Introduction
With the introduction of mammographic screening, the incidence of ductal carcinoma in situ (DCIS) has increased rapidly due to the ability of mammography to identify associated microcalcifications. In the absence of calcifications that are observable by mammography, DCIS is either undetectable or preclinical, detected incidentally during biopsy of a different lesion, or rarely, detected clinically when it induces fibrosis and produces a clinical mass. In the USA, DCIS incidence rate among women older than 40 years increased from 5.6 per 100,000 women in 1990–1994 to 31.6 per 100,000 women in 2010–2014 [
1]. DCIS is regarded as “a neoplastic proliferation of cells within the ductal-lobular structures of the breast that has not penetrated the myoepithelial-basement membrane interface” [
2]. Although DCIS itself is not life-threatening, it can progress to invasive breast cancer (IBC) if left untreated [
3,
4]. The most common treatment options for DCIS are breast-conserving surgery, usually with breast irradiation, and total mastectomy [
5]. Although the detection of DCIS has increased substantially relative to detection of invasive breast cancer, its natural history remains poorly understood. In particular, there is considerable uncertainty about the rates of DCIS progression and regression.
To date, two approaches have been used to characterize DCIS natural history. In the first approach, observational studies have focused on women with a biopsy-confirmed diagnosis of DCIS who did not undergo definitive surgery [
3,
4,
6‐
12]. However, because such observational studies report on outcomes in patients who received core needle or excisional biopsies, the findings do not directly ascertain the unobserved natural history of progression. To address this caveat, a second approach uses mathematical models in conjunction with clinical data and/or data from mammography screening studies to infer latent disease dynamics [
13‐
16]. Estimates of progression risk and mean sojourn time (MST), that is, the time from preclinical DCIS to IBC, vary widely between the two approaches and even between studies of the same approach. For instance, progression risk estimates are generally lower for biopsy-treated women in observational studies (ranging from 12 to 52% [
4,
6‐
12]) than from modeling studies (ranging from 61 to 91% [
13,
14,
17,
18]).
Due to the residual uncertainty about natural history, the extent of breast cancer overdiagnosis and overtreatment are difficult to quantify. Indeed, while researchers generally accept that a fraction of screen-detected DCIS and IBC lesions would not progress to clinical disease if left untreated [
10], estimates of overdiagnosis range from less than 1% [
19] to 37% [
13]. Currently ongoing active monitoring trials for low-risk DCIS patients [
20‐
22] are expected to provide an estimate of the risk of progression from biopsy-confirmed DCIS to invasive disease. However, these trials rely on statistical inference rather than direct observation to provide insight into unobservable natural history dynamics.
Here we developed an alternative modeling approach to study DCIS natural history. Rather than relying on data from screening studies, we employed population-based models of incidence and progression in conjunction with breast cancer incidence data from the Surveillance, Epidemiology and End Results (SEER) program. By comparing two independently developed and validated population models of breast cancer, and evaluating multiple submodels for each, we explored a range of possible natural histories and projected the resulting extent of overdiagnosis.
Discussion
As the natural history of DCIS is mostly unknown, it is challenging to pinpoint a unique DCIS model. This study investigated different DCIS natural history models and selected six plausible models that could explain DCIS and IBC incidence in the USA. Unlike other attempts to model the natural history of DCIS [
18,
19,
38], our extensive work involved two-established modeling groups, several submodels, multiple birth cohorts, and a 40-year time span with the mammogram dissemination patterns observed in the USA [
39]. Our modeling work also showed that several different natural history models fit the observed trends, making any firm conclusions about the DCIS natural history based on observation data difficult.
Most submodels in our study indicated that the majority of unexcised screen-detectable preclinical DCIS lesions progress to IBC. This agrees with previous modeling studies that have estimated progression varying from 61–91% [
13,
14,
17,
18]. Notably, results from our submodels showed that younger women may have a higher proportion of progression from unexcised screen-detectable DCIS to preclinical IBC. A possible explanation is that young women tend to have a more aggressive type of DCIS which is more likely to progress to IBC [
3]. For women older than 50 years, our study found that the proportion of screen-detectable DCIS progressing to preclinical IBC to be between 36 and 99%. Furthermore, our study showed that the proportion of screen-detectable DCIS that could regress varied between 0 and 56%. These results are consistent with previous modeling studies that have estimated the proportion of DCIS regression was between 1 and 37% [
13,
14,
40].
Observational studies of women who did not receive definitive surgery after diagnosis with DCIS have found lower rates of progression invasive cancer, ranging from 12 to 54% [
4,
6‐
12]. However, these studies are not directly comparable to modeling results because they do not capture non-screen detected preclinical DCIS. Indeed, only modeling studies can capture progression rates of lesions uninterrupted by biopsy or other treatment. Lower progression rates in observational studies could be due to a number of factors. First, there is a chance of complete removal of the DCIS lesion during biopsy. This disruption of natural history will bias estimates, resulting in a lower estimated proportion of DCIS progressing to IBC and a non-observed sojourn time. In addition, mammography-detected lesions usually contain calcifications, and it remains unclear whether DCIS with and without calcification have the same natural history. Finally, inflammation of the stroma caused by the biopsy might alter the natural course of DCIS [
41].
An interesting finding from our study was that across all submodels the MSTs were relatively short, in particular when assuming DCIS regression. This agrees with previous modeling studies estimating MSTs between 0.5 months to 2.6 years under the assumption that IBC progresses through screen-detectable DCIS [
13,
17‐
19]. Similar to other modeling studies, MSTs tend to be shorter for preclinical screen-detectable DCIS progressing to preclinical IBC compared to other health states such as clinical DCIS or going into regression [
13,
19]. Although our results show a similar direction as previous studies, all DCIS modeling studies are subject to considerable residual uncertainty of the estimates. Nevertheless, short MSTs for preclinical screen-detectable DCIS progressing to clinical DCIS and IBC could guide discussions on screening intervals in proposing screening guidelines. With regard to treatment, it remains difficult to make suggestive recommendations about treatment for DCIS patients unless a clinical factor and/or molecular signatures can be identified, providing insight into which women can avoid or postpone treatment [
42].
The level of overdiagnosis of breast cancer is challenging to quantify, especially for screen-detected DCIS. Overdiagnosis estimates for DCIS varied between models D and E, with model D showing consistently lower level of DCIS overdiagnosis compared to model E and previous studies [
13,
19,
43]. This difference is probably due to model D having shorter estimated MSTs in preclinical DCIS, and high progression rates to IBC, leading to lower DCIS overdiagnosis estimates. In contrast to model D where the sojourn time is used as an input, model E estimates the mean sojourn times for every submodel. Model D used the Norwegian breast cancer screening data to estimate the DCIS model parameters, while model E was developed on US data only.
Our study showed that DCIS overdiagnosis varied from 13 to 66% when DCIS regression was allowed and 3 to 35% when assuming no DCIS regression. Yen et al. [
13] also estimated that the proportion of the screen-detected DCIS that is non-progressive (not progressing to IBC) varies from 19 to 46% in the prevalence screen and from 3 to 21% in the first subsequent screen. Another study by Seigneurin et al. estimated that 20.3% (95% CI, 3.0–38.9%) of in situ cancer was overdiagnosed, assuming that non-progressive in situ cancer remains in the preclinical phase [
44]. With regard to IBC in our study, overdiagnosis was on average between 1.3 and 2.4% regardless of the assumption of DCIS regression for the age group 30–79 years. This falls in the range of 1–10% for overdiagnosed IBC reported in the systematic review conducted by Puliti et al. based on European studies [
45]. Despite the variation in estimated DCIS overdiagnosis levels by submodel, the overall level of overdiagnosis of DCIS+IBC was not high, ranging 2.5–10.5%.
Most submodels in this study showed a reasonable fit with SEER data, indicating that different sets of parameters can match observed breast cancer rates. The ability to fit population data with varied parameters highlights the difficulty in providing definitive conclusions on the natural history of DCIS. This difficulty makes the pending results of the currently enrolling active monitoring trials LORD, LORIS, and COMET even more critical [
20‐
22]. Although these trials focus on the disease progression of screen-detected, biopsied low-risk lesions, they will significantly enhance our understanding of overall clinical management of DCIS. Further research should focus on the heterogeneity of DCIS in various age groups as younger women may tend to have a more aggressive type of DCIS. Also, future modeling work by DCIS grade, molecular subtype, and inherent factors such as family history of breast cancer will provide more specific information on MSTs and progression of the disease that will contribute to guiding women with specific features on screening and treatment.
Conclusion
Our study suggested that the majority of unexcised screen-detectable preclinical DCIS lesions progress to IBC and that the MSTs are relatively short. Furthermore, our modeling work also showed that several different natural history models fit the observed trends, making any firm conclusions about the DCIS natural history based on observation data difficult. Due to the heterogeneity of DCIS, more research is needed to understand the progression of DCIS by grades, molecular subtypes, and certain inherent factors such as family history of breast cancer.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.