Performance of Diagnostic Breast Imaging in Symptomatic Pregnant and Lactating Patients: Systematic Review and Meta-Analysis

Abstract

Purpose

To perform a systematic review of the literature and meta-analysis to summarize the diagnostic performance of breast imaging modalities for cancer detection in pregnant and lactating patients.

Materials and Methods

A systematic review of the literature in PubMed, Scopus, Web of Science, and Cochrane Library databases published up until March 3, 2023, was conducted. Included studies evaluated patients of any age who underwent breast imaging during pregnancy or lactation. The primary outcome of this review was sensitivity and specificity of each imaging modality. Meta-analysis was performed using a bivariate modeling approach, and summary receiver operating characteristic (ROC) analysis was used to generate a summary area under the ROC curve (AUC).

Results

Twenty-five studies met the eligibility criteria and included 1681 female patients (mean age, 33 years; range, 18–49 years). For US, seven of 24 studies had complete data yielding an AUC of 0.90 (95% CI: 0.85, 0.93), a sensitivity of 81% (95% CI: 56, 94), and a specificity of 85% (95% CI: 71, 92). For mammography, three of 21 studies had complete data yielding an AUC of 0.93 (95% CI: 0.75, 0.97), a sensitivity of 72% (95% CI: 47, 88), and a specificity of 93% (95% CI: 86, 97). For MRI, two of eight studies had complete data yielding an AUC of 95% (95% CI: 59, 96), a sensitivity of 91% (95% CI: 56, 99), and a specificity of 88% (95% CI: 48, 98).

Conclusion

US, mammography, and breast MRI showed high diagnostic performance for detection of pregnancy-associated breast cancer in symptomatic pregnant or lactating patients.

Keywords: Meta-Analysis, Breast, Oncology, Pregnancy, Mammography, MR-Dynamic Contrast Enhanced, Ultrasound

Supplemental material is available for this article.

© RSNA, 2025

Summary

US, mammography, and breast MRI showed high diagnostic performance for pregnancy-associated breast cancer detection in symptomatic patients.

Key Points

• ■ This systematic review identified 25 studies involving 1681 female patients who were pregnant or lactating and evaluated with US, mammography, or breast MRI for breast cancer detection.

• ■ Meta-analysis of the studies with complete data for US, mammography, and breast MRI demonstrated areas under the receiver operating characteristic curve of 0.90 (95% CI: 0.85, 0.93), 0.93 (95% CI: 0.75, 0.97), and 0.95 (95% CI: 0.59, 0.96), respectively.

• ■ Larger, prospective studies that include appropriate follow-up for benign findings are needed to validate these results.

Introduction

Breast cancer is a major international public health concern, despite advancements in early detection and improved treatments (1). Pregnancy-associated breast cancer (PABC) is defined as breast cancer diagnosed during or within 12 months of pregnancy and has an estimated incidence of 26.8 per 100 000 pregnancies (2). As more individuals delay childbearing until later in life, it is expected that this number will grow.

Breast concerns during pregnancy and lactation are a common scenario that radiologists encounter in clinical practice and may be challenging to evaluate (3). Physiologic changes in the breast during pregnancy and lactation alter the typical imaging appearances of normal and benign processes and may mask underlying malignancy. Hormonal changes induce ductal and lobular development as well as milk secretion, causing increased mammographic density, decreased fat distribution, increased vascularity, and increased metabolic rate. Normal physiologic changes in the breast during pregnancy and lactation can also make clinical breast examination more challenging to perform.

The American College of Radiology has developed appropriateness criteria for breast imaging evaluation of pregnant and lactating patients (4). During pregnancy, US is the first-line modality and can be followed by mammography if the US findings are suspicious or show no abnormality. US is also the initial modality for lactating patients younger than 30 years. For lactating patients aged 30 years and older, mammography is performed first followed by US. Dynamic contrast-enhanced breast MRI is not recommended during pregnancy but can be performed postpartum and during lactation for diagnostic evaluation of patients with newly diagnosed breast cancer and for high-risk screening.

Although society guidelines have been developed to assist clinicians in choosing the appropriate imaging evaluation for pregnant or lactating patients presenting with breast concerns, large studies investigating the diagnostic performance of US, mammography, and MRI in those two specific patient populations are lacking. The purpose of this study was to perform a systematic review of the literature and meta-analysis to summarize the diagnostic performance of breast imaging modalities for cancer detection in pregnant and lactating patients.

Materials and Methods

Details of the review protocol are registered with the International Prospective Register of Systematic Reviews (CRD42023398422) (5) and can be accessed at https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=398422. The review was completed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 updated guidelines (6) and with the Preferred Reporting Items for Systematic Review and Meta-Analysis of Diagnostic Test Accuracy Studies (7).

Eligibility Criteria

The literature search was designed and implemented in consultation with an information specialist using a PICO (problem and population, intervention, comparison, outcome) style question for diagnostic test accuracy (8): “In patients who are pregnant or lactating, what is the diagnostic performance of different breast imaging modalities for breast cancer detection?” Inclusion criteria included a study population of individuals of any age who are pregnant or lactating and evaluated using mammography, US, breast MRI, digital breast tomosynthesis, contrast-enhanced mammography, or molecular breast imaging. No publication date restrictions were used. Studies were limited to English language publications. Duplicated studies and society guidelines were removed before review. Exclusion criteria during review included case reports that did not include a systematic analysis of multiple patients, reports that did not present primary data (reviews, book chapters, society guidelines), studies that excluded pregnant and/or lactating patients, reports that described only benign or metastatic findings, and studies using nonclinical or research-based imaging protocols.

Information Sources

The following databases were queried from their inception to March 3, 2023: PubMed, Scopus, Web of Science, and Cochrane Library. Search terms used for each database are provided in Appendix S1. Reference lists from the eligible articles were reviewed to identify any additional studies. Reference lists from relevant narrative review articles were also screened. Study authors were not contacted to identify additional studies.

Study Selection

One author (B.W.W., with 4 years of postgraduate experience) performed the initial assessment for inclusion by screening the titles and abstracts and used the full-text articles to confirm eligibility. The relevant articles were then independently reviewed by two authors (B.W.W. and A.M.F., with 25 years of postgraduate experience). Disagreements on study inclusion were remediated through discussion by the two authors. No automation tools were used in the selection process. Studies with extremely small sample sizes (ie, four or fewer patients) and those that did not report data in a way that could be abstracted were excluded from the meta-analysis but were included in the systematic review.

Data Extraction

One author (B.W.W.) performed the data collection using a structured electronic data collection form. A second author (A.M.F.) independently reviewed the accuracy and completeness of the extracted data. Discrepancies in data collected were remediated by discussion by the two authors. No automation tools were used in the data collection process.

The following variables were extracted: counts of total number of patients, true positives, false positives, true negatives, and false negatives; study design and country of origin; age range and mean age of patient sample; whether the study included pregnant, lactating, or pregnant and lactating patients; modalities assessed; and number of institutions contributing data (Appendix S1).

Risk of Bias and Applicability

Reports included in the review were analyzed using the Quality Assessment of Diagnostic Accuracy Studies-2 tool (9). This tool includes four domains for assessing risk of bias (1: patient selection; 2: index test; 3: reference standard; and 4: flow and timing) and three domains for assessing applicability concerns (1: patient selection; 2: index test; 3: reference standard). Individual domains were assessed as low, unclear, or high for risk of bias and applicability concerns, respectively. All studies were included in the analysis for risk of bias and applicability. Two authors (B.W.W. and A.M.F.) independently assessed each study. Discrepancies were remediated by discussion between the authors. Studies that did not include information about the imaging modalities in their Methods sections were assessed as unclear for applicability. Studies that did not explicitly state whether radiologists were blinded were assessed as unclear for bias.

Diagnostic Accuracy Measures

Reference standards reported by the studies were histopathology and clinical and/or imaging follow-up. True positives, false positives, true negatives, and false negatives were recorded for each study with analysis on a per-lesion basis. For studies reporting Breast Imaging Reporting and Data System (BI-RADS) assessments, BI-RADS 1, 2, and 3 were considered negative index tests. BI-RADS 4, 5, and 6 were considered positive index tests. As descriptive analysis, sensitivity and/or specificity of each primary study, along with 95% CIs, were calculated, provided the study had corresponding data. Sensitivity was calculated as the number of true positives divided by the sum of the true positives and false negatives. Specificity was calculated as the number of true negatives divided by the sum of the true negatives and false positives. Positive likelihood ratio was calculated as sensitivity divided by (1 − specificity). Negative likelihood ratio was calculated as (1 − sensitivity) divided by specificity. Diagnostic odds ratio was calculated as positive likelihood ratio divided by negative likelihood ratio. The primary outcome of this systematic review and meta-analysis was the sensitivity of each imaging modality and, when able to be assessed, the specificity.

Synthesis of Results

Studies were grouped for analysis according to the breast imaging modality or modalities investigated in each article. Forest plots were generated of study-specific sensitivities and specificities for US, mammography, and breast MRI. For each modality, combined sensitivity (and 95% CI) was calculated as the number of true positives divided by the sum of the true positives and false negatives across all studies. Combined specificity (and 95% CI) was calculated as the number of true negatives divided by the sum of the true negatives and false positives across all studies.

Meta-Analysis

Meta-analysis was performed on studies with complete data using the recommended bivariate modeling approach (10) with the R package Mada (Meta-analysis of Diagnostic Accuracy; R Foundation for Statistical Computing) (11). Summary receiver operating characteristic (ROC) analysis was used to combine studies to generate a summary area under the ROC curve (AUC), the 95% CI of which was generated by bootstrap analysis with 1000 resamples. Diagnostic accuracy was considered high for AUC values between 0.9 and 1.0 (12). All analyses were performed in R version 4.3.1 (R Foundation for Statistical Computing).

Additional Analyses

To address study heterogeneity, funnel plots were generated for study-specific sensitivity and specificity log odds ratios versus their standard errors. The Egger test for asymmetry was used to evaluate for potential publication bias (13). P values less than .05 were considered statistically significant.

As a sensitivity analysis to account for missing data, summary ROC analyses were also performed using multiple imputations. In particular, missing specificity data were imputed under a two-step mixed-effects model that captures the general relationship between sensitivity and specificity, while allowing for study-specific variations (more details in Appendix S1).

Results

Study Selection

The initial literature search identified 4261 records, of which 873 were duplicates (Fig 1). Of the 3388 nonduplicate records screened using the title and abstract, 3246 records were removed based on content that was not relevant to breast imaging of pregnant or lactating patients. There were 142 reports retrieved, and their full text was assessed for inclusion. A final number of 25 studies met the eligibility criteria for inclusion in the systematic review (14–38).

Figure 1:

Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram for the study selection process.

Study Characteristics

Study and patient characteristics are summarized in Table 1. Studies were published from 1992 to 2023. All studies were peer reviewed. Most studies (23 of 25) reported data from a single institution. The studies were conducted across 13 different countries with six studies from the United States. The studies included a total of 1681 female patients, with sample sizes from seven to 198. The average patient age was 33 years, ranging from 18 to 49 years. Most studies included both pregnant and lactating patients in their analyses (14 studies), whereas seven studies included only lactating patients and four studies included only pregnant patients. Sixteen studies included only patients with histologically confirmed breast cancer. Nine studies included patients with malignant, negative, and benign findings. Studies involved symptomatic patients—no studies were identified that focused on screening mammography, US, or breast MRI. Data regarding breast cancer risk factors (eg, family history, genetic predisposition) in the study populations were limited, and, when reported, most patients did not have a family history of breast cancer.

Table 1:

Characteristics of Patients Analyzed in Included Studies

All studies used a retrospective design. Retrospective chart reviews were performed on prospectively interpreted imaging that occurred during clinical care. In addition, 10 studies specified that imaging was evaluated retrospectively by one to six radiologists (14,18,21,24,26,28,34–36,38). The index tests assessed included US in 24 studies, mammography in 21 (primarily digital full-field), and MRI (for lactating patients) in eight. There were three studies of US alone, 14 studies that included US and mammography, one study with breast MRI alone, and seven studies with all three modalities (US, mammography, and MRI). No studies were identified that assessed digital breast tomosynthesis, contrast-enhanced mammography, or molecular breast imaging.

Histopathology (including samples from fine-needle aspiration, core biopsy, and surgical excision) was the reference standard for all malignant cases. In the studies by Qian et al (29) and Son et al (33), all breast lesions were pathologically proven to be benign or malignant. Clinical or imaging follow-up was the reference standard for negative and benign imaging findings that did not undergo histopathologic confirmation. The duration of imaging or clinical follow-up was variable: more than 12 months for Robbins et al (31), 1 year for Nissan et al (26), 2 years for Chung et al (16), and 5 years for Bock et al (15). Haliloglu et al (19) reported the duration of follow-up for eight of the 12 patients with BI-RADS 3 lesions but did not specify follow-up for the 61 patients with negative or benign US findings. Obenauer and Dammert (27) indicated that cases without cytologic or histologic examinations were evaluated at follow-up but did not specify a duration. For the two true-negative MRI examinations in the study by Espinosa et al (18) one patient had 18 months of clinical follow-up with no diagnosis of breast malignancy and the other patient had pending clinical and imaging follow-up at the time of publication.

Risk of Bias and Applicability

Table 2 details the quality assessments for risk of bias and applicability concerns for each study. No studies were assessed as high risk of bias or applicability concerns. All studies had low risk of bias regarding patient selection, reference standard, and flow and timing. Although most studies (18 of 25) were low risk for bias with respect to the index test, seven studies had unclear risk of bias for the index test. Five studies with unclear risk of bias only included patients with histologically proven breast cancer and thus readers could not be blinded to the diagnosis of malignancy. All studies ranked low for applicability concerns in patient selection and reference standard, whereas nine studies had unclear applicability concerns for the index test.

Table 2:

Quality Assessments for Risk of Bias and Applicability Concerns of the Included Studies

Results of Individual Studies

Of the 24 studies that included US, 22 studies had evaluable data for sensitivity and seven had evaluable data for specificity (Fig 2). The values for sensitivity ranged from 50% to 99% across 22 studies. The values for specificity ranged from 59% to 98% across seven studies.

Figure 2:

Forest plots for study-specific sensitivities and specificities for US.

Of the 21 studies that included mammography, 18 studies had evaluable data for sensitivity and three had evaluable data for specificity (Fig 3). The values for sensitivity ranged from 50% to 92% across 18 studies. The values for specificity ranged from 92% to 97% across three studies.

Figure 3:

Forest plots for study-specific sensitivities and specificities for mammography.

Of the eight studies that included breast MRI, there were seven studies with evaluable data for sensitivity and two studies for specificity (Fig 4). The values for sensitivity ranged from 90% to 98% across seven studies. The values for specificity ranged from 83% to 92% across two studies.

Figure 4:

Forest plots for study-specific sensitivities and specificities for breast MRI in lactating patients.

Synthesis of Results

Table 3 details the summary performance metrics for the meta-analysis. For US, the combined sensitivity was 92% (95% CI: 89, 94) from 22 studies, and the combined specificity was 76% (95% CI: 73, 80) from seven studies. Summary ROC analysis for the seven studies with complete data (Fig 5A) yielded an AUC of 0.90 (95% CI: 0.85, 0.93), a summary sensitivity of 81% (95% CI: 56, 94), and a summary specificity of 85% (95% CI: 71, 92).

Table 3:

Summary Performance Metrics for the Meta-Analysis

Figure 5:

(A) Confidence regions ellipse plot (left) and summary receiver operating characteristic (ROC) curve (right) for the seven studies on US with complete data. (B) Confidence regions and summary ROC curve for the three studies on mammography with complete data. (C) Confidence regions and summary ROC curve (too few data points) for the two studies on MRI with complete data. AUC = area under the ROC curve.

For mammography, the combined sensitivity was 81% (95% CI: 77, 84) from 18 studies, and the combined specificity was 94% (95% CI: 88, 98) from three studies. Summary ROC analysis for the three studies with complete data (Fig 5B) yielded an AUC of 0.93 (95% CI: 0.75, 0.97), a summary sensitivity of 72% (95% CI: 47, 88), and a summary specificity of 93% (95% CI: 86, 97).

For breast MRI, the combined sensitivity was 99% (95% CI: 95, 100) from seven studies, and the combined specificity was 100% (95% CI: 59, 100) from two studies. Summary ROC analysis for the two studies with complete data (Fig 5C) yielded an AUC of 0.95 (95% CI: 0.59, 0.96), a summary sensitivity of 91% (95% CI: 56, 99), and a summary specificity of 88% (95% CI: 48, 98).

Additional Analyses

Based on the funnel plot analysis of study heterogeneity (Fig 6), there is a potential bias for larger sensitivities and specificities in studies with US (P < .001). There is no visible bias in studies with mammography (P = .38) or breast MRI (P = .15).

Figure 6:

Funnel plots of sensitivity (circles) and specificity (triangles) generated to address study heterogeneity for (A) US, (B) mammography, and (C) breast MRI. Log-OR = odds ratio (log scale), SE = standard error.

With use of multiple imputations analysis, summary AUC, sensitivity, and specificity values were 0.93 (95% CI: 0.90, 0.95), 91% (95% CI: 85, 95), and 79% (95% CI: 60, 91) for US and 0.93 (95% CI: 0.84, 0.97), 79% (95% CI: 74, 83), and 95% (95% CI: 93, 96) for mammography, respectively (Fig S1).

Discussion

Large studies investigating the diagnostic performance of breast imaging modalities in pregnancy and postpartum are lacking. This systematic review of 25 studies involving 1681 female patients summarizes the published literature regarding the diagnostic performance of breast imaging modalities for cancer detection in pregnant and lactating patients presenting with breast symptoms. For studies with complete data, meta-analysis demonstrated summary AUC values of 0.90 for US, 0.93 for mammography, and 0.95 for breast MRI.

Current American College of Radiology Appropriateness Criteria recommend US as the first-line imaging modality for pregnant and lactating patients presenting with a breast concern based on the predominately younger patient age group in which there is decreased sensitivity of mammography due to dense breast tissue (4). If the US is normal or shows suspicious findings, additional imaging with mammography or digital breast tomosynthesis may be indicated. These recommendations are based on a small number of published retrospective studies because no large, prospective multi-institutional studies, systematic reviews, or meta-analyses have been performed to date.

Most of the studies identified in this systematic review included US and mammography. Although diagnostic accuracy values were similar between US and mammography in the meta-analysis, mammography may have an advantage of better specificity than US. However, direct head-to-head comparative diagnostic accuracy studies have not been performed. Because American College of Radiology recommendations support the safety of screening mammography during pregnancy and lactation, the initial diagnostic evaluation for pregnant patients who are at least 30 years old could include both mammography and US similar to patients who are not pregnant. Examples of the potential added value of mammography to improve the specificity of US is through demonstration of pathognomonic fat-fluid level of galactoceles, classically benign layering appearance of milk-of-calcium, or benign dystrophic calcifications of fat necrosis (3). Conversely, the addition of mammography may demonstrate suspicious calcifications potentially occult at US, particularly in cases of ductal carcinoma in situ. It is also possible that the addition of mammography could reduce potential inappropriate use of BI-RADS 3 assessments by revealing suspicious calcifications associated with masses at US that may mimic the appearance of fibroadenomas or lactating adenomas.

The individual studies included in this systematic review and meta-analysis have several limitations. Studies used a retrospective design with relatively small sample sizes and were mostly from single institutions. The lifetime breast cancer risk status of the study patients was not addressed. Furthermore, most studies included patients during pregnancy or lactation and thus represent a heterogeneous study population. Because hormonal changes to the breast evolve during pregnancy and postpartum, these data also do not distinguish differences in performance by trimester or duration of lactation. Some evidence suggests that there are biologic and prognostic differences in breast cancers occurring during pregnancy versus the postpartum period, and there are proposals to change the terminology from PABC to breast cancer diagnosed during pregnancy or breast cancer diagnosed during the postpartum period (39). The studies included in this systematic review and meta-analysis focused on diagnostic evaluation of symptomatic patients, and lesion sizes were not consistently reported or analyzed. This may overestimate the performance of imaging detection of small lesions. Likewise, no studies focused on diagnostic performance in the screening setting—in which small lesions are frequently asymptomatic. Another limitation is that none of the studies with mammography included digital breast tomosynthesis, which is now widely used in clinical practice. It is possible that the diagnostic performance of mammography with digital breast tomosynthesis may be better than that of mammography alone for PABC detection, as supported by increased cancer detection rates observed with digital breast tomosynthesis in the overall population, including younger women with dense breasts (40,41).

This systematic review and meta-analysis had some limitations. Because many studies included only patients with biopsy-proven PABC, specificity data were frequently not available. This limited the number of studies with complete data for the meta-analysis, particularly for mammography and breast MRI. Because gadolinium-based contrast agents are not recommended during pregnancy, the studies of diagnostic performance of breast MRI included only lactating patients, and thus only two studies were included in the meta-analysis. As a sensitivity analysis, the multiple imputations and bootstrap approach can partially mitigate the selection bias that commonly arises when only complete cases are analyzed. However, this approach cannot fully address the challenges posed by limited sample sizes, particularly when data are missing not at random. Another potential methodological limitation is the inclusion of studies that lacked information about blinding. However, retrospective chart review studies used prospective interpretation of the images as part of clinical care and thus reflect actual clinical practice scenarios. Another limitation is potential publication bias for US toward high sensitivity. Furthermore, a limitation that may affect specificity is the variability in follow-up duration across studies with benign and negative imaging results. Last, use of a single author for initial eligibility screening might have introduced selection bias.

In conclusion, the use of American College of Radiology Appropriateness Criteria is supported by the literature for breast imaging evaluation of pregnant and lactating patients. As demonstrated by this systematic review and meta-analysis, US, mammography, and breast MRI showed high diagnostic performance for PABC detection in symptomatic patients. Additional larger studies are needed to more precisely quantify specificity. Future studies may also address the performance of these modalities in the screening setting, especially as patients at increased lifetime risk of breast cancer may be advised to undergo their first screening test during pregnancy or while breastfeeding. The inclusion of breast cancer risk status for study populations is particularly important for subgroup analysis in future research to delineate whether certain modalities are consistently effective across all risk groups or if risk-stratified imaging approaches are needed. Additionally, assessment of the diagnostic performance of digital breast tomosynthesis is needed to better reflect current clinical practice. As the incidence of PABC is expected to rise, imaging evaluation for new breast lumps or other concerning symptoms experienced during pregnancy or breastfeeding should not be deferred to avoid delays in diagnosis.