Impact of the COVID-19 Pandemic on Breast Cancer Mortality in the US

Estimates From Collaborative Simulation Modeling

Oguzhan Alagoz; Kathryn P. Lowry; Allison W. Kurian; Jeanne S. Mandelblatt; Mehmet A. Ergun; Hui Huang; Sandra J. Lee; Clyde B. Schechter; Anna N. A. Tosteson; Diana L. Miglioretti; Amy Trentham-Dietz; Sarah J. Nyante; Karla Kerlikowske; Brian L. Sprague; Natasha K. Stout


J Natl Cancer Inst. 2021;113(11):1484-1494. 

In This Article


The study was determined to be other than human participant research by institutional review boards at each institution.

Model Descriptions

The study included 3 CISNET models:[15] model D (Dana-Farber Cancer Institute),[19] model G-E (Georgetown University-Albert Einstein College of Medicine),[20] and model W-H (University of Wisconsin-Madison and Harvard Medical School).[21] The models are briefly summarized below; details are included in the Supplementary Methods (available online), previous publications, and online.[15,16,18–22] CISNET breast cancer models have been previously used to inform breast cancer–screening guideline development by the US Preventive Services Task Force in 2009 and 2016.[17,18]

Briefly, the models estimate breast cancer incidence and mortality in the absence of screening and treatment and then overlay screening, diagnosis, and treatment effects to replicate US breast cancer trends over time.[15] They account for differences in 4 molecular breast cancer subtypes based on estrogen receptors (ER) and HER2. All 3 models have the ability to follow multiple birth cohorts over time. Although the models share inputs, they utilize different structures, assumptions, and approaches (Supplementary Table 1, available online). For example, in model W-H, earlier diagnosis without a change in cancer staging can reduce mortality, whereas model G-E assumes a mortality benefit only if earlier diagnosis leads to a stage shift. The range of results across the models provides a measure of uncertainty due to unobservable factors such as breast cancer natural history and incidence in the absence of screening.[16]

CISNET models use common inputs for key parameters informed by high-quality data sources (Table 1). The models successfully replicated US breast cancer incidence and mortality trends over time and were recently validated against the UK AGE trial results.[16,18,23] In addition, the models have been validated individually. For example, model W-H was cross-validated against incidence and mortality data from Wisconsin Cancer Reporting System.[21]

Key Inputs and Assumptions

We used the same model inputs describing screening utilization and performance, clinical diagnosis, and treatment dissemination and effectiveness as prior analyses[16,18] to simulate breast cancer mortality from 2020 to 2030. In projecting outcomes for future years, we assumed that current mammography performance and use as well as treatment effectiveness and use remained constant for the 10-year period.

Our scenarios representing pandemic impacts on screening, diagnostic evaluation of breast cancer symptoms, and treatment were based on current literature and expert opinion. In our base case, we assumed a 6-month duration of pandemic-related disruptions (March to September 2020) in screening, diagnosis, and adjuvant chemotherapy, given reports that mammography use recovered to nearly 100% by the end of summer 2020.[2,9,13,14]

Disruptions in screening were informed by data from Epic Health Research Network, which pooled data from 60 health-care organizations representing 10 million women from 306 hospitals in 28 states.[2] Based on their findings, we assumed that 50% of the women scheduled to undergo screening mammography missed their mammograms. Data from 2 Breast Cancer Surveillance Consortium registries (Vermont and San Francisco Bay Area) show that breast imaging volume for "evaluation of a breast problem" (ie, women presenting with symptoms for diagnostic imaging) decreased by 21% and 45%, respectively, during March-June 2020 compared with prepandemic levels in 2019 (Supplementary Figure 1, available online); therefore, we assumed that 25% of women delayed evaluation of breast cancer symptoms, resulting in delayed diagnosis and treatment. Finally, because there exist very limited data regarding COVID-19's impact on breast cancer therapy, we based our treatment-related inputs on expert opinion. We base our assumption on treatment disruptions on the results of clinical trials including TAILORx and RxPonder that showed greater chemotherapy benefit for younger than older women, and of population-based studies showing that prepandemic treatment use coincided with these trial results.[24–26] We assumed that oncologists would be more likely to recommend against cytotoxic chemotherapy for older women given their higher mortality rates from COVID-19 infection and concerns about treatment-related immunosuppression.[27–29] We did not assume reductions in other systemic treatments (eg, endocrine therapy) because they are not immunosuppressive and, for certain medications, are taken at home and thus would be unlikely to be withheld for infection concerns. We modeled no chemotherapy reduction for patients with ER-negative and/or HER2-positive disease or for patients with stage IIB or higher cancer of any subtype because we assumed that oncologists recognize the more favorable risk–benefit ratio of chemotherapy for these higher-risk patients and recommend it despite the pandemic.[30–32]

Pandemic Impact Scenarios

We simulated 6 scenarios (Table 2). Scenario 1 (no COVID-19 impact) assumed that the patterns in screening, diagnosis, and treatment between 2020 and 2030 would remain the same as in 2019. Scenario 2 represents the reduced screening scenario. Because it is not yet known how long women who missed screening during the pandemic will delay their screening, we simulated 3 different subscenarios for varying the time to return to screening (scenarios 2a-2c). In scenarios 2a-2c, women who missed their screening exams could be detected via clinical presentation and could start treatment during the pandemic period. Under scenario 2a (delayed screening), women who missed their screening exam resume screening 6 months after the missed mammogram. Under scenario 2b (skipped screening), women who missed their screening exam do not return until their next scheduled mammogram. Under scenario 2c (hybrid delayed/skipped screening), one-half of women who missed their screening mammogram resume screening 6 months after the missed mammogram, and one-half do not return until their next scheduled mammogram.

Scenario 3 represents the delayed diagnosis of symptomatic cases in which women who delayed evaluation of symptoms experienced a 6-month delay in breast cancer diagnosis relative to their expected diagnosis in the absence of the pandemic (Supplementary Table 2, available online). Scenario 4 represents reduced chemotherapy treatment. Under scenario 4, among women diagnosed with ER+ and HER2− tumors in stages I and IIA who would have received chemotherapy if not for the pandemic, 25% of those younger than age 70 years and 50% of those older than age 70 years did not receive clinically indicated adjuvant chemotherapy. Scenario 5 represents reduced screening and delayed diagnosis and therefore jointly models scenarios 2 and 3. Finally, scenario 6 represents reduced screening, delayed diagnosis, and reduced chemotherapy treatment and hence jointly models scenarios 2, 3, and 4.


Each model estimated the effect of COVID-19 disruptions on breast cancer deaths among all women aged 30 to 84 years between 2020 and 2030 in the United States. We modeled each disruption independently (eg, screening only, diagnosis only, treatment only) and combinations of disruptions (Table 2). For all analyses, results were age adjusted to the US standard population.[33,34] We calculated the cumulative number of excess breast cancer deaths from the pandemic as the difference between deaths in analysis vs usual care with no COVID impact. We also calculated the percent increase in cumulative number of additional breast cancer deaths in each scenario vs usual care. Results are reported as the median and range across the 3 models.

Sensitivity Analysis

We conducted a sensitivity analysis on the proportion of women who delayed (scenario 2a) vs skipped (scenario 2b) their mammography exams during the pandemic (ranging from 0% to 100%), who missed screening exams during the pandemic period, who experienced reduced chemotherapy, who had delays in diagnosis, and on the impact of COVID-19 on other-cause mortality. We also conducted a sensitivity analysis on the duration of the pandemic, given the pandemic situation is still evolving, in which we assumed a 12-month duration of pandemic-related disruptions instead of 6 months. In doing so, we extended the baseline assumptions on the effects of disruptions on screening, diagnosis, and treatment. We used only 1 model (model W-H) in the sensitivity analyses, which provided the median estimates by 2030 for the base case.