In order to ensure evaluability and comparability, the studies were required to contain the following information to be included in the present analysis:

- breast density according to the BI-RADS ACR categories and/or quantification; ACR 1: almost entirely fat (low density, up to 25% mammary gland parenchyma), ACR 2: scattered fibroglandular densities (average density, 26-50% gland parenchyma), ACR 3: heterogeneously dense (high density, 51%-75% gland parenchyma), ACR 4: extremely dense (very high density, more than 75% gland parenchyma) 8.

For inclusion, all of the following criteria were required to be satisfied:

1. study/review deals with the questions under investigation

2. adequate type of study

3. adequate study population

4. intervention complies with technical standards (transducer frequency > 5 MHz)

5. required data are provided

Publications were excluded if any of the following criteria were met:

1. redundancy: multiple publications of the same data

2. methods: case reports, expert opinions, or poor-quality case-control studies

The systematic literature search covered the period from January 1^st, 2000 up to August 30^th, 2008 in the following databases:

1. PubMed (Internet portal of the National Library of Medicine) http://www.ncbi.nlm.nih.gov/sites/entrez?db=pubmed

2. Database of Abstracts of Reviews of Effects (DARE), of the Centre for Reviews and Dissemination (CRD) http://www.york.ac.uk/inst/crd/crddatabases.htm#DARE

3. Cochrane-database 'Cochrane Reviews' and 'Clinical Trials' http://www3.interscience.wiley.com/cgi-bin/mrwhome/106568753/HOME

The following search words were used:

PubMed: screening AND (mammography OR mammogram) AND breast neoplasms

AND (ultrasonography, mammary OR breast echography)

Cochrane und DARE: mammography screening AND ultrasound

The abstracts found were checked for relevance of content. Abstracts that did not prove to be relevant were excluded. Full-text versions of the abstracts deemed to contain relevant information were checked as to the criteria for inclusion and exclusion. The reasons for excluding a full text were given in each case.

Since none of the identified studies addressed the mortality rate or the difference in mortality rate contributed by supplemental breast ultrasound, surrogate parameters, namely the assessment criteria of diagnostic tests, the additional relative and absolute rates of detected cancers compared to mammography and tumor size and lymph node negative status were used as target values. Negative predictive value, sensitivity and specificity of breast ultrasound were only computed if the follow-up period continued as long as the screening interval, thus permitting the identification of false-negative findings.

The risk of adverse impacts for the women examined was quantified using the number of biopsies performed due to breast ultrasound findings. The positive predictive value for biopsies with malignant histological findings characterizes the number of unnecessary biopsies induced by false positives.

It could be inferred from the information given in five of the six studies that women had been enrolled consecutively. The reference standard, i.e. histological confirmation of the findings, was applied in all studies for those results classified as malignant or suspicious or indeterminate.

Only Kolb et al., 2002 48 reported a follow-up period of at least one year (up to the next screening exam) for all benign, but not for the negative findings. Kaplan, 2001 49 mentioned a follow-up period for a part of the results classified as benign. None of the other studies provided any information on follow-up.

Due to non-consecutive inclusion of patients and/or an inconsistently applied reference standard, all studies were ultimately assigned to the level of evidence category 3b 31.

As for publication quality, it is worth noting that false positives were expressed in such a way that the positive predictive values for ultrasound examination and the biopsy rate could be calculated in only five of the six studies. Information on mean or median age was lacking in one study, while the relative rate of cancers detected by procedures done as a complement to mammography was reported in five of the six studies.

The systematic search yielded studies in which breast ultrasound was used as a supplemental examination following mammographic interpretation. Moreover, of the group of asymptomatic women with negative mammographic results, women with breast tissue density (ACR 2-4) went on to be examined by ultrasound. The only exception is the study performed by Leconte et al., 2003 50, where 3% had palpable findings (Additional file 2, Part I).

The fraction of these women relative to all women screened within the specified period was reported in two studies at 36% (Kaplan, 2001 49) and 35.8% (Corsetti et al., 2008 51). The size of the study populations ranged from n = 1517 to n = 13547, with mean n = 5118.

The median age was reported in four studies, ranging from 47.6 years to 60.7 years, with overall age ranges of more than 30 years in each study. One study provided information on mean age (52 years) 51 and one study just reported age ranges from 35-87 years (Kaplan., 2001 49).

Two studies analyzed mammographic results of breasts in categories ACR 3 to ACR 4 5149, and the other studies evaluated the mammograms of ACR 2 to ACR 4 breast tissue 48505253.

Women with breasts of types ACR 3 or ACR 4 proved to have the highest proportion of breast cancers diagnosed by ultrasound screening.

Leconte et al. 50 diagnosed 16 carcinomas in all; 11 of which were detected in ACR 3 and ACR 4 breast tissue and five in women with breasts in categories ACR 1 and 2. Buchberger et al. 53 diagnosed 36 malignancies in breast tissue of ACR types 3 and 4 (0.3% cancer detection rate ACR 3, 1.1% cancer detection rate in ACR 4), while two carcinomas were found in the ACR category 2 breasts (0.4%). Crystal et al. 52 found no carcinomas in ACR-2 women, and 0.4% and 0.3% in breast-density categories ACR 3 and ACR 4, respectively. Kaplan 49 diagnosed cancer at the rate of 0.11% in ACR 2 breast tissue and 0.27% and 0.25% in ACR categories 3 and 4, in the one series where technologists performed the screening ultrasound.

The relative percentage of carcinomas found in supplemental breast ultrasound examinations as a fraction of the total number of detected cancers was reported in four studies, with a mean percentage of 22.5% (15%-34%) 53514850. The percentage as a fraction of the total population screened was calculated from all studies, with a mean value of 0.32% (0.23%-0.41%).

The mean size of the detected carcinomas was 9.9 mm (median size range 9.0 mm to 11.0 mm) in five of the six studies 5352494850. The mean percentage of invasive cancer detected was 94% (81%-100%) compared to noninvasive cancer (ductal carcinoma in situ, DCIS) with a mean of 6% (0%-19%) in all studies. Lymph node status was reported in four studies with negative lymph nodes in 90% (86%-100%) 51524948.

The positive predictive value with regard to the detection of additional malignancies for ultrasound prompted biopsies was reported or was able to be inferred from the given data in four of the six studies. The mean positive predictive value was 15% (2%-28%) 53524948. In other words, the percentage of positively classified findings for which no carcinoma was subsequently found ranged from 72% to 98%. This large variation of positive predictive values can mainly be attributed to the application of different sonographic criteria for malignancy and different assessment categories (see next section). Only one study reported that all findings classified as benign were followed up in the interval between two screenings (Kolb et al., 2002 48). However, Kolb et al. 48 do not mention a follow-up for those patients with negative results. Kolb states a sensitivity of 75.3%, a specifity of 96.8% and a negative predictive value of 99.7% 48.

The positive predictive value of detecting carcinomas by breast ultrasound mainly derives from the sonographic assessment criteria and categories applied. Overall, the range is still low relative to mammography.

Kaplan, 2001 49 used a two-armed categorization approach, namely a simple subdivision into negative and positive. All positive results were considered to be potentially suspicious. In contrast to the other studies, Kaplan 49 also deemed cysts positive if they exceeded a size of 1 cm. The author had a low positive predictive value of 2% 49. Three studies (Buchberger et al., 2000 53; Kolb et al., 2002 48 and Leconte et al., 2003 50) subdivided their findings into three categories, but applied different definitions of these categories. Categories 1, 2 and 3 were called 'normal', 'benign', and 'suspicious' (Kolb et al. 48) or 'benign', 'indeterminate' and 'malignant' (Buchberger et al. 53). These authors found positive predictive values of biopsy of 10.3% (Kolb et al. 48) and 28% (Buchberger et al. 53). The values for Leconte's study could not be ascertained 50. Crystal et al., 2003 52 used four arms of classification, thereby reaching a positive predictive value for malignancy of 20%. Other than Buchberger et al., Crystal et al. included the 'indeterminate' findings in their calculation of suspicious findings. Only one study used the five categories proscribed by the BI-RADS classification for breast ultrasound scans (Corsetti et al., 2008 51). In this study, however, the positive predictive value could not be determined.

The rate of biopsies resulting from suspicious findings detected by breast ultrasound ranged from 2.3% (Crystal et al. 52) to 4.7% (Corsetti et al., 2008 51) as a fraction of the total population of women screened. The positive predictive value of biopsies ranged from 8.4% to 13.7%. The difference compared to the positive predictive value of the breast ultrasound categorization may be explained by the fact that biopsies were sometimes also performed when findings were classified as 'indeterminate'. Clinically speaking, this means that 86.3% up to 91.6% of the invasive interventions done to confirm the diagnosis did not lead to a diagnosis of cancer.

Despite choosing a sensitive search strategy and a very broad definition of "breast cancer screening", only few single-center cohort studies analyzing breast ultrasound in the framework of breast cancer screening could be identified. All of these studies, however, were planned and conducted prospectively. The study populations were characterized by wide age ranges. For this reason, the results cannot be directly applied to women asked to participate in an organized population-based mammography screening program while aged between 50 and 69 years.

Since the studies applied different assessment criteria and classifications with regard to ultrasound morphology, their results are comparable only to a limited extent. As most studies lacked the requisite follow-up needed for computing sensitivity, specificity and negative predictive value, a comprehensive appraisal of the test quality of the intervention breast ultrasound could not be undertaken. None of the studies of US in breast screening were RCTs; therefore studies were not designed to provide evidence on screening benefit in terms of mortality reduction.

The effectiveness of breast ultrasound as a screening tool was mainly studied in women with breast tissue of the categories ACR 2 to ACR 4 and invariably only after previous mammography screening had yielded a negative result. Women with dense breasts run a four- to six-fold higher risk of developing breast cancer than other women 2627. The results revealed that supplemental breast ultrasound after negative mammographic screening permits the diagnosis of primarily invasive carcinomas in this risk group of women in an absolute mean of 0.32% of the cases. For comparative purposes: In population-based screening using mammography alone a mean of 0.4%-0.9% of all women screened were diagnosed with cancer 54.

The majority of cancers were detected in breast tissue of ACR types 3 and 4. According to these studies, the mean size of the cancers diagnosed additionally by breast ultrasound was not larger than the size of those visualized in mammography screening.

Berg (2004 39) reported similar results in a review of the application of breast ultrasound in women with dense breasts. The review was excluded because no systematic search strategy was specified, but Berg analyzed five of the six studies identified (Buchberger et al. 53; Kaplan 49; Kolb et al. 48; Crystal et al. 52; and Leconte et al. 50) and, in addition, a study by Gordon et al., 1995 55. She calculated a pooled value for the additional detection of cancers in the study populations of 0.35% and reported that 94% of the cancers detected in breast ultrasound were invasive with a mean diameter of 9-11 mm 39. More than 90% of the women in whom cancer was detected by breast ultrasound had breast density corresponding to ACR categories 3 or 4 39.

Hence, the contribution of supplemental breast ultrasound for the detection of breast cancer is primarily relevant for women with ACR 3 or ACR 4 tissue density. The question of whether improved detection of carcinomas leads to decreased breast cancer mortality cannot be answered on the basis of the available studies, although mathematical models allow to relate tumor size, lymph node involvement and death rates 185657.

Concerning possible adverse impacts of additional breast ultrasound, the biopsy rates in the evaluated studies of 2.3%-4.7% were significantly higher than the biopsy rates of about 1%-2% resulting from mammographic screenings 585960. Moreover, the positive predictive value of biopsies (mean 10.3%) was significantly lower than that of mammography screening (mean approximately 38%) 596061, i.e. three times more women need to undergo a biopsy per carcinoma detected. Although findings visualized by ultrasound examination can easily be biopsied in a minimally-invasive procedure, risk assessment should still take into consideration the psychological strain that women experience before and during biopsy performed because of a false-positive ultrasound result 6263.

The heterogeneous nature of the assessment criteria for breast ultrasound examinations, which in turn influence biopsy rates, is quite striking. Due to this heterogeneity, comparability of the studies is limited, and the effect of differences in experience of examiners is difficult to quantify. Since few studies included adequate follow-up of subjects with negative or benign findings, the false-negative rate in women with high breast density -- and hence the proportion of cancers missed by mammography and also missed by the addition of breast ultrasound in this group -- cannot be reliably estimated.

In 2003, Berg 61 initiated a prospective multicenter trial, randomized to the sequence of performance of mammography and ultrasound in women with intermediate and high breast cancer risk. Adding a single screening ultrasound to mammography yielded additional 1.1 to 7.2 cancers per 1000 high-risk women, but also substantially increased the number of false positives 37. The study provided evidence for the importance of supplemental breast ultrasound screening. Furthrmore the standardized scanning and interpretive criteria proved to be practicable for independent performance and intrepretation and could be used for further implementation. However, in view of the selected study cohort the results cannot be applied directly to a population-based screening program for asymptomatic women aged 50-69 years without personal history evaluation prior performing mammography.