The Early Detection of Breast Cancer Using Liquid Biopsies: Model Estimates of the Benefits, Harms, and Costs

Simple Summary Breast cancer screening is associated with benefits, such as mortality reduction and improved quality of life, and harms, such as false-positive results, overdiagnoses, and costs. Novel screen tests could be considered to reduce the harms and increase the benefits of screening. Liquid biopsies have been proposed as a novel method for the early detection of breast cancer. However, studies show that liquid biopsies based on cell-free DNA have a low sensitivity for early-stage breast cancer. Using the microsimulation model MISCAN-Fadia, we model the benefits, harms, and costs of the early detection of breast cancer using liquid biopsies for varying levels of liquid biopsy sensitivity and specificity. We found that liquid biopsies are unlikely to be an alternative to digital mammography, given the test performance based on a CCGA substudy. When liquid biopsies are unable to detect the precursor lesion of breast cancer—ductal carcinoma in situ (DCIS)—they need to be able to detect small, early-stage tumors, with high specificity, at low costs in order to be an alternative to digital mammography. We estimated a maximum liquid biopsy price of USD 187, which is substantially lower than currently listed prices. Abstract Breast cancer screening is associated with harms, such as false-positives and overdiagnoses, and, thus, novel screen tests can be considered. Liquid biopsies have been proposed as a novel method for the early detection of cancer, but low cell-free DNA tumor fraction might pose a problem for the use in population screening. Using breast cancer microsimulation model MISCAN-Fadia, we estimated the outcomes of using liquid biopsies in breast cancer screening in women aged 50 to 74 in the United States. For varying combinations of test sensitivity and specificity, we quantify the impact of the use of liquid biopsies on the harms and benefits of screening, and we estimate the maximum liquid biopsy price for cost-effective implementation in breast cancer screening at a cost-effectiveness threshold of USD 50,000. We investigate under what conditions liquid biopsies could be a suitable alternative to digital mammography and compare these conditions to a CCGA substudy. Outcomes were compared to digital mammography screening, and include mortality reduction, overdiagnoses, quality-adjusted life-years (QALYs), and the maximum price of a liquid biopsy for cost-effective implementation. When liquid biopsies are unable to detect DCIS, a large proportion of overdiagnosed cases is prevented but overall breast cancer mortality reduction and quality of life are lower, and costs are higher compared to digital mammography screening. Liquid biopsies prices should be restricted to USD 187 per liquid biopsy depending on test performance. Overall, liquid biopsies that are unable to detect ductal carcinoma in situ (DCIS) need to be able to detect small, early-stage tumors, with high specificity, at low costs in order to be an alternative to digital mammography. Liquid biopsies might be more suitable as an addition to digital mammography than as an alternative.


Introduction
Breast cancer screening with digital mammography is associated with benefits and harms. Routine screening has been shown to reduce the risk of death by breast cancer up to 40% [1]. However, under the current strategy recommended by the United States Preventive Service Task Force (USPSTF), there are still 953 false-positive screens per 1000 women who are screened biennially from ages 50 to 74 in the U.S., meaning women are likely to receive 1 false-positive result in 13 rounds of screening during their lifetime. In addition, 19 women per 1000 women screened are over-diagnosed with breast cancer [2]. Overdiagnosis is defined as a cancer diagnosis that would otherwise not have caused symptoms or would not have caused death [3]. To overcome the harms of breast cancer screening with digital mammography, novel screening methods should be considered [4].
Liquid biopsies have been proposed as a novel method for the early detection of cancer [5]. Liquid biopsies offer a real-time, minimal-invasive method to detect cancer in the blood through circulating tumor biomarkers [6,7]. Currently, the Circulating Cell-Free Genome Atlas (CCGA) (NCT02889978) study is conducted to develop a blood-based assay based on the sequencing of circulating cell-free DNA (cfDNA) in combination with machine learning to detect multiple cancers early [8]. Two modeling studies show that multi-cancer early detection (MCED) may have a significant impact on population health, especially in cancers without a routine screening program [9,10].
However, due to the low tumor fraction of cfDNA in early-stage cancers [11][12][13][14][15], the suitability of liquid biopsies for early detection is debated [16,17]. There are currently no identified clinical tumor markers for ductal carcinoma in situ (DCIS) [18], the nonobligate precursor lesion of invasive breast cancer. In addition, CCGA and STRIVE (NCT03085888) results for early detection show an average sensitivity of 0% in stage I breast cancer [19] and show mostly detection of tumors with a poor prognosis [20].
In this study, we model the use of liquid biopsies based on cell-free DNA in routine breast cancer screening. We quantify the impact of the use of liquid biopsies with varying combinations of sensitivity and specificity on the harms and benefits of screening, and we estimate the maximum liquid biopsy price for cost-effective implementation in breast cancer screening. In addition, we investigate under what conditions liquid biopsies could be a suitable alternative to digital mammography and compare these conditions to a CCGA substudy [21].

Materials and Methods
We model the use of liquid biopsies in breast cancer screening in the United States according to three scenarios: (1) liquid biopsies are able to detect ductal carcinoma in situ (DCIS), (2) liquid biopsies are unable to detect DCIS, and (3) liquid biopsies have perfect sensitivity. We compare the harms, benefits, and cost-effectiveness of liquid biopsy screening to screening with digital mammography according to the 2016 U.S. Preventive Service Task Force Recommendation [22].

Model Overview
This study is performed using the breast cancer model of the Erasmus Medical Center, more commonly referred to as MISCAN-Fadia or model E [23], developed within CISNET. The Cancer Intervention and Surveillance Modeling Network (CISNET) aims to estimate the effect of cancer-control interventions on the incidence and risk of death from cancer in the general population [24,25]. The MISCAN-Fadia model, and all other CISNET breast cancer models for that matter, use data from the Surveillance, Epidemiology, and End Results Program and the Breast Cancer Surveillance Consortium [26].
MISCAN-Fadia, short for Microsimulation Screening Analysis-Fatal Diameter, is a discrete event-driven microsimulation model that uses a parallel universe approach. This approach mimics a randomized controlled trial by running scenarios, with and without intervention. The MISCAN-Fadia model has four principal input components: population demographics, screening, the natural history of breast cancer, and treatment. The screening module simulates screening strategies and the proportion of DCIS that is detected by screening, referred to as DCIS sensitivity (%) in this study. The natural history of breast cancer component simulates continuous tumor growth of invasive breast cancer based on several input characteristics, such as tumor biology and age. At tumor onset, tumors have a diameter of 0.01 cm. Each simulated tumor has unique sizes for screen detectability, clinical diagnosis, and fatality. The median size (diameter in cm) of a tumor at screen detectability determines the sensitivity for invasive breast cancer of a screen test. For example, a screen test that detects a tumor at a diameter of 0.01 cm has a 100% sensitivity for invasive BC. The DCIS sensitivity and median tumor size at screen detection together determine in what stage a tumor is detected. The model and the other principal input components have previously been described in more detail [23,27].
For this analysis, we adhered to the current USPSTF strategy where women are screened biennially from ages 50 to 74, with on average 13 screening rounds during their lifetime. We simulated a cohort of 10 million women born in 1970. We counted the outcomes over their lifetime, starting at age 30, assuming that their adherence to screening was 100% and all women diagnosed with breast cancer received treatment.

Liquid Biopsy Test Characteristics
Currently, digital mammography has a sensitivity of 87% [28], a specificity of 88%, and a cost of USD 149 per mammogram in the U.S. [29]. We ranged sensitivity and specificity to generate several liquid biopsies with combinations of hypothetical test characteristics, summarized in Table 1. We assume liquid biopsies are either able to detect DCIS similar to digital mammography (91% DCIS sensitivity) or not at all (0% DCIS sensitivity). CCGA and STRIVE results show an average sensitivity of 0% in stage I breast cancer [19] and there are currently no identified clinical tumor markers for DCIS [18]. We ranged the sensitivity for invasive breast cancer relative to digital mammography by varying the median tumor size for screen detection. MISCAN-Fadia simulates the natural growth of invasive breast cancer tumors and the size at which a tumor is screen-detected according to Weibull distributions. The median tumor size for screen detection refers to the median of the Weibull distribution at which 50% of tumors are screen detectable [27].
CCGA and STRIVE report a high specificity for ctDNA liquid biopsies of~99% [19]. We thus investigated the effect of increasing the specificity from 88% to 96% and 100%. We assumed that liquid biopsy sensitivity and specificity were dependent on age similar to digital mammography. To estimate the impact of a hypothetical perfect test, we investigated the effect of increasing sensitivity for invasive breast cancer to up to 100% (median tumor size of 0.61 and 0.01 cm). In total, we model 46 different combinations of test characteristics.

Health Effects
Using the MISCAN-Fadia model, we estimated the benefits: number of breast cancer deaths and life-years, and harms: false-positives and overdiagnoses. Additionally, we derived disaggregated life-years gained for several components in breast cancer screening, including screening itself, true-positive follow-up, false-positive results, clinical detection, and breast cancer care. Consequently, quality of life adjustments (Table 2) were used to calculate quality-adjusted life-years (QALYs) for each of these components. All costs were inflated to 2020 US dollars. Abbreviations: QoL, quality of life; OCD, other cause of death. * All women with a false-positive result were assumed to have follow-up imaging. In 10.6% of women with a falsepositive result, tissue biopsy was performed in addition to imaging; ** The initial phase of care consists of the first 12 months after diagnosis. The terminal phase of care covers the last 12 months of the life of a woman who has breast cancer. When a woman died of another cause of death (OCD) but had a breast cancer diagnosis during her lifetime, QoL adjustments and costs were applied. The continuous phase of care constitutes the time between the initial and terminal phase.

Costs
Costs of screening, diagnostics, and treatments were based on published estimates (Table 2) [31,32]. The model produced disaggregated results for the same components as under health effects. We converted all costs to 2020 US dollars.

Analysis
To convert the model inputs to the sensitivities of the modeled liquid biopsies, we estimated the proportion of DCIS and invasive breast cancer that were detected by screening in women aged 50 to 74. This conversion allowed us to compare our model estimates to the results to the CCGA substudy.
We performed the cost-effectiveness analysis using a federal payer perspective, a lifetime horizon, and discounted costs and health effects at 3%, as recommended by the U.S. Public Health Service [33]. Using health utilities for specific health states and costs for specific events, the total costs and QALYs can be derived to calculate cost-effectiveness [27]. The difference in costs divided by the difference in QALYs between a scenario with liquid biopsy and routine screening with digital mammography results in an ICER. The marginal difference in costs between the ICERs and a cost-effectiveness threshold of USD 50,000 was used to calculate the liquid biopsy maximum price, as USD 50,000 is a common cost-effectiveness threshold used in the U.S. [34].

Proportion of Breast Cancers Detected by Screening
A liquid biopsy with a DCIS sensitivity of 91% and a tumor size of 1.39 cm for screen detection, detected 499,859 cases of invasive breast cancer and 211,799 cases of DCIS in routine screening in women aged between 50 and 74. This resulted in a combined sensitivity for invasive BC and DCIS of 70%. In the scenario where liquid biopsies were unable to detect DCIS, the overall proportion of breast cancers detected by screening was lower. A liquid biopsy with a DCIS sensitivity of 0% and a tumor size of 1.39 cm for screen detection had a combined sensitivity of 62%. In the scenario of a perfect screen test with up to 100% combined sensitivity, considerably more breast cancers were detected by screening and fewer breast cancers were interval detected (Table 3).  Table 4 presents the modeled liquid biopsies that yielded similar or more QALYs than digital mammography screening and for which a maximum price could be estimated. Overall, our modeling results show that both the benefits and harms of screening in terms of mortality reduction and overdiagnoses increase with a higher sensitivity for invasive breast cancer.  A higher specificity positively influenced QALYs and lowered costs at all levels of sensitivity. In MISCAN-Fadia, specificity does not influence the morality-reduction or the number of overdiagnoses, but does influence the amount of false-positive screens. Falsepositive screens negatively affect quality-adjusted life years (QALYs) gained and increase the total costs of screening, thereby increasing the eventual liquid biopsy maximum price. Disaggregated costs and QALY decrements per scenario are presented in Tables A2 and A3 in Appendix A.

Outcomes of Liquid Biopsy Screening
In the scenario where liquid biopsies are unable to detect DCIS, there were fewer overdiagnoses and a slightly smaller mortality reduction compared to liquid biopsies that are able to detect DCIS. For example, a liquid biopsy that is unable to detect DCIS and has a combined sensitivity of 67%, resulted in a mortality reduction of 23.1%, and 1.9 overdiagnoses per 1000 women compared to a mortality reduction of 25.1% and 18.1 overdiagnoses per 1000 women for a liquid biopsy that is able to detect DCIS. Both liquid biopsies had the same model input for invasive breast cancer. At no DCIS detection, only a combined sensitivity of 73% yielded an increased mortality reduction (26.1%) compared to digital mammography screening. When liquid biopsies are unable to detect DCIS, sensitivity and specificity had to be high for liquid biopsies to be acceptable to digital mammography. We estimated maximum prices between USD 156 and USD 187 for liquid biopsies with a DCIS sensitivity of 0% (Table 4, Figure 1B).

Perfect Screen Test
Additionally, we estimated maximum prices for liquid biopsies approaching 100% sensitivity for invasive breast cancer. At a combined sensitivity of 90% and 87%, the estimated maximum liquid biopsy price ranged from USD 183 to USD 303 depending on the test characteristics. At 99% combined sensitivity, the total costs increased steeply due to the high invasive breast cancer sensitivity. Here, liquid biopsies were less effective and more costly than screening with digital mammography and are therefore not reported in Table 4. (Table A1).

Comparison to a CCGA Substudy
We plotted the true positive rates and false positive rates of the modeled liquid biopsies in our study on the ROC curve of the breast cancer prediction model of a CCGA substudy [21] (Figure 1A) to compare their test performance. The scenario in which liquid biopsies are unable to detect DCIS and have a specificity of 88% seems most concurrent with the predicted results by the CCGA substudy. Our modeled estimates provide threshold values at which a novel screen test would be an acceptable alternative to digital mammography, in terms of the benefits, harms, and cost-effectiveness of screening (Table 4, Figure 1A). The liquid biopsies with a maximum price in Figure 1B yield equal or greater QALYs than digital mammography. The filled labels are the liquid biopsies that also yield a greater mortality reduction than digital mammography screening.

Discussion
Due to the low tumor fraction of cfDNA in early-stage cancers [11][12][13][14][15] it is questionable whether liquid biopsies can be used in population screening [16]. In this study, we modeled the early detection of breast cancer using liquid biopsies that were either able or unable to detect DCIS, the nonobligate precursor lesion of invasive breast cancer. We estimated the benefits, harms, and costs for different combinations of liquid biopsy sensitivity and specificity. Our findings show that no DCIS detection reduces the number of overdiagnoses by~90% but also results in a slightly lower mortality reduction. We found that when liquid biopsies are unable to detect DCIS, they need to be able to detect small, early-stage tumors, with high specificity, at low costs in order to be an alternative to digital mammography. We estimated a maximum liquid biopsy price of USD 187 for liquid biopsies fulfilling these requirements. When comparing our model estimates to a CCGA substudy, liquid biopsies seem unlikely to replace digital mammography in routine breast cancer screening given their current test performance.
Our modeling results provide threshold test characteristics for novel tests for breast cancer screening and show how test characteristics affect the harms and benefits of screening. In practice, there is generally a trade-off between sensitivity and specificity. We show that higher specificity increased QoL and decreased costs because of the large number of false-positive screens that were prevented compared to digital mammography screening. For breast cancer sensitivity, MISCAN-Fadia distinguishes between sensitivity for DCIS and sensitivity for invasive breast cancer. A majority of overdiagnosed cases by digital mammography consists of DCIS [2], and, therefore, the number of overdiagnoses decreased when liquid biopsies were unable to detect DCIS but at a lower mortality reduction. In the natural history of breast cancer module of MISCAN-Fadia, pre-clinical detectable DCIS may either develop to clinical DCIS with symptoms, progress to invasive breast cancer, or regress, resulting in no breast cancer [27]. Consequently, women who develop preclinical progressive DCIS or DCIS with symptoms are identified in a later stage of breast cancer, resulting in a higher risk of mortality, lower QoL, and higher costs of breast cancer care. Liquid biopsies that are unable to detect do DCIS should thus have a high sensitivity for invasive breast cancer to be an acceptable, cost-effective alternative to digital mammography. Still, the perfect test scenario shows that even though~100% sensitivity for invasive breast cancer resulted in a mortality reduction of 70%, this was not cost-effective due to the large number of cases overdetected and the consequent quality of life reductions and costs related to breast cancer care. This emphasizes that a high sensitivity and the consequent greater detection of smaller invasive tumors is not always beneficial in breast cancer screening.
One of the benefits of liquid biopsies as a screen test is the possibility for multi cancer early detection (MCED), especially for cancers with no current screening program [9,10]. However, the use of liquid biopsies in existing screening programs raises a number of challenges. First of all, imaging will always be required after a positive liquid biopsy result for accurate diagnosis and treatment. We take this into account in our study in the diagnostic work-up. In the hypothetical perfect test scenario, liquid biopsies could detect very small tumors that may not yet be detectable by imaging. It will be challenging how women in this instance should be counseled and followed up, e.g., by offering them intense active surveillance and additional imaging until a tumor becomes visible. Second of all, liquid biopsies may not be acceptable as a replacement of the current screen test because of insufficient performance and/or cost-effectiveness. The exact costs of liquid biopsies in a routine screening setting are unclear. In our analysis, we estimate the maximum price for liquid biopsy implementation at a cost-effectiveness threshold of USD 50,000 to be USD 149 to USD 187 per liquid biopsy depending on test performance. Liquid biopsies with comprehensive gene panels, which would be required for population screening, cost USD 1500 to USD 1750 [12]. At that price, liquid biopsies are not suitable for routine testing, although liquid biopsy prices are expected to drop quickly. Analytical methods with a higher sensitivity and lower cost should be considered, such as fragment size analysis and low coverage whole genome sequencing of ctDNA [35].
Although liquid biopsies may not be suitable as a sole screen test, they might be used as an alternative or additional screen test. Liquid biopsies may result in a lower pain experience for women and increase adherence, can be used for tumor profiling and the identification of high risk tumors [20,21], and the identification of individuals with a predisposition for breast cancer [36]. In the STRIVE study, blood samples are taken within 28 days of a screening mammogram, offering the possibility to study the performance of liquid biopsies compared to digital mammography in the future.
There are several limitations to this study that should be considered. First of all, there is no data available on liquid biopsy test performance in a screening setting and, therefore, we had to vary test performances relative to current routine screening. As a consequence, our results should be considered as threshold estimates for the future implementation of liquid biopsies or other novel screening methods but no economic evaluation of an actual, available liquid biopsy. Second, to isolate optimal test performance we assumed adherence to screening and treatment to be 100% in our primary analyses, which most likely will not be attainable in reality. Third, our cost-effectiveness analysis adheres to a federal payer perspective indicating that we did not include all costs from a societal perspective, such as patient time costs. Last, we had to make numerous assumptions on liquid biopsy test characteristics, e.g., the effect of age and tumor biology on the efficacy of liquid biopsies, which might prove different in the future.
In this study, we compared our results to the current USPSTF strategy recommending biennial screening between ages 50 and 74. This guideline factors in the effect of age on breast cancer incidence and digital mammography sensitivity [22]. However, other guidelines recommend different strategies. The American Cancer Society recommends annual screening between ages 45 and 54, followed by biennial screening from age 55 onwards [37], and the American College of Radiology recommends annual screening between ages 40 and 74 [38]. We expect that the impact of liquid biopsies versus digital mammography does not vary much by screening strategy, as the major assumed difference between the two screening modalities is no sensitivity for DCIS. Although the magnitude of benefits and harms may vary slightly, we expect the relative impact on overdiagnoses, breast-cancer mortality, and cost-effectiveness to be similar at earlier starting ages and annual screening intervals.

Conclusions
This study estimates the benefits, harms, and costs of the early detection of breast cancer using liquid biopsies with varying combinations of test sensitivity and specificity.
When liquid biopsies are unable to detect DCIS, a large proportion of overdiagnosed cases is prevented but overall breast cancer mortality reduction and quality of life are lower, and costs are higher compared to digital mammography screening. In case of no DCIS detection, liquid biopsies should meet to the following requirements in order to be an alternative to digital mammography: sensitivity for detecting invasive disease needs to be high, i.e., liquid biopsies need to be able to also detect small, early-stage tumors, with high specificity, and at low costs. Our study provides threshold values for the use of novel screen tests in breast cancer screening and may guide future evaluation of liquid biopsies as an additional test in breast cancer screening.

Data Availability Statement:
The data presented in this study are available upon request to the corresponding author. The data are not publicly available due to privacy.

Conflicts of Interest:
The authors declare no conflict of interest. Table A1. Outcomes, cost-effectiveness, and threshold costs for routine screening with liquid biopsies per 1000 women screened over their lifetime.