Cancer outcomes in women without upfront surgery for ductal carcinoma in situ: observational cohort study
admin
Abstract
Objective To determine the risk of subsequent ipsilateral invasive breast cancer in women who do not receive upfront surgery on diagnosis of ductal carcinoma in situ (DCIS).
Design Observational cohort study using data abstracted directly from patients’ medical records and from a national cancer registry in patients with primary DCIS diagnosed between 2008 and 2015.
Setting Commission on Cancer accredited facilities (n=1330) in the US.
Participants 1780 women with diagnosis of primary DCIS on needle biopsy who were alive and free of invasive breast cancer at 6 months after diagnosis.
Interventions No surgery within 6 months of diagnosis.
Main outcome measures Primary outcome: ipsilateral invasive breast cancer; secondary outcome: death due to breast cancer. Subgroup analysis by risk status, based on eligibility criteria of ongoing active monitoring trials: low risk if aged ≥40 years at diagnosis of an imaging detected, nuclear grade I/II, and hormone receptor positive DCIS; high risk otherwise.
Results Median age at diagnosis was 63 years, and median follow-up was 53.3 months. Among all 1780 women, the number of ipsilateral invasive breast cancer events was 115 (6.5%) and the number of deaths from breast cancer was 29 (1.6%). The 8 year cumulative incidence of ipsilateral invasive breast cancer was 10.7% (95% confidence interval (CI) 8.4% to 12.8%). Incidence of invasive cancer differed by both disease and patient related factors, with 8 year cumulative incidences of ipsilateral invasive breast cancer ranging from 8.5% (95% CI 4.7% to 12.1%) among women at low risk (n=650) to 13.9% (10.5% to 17.2%) among those at high risk (n=833). The 8 year disease specific survival probability was 96.4% (95% CI 95.0% to 97.9%) overall and 98.1% (96.7% to 99.6%) among women at low risk.
Conclusions In a cohort of patients who did not receive initial surgery for DCIS, the 8 year cumulative incidence of invasive cancer in the same breast varied between 8% and 14%. Effective risk stratification tools and shared decision making are essential for this patient population.
Introduction
Ductal carcinoma in situ (DCIS) is a non-obligate pre-cursor lesion of invasive breast cancer that is primarily detected by screening.1 Each year, more than 50 000 women in the US receive a diagnosis of DCIS,2 accounting for 30% of mammographically detected breast cancers.3 Guideline concordant treatment for DCIS consists of a mastectomy or breast conserving surgery, often followed by radiation treatment and endocrine therapy.4 Survival after diagnosis of DCIS is excellent, with 10 year disease specific survival of 98% or higher across treatment options.5 These exceptionally favorable outcomes have raised concerns that many patients with DCIS may be over-treated by being exposed to the physical and psychological harms of therapy without deriving clinical benefit.6
These concerns have spurred efforts to de-escalate treatment through active monitoring, whereby patients undergo regular imaging surveillance, with or without endocrine therapy, and receive surgery only in the event of invasive progression. Four clinical trials (COMET, LORIS, LORD, and LORETTA) are evaluating surgery versus active monitoring in women with low risk DCIS, broadly defined as screen detected, grade I/II, and hormone receptor positive disease diagnosed in women aged 40 years and older.78910 Because trial eligible, low risk DCIS lesions are expected to grow slowly, conclusive long term results are not anticipated for many years.
In the meantime, observational cohorts of women who did not receive locoregional therapy after diagnosis of DCIS provide the best available evidence to inform decision making for the growing number of women who are interested in active monitoring. Despite known limitations, analyses from such cohorts provide a valuable complement to the clinical trials in three important ways: they can be used to assess the generalizability of future trial results in diverse populations; they provide an opportunity to potentially broaden current eligibility criteria for active monitoring; and they possess more power to assess disease specific survival.
We conducted in-depth abstraction of original source documentation for outcomes in a US based cohort of women with a diagnosis of primary DCIS who did not receive upfront surgery. We did descriptive analyses to estimate the risk of subsequent invasive cancer and to identify factors associated with this risk.
Methods
Data source
The data for this study were collected through the 2018 Commission on Cancer Special Study on DCIS, in collaboration with the American College of Surgeons. Through the Commission on Cancer Special Study mechanism, detailed longitudinal patient data were collected from original source documentation at each site in a stratified random sample of patients with a diagnosis of primary DCIS.
The sampling frame for the special study was based on the 2015 NCDB Participant User File.11 The main eligibility criteria were female sex, age 18 years or older, and a pathology confirmed diagnosis of DCIS without invasive cancer between 1 January 2008 and 31 December 2015 (supplementary table A). Eligible patients were divided into two groups: those who had received upfront surgery, either with or without radiation, within six months of diagnosis, and those who did not (supplementary table B). A treatment stratified random sample of up to 20 eligible patients from each of the 1330 participating Commission on Cancer facilities was identified; women without upfront surgery were oversampled, accounting for up to 10 selected patients per facility.
Between 16 July 2018 and 26 October 2018, cancer registrars at each participating site abstracted in-depth data about diagnosis, treatment, subsequent breast cancer events (DCIS, invasive breast cancer, lymph node cancer, distant breast cancer metastasis), and survival (including cause of death) for up to 10 years after diagnosis or date of abstraction, whichever came first. Data abstraction was based on primary chart review at treating facilities; when necessary, registrars requested the relevant information from outside facilities.
Study population
On the basis of the records abstracted through the Commission on Cancer Special Study on DCIS, we made the following exclusions for our study: women who received any surgery or radiation treatment within six months of diagnosis; women who died or experienced a breast cancer event within six months of diagnosis; women with multiple primaries; and women for whom registrars were not able to assess whether a subsequent breast event took place during follow-up (fig 1).
Study cohort. NCDB=National Cancer Database
To assess the quality of data abstraction, we considered women in the Commission on Cancer Special Study who had received breast conserving surgery, with or without radiation treatment, within six months of diagnosis (fig 1). For this cohort, we included only women who had pure DCIS on surgical pathology.
Our choice of using the six month mark to define the study cohort was motivated by the distribution of the time to first surgery among all women in the special study who received breast conserving surgery within one year of diagnosis. In this cohort, 95% of women had surgery in the first four months, identifying six months as a conservative cut-off point. We did sensitivity analyses around this choice (see Statistical analysis).
Primary and secondary outcomes
The primary outcome was subsequent diagnosis of ipsilateral invasive breast cancer. The secondary outcome was death from breast cancer. Additional outcomes were death from causes unrelated to breast cancer (other cause death) and time to and type of first ipsilateral surgery after the six month time point (where applicable). For consistency, the following events were also included in the definition of an ipsilateral invasive breast cancer: ipsilateral lymph node metastasis in the absence of a preceding ipsilateral invasive breast cancer; distant breast cancer metastasis in the absence of a preceding invasive breast cancer or lymph node metastasis; and death from breast cancer in the absence of a preceding invasive breast cancer, lymph node metastasis, or distant breast cancer metastasis.
Covariates
Baseline covariates at the time of diagnosis were either derived from the NCDB Participant User File11 or collected through the special study (supplementary table C). Hormone receptor status was coded as positive if the tumor was estrogen and/or progesterone receptor positive. Receipt of endocrine therapy was coded as a time dependent binary variable, characterized by start and end date of the first course. Because duration of endocrine therapy was originally abstracted as a categorical variable (5 years, and unknown duration), we combined the start time of therapy together with the midpoint of the duration category (6, 18, 42, 90, and 60 months) to construct a binary time dependent covariate. We assessed robustness to this midpoint rule through sensitivity analyses (see Statistical analysis).
Statistical analysis
We estimated the cumulative incidence of ipsilateral invasive breast cancer and the disease specific survival probability by using Kaplan-Meier estimators, including pointwise 95% confidence intervals. Women were censored at time of data abstraction, loss to follow-up, or death (other cause death for disease specific survival), whichever came first. We reported cumulative incidence and disease specific survival probability estimates up to eight years from diagnosis, after which the number of women at risk for the primary outcome dropped below 10% of the initial study population.12 We reported point estimates and 95% confidence intervals at two, five, and eight years after diagnosis.
We assessed the cumulative incidence of other cause death in a competing risk analysis,13 accounting for the competing risk of death from breast cancer. Women were censored at time of data abstraction or loss to follow-up, whichever came first, and estimates were reported until eight years from diagnosis. To place the competing risk analysis in the context of the general population, we used published age and cohort specific rates of non-breast cancer death among US women to estimate the cumulative incidence of other cause death in a cohort with the same age distribution as our study cohort.14
In a subgroup analysis, we repeated the ipsilateral invasive breast cancer, disease specific survival, and other cause mortality analyses by stratifying participants into risk groups. Risk categorization was based on the inclusion criteria of ongoing active monitoring trials,78910 and we classified women as being at low risk if aged ≥40 years at the time of diagnosis with an imaging detected, nuclear grade I/II, and hormone receptor positive DCIS. Women who did not meet all four criteria were classified as being at high risk. For ipsilateral invasive breast cancer and disease specific survival, we used the log-rank test to compare the time-to-event distributions.
In a sensitivity analysis, we assessed the robustness of our findings to the choice of a six month surgery-free period for cohort inclusion. More precisely, we estimated the cumulative incidence of ipsilateral invasive breast cancer, overall and by risk status, among women who did not have surgery within nine and 12 months of diagnosis, respectively.
We used univariate Cox models to describe the associations between the risk of ipsilateral invasive breast cancer and age at diagnosis (continuous), race (white, black, other), mode of detection (imaging, symptomatic), nuclear grade (grade I/II, grade III), hormone receptor status (positive, negative), and endocrine therapy (binary, time dependent). We reported the strength of association as a hazard ratio with 95% confidence interval. To determine an appropriate functional shape of the age dependence, we used the Akaike information criterion. For baseline covariates, we verified the proportional hazards assumption by using Schoenfeld residuals.15 To handle missing baseline covariates, we did multiple imputation by chained equations using the mice package in R16; the final estimates were pooled across 20 datasets to account for variability due to imputation. For endocrine therapy, we applied Monte Carlo simulations to assess the robustness of the results to the midpoint rule (see Covariates) used for converting categorical duration of endocrine therapy to continuous time intervals (supplementary methods). All tests were two sided, with α=0.05. We used SAS software (version 9.4) and R Statistical Software (v4.2.2) for all analyses.
Data abstraction quality
Accurate estimation of risk of ipsilateral invasive breast cancer depends on capturing all subsequent breast cancer events during data abstraction. Given the scarcity of comparable studies in women without upfront surgery, we benchmarked the quality of abstraction in women who did have upfront surgery. To do this, we first estimated the cumulative incidence of ipsilateral breast cancer events (including in situ tumors) among women in the Commission on Cancer Special Study who underwent breast conserving surgery within six months of diagnosis. We then compared these estimates with those from contemporary clinical trials to assess the potential for under-reporting of cancer events.
As an additional quality control measure, we asked registrars to report their level of confidence (on a scale from 0 to 100%) in having had access to all relevant medical records related to subsequent breast events, and we repeated the primary outcome analysis in the subset of women for whom the registrar reported ≥95% confidence of having had access to all relevant medical records.
Patient and public involvement
Patient advocates (DC and colleagues) were involved in all aspects of the study, including providing input on study design, development of outcomes and covariates, and providing editorial assistance. They will also participate in dissemination of study results on publication.
Results
Study population
In the Commission on Cancer Special Study on DCIS cohort, a total of 1780 women from 751 sites did not receive surgical or radiation treatment within six months of diagnosis (table 1). Median age at diagnosis was 63 years, and median follow-up after diagnosis was 53.3 months. Most women were of white (n=1240; 70%) or black (n=383; 22%) race, and 588 (33%) had one or more comorbidities at the time of diagnosis. The primary mode of DCIS detection was imaging (n=1527; 86%). Among 1533 women with a tumor of known grade, 898 (59%) had a tumor of low to intermediate grade; among 1530 women with a tumor of known hormone receptor status, 1342 (88%) had a receptor positive tumor.
Characteristics of patients, tumors, and treatment. Values are numbers (percentages) unless stated otherwise
A total of 675 (38%) women underwent at least one ipsilateral surgical procedure after the initial six month period. Among the 428 lumpectomies and 247 mastectomies received as first ipsilateral surgery after diagnosis, most (n=422; 63%) occurred between six and 12 months (supplementary figure A); 437 (25%) women received endocrine therapy during follow-up.
Ipsilateral breast cancer
Among all 1780 women, the number of ipsilateral invasive breast cancer events was 115 (6.5%), and the median time from diagnosis to ipsilateral invasive breast cancer event was 33.7 months (supplementary table D). At eight years after diagnosis, the cumulative incidence of ipsilateral invasive breast cancer was 10.7% (95% confidence interval (CI) 8.4% to 12.8%) (fig 2 (top); table 2). At eight years after diagnosis, the cumulative incidence of ipsilateral invasive breast cancer among the 650 women with low risk DCIS was 8.5% (95% CI 4.7% to 12.1%), compared with 13.9% (10.5% to 17.2%) among the 833 women with high risk DCIS (log-rank test Pfig 2 (bottom); table 2).
Cumulative incidence of ipsilateral invasive breast cancer (iIBC) in women with diagnosis of ductal carcinoma in situ (DCIS) who did not receive surgery within 6 months of diagnosis. Top: Kaplan-Meier cumulative incidence (with 95% confidence band) of iIBC among all women in cohort. Only women who were alive and free of invasive breast cancer at 6 month mark (vertical dashed line) were included in analysis; numbers at risk and cumulative number of events are tabulated. Bottom: Kaplan-Meier cumulative incidence (with 95% confidence bands) of iIBC among women with diagnosis of low risk (n=650) and high risk (n=833) DCIS. Only women who were alive and free of invasive breast cancer at 6 month mark (vertical dashed line) were included in analysis; P value calculated using log-rank test; numbers at risk and cumulative number of events are tabulated
Cumulative incidence of ipsilateral invasive breast cancer (iIBC) and disease specific survival in women with diagnosis of ductal carcinoma in situ who did not receive upfront surgery within 6 months of diagnosis. Values are percentages (95% confidence intervals)
In univariate analyses (supplementary table E), the risk of ipsilateral invasive breast cancer was associated with symptomatic detection of disease, negative hormone receptor status, high nuclear grade, and the presence of comedonecrosis; its relation with the age at diagnosis was non-linear (supplementary figure B). Taking account of uncertainty around its exact duration, endocrine therapy did not show a consistent association with the risk of ipsilateral invasive breast cancer.
Disease specific survival and other cause mortality
Among the 1780 women in the study, 29 (1.6%) died from breast cancer. At eight years after diagnosis, disease specific survival was 96.4% (95% CI 95.0% to 97.9%) (fig 3 (top); table 2). Among the 650 women with low risk DCIS, eight year disease specific survival was 98.1% (95% CI 96.7% to 99.6%), compared with 94.4% (91.6% to 97.2%) among the 833 women with high risk DCIS (fig 3 (bottom); table 2).
Disease specific survival in women with diagnosis of ductal carcinoma in situ (DCIS) who did not receive surgery within 6 months of diagnosis. Top: Kaplan-Meier disease specific survival probability (with 95% confidence band) among all women in cohort. Only women who were alive and free of invasive breast cancer at 6 month mark (vertical dashed line) were included in analysis; numbers at risk and cumulative number of events are tabulated. Bottom: Kaplan-Meier disease specific survival probability (with 95% confidence band) among women with diagnosis of low risk (n=650) and high risk (n=833) DCIS. Only women who were alive and free of invasive breast cancer at 6 month mark (vertical dashed line) were included in analysis; P value calculated using log-rank test; numbers at risk and cumulative number of events are tabulated
In a competing risk analysis, the eight year cumulative incidence of death from a cause unrelated to breast cancer in the cohort was 27.8%, compared with an expected incidence of other cause death of 19.0% in the age adjusted general population (supplementary figure C).
Sensitivity analysis
To assess the sensitivity of the estimates of cumulative incidence of ipsilateral invasive breast cancer to our choice of a six month surgery-free period for cohort eligibility, we repeated the analyses among women who did not have initial surgery within nine months (n=1422) and 12 months (n=1291) (supplementary table F). The corresponding estimates of the eight year cumulative incidence of ipsilateral invasive breast cancer were 11.6% (95% CI 9.0% to 14.2%) and 12.1% (9.2% to 14.8%) for nine and 12 months, respectively, compared with 10.7% (8.4% to 12.8%) for six months.
Data abstraction quality
To assess the quality of cancer event abstraction in the Commission on Cancer Special Study, we estimated the risk of ipsilateral cancer among women who had received breast conserving surgery within six months of diagnosis. In this patient group (n=11 783), the five year cumulative incidence was 1.4% (95% CI 1.1% to 1.6%) for ipsilateral invasive breast cancer alone and 3.4% (3.1% to 3.8%) for any ipsilateral cancer, including DCIS (supplementary table G).
The registrars’ self-reported confidence in having had access to relevant medical records was high (median 95%). When we restricted the primary analysis to the 974 women for whom the registrar reported a confidence of 95% or higher for having had access to all relevant medical records, the eight year cumulative incidence of ipsilateral invasive breast cancer was 12.1% (95% CI 8.9% to 15.3%) (supplementary table F).
Discussion
Surgery remains universally recommended for patients with a new diagnosis of DCIS to prevent future invasive cancer. As a result, the natural history of DCIS in the absence of excision is largely unknown. Here, we report on the outcomes of 1780 women who did not receive upfront surgery after primary diagnosis of DCIS. This is a highly unique cohort given that 98% of women given a diagnosis of DCIS in the US undergo immediate surgery5 and that most cancer registries do not capture long term cancer recurrence and progression data. The eight year cumulative incidence of diagnosis with ipsilateral invasive breast cancer was 10.7% overall and 8.5% among women who met the main eligibility criteria for ongoing active monitoring trials (low risk DCIS). For context, the 10 year cumulative incidence of ipsilateral invasive cancer in women with atypia—a benign diagnosis for which the treatment recommendation is close monitoring—has been reported to be between 6% and 13%.171819
Comparison with other studies
To date, the only randomized controlled trial comparing women with and without upfront surgery is the ongoing US based COMET (Comparing an Operation to Monitoring, With or Without Endocrine Therapy for Low-Risk DCIS; AFT-25) study.7 In COMET, 957 women were randomized to either active monitoring (follow-up every six months with breast imaging and physical examination) or guideline concordant care. In the two year interim analysis of COMET, active monitoring was non-inferior to guideline concordant care with respect to the cumulative incidence of ipsilateral invasive breast cancer (active monitoring 4.2%; guideline concordant care 5.9%).20 In a secondary analysis, and on the basis of patient reported outcomes, the COMET study found that the lived experiences of women in both trial arms were similar.21 These results are an important first step that supports the short term safety and acceptability of active monitoring among patients at low risk. However, long term follow-up data are needed to derive definitive conclusions about the viability of omitting upfront surgery. In the meantime, observational studies such as this one provide the best available evidence.
Two contemporary observational studies have produced comparable long term estimates of the cumulative incidence of ipsilateral invasive breast cancer in women without upfront surgery after DCIS. A Surveillance, Epidemiology, and End Results (SEER) based study of 1286 women in the US estimated the five year cumulative incidence to be 7.3% (95% CI 5.8 to 9.1%),22 similar to our corresponding estimate of 7.2% (5.8% to 8.7%). By contrast, a UK based study of 311 women reported a substantially higher five year incidence of 16% (95% CI 12% to 22%)23; although the study’s small size precludes definitive conclusions, its higher incidence of ipsilateral invasive breast cancer could be due to lower use of endocrine therapy and higher prevalence of occult invasion missed on biopsy.2425
A recent National Cancer Database based study focused on all women given a diagnosis of low risk DCIS between 2004 and 2017.26 The study reported an increasing trend in the proportion of women without upfront surgery over time, reaching 4.2% in 2017. Interestingly, among patients at low risk who received a diagnosis below the age of 50 years, omission of upfront surgery was not associated with adverse overall survival, suggesting that potential differences in invasive progression between women with and without upfront surgery may not translate into survival differences for these patients.
Strengths and limitations of study
Notable strengths of our study pertain to the size of the cohort and the granularity of the collected data. With 1780 women who did not receive upfront surgery, we assembled the largest such cohort to date. By augmenting cancer registry data from the National Cancer Database with de novo abstraction of longitudinal medical records, we were able to assess the risk of ipsilateral invasive breast cancer and disease specific survival, both outcomes that are not captured in the National Cancer Database itself.
Outcomes among the 11 783 women in the Commission on Cancer Special Study on DCIS who did receive upfront surgery provide a benchmark for the data abstraction process. In this cohort, the five year cumulative incidence of ipsilateral recurrence (DCIS or invasive cancer) was 3.4%, similar to estimates from the contemporary IBIS-II (2.5-3.0%) and NSABP B-35 (3.7%) trials.2728 Together with the registrars’ high confidence in having had access to relevant medical records, these findings suggest a limited risk of systematic under-reporting of subsequent breast events in our study.
This study has limitations. Firstly, data abstraction by local cancer registrars at participating sites may have introduced variability in the overall quality of records. We incorporated a series of measures to minimize these risks by providing registrars with a detailed instruction manual, doing a comprehensive pilot study at 32 participating sites, offering a virtual helpdesk, and answering questions during weekly webinars. Secondly, the relatively high prevalence of comorbid conditions and deaths from other causes suggests that the study cohort may not fully represent the generally healthy DCIS patient population,29 and thus the study is susceptible to confounding by indication. We note, however, that comorbidity status was not associated with the risk of invasive progression in univariate analyses. Therefore, we expect the estimates to be broadly generalizable, particularly when considering patients’ age and tumor pathology. Thirdly, owing to the observational nature of the study, cohort allocation was determined at six months after diagnosis, and only events occurring thereafter were included in the analyses. However, because DCIS is a slowly growing disease,130 we expect the immortal time bias to be limited.31 Fourthly, the women in our cohort may differ from those who would participate in a prospective active monitoring protocol. Given that immediate surgery remains the standard of care, the decision to forego upfront surgery in this cohort may have been influenced by factors such as overall health status and the perceived benefit of surgery relative to existing comorbidities and individual healthcare goals. Finally, the women in our cohort were unlikely to have been monitored as closely as would be required in an active surveillance protocol.7
Meaning of study
The variability in outcomes between women with low risk and high risk DCIS emphasizes the need for accurate risk stratification tools in women with a diagnosis of DCIS.3233 For patients at an elevated risk of invasive progression, offering them timely surgery and adjuvant therapy is important. At the other end of the risk spectrum, select patients may benefit from de-escalation of treatment.20
Our definition of low risk DCIS mirrors the main eligibility criteria of ongoing active monitoring trials78910—that is, screen detected, non-high grade, hormone receptor positive DCIS diagnosed in women aged 40 years and older. In univariate analyses, these same features were associated with the risk of invasive progression, thus providing additional rationale for this definition of low risk DCIS. In our study, the cumulative incidence of ipsilateral invasive breast cancer in women with low risk DCIS was similar to corresponding estimates in patients with atypia,171819 which highlights the importance of ongoing efforts to assess the viability of treatment de-escalation.78910
As with all interventions for asymptomatic pre-cancers, the benefit-harm trade-offs for an individual patient must be considered in the context of the patient’s ability to accrue the long term oncologic benefits from treatment. Previous studies show that patients report long term pain and neuropathy after breast conserving surgery, especially when it is combined with radiation.343536 Generally, however, breast conserving surgery has few morbidities and is well tolerated by patients. In addition, both the practical and psychosocial burdens of ongoing surveillance must be considered. Although the optimal surveillance strategy in patients without upfront surgery is unknown, it may have different burdens from those in patients undergoing initial surgery, with the associated potential for worry and anxiety.
Although this descriptive study alone cannot change practice, the data presented here provide two important perspectives for patients and their healthcare providers. Firstly, DCIS should not be categorically regarded as “cancer,” with the attendant fears of cancer dissemination and death.37 Especially in those patients with low risk disease who may have competing health risks requiring a delay to surgery, an active monitoring approach is not likely to pose undue oncologic dangers.20 Secondly, the eight year risk of a cancer event following a diagnosis of DCIS does not exceed 11% and is comparable to the magnitude of risk conferred by other high risk histologies such as atypia. This level of risk should be part of the conversation as patients consider their treatment options and underscores the central role of patient education and shared decision making, particularly in the setting of low risk DCIS. The findings from ongoing prospective clinical trials that collect patient reported outcomes will inform how any perceived benefit to the patient or population more broadly should be balanced against the side effects of treatment and the resource utilization expended for surgery, potential reconstruction, and radiation in this favorable risk population.21
Conclusions
In women who did not receive initial surgery on diagnosis of DCIS, the eight year cumulative incidence of invasive cancer in the same breast was 10.7% overall and 8.5% in women with low risk DCIS. Effective risk stratification tools and shared decision making are essential in this patient population.
