Digital breast tomosynthesis for breast cancer diagnosis in women with dense breasts and additional breast cancer risk factors: A systematic review

Abstract

Introduction

Digital breast tomosynthesis (DBT) may improve sensitivity in population screening. However, evidence is currently limited on the performance of DBT in patients at a higher risk of breast cancer. This systematic review compares the clinical effectiveness and cost-effectiveness of DBT, digital mammography (DM), and ultrasound, for breast cancer detection in women with dense breasts and additional risk factors.

Methods

Medline, Embase, and Evidence-Based Medicine Reviews via OvidSP were searched to identify literature from 2010 to August 21, 2023. Selection of studies, data extraction, and quality assessment (using QUADAS-2 and CHEERS) were completed in duplicate. Findings were summarised descriptively and narratively.

Results

Twenty-six studies met pre-specified inclusion criteria. In women with breast symptoms or recalled for investigation of screen-detected findings (19 studies), DBT may be more accurate than DM. For example, in symptomatic women, the sensitivity of DBT + DM ranged from 82.8 % to 92.5 % versus 56.8 %–81.3 % for mammography (DM/synthesised images). However, most studies had a high risk of bias due to participant selection. Evidence regarding DBT in women with a personal or family history of breast cancer, for DBT versus ultrasound alone, and cost-effectiveness of DBT was limited.

Conclusions

In women with dense breasts and additional risk factors for breast cancer, evidence is limited about the accuracy of DBT compared to other imaging modalities, particularly in those with personal or family history of breast cancer. Future research in this population should consider head-to-head comparisons of imaging modalities to determine the relative effectiveness of these imaging tests.

Systematic review registration

PROSPERO registration number CRD42021236470.

Highlights statements

• •Evaluation of evidence on accuracy of tomosynthesis (vs mammography, ultrasound) in women at increased risk.
• •Focus on subgroups with risk defined by dense breasts and additional risk factors.
• •Tomosynthesis may be more accurate than mammography in those with symptoms or recalled for assessment.
• •Limited evidence on accuracy of tomosynthesis in women with personal or family history of breast cancer.

1. Introduction

Digital breast tomosynthesis (DBT), or three-dimensional (3D) mammography, is a radiographic test involving the acquisition of a series of low-dose X-rays over a range of angles, which are reconstructed to create a pseudo-3D image of the breast [1]. In practice, DBT is used with either acquired (standard 2D) digital mammography (DM), or with synthesised mammography (SM), where 2D images are constructed from the 3D acquisition to limit radiation [2].

DBT imaging can potentially overcome some disadvantages of digital mammography by enhancing visualization through overlapping breast tissue. Further, DBT may allow better detection in breast tissue with high density, overcoming some of the masking effect of density, which reduces mammography sensitivity [3]. In addition, high breast density is an independent risk factor for breast cancer [4].

While there is a growing body of research on DBT applied in general population screening [5], the evidence on the effectiveness of DBT in specific patient groups with additional risk factors for breast cancer, including those with signs or symptoms of breast cancer, or a personal or family history of breast cancer, is limited.

The objective of this systematic review was to compare the evidence on the clinical effectiveness and cost-effectiveness of DBT, digital mammography or ultrasound (US) in identifying breast cancer in women with dense breasts who had additional risk factors for breast cancer. This work was undertaken to provide evidence to inform Australia's national policy on breast imaging tests [6] and has global relevance for breast screening and diagnostic practice.

2. Methods

The reporting of this systematic review was guided by the standards of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [7] (Supplementary Table 1). This systematic review was prospectively registered in PROSPERO, registration number CRD42021236470. The authors can provide a protocol on request.

2.1. Data sources and searches

On August 21, 2023, Medline, Embase, and Evidence-Based Medicine Reviews (EBMR) databases via OvidSP were searched to identify relevant literature. To align with DBT use in clinical practice, only studies published since 2010 were included. Details of the search strategy are in Supplementary materials.

2.2. Eligibility criteria

2.2.1. Study inclusion criteria

• 1)
  Women with heterogeneously or extremely dense breasts (assessed by visual classification or automated software), and additional risk factors for breast cancer. Aligning with evidence gaps relevant to national policy in Australia [6], three subpopulations with additional specific risk factors may be considered for breast imaging referral:

  • i.
    Women with symptoms or signs of possible breast cancer. This group included those recalled from screening; subpopulations were defined as symptomatic and recalled, respectively.
  • ii.
    Women with a personal history of breast cancer (with or without symptoms), and
  • iii.
    Women with a family history of breast cancer (with or without symptoms).

Studies reporting a combination of these three populations were included.

• 2)Studies comparing DBT with either DM or US; and
• 3)Studies reporting at least one of the outcomes: cancer detection rates, incremental cancer yield, test accuracy (sensitivity, specificity, predictive values), recall rates or false-positive recall rates, characteristics of detected cancers, or cost-effectiveness.

2.3. Study selection

A single reviewer screened titles and abstracts against the inclusion criteria. For full-text screening, one reviewer assessed the relevance of each identified article. For consistency, two reviewers independently conducted a pilot screening of the first 10 % of references. Non-systematic reviews, letters, editorials, conference abstracts and technical reports, animal, in vitro and laboratory studies, and articles published in languages other than English were excluded. Reference lists of included studies and relevant systematic reviews were checked to identify any articles missed in the literature search.

2.4. Data extraction and quality appraisal

Data was extracted from each study in a pre-designed data-extraction form. The following details were extracted: author names, publication year, study design, study location, setting, length of patient follow-up, study population characteristics, description of breast density classification used, description of the intervention, description of the comparator, and the relevant outcomes (described under eligibility criteria) assessed. Data extraction was divided among four authors (SR, SB, VBN, AL), so that data from each article was extracted independently and in duplicate. Disagreements were resolved through discussion, and then arbitrated by a senior author (NH).

Quality of the included studies was assessed using a modified QUADAS-2 tool for quality assessment of diagnostic accuracy studies [8] adapted to better reflect the context of cancer screening studies [9], and the CHEERS checklist [10] for the one cost-effectiveness study. As with data extraction, the quality of each study was assessed by two reviewers independently, and discrepancies were resolved through discussion.

2.5. Evidence synthesis

Accuracy measures (sensitivity and specificity) were presented in a forest plot, where there were sufficient studies, without pooling the quantitative results. For other measures, a narrative synthesis is provided. A quantitative synthesis of results (meta-analysis) was unfeasible due to diversity in comparisons, outcomes, and reporting across studies.

3. Results

3.1. Study inclusion

Fig. 1 presents the search results as per the PRISMA reporting guidelines. Our database searches yielded 2987 articles, with 26 studies meeting eligibility. Twenty studies compared DBT with mammography (DM or SM), seven studies compared DBT with US (two also compared DBT with mammography), and one study reported cost-effectiveness outcomes [11,12].

Fig. 1.

PRISMA diagram

*Two of the seven studies reporting this comparison also compared DBT with mammography [11,12].

Of the 26 included studies, 19 were in women with dense breasts and symptoms or signs of possible breast cancer (Table 1, Supplementary Table 2). Equal proportions of these studies included women who were recalled from screening (recalled group) for further assessment, women who had breast symptoms (symptomatic group) or both (mixed group). Four studies included women with dense breasts and a personal history of breast cancer, one study had women with dense breasts and a family history of breast cancer, and two studies included a mixture of these populations. Studies differed in the eligibility criteria, breast density classification, and number of DBT views. Seventeen studies compared DBT with DM alone, and three studies had SM, or “DM or SM” as the comparator. Regarding outcomes, 15 reported diagnostic accuracy (sensitivity and specificity), eight reported recall rates, and five reported cancer detection rates. All studies comparing DBT with US (n = 7) reported diagnostic accuracy, one reported cancer detection rate and one cost-effectiveness. Eight studies did not specify whether DBT was acquired with DM or used with SM and are annotated as DBT (nfs: not further specified); however, it is acknowledged in practice that DBT is frequently completed with SM or DM.

Table 1.

Characteristics of the included studies in women with dense breasts and at high risk of breast cancer.

STUDY NAME	POPULATION	INTERVENTION	COMPARATOR	DBT VIEWS	OUTCOMES REPORTED	STUDY DESIGN, N (WOMEN WITH DENSE BREASTS)	DENSE BREAST CLASSIFICATION	RISK OF BIAS
ARSLAN 2022 [13]	Recalled	DBT (nfs)	US	2-view	Diagnostic accuracy	Retrospective cohort, 32	BI-RADS C & D	High risk of bias, unclear risk of applicability
AZZAM 2020 [11]	Mixed (recalled or symptomatic)	DBT (nfs)	DM, US	1-view	Diagnostic accuracy, Cancer detection rate	Prospective cohort, 37 (63 lesions)	BI-RADS C & D	Unclear risk of bias, Low risk of applicability
BIAN 2016 [14]	Symptomatic	DBT (nfs)	DM	2-view	Diagnostic accuracy, Recall rates, Cancer detection rate	Retrospective cohort, 631	BI-RADS C & D	High risk of bias, high risk of applicability
BLANKENBURG 2023 [15]	Family history	DBT + US	US + DM	NR	Cost-effectiveness	NR	NR	NA
CARBONARO 2016 [16]	Recalled	DBT + DM	DM	2-view	Recall rates	Prospective cohort, 142	BI-RADS C & D	Unclear risk of bias, high risk of applicability
CHAE 2016 [17]	Mixed (recalled or symptomatic)	DBT (nfs)	DM	1-view	Diagnostic accuracy	Prospective cohort, 472	BI-RADS C & D	High risk of bias, high risk of applicability
CHIKARMANE 2020 [18]	Personal history	DBT+(DM or SM)	DM	2-view	Recall rates, Cancer detection rate	Retrospective cohort, 4609 examinations	BI-RADS C & D	High risk of bias, low risk of applicability
GILBERT 2015 [19]	Recalled, Family history	DBT + DM and DBT + SM	DM	2-view	Diagnostic accuracy	Retrospective cohort, 2096	Unclear	High risk of bias, high risk of applicability
HADADI 2022 [20]	Recalled	DBT (nfs)	US	2-view	Diagnostic accuracy	Retrospective cohort, 282 breasts	BI-RADS C & D	High risk of bias, low risk of applicability
HUANG & LIN 2022 [21]	Symptomatic	DBT + DM	DM	2-view	Diagnostic accuracy	Retrospective cohort, 710	BI-RADS C & D	High risk of bias, low risk of applicability
KIM 2017 [22]	Symptomatic, Mixed (recalled or symptomatic)	DBT + DM	US + DM	2-view	Diagnostic accuracy	Retrospective cohort, 774 (214 symptomatic)	BI-RADS C & D	High risk of bias, low risk of applicability
LEE 2016 [23]	Recalled	DBT (nfs)	US	2-view	Diagnostic accuracy	Retrospective cohort, 108	BI-RADS C & D	High risk of bias, high risk of applicability
LEE 2019 [24]	Recalled, Personal history, Family history	DBT + SM	DM	2-view	Recall rates	Prospective cohort, 121 (36 % family and 21 % pers hist, 44 % both)	Unclear	High risk of bias, high risk of applicability
MALL 2018 [25]	Recalled	DBT (nfs)	DM	2-view	Recall rates	Retrospective cohort, 144a	Unclear	High risk of bias, low risk of applicability
MICHELL 2012 [26]	Recalled	DBT + DM	DM	2-view	Diagnostic accuracy	Prospective cohort, 644 lesions	The RCR dense and glandular	High risk of bias, high risk of applicability
NIA 2023 [27]	Personal history	DBT + DM	DM	2-view	Recall rates, Cancer detection rate	Retrospective surveillance cohort, 10,981	Clinician decision	High risk of bias, unclear risk of applicability
OHASHI 2018 [28]	Mixed (recalled or symptomatic)	DBT + DM	DM	1-view	Diagnostic accuracy	Retrospective cohort, 1064 breastsb	BI-RADS C & D	High risk of bias, high risk of applicability
SHIN 2015 [29]	Mixed (recalled or symptomatic)	DBT + DM	DM	1-view	Diagnostic accuracy	Retrospective cohort, 115	BI-RADS C & D	High risk of bias, high risk of applicability
SINGLA 2018 [30]	Personal history	DBT + DM	DM	1-view	Diagnostic accuracy, Recall rates	Prospective cohort, 66	BI-RADS C & D	High risk of bias, low risk of applicability
SUDHIR 2021 [12]	Symptomatic	DBT + SM	SM, DBT + US	2-view	Diagnostic accuracy, Cancer detection rate	Prospective cohort, 166 lesions	BI-RADS C & D	High risk of bias, low risk of applicability
WALDHERR 2013 [31]	Mixed (recalled or symptomatic)	DBT (nfs)	DM	1-view	Diagnostic accuracy	Retrospective cohort, 144 (66 recalled, 78 symptomatic)c	BI-RADS C & D	High risk of bias, low risk of applicability
WANG 2022 [32]	Recalled	DBT + DM	DM	NR	Diagnostic accuracy	Retrospective cohort, 460	BI-RADS C & D	High risk of bias, high risk of applicability
WHELEHAN 2021 [33]	Symptomatic	DBT + DM	DM	NR	Diagnostic accuracy	Retrospective cohort, 159	BI-RADS C & D, VAS 0–100: Q3, Q4 and Volpara: Q3, Q4	High risk of bias, low risk of applicability
YANG & ZHOU 2019 [34]	Symptomatic	DBT + DM	US	2-view	Diagnostic accuracy	Retrospective cohort, 102	BI-RADS C & D	High risk of bias, low risk of applicability
YOON 2022 [35]	Personal history	DBT + DM	DM	NR	Diagnostic accuracy, Recall ratesd	Retrospective surveillance cohorte, 314f	BI-RADS C & D	High risk of bias, low risk of applicability
YOU 2020 [36]	Symptomatic	DBT + DM and DBT + SM	DM or SM	NR	Diagnostic accuracy	Prospective cohort, 158 lesions	BI-RADS C & D	High risk of bias, low risk of applicability

BI-RADS, American College of Radiology Breast Imaging Reporting & Data System; Cat., Categories; DBT, digital breast tomosynthesis; DM, digital mammography (includes full-field digital mammography); NA, not applicable; nfs, not further specified whether DBT alone or with DM or SM; NR, not reported; RCR, The Royal College of Radiologists; SM, synthetic mammography; US, ultrasound; VAS, visual analogue scale.

^a46 % of women had dense breasts.

^b85 % had dense breasts.

^cProportion of women with dense breasts was not reported.

^dOutcomes were reported separately for ipsilateral and contralateral breast.

^eStudy outcomes were based on early post-treatment surveillance.

^f85 % of women had dense breasts.

3.2. Quality assessment, risk of bias, applicability

All the included studies presented a potential for bias (Supplementary Table 3), with the ‘patient selection’ domain consistently raising concerns about selection bias. These concerns included using non-random or non-consecutive patient recruitment, and inappropriate exclusions. Additionally, several studies had a potential bias in the ‘index test’ domain due to a lack of reader blinding to findings from comparison tests. Twelve studies did not specify the method used to classify breast density or included a population with previously pathologically proven breast cancer in a non-random sample. The quality of the cost-effectiveness study was good, but did not report the rationale for the perspective, time horizon and discount rate, and the public availability of the model (Supplementary Table 4).

3.3. Comparative diagnostic accuracy

Diagnostic accuracy estimates are reported in the text, Fig. 2, Fig. 3 and Supplementary Table 5.

Fig. 2.

Sensitivity and specificity of DBT compared with DM/SM

DBT, Digital Breast Tomosynthesis; DM, digital mammography; nfs, not further specified whether DBT alone or with DM or SM; SM, Synthetic or 2-dimensional images reconstructed from the DBT acquisition.

Mixed population refers to the inclusion of both symptomatic and recalled populations.

Bian 2016, Shin 2015 and Whelehan 2021 studies did not report 95 % confidence intervals.

You 2020 study reported two interventions: DBT + DM (You 2020a) and DBT + SM (You 2020b).

Gilbert 2015 study reported two comparators: DM (Gilbert 2015a) and SM (Gilbert 2015b).

Fig. 3.

Sensitivity and specificity of DBT compared with ultrasound

DBT, Digital Breast Tomosynthesis; DM, digital mammography; nfs, not further specified whether DBT alone or with DM or SM; SM, Synthetic or 2-dimensional images reconstructed from the DBT acquisition.

Mixed population refers to the inclusion of both symptomatic and recalled populations.

3.3.1. DBT (nfs) versus DM/SM alone

3.3.1.1. Women presenting with symptoms

One study included symptomatic women who reported a higher sensitivity as well as specificity of DBT (nfs) than DM alone (Fig. 2); however, the statistical significance of the difference was not reported and could not be calculated due to a lack of reported data [14]. Three studies reported sensitivity and specificity estimates among mixed cohorts (symptomatic and recalled) did not find statistically significant differences between DBT (nfs) and DM/SM alone [11,17,31] (Fig. 2).

3.3.2. DBT + DM/SM versus DM/SM alone

3.3.2.1. Women presenting with symptoms

In symptomatic women, the sensitivity of DBT + DM/SM ranged from 82.8 % to 92.5 % compared to 56.8% to 81.3 % for DM/SM alone (Fig. 2). Of the three studies that reported 95 % confidence intervals (CIs), two reported a statistically significant difference in sensitivity of DBT + DM/SM than DM/SM alone [21,36]. The specificity of DBT in symptomatic women ranged from 84.8 to 87.3 %, while that of DM/SM was between 63.3% and 88.6 % (Fig. 2). Two studies noted a statistically significant difference between the specificity of DBT and DM/SM [12,21]. One study [12] reported higher specificity for DBT + SM than SM alone (84.8 %, 95 % CI 75.0–91.9 % versus 63.3 %, 95 % CI 51.7−73.9 %), and another [21] reported higher specificity for DBT + DM than DM alone (83.7 %, 95 % CI 80.3−86.8 %, versus 70.6 %, 95 % CI 66.4−74.4 %).

Of the two studies on women recalled from screening, one study reported sensitivity and specificity estimates, while one reported the area under the curve (AUC). In the former, the sensitivity of DBT + DM was significantly higher (92 %, 95 % CI 88.5−94.2 %) than DM alone (59 %, 95 % CI 54.3−64.2 %) [32], but there was no difference in the specificity between the two imaging modalities (Fig. 2). In the second study, the AUC was significantly larger for DBT + DM than DM alone (AUC 96.2 %, 95 % CI 95.0−97.3 % versus 88.6 %, 95 % CI 85.8−91.4 %) [26] (Supplementary Table 5). Two studies of mixed populations (symptomatic and recalled patients) did not find a statistically significant difference between the specificity or sensitivity of DBT + DM versus DM alone [28,29].

3.3.2.2. Women with a personal history of breast cancer

Two studies of women with a personal history of breast cancer noted that specificity was significantly higher for DBT than for DM [30,35] (Supplementary Table 5). Specificity was higher in women with breast density Category C (71.4 % versus 37.1 %) or Category D (100 % versus 50 %) [30], and in women with an ipsilateral (96.1 % versus 89.3 %) or contralateral recurrence of breast tumours (97.5 % versus 93.7 %) [35]. Further, neither study reported a statistically significant difference between the sensitivities. The sensitivity of DBT ranged from 22.0% to 100 %, and the sensitivity of DM was 33.3% to 100 %.

3.3.2.3. Women with a family history of breast cancer

One study in women with a family history of breast cancer reported no difference between the sensitivity of DBT + DM/SM and DM alone (86.9−90.9 % versus 85.6 %) [19]. However, this study reported a significantly higher specificity of DBT + SM (72 %) and DBT + DM (69.7 %) than DM alone (57.0 %) [19].

3.3.3. DBT (nfs) versus ultrasound (US) alone

3.3.3.1. Women presenting with symptoms

One study of symptomatic women [34] reported a higher AUC for DBT in combination with DM than US alone (91.5 % versus 79.2 %) but did not report the statistical significance of findings (Supplementary Table 5).

In women recalled from screening, the sensitivity of DBT ranged from 85.7 % to 100 %, and that of US was 71.4% to 100 % (Fig. 3). One study [20] reported a higher sensitivity of DBT (98.2 %, 95 % CI 94.8−99.6 %) than US (80.0 %, 95 % CI 73.0−85.6 %); another [23] reported a sensitivity of 100 % for both modalities. The specificity of DBT in recalled cohorts ranged from 15.4 % to 96.0 %, with US ranging from 55 % to 100 % (Fig. 3). While one study did not find a difference between the specificities [13], two studies reported conflicting results—the specificity of DBT was lower than US in one study (15.4 %, 95 % C.I. 9.6–23.0 % versus 55 %, 95 % CI 44.8−63.0 %) [20], but higher than US in another (81.3 % 95 % CI 71.8−88.7 % versus 53.9 %, 95 % CI 43.1−64.4 %) [23]. One study included mixed cohorts (symptomatic and recalled patients); but did not report differences between sensitivity or specificity of DBT versus DM (86 % versus 97 % and 81 % versus 85 %, respectively) [11].

3.3.4. DBT + DM/SM versus ultrasound (US)

3.3.4.1. Women presenting with symptoms

The only study comparing DBT + DM with US + DM [22] found no significant difference between sensitivity or specificity in either symptomatic or mixed cohorts (symptomatic and recalled patients) (Fig. 3).

3.3.4.2. Women with a personal history of breast cancer

One study comparing DBT + SM with DBT + US [12] found no difference between the sensitivity or specificity in symptomatic cohorts (Fig. 3).

3.4. Other outcomes

Some studies reported recall rates, cancer detection rates and positive and negative predictive values (Supplementary Table 6).

3.4.1. Recall rates and cancer detection rates

Only a few studies reported the recall rates (n = 8/26) [14,16,18,24,25,27,30,35] and cancer detection rates (n = 5/26) [11,12,14,18,27]. Generally, DBT had lower recall rates than DM or SM in most of the subpopulations (Supplementary Table 6). Most studies did not report the statistical significance of the differences in recall rates, and the studies examining US as a comparator did not report recall rates.

In women with a personal history of breast cancer, two studies noted lower recall rates for DBT + DM/SM than DM/SM [30,35], with two other studies finding similar recall rates between the imaging modalities [18,27]. Several studies reported cancer detection rates; however, no notable differences were observed for any population or comparison. None of the studies reported comparative biopsy rates (Supplementary Table 6).

3.4.2. Positive and negative predictive values

Fifteen studies reported positive predictive values (PPV), negative predictive values (NPV) or overall accuracy [[11], [12], [13], [14],18,[20], [21], [22], [23],27,28,[30], [31], [32],35]. Studies comparing DBT with DM/SM consistently reported a higher PPV, NPV, and overall accuracy of DBT than DM; statistical significance was reported in only one study;however, the difference was not significant (Supplementary Table 6). Studies comparing DBT with US reported inconsistent results in predictive values, with three of the six studies reporting a lower PPV for DBT than for US [11,13,20], and three studies reporting a higher PPV for DBT [12,22,23].

3.4.3. Cost-effectiveness of DBT + DM versus DBT + US

One American study reported the cost-effectiveness of adding other breast imaging modalities to DM or DBT [15]. This study included women with a personal history or first-degree family history, or who had a 15–20 % lifetime risk of breast cancer. The cost-effectiveness analysis used a decision tree linked to a Markov model to cost from a healthcare perspective with a willingness-to-pay threshold of US$100,000. The study found that DBT + US was more cost-effective than DBT + DM, reporting an ICER of US$9415 and US$16,322, respectively, per quality-adjusted life year.

4. Discussion

This systematic review of 26 studies, compared the diagnostic accuracy of DBT with DM or ultrasound in women with dense breasts who had specific risk factors for breast cancer. Risk factors included having symptoms (or signs) of breast cancer, a personal history, or a family history of breast cancer, representing population subgroups at an increased breast cancer risk. The performance of DBT compared to standard imaging has mostly been studied in the screening of average-risk women, with minimal evidence in the included populations [6].

This review identified some evidence in women with breast cancer symptoms or those recalled for further investigation of imaging findings, suggesting DBT may perform better than DM for accuracy measures. However, most studies had a high risk of bias regarding study participant selection. Therefore, findings of this review may not be generalisable to all women with additional breast cancer risk factors. Furthermore, findings were inconsistent across studies or did not represent statistically significant differences for DBT versus DM. Given the limited findings, it is difficult to draw meaningful conclusions regarding the superiority of diagnostic accuracy of one technology over the other, specifically for the population of interest, from the available evidence due to heterogeneity within the studies. Differences included the comparator technologies and variation in the use of DM or SM technologies along with DBT, the number of DBT views used (single view, double view, or not reported), and varied study designs. In addition, some studies did not find statistically significant differences between DBT and DM (or US) for diagnostic accuracy measures. The study sample sizes were modest, highlighting the possibility that studies may not have had sufficient power to identify significant differences in key outcomes such as sensitivity or specificity. Overall, given the heterogeneity between the included studies, it may be difficult to draw conclusions regarding diagnostic performance of DBT compared with other imaging modalities for the subpopulations of interest.

The findings of this review do not provide sufficient data to make inferences regarding the cost-effectiveness of DBT. The review identified only one study that evaluated the cost-effectiveness of DBT + US and DBT + DM in women with a personal or family history of breast cancer. Further economic evaluations that compare DBT with DM or US in women with additional risk factors, are required to meaningfully inform policy decisions regarding the most cost-effective breast imaging modality in these populations.

Other systematic reviews of DBT in population screening have shown the consistent finding that DBT increases breast cancer detection (particularly in 2-year screening programs) compared to DM [5,37]. Evidence shows enhanced detection on mammography in women who have dense breasts [5,37]. As population screening was not the focus of this review but rather sub-groups of the population with dense breasts and other risk factors, it is difficult to compare our work to existing reviews. However, it is plausible that evidence from general screening may not be generalisable to groups at increased risk of breast cancer. As such, this highlights the need for primary studies of DBT, especially in those with a personal or family history of breast cancer, for whom there was little evidence.

4.1. Strengths and limitations

To the authors' knowledge, this paper is the first systematic review of studies reporting the effectiveness of DBT in women with dense breasts with additional risk factors for breast cancer, including those with signs or symptoms of breast cancer or a personal or family history of breast cancer. The synthesis included studies across all languages and countries, where papers were published in English. One limitation in studies assessing recalled cohorts arises from study participants being recalled based on DM, meaning that the DM was already ‘positive’. Therefore, paired comparisons between DBT and DM may be inappropriate, and this study design may have led to biased estimates of the specificity of DBT relative to DM in either direction (under- or overestimation). Another caveat for interpreting the findings was the high ‘prevalence’ of breast cancer (range, 9−74 %) across studies. This would likely inflate specificity and sensitivity if studies included a greater proportion of participants with cancer. A further concern was regarding the independent interpretation (blinded) of US, as it often occurred after the DBT/digital mammography findings (unblinded interpretation). Whilst it is acknowledged that in clinical practice, US is usually reported with the knowledge of DBT/mammography findings, and the studies follow clinical practice, meaning that the independent accuracy of the US cannot be estimated from current studies. Additionally, there was limited evidence in the populations with a personal history, or a family history of breast cancer. None of the included studies randomly assigned participants to imaging modalities;most studies were retrospective.

This review highlights the methodological limitations identified in the published studies and the paucity of evidence for subgroups of women with specific risk factors for breast cancer. The results may inform future research on the diagnostic performance of DBT. Future studies could use a prospective design, include consecutive participants in defined subgroups to minimise selection bias, and use larger cohorts of women with a family history or personal history of breast cancer to improve the precision of results and enable more robust comparisons. Studies comparing DBT and ultrasound in women with dense breasts could also consider independent interpretation of the tests, where feasible, as previously done in the ASTOUND trial in women with dense breasts [38].

5. Conclusion

In this systematic review of 26 studies on women with dense breasts and additional risk factors (family history, personal history, symptoms), findings suggest that DBT may be more accurate than digital mammography in some clinical settings. However, the included studies had limitations, including a high risk of bias in participant selection. The authors note there was minimal evidence of DBT in women with a personal or family history of breast cancer, and this warrants further research. Further, it is impossible to reliably compare the performance of DBT and US alone because US is usually conducted with knowledge of DBT in clinical practice. The included studies had heterogeneous (within and across the study) populations and used various modalities and comparisons. Further, very few cost-effectiveness studies were identified. Therefore, future studies should address these limitations to improve the quality of the evidence in population subgroups at increased risk of breast cancer.

Statements and declarations

Funding

This work was funded by an MRFF Preventive and Public Health (Targeted Health System Research) grant (#1199927). N. Houssami declares receiving funding from an NBCF Chair in Breast Cancer Prevention grant (#EC-21-001). The funding sources listed were not involved in the study design, data collection, analysis and interpretation, report writing, or decision to submit the article for publication.

Authors contributions

Smriti Raichand: Data curation, Visualization, Writing–Original draft; Vendula Blaya-Novakova: Methodology, Investigation, Data curation, Project administration, Writing– Review & Editing; Slavica Berber: Methodology, Investigation, Data curation, Writing– Review & Editing; Ann Livingstone: Investigation, Data curation, Writing–Review & Editing; Naomi Noguchi: Methodology, Writing−reviewing and editing, Supervision; Nehmat Houssami: Funding acquisition, Conceptualisation, Methodology, Writing−reviewing and editing, Supervision.

Data availability

The datasets generated during and/or analysed during the current study are available from the corresponding author upon reasonable request.

Ethics approval

Ethics approval to conduct this research is not required because this systematic review uses aggregate data from published literature.

CRediT authorship contribution statement

Smriti Raichand: Writing – review & editing, Writing – original draft, Methodology, Investigation, Data curation. Vendula Blaya-Novakova: Writing – review & editing, Writing – original draft, Methodology, Investigation, Data curation. Slavica Berber: Writing – review & editing, Writing – original draft, Methodology, Investigation, Data curation. Ann Livingstone: Writing – review & editing, Writing – original draft, Investigation, Data curation. Naomi Noguchi: Writing – review & editing, Methodology. Nehmat Houssami: Writing – review & editing, Supervision, Methodology, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors have no relevant financial or non-financial interests to disclose.

Appendix A. Supplementary data

The following is the Supplementary data to this article.