness of AS, thereby further narrowing the tradeoffs of AS compared with UC (41). In addition, the model does not include potential systemic risk-reducing measures such as endocrine therapy, which may further reinforce our observation that AS may be a rational strategy for select patient groups. Moreover, ongoing work to improve risk stratification for invasive cancer progression will allow better patient selection for AS, further mitigating the mortality risk of this approach (Supplementary Methods and Supplementary Table 4, available online).

In summary, our model shows that AS could be a safe and viable management strategy for carefully selected populations of DCIS patients. Based on NNT estimates and mortality risks from competing causes, we found AS to be most effective in older patients and those with substantial competing mortality risks. Improved diagnostic technologies that minimize the risk of understaging at diagnosis are necessary for a further reduction in breast cancer mortality with AS. These findings can motivate and inform future AS trials for DCIS with the goal of reducing overtreatment in the DCIS patient population.

Funding

National Institutes of Health (R01-GM096190 to MDR, RD); Swiss National Science Foundation (P300P2_154583 to MDR); The National Cancer Institute Early Detection Research Network (UO1 CA084955 to JRM).

Supplementary Material

Supplementary Data:

Notes

The authors declare no relevant conflicts of interest.

References

1. Ernster VL, Ballard-Barbash R, Barlow WE, et al. Detection of ductal carcinoma in situ in women undergoing screening mammography. J Natl Cancer Inst. 2002;94(20):1546–1554. [PubMed]
2. American Cancer Society. Cancer Facts and Figures 2015. Atlanta, GA.
3. Erbas B, Provenzan E, Armes J, et al. The natural history of ductal carcinoma in situ of the breast: a review. Breast Cancer Res Treat. 2006;97(2):135–144. [PubMed]
4. Ozanne EM, Shieh Y, Barnes J, et al. Characterizing the impact of 25 years of DCIS treatment. Breast Cancer Res Treat. 2011;129(1):165–173. [PubMed]
5. National Comprehensive Cancer Network. Stage 0 Breast Cancer (Version 1.2014). http://www.nccn.org/patients/guidelines/stage_0_breast/index.html. Accessed November 23, 2015.
6. Ganz PA. Quality-of-life issues in patients with ductal carcinoma in situ. J Natl Cancer Inst Monogr. 2010;2010(41):218–222. [PMC free article] [PubMed]
7. Lauzier S, Maunsell E, Levesque P, et al. Psychological distress and physical health in the year after diagnosis of DCIS or invasive breast cancer. Breast Cancer Res Treat. 2010;120(3):685–691. [PubMed]
8. Harding C, Pompei F, Burmistrov D, et al. Breast Cancer Screening, Incidence, and Mortality Across US Counties. JAMA Intern Med. 2015;175(9):1483–1489. [PubMed]
9. Gulati R, Inoue LYT, Gore JL, et al. Individualized Estimates of Overdiagnosis in Screen-Detected Prostate Cancer. J Natl Cancer Inst. 2014; In press. [PMC free article] [PubMed]
10. Hall SF, Irish J, Groome P, et al. Access, excess, and overdiagnosis: the case for thyroid cancer. Cancer Med. 2014;3(1):154–161. [PMC free article] [PubMed]
11. Loeb S, Bjurlin MA, Nicholson J, et al. Overdiagnosis and Overtreatment of Prostate Cancer. Eur Urol. 2014;65(6):1046–1055. [PMC free article] [PubMed]
12. Patz EF, Pinsky P, Gatsonis C, et al. Overdiagnosis in Low-Dose Computed Tomography Screening for Lung Cancer. JAMA Int Med. 2014;174(2):269–274. [PMC free article] [PubMed]
13. Welch HG, Passow HJ. Quantifying the Benefits and Harms of Screening Mammography. JAMA Int Med. 2014;174(3):448–453. [PubMed]
14. Bleyer A, Welch HG. Effect of screening mammography on breast cancer incidence. N Engl J Med. 2013;368(7):679. [PubMed]
15. Etzioni R, Gulati R, Mallinger L, et al. Influence of Study Features and Methods on Overdiagnosis Estimates in Breast and Prostate Cancer Screening. Ann Int Med. 2013;158(11):831–838. [PMC free article] [PubMed]
16. SEER Stat Fact Sheets: Breast Cancer http://seer.cancer.gov/statfacts/html/breast.html Accessed March 23, 2015.
17. Worni M, Akushevich I, Greenup R, et al. Trends in treatment patterns and outcomes for ductal carcinoma in situ (DCIS). J Natl Cancer Inst. 2015;107(12):djv263 [PMC free article] [PubMed]
18. Surveillance E, and End Results (SEER) database of the United States Natial Cancer Institute 1991–2011. http://seer.cancer.gov/ Accessed November 23, 2015.
19. R Core Team. R: A language and environment for statistical computing. Foundation for Statistical Computing, Vienna, Austria; 2013. http://www.R-project.org/ Accessed November 23, 2015.
20. Chen HH, Duffy SW, Tabar L. A Markov chain method to estimate the tumour progression rate from preclinical to clinical phase, sensitivity and positive predictive value for mammography in breast cancer screening. Statistician. 1996;45(3):307–317.
21. Brennan ME, Turner RM, Ciatto S, et al. Ductal carcinoma in situ at core-needle biopsy: meta-analysis of underestimation and predictors of invasive breast cancer. Radiology. 2011;260(1):119–128. [PubMed]
22. Gray RJ. A Class of K-Sample Tests for Comparing the Cumulative Incidence of a Competing Risk. Ann Stat. 1988;16(3):1141–1154.
23. Nystrom L, Rutqvist LE, Wall S, et al. Breast cancer screening with mammography: overview of Swedish randomised trials. Lancet. 1993;341(8851):973–978. [PubMed]
24. Liu Y, Perez M, Schootman M, et al. Correlates of fear of cancer recurrence in women with ductal carcinoma in situ and early invasive breast cancer. Breast Cancer Res Treat. 2011;130(1):165–173. [PMC free article] [PubMed]
25. Clarke LD, Plevritis SK, Boer R, et al. A comparative review of CISNET breast models used to analyze U.S. breast cancer incidence and mortality trends. J Natl Cancer Inst Monogr. 2006;(36):96–105. [PubMed]
26. Tan SY, van Oortmarssen GJ, de Koning HJ, et al. The MISCAN-Fadia continuous tumor growth model for breast cancer. J Natl Cancer Inst Monogr. 2006;(36):56–65. [PubMed]
27. Plevritis SK, Sigal BM, Salzman P, et al. A stochastic simulation model of U.S. breast cancer mortality trends from 1975 to 2000. J Natl Cancer Inst Monogr. 2006;(36):86–95. [PubMed]
28. Berry DA, Inoue L, Shen Y, et al. Modeling the impact of treatment and screening on U.S. breast cancer mortality: a Bayesian approach. J Natl Cancer Inst Monogr. 2006;(36):30–6. [PubMed]
29. Lee S, Zelen M. A stochastic model for predicting the mortality of breast cancer. J Natl Cancer Inst Monogr. 2006;(36):79–86. [PubMed]
30. Mandelblatt J, Schechter CB, Lawrence W, et al. The SPECTRUM population model of the impact of screening and treatment on U.S. breast cancer trends from 1975 to 2000: principles and practice of the model methods. J Natl Cancer Inst Monogr. 2006;(36):47–55. [PubMed]
31. Hanin LG, Miller A, Zorin AV, et al. The University of Rochester model of breast cancer detection and survival. J Natl Cancer Inst Monogr. 2006;(36):66–78. [PubMed]
32. Fryback DG, Stout NK, Rosenberg MA, et al. The Wisconsin Breast Cancer Epidemiology Simulation Model. J Natl Cancer Inst Monogr. 2006;(36):37–47. [PubMed]
33. Warren JL, Klabunde CN, Schrag D, et al. Overview of the SEER-Medicare data - Content, research applications, and generalizability to the United States elderly population. Med Care. 2002;40(8):3–18. [PubMed]
34. Yu JB, Gross CP, Wilson LD, et al. NCI SEER Public-Use Data: Applications and Limitations in Oncology Research. Oncology (Williston Park). 2009;23(3):288–295. [PubMed]
35. Halasz LM, Sreedhara M, Chen YH, et al. Improved outcomes of breast-conserving therapy for patients with ductal carcinoma in situ. Int J Radiat Oncol Biol Phys. 2012;82(4):e581–e586. [PubMed]
36. Cossetti RJD, Tyldesley SK, Speers CH, et al. Comparison of Breast Cancer Recurrence and Outcome Patterns Between Patients Treated From 1986 to 1992 and From 2004 to 2008. J Clin Oncol. 2015;33(1):65–U106. [PubMed]
37. Allred DC, Anderson SJ, Paik S, et al. Adjuvant tamoxifen reduces subsequent breast cancer in women with estrogen receptor-positive ductal carcinoma in situ: a study based on NSABP protocol B-24. J Clin Oncol. 2012;30(12):1268–1273. [PMC free article] [PubMed]
38. Cuzick J, Sestak I, Pinder SE, et al. Effect of tamoxifen and radiotherapy in women with locally excised ductal carcinoma in situ: long-term results from the UK/ANZ DCIS trial. Lancet Oncol. 2011;12(1):21–29. [PMC free article] [PubMed]
39. Wapnir IL, Dignam JJ, Fisher B, et al. Long-term outcomes of invasive ipsilateral breast tumor recurrences after lumpectomy in NSABP B-17 and B-24 randomized clinical trials for DCIS. J Natl Cancer Inst. 2011;103(6):478–488. [PMC free article] [PubMed]
40. Sagara Y, Mallory MA, Wong S, et al. Survival Benefit of Breast Surgery for Low-Grade Ductal Carcinoma In Situ: A Population-Based Cohort Study. JAMA Surg. 2015;150(8):739–745. [PubMed]
41. Wisner DJ, Hwang ES, Chang CB, et al. Features of occult invasion in biopsy-proven DCIS at breast MRI. Breast J. 2013;19(6):650–658. [PMC free article] [PubMed]

Articles from JNCI Journal of the National Cancer Institute are provided here courtesy of Oxford University Press