Systematic Review and Meta-Analysis of the Diagnostic Accuracy of Ductoscopy in Patients with Pathological Nipple Discharge

Systematic review

Systematic review and meta-analysis of the diagnostic accuracy of ductoscopy in patients with pathological nipple discharge

L. Waaijer1,J.M.Simons1,I.H.M.BorelRinkes1,P.J.vanDiest2,H.M.Verkooijen3 and A. J. Witkamp1

Departments of 1Surgery and 2Pathology and 3Imaging Division, University Medical Centre Utrecht, Utrecht, The Netherlands Correspondence to: Ms L. Waaijer, Department of Surgery, University Medical Centre Utrecht, PO Box 85500, G04.228, 3508 GA Utrecht, The Netherlands (e-mail: [email protected])

Background: Invasive surgery remains the standard for diagnosis of pathological nipple discharge (PND). Only a minority of patients with nipple discharge and an unsuspicious finding on conventional breast imaging have cancer. Ductoscopy is a minimally invasive alternative for evaluation of PND. This systematic review and meta-analysis was designed to evaluate the diagnostic accuracy of ductoscopy in patients with PND. Methods: A systematic search of electronic databases for studies addressing ductoscopy in patients with

PND was conducted. Two classification systems were assessed. Forany DS , all visualized ductoscopic

abnormalities were classified as positive, whereas forsusp DS , only suspicious findings were considered

positive. After checking heterogeneity, pooled sensitivity and specificity of DSany and DSsusp were calculated. Results: The search yielded 4642 original citations, of which 20 studies were included in the review. Malignancy rates varied from 0 to 27 per cent. Twelve studies, including 1994 patients, were eligible for

meta-analysis. Pooled sensitivity and specificity of DSany were 94 (95 per cent c.i. 88 to 97) per cent and

47 (44 to 49) per cent respectively. Pooled sensitivity and specificity of DSsusp were 50 (36 to 64) and 83 (81 to 86) per cent respectively. Heterogeneity between studies was moderate to large for sensitivity 2 = ⋅ 2 = ⋅ 2 = ⋅ (DSany: I 17 5 per cent; DSsusp: I 37 9 per cent) and very large for specificity (DSany: I 96 8per 2 = ⋅ cent; DSsusp: I 92 6 per cent). Conclusion: Ductoscopy detects about 94 per cent of all underlying malignancies in patients with PND, but does not permit reliable discrimination between malignant and benign findings.

Paper accepted 6 January 2016 Published online 23 March 2016 in Wiley Online Library (www.bjs.co.uk). DOI: 10.1002/bjs.10125

Background with MRI, as well as galactography and/or cytology of nipple aspiration fluid (NAF) or ductal lavage. In patients Pathological nipple discharge (PND), defined as spon- with PND with no signs of malignancy on conventional taneous, unilateral, haemorrhagic or serous discharge, is diagnostic examination by mammography and ultrasono- responsible for approximately 5 per cent of surgical refer- graphy, the added value of breast MRI is limited4,12. rals to the breast clinic1. Malignancy rates of up to 23⋅9 per cent have been reported in patients with PND2, Investigations by galactography and cytological analysis but these studies are outdated, contained only surgically of NAF or ductal lavage lack sensitivity and usually do 2,4,13–15 treated patients or included patients with suspect findings not reveal the underlying pathology .Useofthese on imaging2–7. Modern studies on patients with PND diagnostic modalities therefore often results in a series of and negative conventional imaging by mammography and negative results. ultrasound imaging have shown considerably lower malig- There is no consensus on the diagnostic approach nancy rates of 3–7 per cent3,8–11. to PND, but surgery, by either selective duct excision Toevaluate the nature of PND, mammography and ultra- (microdochectomy) or major duct excision, is considered sonography are commonly used, sometimes supplemented the standard16. These procedures are invasive with a

concomitant risk of complications17, and possible effects Original studies performing ductoscopy were eligible for on cosmesis, breastfeeding potential and sensitivity of the inclusion in the systematic review. Studies had to report nipple. ductoscopic outcome in patients with PND or a hetero- Ductoscopy is a minimally invasive procedure that visu- genic cohort in which test performance was documented alizes the ductal epithelium of the breast via the nipple14,18. separately for each group, and use histological assessment It can be performed under local anaesthesia in the out- (following surgical excision or endoscopic tissue collection) patient clinic. Although ductoscopy is frequently used for and/or follow-up as a reference test. evaluating PND19, its accuracy in patients with symp- Studies reporting small series (fewer than 25 patients), toms of PND remains a matter of debate. Several studies studies including patients with preoperative diagnosis of have reported promising results, but included hetero- breast cancer, overlapping studies (only the most recent geneous patient populations, rating systems and previous article or the article reporting the data of interest was diagnostics20. Although some studies21,22 reported a sig- included) and studies other than original clinical trials nificant correlation between intraductal morphological (for instance congress abstracts, reviews, editorials, case findings and histological diagnosis, others23–26 found this reports, ex vivo studies, protocols) were excluded. only for papillomas. A cross-reference check was performed to assess the The aim of the present study was to carry out a sys- quality of the search and to identify eligible additional tematic review of the literature, and if possible to per- studies not identified by the primary search. form a meta-analysis to determine the diagnostic accuracy Two authors performed study selection, quality assess- of ductoscopy in the diagnosis of malignancy in patients ment and data extraction independently. Any differences with PND. were resolved by mutual agreement. Any disagreement was resolved through re-evaluation by a third author, blinded to the outcome of previous assessment. Methods

Search strategy Quality assessment A systematic search of the PubMed, Embase and Cochrane The quality of eligible studies was assessed using the Qual- Library databases was performed according to the Pre- ity Assessment of Diagnostic Accuracy Studies (QUADAS) ferred Reporting Items for Systematic Reviews and 2 questionnaire28. The QUADAS-2 was tailored to the Meta-Analyses (PRISMA) guidelines27. The search query present analysis, as described in the guideline (Appendix S2, was limited to 2 February 2015 and included synonym supporting information)28, and applied to a random terms within the index test (ductoscopy) (Appendix S1, sample from the available studies to evaluate inter-rater supporting information). To minimize the risk of miss- agreement. Finally, all included studies were evaluated ing relevant studies, synonyms for the target condition (Table S1, supporting information). or reference standard were not included in the search strategy. Data extraction and management The following data were extracted for each study: study Selection of studies period, inclusion criteria and sample size, together with After removal of duplicates, all identified studies were patients’ ages, and technical and procedural characteristics. screened by title and abstract. Full-text papers were For meta-analysis, two different definitions were used retrieved when studies evaluated ductoscopy, reported to classify ductoscopic findings. For DSany,anyvisu- original data, and were written in English, German, alized finding at ductoscopy was classified as positive, French or Dutch. and normal ducts in ductoscopy were classified as nega-

Studies reporting ductoscopic findings in patients pre- tive. For DSsusp, ductoscopically suspicious findings were senting with PND were included. Studies that reported classified as positive29–31. Suspicious ductoscopic findings on ductoscopic findings combined with histopathological were defined as duct wall irregularities, epithelial thick- outcome after surgery were classified as ‘surgical refer- ening, inflammatory changes, web, stricture, red patches ence studies’. In some studies, negative or non-suspect or fronds. Normal ducts, intraductal debris or polypoid ductoscopic findings were not followed by surgery. In lesions were defined as non-suspicious/negative. these patients, clinical follow-up was used as a surrogate Malignancy at histopathology or during follow-up was reference standard and these studies were classified as considered a positive outcome of the reference test. Malig- ‘selective follow-up studies’. nancy was defined as ductal carcinoma in situ (DCIS)

Records identified through database searching n = 4811 PubMed n = 2893 Embase n = 1350 Cochrane Library n = 568

Duplicates removed n = 169

Title/abstract screened n = 4642

Records excluded n = 4484

Full-text articles assessed n = 158

Papers excluded based on study type n = 90 Review n = 31 Unpublished study n = 1 Language n = 25 Not about ductoscopy n = 4 Not original research n = 30 n Articles included in systematic review = 20 Original articles excluded n = 49< n Surgical reference = 13 Inclusion criteria n = 24 n Selective follow-up = 7 No results of ductoscopy n = 3 Overlapping data n = 10 Study size < 25 n = 12 Articles included in meta-analysis Reference test inadequate n = 11 n DSany = 12 n DSsusp = 8

Fig. 1 Flow chart showing selection of articles for review. *Original articles could be excluded for more than one reason or invasive breast carcinomas. Intraductal (solitary or Statistical analysis multiple) papillomas, atypical ductal hyperplasia, fibroade- Sensitivity, specificity and corresponding 95 per cent nomas, mastopathy, cysts and fibrosis were considered confidence intervals were calculated for DS and DS inconspicuous findings. any susp of individual studies from the contingency tables. Het- The primary endpoints for this systematic review were erogeneity was assessed by the I2 statistic (estimated the sensitivity and specificity of DS and DS .Sensi- any susp proportion of unexplained interstudy variance) before tivity of DS was defined as the proportion of women any pooling. Random-effects models, using a restricted maxi- with a positive ductoscopy among all women subsequently mum likelihood estimator, were used when there was large diagnosed with malignancy at histological assessment or interstudy variance (I2 greater than 25 per cent), to calcu- follow-up. Specificity of DS was defined as the pro- any late a pooled estimate. Otherwise, fixed-effect modelling portion of women with a negative ductoscopy among all was applied. Results were reported in accordance with the women with benign outcome at histology or follow-up. PRISMA recommendations32. For DSsusp, sensitivity was defined as the proportion of women with suspect findings at ductoscopy among all women diagnosed with malignancy at histology or Results follow-up. Specificity of DSsusp was defined as the proportion of women with non-suspect ductoscopy among all The search query resulted in 4811 citations. After filtering women with benign outcome at histology or follow-up. duplicates, 4484 citations were excluded based on title and To construct 2 × 2 contingency tables, data on the abstract (Fig. 1). The full text of 158 studies was retrieved number and morphological characteristics of ductoscopic for detailed review. Cross-reference checking of these findings and outcomes of histology or follow-up were studies identified no additional studies. Ninety studies extracted. If, owing to the specific classification system were excluded after full-text assessment owing to language used or the lack of a detailed description, data extraction (25) and type of article (31 reviews, 13 editorials, letters was not possible, the study was excluded from the analysis. or notes, 8 congress abstracts, 5 case reports, 3 ex vivo A stratified analysis of surgical reference studies versus studies, 1 study protocol), and four studies did not address selective follow-up studies was performed to evaluate the ductoscopy. Forty-nine original studies were excluded impact of differential outcome on sensitivity and specificity after full-text assessment, some for more than one reason: of DSany. 24 studies included a different population (women without

Table 1 Study characteristics

Composite Study Study Study Ductoscope Reference reference Reference Country period population design size (mm)* standard standard

Surgical reference studies Fisher and USA 2006–2007 PND Retrospective 0⋅9 Microdochectomy n.a. Margenthaler33 Hahn et al.34 Germany 2007–2008 PND without abnormal Prospective 0⋅9–1⋅1 Microdochectomy n.a. imaging findings amenable to biopsy Denewer et al.22 Egypt 2005–2006 PND without palpable Prospective n.s. Microdochectomy n.a. lesions or imaging abnormalities Dietz et al.24 USA 1999–2001 PND Retrospective 0⋅9–1⋅2 Microdochectomy n.a. Moncrief et al.26 USA 1996–2003 PND Retrospective 0⋅9 Microdochectomy n.a. Simpson et al.35 Canada 2004–2008 PND Prospective 0⋅7 Microdochectomy n.a. Deshmane36 India 2004–2007 PND undergoing DS Retrospective 1⋅1 Microdochectomy n.a. and DS-directed surgery Dooley37 USA 2000–2001 PND n.s. 0⋅9 Microdochectomy n.a. Ohlinger et al.38 Germany 2006–2009 PND Prospective 0⋅9 Microdochectomy n.a. Tekin et al.39 Turkey 2002–2007 PND without imaging n.s. 1 Microdochectomy n.a. abnormalities Albrecht et al.40 Germany 2002–2006 PND and/or abnormal Retrospective 0⋅55–1⋅30 Open biopsy or n.a. diagnostic findings microdochectomy Beechey-Newman England n.s. PND undergoing Prospective 0⋅95–1⋅1 Microdochectomy or n.a. et al.41 diagnostic surgery major duct excision Sauter et al.42 Canada n.s. PND with/without Prospective 0⋅9 n.s. n.a. (suspect) imaging abnormalities Selective follow-up studies Khan et al.43 USA 2005–2009 PND Prospective 0⋅9 Polypoid lesion: No intraductal lesion: FU microdochectomy Kamali et al.44 Turkey 2005–2010 PND Retrospective 0⋅55–1⋅1 Polypoid lesion: DP Negative DS and Unsuccessful DP and/or cytology: FU positive cytology: microdochectomy Fackler et al.45 USA n.s. PND or atypical lavage Prospective 0⋅9 Positive DS: Negative DS: FU cytology† microdochectomy Bender et al.46 Turkey 2006–2008 PND Prospective 0⋅55–1⋅1 Positive cytology: Negative DS and microdochectomy negative cytology: FU Non-sessile polypoid lesions: DP Sessile polypoid lesions/MPs and negative cytology: microdochectomy Liu et al.47 China 1997–2005 PND n.s. 0⋅72 Abnormal DS, positive Negative DS, cytology 0⋅45–0⋅8 cytology, and imaging: FU clinical/radiological abnormalities: microdochectomy or lumpectomy Waaijer et al.11 The Netherlands 2010–2014 PND, BI-RADS ≤ 3or Prospective 1⋅15 Suspect DS or patient’s Non-suspicious findings proven benign request: or papillomas: FU histologically microdochectomy or major duct excision Non-sessile polypoid lesions: DP Matsunaga et al.48 Japan 1993–2007 PND without specific n.s. 0⋅72 Surgery or intraductal No surgery or biopsy: findings on biopsy FU mammography or US‡

*Manufacturer details are given in parentheses. †Women in the Program for Breast Cancer Early Detection and Prevention who had symptoms of pathological nipple discharge (PND) or atypical findings on lavage cytology. ‡Indication for ductoscopy (DS) based on cytological and galactographic findings, which were not described in further detail. n.a., Not available; n.s., not stated; FU, follow-up; DP, ductoscopic papillomectomy; MP,multiple papilloma; BI-RADS, Breast Imaging Reporting and Data System; US, ultrasonography.

Flow and timing Low High Unclear Intermediate

Reference standard Reference standard

Index test Index test QUADAS-2 domain QUADAS-2

Patient selection domain QUADAS-2 Patient selection

0 20 40 60 80 100 0 20 40 60 80 100 Risk of bias (~) Applicability concerns (~) a Risk of bias b Applicability concerns

Fig. 2 Proportion of studies with low, intermediate, high or unclear risk of bias (a) and concerns regarding applicability (b) determined with the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) 2 tool

PND, such as those with malignancies or high risk of risk of bias in flow and timing was assessed in 55 per cent breast cancer) or the results of patients with PND in a of studies (11 of 20), mostly due to exclusion of unsuccess- heterogeneous cohort were not (separately) reported; 12 fully cannulated patients or to an inadequate or unclearly studies included fewer than 25 patients; 11 used a different described study flow. High risk of bias in patient selection, or inadequate reference test; three did not report results of index test and reference standard was found in three36,40,45, ductoscopic findings; and ten studies used the same patient two38,40 and one46 respectively of the 20 included studies. cohort. Regarding applicability, concerns were raised mostly by Finally, 13 surgical reference studies (with histology studies that included only a selected group of patients as the reference standard in all patients)22,24,26,33–42,and who were already referred for surgery or, in the case of six selective follow-up studies (with surgical confirma- retrospective studies, who had undergone both ductoscopy tion for patients with suspicious ductoscopic findings and surgery. The index test was clearly described in most and follow-up in patients with normal/unsuspicious studies, but few data were supplied regarding its conduct ductoscopic results)43 –48 (Ta b l e 1 ) were included. Results and interpretation. The histopathological outcome was of a recent prospective study by the authors themselves mostly described clearly, but data on conduct of follow-up were also included11. were limited.

Quality assessment Characteristics of the included studies The assessment of methodological quality showed that the Ten studies were prospective cohort studies (Oxford overall quality of the included studies was acceptable (Fig. 2; Centre for Evidence-Based Medicine (OCEBM) level 2), Table S1, supporting information). The inter-rater agree- six were retrospective cohort studies (OCEBM level 3) ment of the modified QUADAS-2 tool was fair, with an and the study design of four studies was not mentioned agreement rate of 0⋅69 (95 per cent c.i. 0⋅59 to 0⋅79). Sev- (OCEBM level unknown) (Ta b l e 1 ). enty per cent of studies (14 of 20) were classified as having Tissue for histological examination was acquired mostly low risk of bias in patient selection. Other studies included by surgery (microdochectomy or major duct excision). patients with different or poorly described previous diag- Ductoscopic collection of histological tissue was performed nostic investigation. Low risk of bias in the index test was in four studies11,44,46,48. Three studies11,44,46 used a ‘basket’ found in 65 per cent of studies (13 of 20). Undefined, device (3-wire basket; PolyDiagnost, Pfaffenhofen, Ger- or even unreported, ductoscopic scoring systems were the many), which is feasible for non-sessile polypoid lesions. most common reason for risk of bias in index test. For ref- In selective follow-up studies11,43,45,48, patients were erence standard, a low risk of bias was found in 70 per selected for follow-up based on negative ductoscopy cent of the studies (14 of 20). Higher risks were caused findings. In three studies44,46,47, negative cytology results mainly by unreported methods of tissue acquisition or for either NAF and/or ductal lavage were a second poorly reported follow-up. A high, intermediate or unclear prerequisite.

Table 2 Patient characteristics and outcome Follow-up Missing Malignancy# Age No. of Successful Histology (as reference reference at initial Malignancy# Total Reference (years)* patients‡ cannulation obtained test) test Papilloma histology at follow-up malignancy** Surgical reference studies Fisher and 48 (22–80)† 121 119 (98⋅3) 121 (100) 0 (0) 0 (0) 64 (52.9) 7 (5⋅8) n.a. 7 (5⋅8) Margenthaler33 Hahn et al.34 51⋅7 (20–71) 33 33 (100) 33 (100) 0 (0) 0 (0) 22 (67) 2 (6) n.a. 2 (6) Denewer et al.22 46 (20–75) 54 54 (100) 54 (100) 0 (0) 0 (0) 18 (33) 6 (11) n.a. 6 (11) Dietz et al.24 52 (14–82) 121§ 105 (88⋅2)†† 121 (100) 0 (0) 0 (0) 83 (68⋅6) 5 (4⋅1) n.a. 5 (4⋅1) Moncrief et al.26 48⋅7 (28–83) 68§ 59 (87)§ 68 (100) 0 (0) 0 (0) 37 (54) 6 (9) n.a. 6 (9) Simpson et al.35 (25–76) 65 63 (95)§ 62 (95) 0 (0) 3 (5) 52 (80) 3 (5) n.a. 3 (5) Deshmane36 n.s. 26 24 (92) 24 (92) 0 (0) 2 (8) 11 (42) 7 (27) n.a. 7 (27) Dooley37 (18–27) 27 26 (96) 26 (96) 0 (0) 1 (4) n.s. 2 (7) n.a. 2 (7) Ohlinger et al.38 52⋅2 (19–86) 247 214 (86⋅6) 214 (86⋅6) 23 (9⋅3) 33 (13⋅4) 110 (44⋅5) 11 (4⋅5) n.a. 11 (4⋅5) Tekin et al.39 45⋅3 (17–71) 34 30 (88) 20 (59) 0 (0) 14 (41) 9 (26) 3 (9) n.a. 3 (9) Albrecht et al.40 53⋅3 (21–78) 100§ 98 (98⋅0)§ 100 (100)§ 0 (0) 0 (0) 44 (44⋅0) 11 (11⋅0) n.a. 11 (11⋅0) Beechey-Newman n.s. 50 50 (100) 33 (66) 0 (0) 17 (34) 18 (36) 0 (0) n.a. 0 (0) et al.41 Sauter et al.42 48⋅4 (25–81) 40 40 (100) 39 (98) 0 (0) 1 (3) 18 (45) 3 (8) n.a. 3 (8) Overall (%)† 94 98 0 3 45 7 7 Selective follow-up studies Khan et al.43 47⋅5 59 50 (85) 38 (64) 19 (32) 2 (3) 24 (41) 6 (10) 0 (0)§§ 6 (10) Kamali et al.44 42⋅7 (14–83)† 456§ 383 (84⋅0)§ 157 (34⋅4) 196 (43⋅0) 103 (22⋅6) 97 (21⋅3) 21 (4⋅6) 0 (0)¶¶ 21 (4⋅6) Fackler et al.45 48 (26–75) 64¶ n.s. 36 (56) 28 (44) 0 (0) 27 (42) 7 (11) 0 (0)## 7 (11) Bender et al.46 49⋅2 (24–80)† 126¶¶¶ 102 (81⋅0) 32 (25⋅4) 71 (56⋅3) 24 (19⋅0) 30 (23⋅8) 1 (0⋅8) 0 (0)*** 1 (0⋅8) Liu et al.47 43 (13–80)† 1093§ n.s. 436 (39⋅9) 489 (44⋅7) 168 (15⋅4)‡‡ 232 (21⋅2) 51 (4⋅7) 1 (0⋅1)††† 52 (4⋅8) Waaijer et al.11 47 (20–72) 82 71 (87) 36 (44) 45 (55) 1 (1) 23 (28) 3 (4) 1 (1)‡‡‡ 4(5) Matsunaga et al.48 n.s. 323 323 (100) 235 (72⋅8) 88 (27⋅2) 0 (0) 155 (48⋅0) 80 (24⋅8) n.s.§§§ 80 (24⋅8) Overall (%)† 85 44 44 3 28 5 5

Values in parentheses are percentages unless indicated otherwise; *values are mean (range), except †median (range). ‡Number in whom ductoscopy was performed, including unsuccessful cannulation. §Outcome reported in number of ductoscopies; ¶outcome reported in number of ducts. #Invasive carcinoma or ductal carcinoma in situ. **Malignancy outcome at histology and follow-up. ††Cannulation given in number of patients, whereas other rates in this study are reported in number of ductoscopies. ‡‡All 157 patients with positive ductoscopic findings were lost to follow-up. §§Follow-up by repeat physical examination at 3, 6 and 12 months, mammography at 12 months followed by annual examination and mammography; median follow-up 38 (range 7–63) months. ¶¶Follow-up by physical examination, ultrasonography, mammography and other diagnostic procedures. ##Minimum follow-up of 2 years. ***Duration of follow-up not given. †††Mean follow-up 19 months. ‡‡‡Median follow-up 17 (range 3–45) months. §§§Minimum follow-up 5years.¶¶¶From the numbers of patients supplied in the article, a total of 126 can be calculated. n.s., Not stated; n.a., not applicable.

Twenty studies included in the systematic review 50 (95 per cent c.i. 36 to 64) and 83 (81 to 86) per cent reported on 3189 ductoscopies in 3144 patients with respectively (Fig. 4). Heterogeneity analyses indicated PND; the number of analysed ductoscopies ranged from degrees of between-study heterogeneity for pooled sensi- 26 to 1093 (Ta b l e 2 ). The cannulation success rate ranged tivity (I2 = 37⋅9 per cent) and pooled specificityI ( 2 = 92⋅6 from 81⋅0 to 100 per cent. Malignancy rates ranged from per cent). 0 to 27 per cent. The proportion of patients with papillomas, calculated only from surgical reference studies, ranged from 26 to 80 per cent. Stratified analysis (surgical reference versus Twelve studies including 1994 ductoscopies were selective follow-up) included in the meta-analysis. In total, 151 malignancies (7⋅6 per cent) were found. The respective pooled sensitivity Studies were stratified by differential outcome (surgical and specificity of DSany were 94 (95 per cent c.i. 88 to 97) reference versus selective follow-up). Follow-up ranged and 47 (44 to 49) per cent respectively (Fig. 3). Statistical from a median of 17 months to a minimum follow-up of analyses of heterogeneity revealed I2 values of 17⋅5and 5 years (Ta b l e 2 ). Median successful cannulation in selec- 96⋅8 per cent respectively. tive follow-up studies was 85 (range 81⋅0–100) per cent, Eight studies11,22,24,26,33,35,44,46 with 933 ductoscopies compared with 94 (range 86⋅6–100) per cent in the surgi- including 52 malignancies (5⋅6 per cent) were eligible cal reference studies. In selective follow-up studies, histo- for the DSsusp analysis (Table S2, supporting informa- logical assessment was obtained for a median of 44 (range ⋅ ⋅ tion). Pooled sensitivity and specificity of DSsusp were 25 4–72 8) per cent of patients, and follow-up for a median

Reference Sensitivity

Bender et al.46 1·00 (0·03, 1·00) Denewer et al.22 1·00 (0·54, 1·00) Deshmane36 1·00 (0·59, 1·00) Dietz et al.24 1·00 (0·29, 1·00) Fackler et al.45 1·00 (0·59, 1·00) Fisher and Margenthaler33 1·00 (0·66, 1·00) Kamali et al.44 0·76 (0·53, 0·92) Khan et al.43 1·00 (0·54, 1·00) Liu et al.47 0·94 (0·84, 0·99) Moncrief et al.26 1·00 (0·54, 1·00) Simpson et al.35 1·00 (0·29, 1·00) Waaijer et al.11 1·00 (0·29, 1·00)

Pooled sensitivity = 0·94 (0·88, 0·97) 2 = 13·33, 11 d.f., P = 0·272 0 0·2 0·4 0·610·8 Inconsistency (I2) = 17·5~ Sensitivity

a Sensitivity of DSany

Reference Specificity

Bender et al.46 0·70 (0·60, 0·78) Denewer et al.22 0·27 (0·15, 0·42) Deshmane36 0·29 (0·10, 0·56) Dietz et al.24 0·06 (0·02, 0·12) Fackler et al.45 0·49 (0·36, 0·63) Fisher and Margenthaler33 0·10 (0·05, 0·17) Kamali et al.44 0·63 (0·57, 0·68) Khan et al.43 0·33 (0·20, 0·50) Liu et al.47 0·56 (0·52, 0·59) Moncrief et al.26 0·04 (0·00, 0·13) Simpson et al.35 0·19 (0·10, 0·31) Waaijer et al.11 0·26 (0·17, 0·39)

Pooled specificity = 0·47 (0·44, 0·49) 2 = 340·88, 11 d.f., P < 0·001 2 0 0·2 0·4 0·610·8 Inconsistency (I ) = 96·8~ Specificity

b Specificity of DSany

Fig. 3 Forest plot and pooled estimates of DSany. a Sensitivity of DSany was defined as the proportion of women with a positive finding at ductoscopy among all women subsequently diagnosed with malignancy at histology or follow-up. b Specificity of DSany was defined as the proportion of women with a negative finding at ductoscopy among all women with a benign outcome at histology or follow-up. Sensitivity was pooled by a fixed-effect model (Mantel–Haenszel), and specificity by a random-effects model (DerSimonian–Laird). The diamond represents the pooled estimate, with its 95 per cent confidence interval depicted by dotted lines. Values in parentheses are also 95 per cent confidence intervals of 44 (27⋅2–56⋅3) per cent. In surgical reference stud- selective follow-up studies (median 5 (range 0⋅8–24⋅8) per ies, histological outcome was acquired for a median of 98 cent) were diagnosed at initial histological assessment.

(range 59–100) per cent of patients. Stratified analysis did not modify the sensitivityany ofDS . Malignancy was diagnosed in a median of 7 (range The pooled sensitivity of surgical reference studies was 0–27) per cent of all patients in surgical reference stud- 100 (95 per cent c.i. 90 to 100) per cent (I2 = 0 per cent) ies, compared with 5 (0⋅8–24⋅8) per cent of all patients versus 91 (83 to 96) per cent (I2 = 37⋅5 per cent) of selec- in selective follow-up studies. Most malignancies in the tive follow-up studies. In contrast, the pooled specificity

Reference Sensitivity

Bender et al.46 1·00 (0·03, 1·00) Denewer et al.22 0·33 (0·04, 0·78) Dietz et al.24 0·00 (0·00, 0·71) Fisher and Margenthaler33 0·67 (0·30, 0·93) Kamali et al.44 0·57 (0·34, 0·78) Moncrief et al.26 0·17 (0·00, 0·64) Simpson et al.35 0·67 (0·09, 0·99) Waaijer et al.11 0·67 (0·09, 0·99)

Pooled sensitivity = 0·50 (0·36, 0·64) 2 = 11·26, 7 d.f., P = 0·128 2 0 0·2 0·4 0·610·8 Inconsistency (I ) = 37·9~ Sensitivity

a Sensitivity of DSsusp

Reference Specificity

Bender et al.46 1·00 (0·96, 1·00) Denewer et al.22 0·81 (0·67, 0·91) Dietz et al.24 0·62 (0·53, 0·71) Fisher and Margenthaler33 0·71 (0·61, 0·79) Kamali et al.44 0·91 (0·87, 0·94) Moncrief et al.26 0·91 (0·79, 0·97) Simpson et al.35 0·76 (0·63, 0·86) Waaijer et al.11 0·79 (0·68, 0·88)

Pooled specificity = 0·83 (0·81,0·86) 2 = 91·92, 7 d.f., P < 0·001 I2 0 0·2 0·4 0·610·8 Inconsistency ( ) = 92·6~ specificity

b Specificity of DSsusp

Fig. 4 Forest plot and pooled estimate of DSsusp. a Sensitivity of DSsusp was defined as the proportion of women with suspicious findings at ductoscopy among all women diagnosed with malignancy at histology or follow-up. b Specificity of DSsusp wasdefinedasthe proportion of women with a non-suspicious finding at ductoscopy among all women with benign outcome of histology or follow-up. Sensitivity and specificity were both pooled by a random-effects model (DerSimonian–Laird). The diamond represents the pooled estimate, with its 95 per cent confidence interval depicted by dotted lines. Values in parentheses are also 95 per cent confidence intervals of surgical reference studies was 12 (9 to 16) per cent The search yielded 4642 original citations, of which 20 (I2 = 79⋅2 per cent) versus 56 (53 to 58) per cent (I2 = 89⋅6 studies were eligible for the review, including a total of 3144 per cent) in selective follow-up studies. patients. Meta-analysis classifying any intraductal lesion

visualized by ductoscopy as a positive finding (DSany)was Discussion possible in 12 studies, including a total of 1994 patients, and resulted in a pooled sensitivity of 94 (95 per cent c.i. 88 to Currently there is no conformity on the assessment of 97) per cent and specificity of 47 (44 to 49) per cent. To patients with PND and the role of ductoscopy within this find the optimal cut-off point a second classification scale approach. High-quality studies on techniques that evolved was analysed, in which only suspicious ductoscopic findings since the introduction of ductoscopy remain scarce. The were classified as a positive finding (DSsusp).Forthisanaly- aim of this study was to assess the diagnostic accuracy of sis a total of eight studies including 933 ductoscopies were ductoscopy in patients with PND. eligible, resulting in pooled sensitivity and specificity of 50

(36 to 64) and 83 (81 to 86) per cent respectively. Statistical Toinvestigate this possible distorted selection of patients, investigation revealed that heterogeneity in both DSany and the present review also included studies using follow-up as DSsusp was moderate to large for sensitivity and very large a reference standard (selective follow-up studies). Selective for specificity. follow-up was described for patients with negative ducto- Although ductoscopy was capable of visualizing about scopic (and cytological) findings and those with endo- 94 per cent of lesions, the specificity was low. The cut-off scopically acquired histological tissue diagnosis of benign point assessed in the DSsusp classification resulted in intraductal papilloma. Stratified analysis of the two dif- increased specificity at the cost of unacceptably low sensi- ferential reference designs (surgical reference versus tivity. These results imply that, to exclude malignancy in selective follow-up) showed a difference in cannulation patients with PND and positive ductoscopy, histological rate, which indicates a possible selection bias in the diagnosis remains necessary. studies describing surgical reference in all patients. No To minimize the number of surgical interventions, significant difference in median malignancy rate was acquisition of histological tissue is preferably done endo- found (7 per cent for surgical reference versus 5 per cent scopically. In addition, for a precise histological correlation for selective follow-up studies). The selective follow-up of lesion morphology, endoscopic biopsy is the preferred studies showed low rates of malignancy in the conser- method. Currently, tissue collection of polypoid lesions vatively treated cohort. These findings suggest that the can be performed with a commercially available ‘grabbing’ selection criteria employed in the selective follow-up biopsy clip or a so-called ‘basket’ device (PolyDiagnost), studies might indeed be capable of triaging patients for resulting in reliable histopathological diagnosis11,49,50. surgery or conservative treatment. Stratified analysis Unfortunately this device is not yet available for the US showed an increased specificity in the selective follow-up market. Intraductal needle biopsy devices (for example studies of 56 per cent, compared with 12 per cent in JN Biopsy Needle; Hakko Company, Tokyo, Japan) can surgical reference studies. As the overall rate of diag- be used for flat epithelial lesions and have been used in nosed malignancy is comparable, missed malignancies in research, but are relatively short (45⋅5 mm) and are not yet the follow-up group are not likely. In the study design commercially available in either Europe or the USA48,51. where selected criteria will lead to surgery (selective For intraductal lesions in which endoscopic tissue diagnosis follow-up studies), the physician might use this crite- is not possible, surgery therefore remains inevitable. rion with more caution, leading to a relatively high rate Owing to the high sensitivity and low incidence of malig- of false-positives in surgical studies where all patients nancy in patients with PND, the negative predictive value undergo operation. of DSany ranged from 98 to 100 per cent. The authors The main limitation of this review is the heterogeneity therefore conclude that a negative finding on ductoscopy in interpretation of the index test (ductoscopy) between makes subsequent surgery redundant. This implies that the different studies. The employed morphological cat- ductoscopy can help in preventing unnecessary explorative egory scales vary, and are likely to be subjective and surgery for benign disease, and in avoiding its concomitant observer-dependent. This impairs the comparison and risk of complications and effects on cosmesis. pooling of the various studies. Owing to this wide range The malignancy rate of the 20 included studies ranged or variability, two arbitrary choices for cut-off values of from 0 to 27 per cent, but only two studies demonstrated positive ductoscopy, DSany and DSsusp, were analysed. malignancy rates above 11 per cent36,48. These studies Endoscopic appearances should be classified according to both scored high regarding risk of bias and applicability one universal scale, as has been proposed and validated concerns (Table S2, supporting information). Deshmane36 previously: the superficial, polypoid solitary, polypoid included 26 patients with PND without mentioning previ- multiple, and combined types29,30. Additionally, the depth ous diagnostic investigations. Matsunaga and co-workers48 of lesions should be reported to provide improved ability included patients with bloody nipple discharge without to detect more peripherally located lesions. This is impor- specific findings on mammography or ultrasound imag- tant, as these are the lesions that could be missed by blind ing, but used cytology and galactography as an indication surgical excision52. for ductoscopy, without further defining these indications. A further limitation is the poorly reported information on As cytology and galactography in PND have reported low inclusion methods, previous diagnostic investigations and sensitivity with higher specificity2, the high malignancy patient characteristics, raising concerns regarding applica- rates of these two studies probably relate to the selection bility and hampering quality assessment and data extraction of the included patient population, which is possibly not (Fig. 2). Despite this, only two studies36,48 showed outliers applicable to the present research question. concerning malignancy prevalence.

For the present study, only DCIS and invasive breast with PND each year is significant, implementation of cancer were classified as a positive (malignant) outcome. this procedure in specialized breast clinics can be rec- As a consequence, diagnoses of atypical ductal hyperpla- ommended. Future research and investment should focus sia, atypical lobular neoplasia and papilloma (with or with- on additional techniques, such as microendoscopic biopsy out atypia) were left out of the analysis, despite their devices, to improve diagnostic accuracy further. impact regarding breast cancer risk. However, the risk and management of these findings when visible on conven- Disclosure tional imaging is not unambiguous. The long-term risk of such lesions without concurrent abnormalities on imaging The authors declare no conflict of interest. remains to be clarified. Evaluation of other diagnostic techniques was not an References 4,15,16 aim of this review. Previous studies in women with 1 Dixon J, Mansel R. ABC of Breast Diseases. Symptoms PND have shown that radiological investigations, includ- assessment and guidelines for referral. BMJ 1994; 309: ing ductography and MRI, lack sensitivity. Cytological 722–726. examination of nipple fluid has proven to be unhelpful53, 2 Montroni I, Santini D, Zucchini G, Fiacchi M, Zanotti S, although Liu and colleagues47 reported slight improve- Ugolini G et al. Nipple discharge: is its significance as a risk ment in the diagnostic performance of ductoscopy when factor for breast cancer fully understood? Observational cytological findings were included. In addition, duc- study including 915 consecutive patients who underwent selective duct excision. Breast Cancer Res Treat 2010; 123: tal lavage fails to yield adequate specimens for reliable 895–900. 54 cytological diagnosis . However, the use of proteomic 3 Dillon MF, Mohd Nazri SR, Nasir S, McDermott EW, biomarkers in serum or methylation in nipple aspira- Evoy D, Crotty TB et al. The role of major duct excision tion fluid or ductal lavage appears to be a promising and microdochectomy in the detection of breast carcinoma. , alternative45 55–59. Imaging enhancement techniques such BMC Cancer 2006; 6: 164. autofluorescence might also assist in optimization of 4 Morrogh M, Morris EA, Liberman L, Borgen PI, King TA. diagnostic accuracy60,61. The predictive value of ductography and magnetic By the introduction of transductal intervention devices, resonance imaging in the management of nipple discharge. ductoscopy may be used both as a diagnostic technique Ann Surg Oncol 2007; 14: 3369–3377. 5 Kocdor MA, Sevinc AI, Canda T, Balci P, Saydam S, and to attain symptom resolution11,50. In a recent study11 Cavdaroglu O et al. Pathologic nipple discharge in patients from the present authors on interventional ductoscopy a with radiologically invisible mass: review of 28 consecutive therapeutic effect was found in 70 per cent of all successful sub-areolar explorations. Breast J 2009; 15: 230–235. ductoscopic lesion removals. In a subgroup of patients 6 Cabioglu N, Hunt KK, Singletary SE, Stephens TW, Marcy with negative findings on ductoscopy, nipple discharge S, Meric F et al. Surgical decision making and factors resolved spontaneously, suggesting either a self-limiting determining a diagnosis of breast carcinoma in women cause or a therapeutic effect of ductal saline lavage alone. presenting with nipple discharge. J Am Coll Surg 2003; Of the 82 patients in that study, 40 (49 per cent) no longer 7515: 354–364. experienced symptoms of PND with no need for surgery. 7 Gülay H, Bora S, Kìlìçturgay S, Hamaloglu˘ E, Göksel H. Management of nipple discharge. J Am Coll Surg 1994; 178: Thus, ductoscopy may be able to lead to a reduction in 471–474. surgical excision rates for a benign cause or symptom 8 Vargas HI, Vargas MP, Eldrageely K, Gonzalez KD, resolution. It is hypothesized that this would be of benefit Khalkhali I. Outcomes of clinical and surgical assessment of in terms of hospital costs, as an invasive procedure can women with pathological nipple discharge. Am Surg 2006; be omitted; however, a cost–benefit analysis is required to 72: 124–128. determine the extent of this potential benefit. 9 Seltzer MH. Breast complaints, biopsies, and cancer Currently, the use of ductoscopy as part of standard correlated with age in 10 000 consecutive new surgical care for patients presenting with PND is limited to a referrals. Breast J 2004; 10: 111–117. few medical centres. Ductoscopy has evolved over recent 10 Gray RJ, Pockaj BA, Karstaedt PJ. Navigating murky waters: a modern treatment algorithm for nipple discharge. decades, with additional techniques becoming available. Am J Surg 2007; 194: 850–854. Even though this development is ongoing, the literature on 11 Waaijer L, van Diest PJ, Verkooijen HM, Dijkstra NE, van ductoscopy remains limited and most clinicians are unfa- der Pol CC, Borel Rinkes IH et al. Interventional miliar with its use. However, ductoscopy, in its current ductoscopy reduces the amount of surgical procedures in form, is already a helpful technique for evaluating and patients with pathological nipple discharge. Br J Surg 2015; treating PND, and, as the number of patients presenting 102: 1639–1648.

12 Van Gelder L, Bisschops RH, Menke-Pluymers MB, 28 Whiting PF, Rutjes AWS, Westwood ME, Mallet S, Deeks Westenend PJ, Plaisier PW. Magnetic resonance imaging in JJ, Reitsma JB et al.; QUADAS 2 Group. QUADAS 2: a patients with unilateral bloody nipple discharge; useful when revised tool for the quality assessment of diagnostic accuracy conventional diagnostics are negative? World J Surg 2015; studies. Ann Intern Med 2011; 155: 529–536. 39: 184–186. 29 Makita M, Akiyama F, Gomi N, Ikenaga M, Yoshimoto M, 13 Mokbel K, Elkak A. The evolving role of mammary Kasumi F et al. Endoscopic classification of intraductal ductoscopy. Curr Med Res Opin 2002; 18: 30–32. lesions and histological diagnosis. Breast Cancer 2002; 9: 14 Okazaki A, Okazaki M, Asaishi K, Satoh H, Watanabe Y, 220–225. Mikami T et al. Fiberoptic ductoscopy of the breast: a new 30 Endoscopic classification of intraductal lesions. In diagnostic procedure for nipple discharge. Jpn J Clin Oncol Proceedings of 6th Annual Meeting of Japanese Association of 1991; 21: 188–193. Mammary Ductoscopy. Japanese Association of Mammary 15 Dawes LG, Bowen C, Venta LA, Morrow M. Ductography Ductoscopy, 2001. for nipple discharge: no replacement for ductal excision. 31 Okazaki A, Hirata K, Okazaki M, Svane G, Azavedo E. Surgery 1998; 124: 685–691. Nipple discharge disorders: current diagnostic management 16 Simmons R, Adamovich T, Brennan M, Christos P, Schultz and the role of fiber-ductoscopy. Eur Radiol 1999; 9: M, Eisen C et al. Nonsurgical evaluation of pathologic 583–590. nipple discharge. Ann Surg Oncol 2003; 10: 113–116. 32 Liberati A, Altman DG, Tetzlaff J, Mulrow C, Gøtzsche PC, 17 Sharma R, Dietz J, Wright H, Crowe J, Dinunzio A, John PA. The PRISMA statement for reporting systematic Woletz J et al. Comparative analysis of minimally invasive reviews and meta-analyses of studies that evaluate health microductectomy versus major duct excision in patients with care interventions: explanation and elaboration. JClin pathologic nipple discharge. Surgery 2005; 138: 591–597. Epidemiol 2009; 62: 1–34. 18 Teboul M. A new concept in breast investigation: 33 Fisher CS, Margenthaler JA. A look into the ductoscope: its echo-histological acino-ductal analysis or analytic role in pathologic nipple discharge. Ann Surg Oncol 2011; echography. Biomed Pharmacother 1988; 42: 18: 3187–3191. 289–295. 34 Hahn M, Fehm T, Solomayer EF, Siegmann KC, 19 Tang SSK, Twelves DJ, Isacke CM, Gui GP. Mammary Hengstmann AS, Wallwiener D et al. Selective ductoscopy in the current management of breast disease. microdochectomy after ductoscopic wire marking in women Surg Endosc 2011; 25: 1712–1722. with pathological nipple discharge. BMC Cancer 2009; 9: 20 Kapenhas-Valdes E, Feldman SM, Boolbol SK. The role of 151. mammary ductoscopy in breast cancer: a review of the 35 Simpson JS, Connolly EM, Leong WL, Escallon J, literature. Ann Surg Oncol 2008; 15: 3350–3360. McCready D, Reedijk M et al. Mammary ductoscopy in 21 Makita M, Akiyama F, Gomi N, Iwase T, Kasumi F, the evaluation and treatment of pathologic nipple Sakamoto G. Endoscopic and histologic findings of discharge: a Canadian experience. Can J Surg 2009; 52: intraductal lesions presenting with nipple discharge. Breast J E245–E248. 2006; 12: S210–S217. 36 Deshmane V. Intraductal approach to breast cancer: the role 22 Denewer A, El-Etribi K, Nada N, El-Metwally M. The role of mammary ductoscopy. Indian J Surg Oncol 2010; 1: and limitations of mammary ductoscope in management of 228–231. pathologic nipple discharge. Breast J 2008; 14: 442–449. 37 Dooley WC. Routine operative breast endoscopy for bloody 23 Louie LD, Crowe JP, Dawson AE, Lee KB, Baynes DL, nipple discharge. Ann Surg Oncol 2002; 9: 920–923. Dowdy T et al. Identification of breast cancer in patients 38 Ohlinger R, Stomps A, Paepke S, Blohmer JU, Grunwald S, with pathologic nipple discharge: does ductoscopy predict Hahndorf W et al. Ductoscopic detection of intraductal malignancy? Am J Surg 2006; 192: 530–533. lesions in cases of pathologic nipple discharge in comparison 24 Dietz JR, Crowe JP, Grundfest S, Arrigain S, Kim JA. with standard diagnostics: the German multicenter study. Directed duct excision by using mammary ductoscopy in Oncol Res Treat 2014; 37: 628–632. patients with pathologic nipple discharge. Surgery 2002; 39 Tekin E, Akin M, Kurukahvecioglu O, Tezcaner T, Gulen 132di: 582–588. M, Anadol AZ et al. The value of breast ductoscopy in 25 Kamali S, Bender O, Aydin MT, Yuney E, Kamali G. radiologically negative spontaneous/persistent nipple Ductoscopy in the evaluation and management of nipple discharge. Breast J 2009; 15: 329–332. discharge. Ann Surg Oncol 2010; 17: 778–783. 40 Albrecht C, Thele F, Grunwald S, Kohlmann T, 26 Moncrief RM, Nayar R, Diaz LK, Staradub VL, Morrow Hegenscheid K, Utpatel K et al. Nipple discharge: role of M, Khan SA. A comparison of ductoscopy-guided and ductoscopy in comparison with standard diagnostic tests. conventional surgical excision in women with spontaneous Onkologie 2013; 36: 12–16. nipple discharge. Ann Surg 2005; 241: 575–581. 41 Beechey-Newman N, Kulkarni D, Kothari A, D’Arrlgo C, 27 Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred Culora G, Hamed H et al. Breast duct microendoscopy in reporting items for systematic reviews and meta-analyses: nipple discharge: microbrush improves cytology. Surg Endosc the PRISMA statement. BMJ 2009; 339: b2535. 2005; 19: 1648–1651.

42 Sauter ER, Ehya H, Klein-Szanto AJP, Wagner-Mann C, 52 Kapenhas-Valdes E, Feldman SM, Cohen JM, Boolbol SK. MacGibbon B. Fiberoptic ductoscopy findings in women Mammary ductoscopy for evaluation of nipple discharge. with and without spontaneous nipple discharge. Cancer Ann Surg Oncol 2008; 15: 2720–2727. 2005; 103: 914–921. 53 Kooistra BW, Wauters C, van de Ven S, Strobbe L. The 43 Khan SA, Mangat A, Rivers A, Revesz E, Susnik B, Hansen diagnostic value of nipple discharge cytology in 618 N. Office ductoscopy for surgical selection in women with consecutive patients. Eur J Surg Oncol 2009; 35: 573–577. pathologic nipple discharge. Ann Surg Oncol 2011; 18: 54 Loud JT, Thiébbaut AC, Abati AD, Filie AC, Danforth D, 3785–3790. Giusti R et al. Ductal lavage in women from BRCA 1/2 44 Kamali S, Harman GK, Akan A, Sim¸sekS,BenderO.Useof families: is there a future for ductal lavage in women at ductoscopy as an additional diagnostic method and its increased genetic risk of breast cancer? Cancer Epidemiol applications in nipple discharge. Minerva Chir 2014; 69: Biomarkers Prev 2010; 18: 1243–1251. 65–73. 55 Antill YC, Mitchell G, Johnson SA, Devereux L, Milner A, 45 Fackler MJ, Rivers A, Teo WW, Mangat A, Taylor E, Zhang Di Iulio J et al. Gene methylation in breast ductal fluid from Z et al. Hypermethylated genes as biomarkers of cancer in BRCA1 and BRCA2 mutation carriers. Cancer Epidemiol women with pathologic nipple discharge. Clin Cancer Res Biomarkers Prev 2010; 19: 265–274. 2009; 15: 3802–3811. 56 Suijkerbuijk KPM, van der Wall E, Vooijs M, van Diest PJ. 46 Bender O, Balci FL, Yüney E, Akbulut H. Scarless Molecular analysis of nipple fluid for breast cancer endoscopic papillomectomy of the breast. Onkologie 2009; screening. Pathobiology 2008; 75: 149–152. 32: 94–98. 57 De Noo ME, Deelder A, van der Werff M, Ozalp A, 47 Liu GY, Lu JS, Shen KW, Wu J, Chen CM, Hu Z et al. Mertens B, Tollenaar R. MALDI-TOF serum protein Fiberoptic ductoscopy combined with cytology testing in profiling for the detection of breast cancer. Onkologie 2006; the patients of spontaneous nipple discharge. Breast Cancer 29: 501–506. Res Treat 2008; 108: 271–277. 58 Mertens BJ, De Noo ME, Tollenaar RA, Deelder AM. Mass 48 Matsunaga T, Misaka T, Hosokawa K, Taira S, Kim K, spectrometry proteomic diagnosis: enacting the double Serizawa H et al. Intraductal approach to the detection of cross-validatory paradigm. J Comput Biol 2006; 13: intraductal lesions of the breast. Breast Cancer Res Treat 1591–1605. 2009; 118: 9–13. 59 Suijkerbuijk KP, van Diest PJ, van der Wall E. Improving 49 Ling H, Liu G, Lu J, Love S, Zhang J, Xu X et al. Fiberoptic early breast cancer detection: focus on methylation. Ann ductoscopy-guided intraductal biopsy improve the diagnosis Oncol 2011; 22: 24–29. of nipple discharge. Breast J 2009; 15: 168–175. 60 Jacobs VR, Paepke S, Schaaf H, Weber BC, Kiechle-Bahat 50 Balci FL, Feldman SM. Interventional ductoscopy for M. Autofluorescence ductoscopy: a new imaging technique pathological nipple discharge. Ann Surg Oncol 2013; 20: for intraductal breast endoscopy. Clin Breast Cancer 2007; 7: 3352–3354. 619–623. 51 Shen KW, Wu J, Lu JS, Han QX, Shen ZZ, Nguyen M et al. 61 Douplik A, Leong WL, Easson AM, Done S, Netchev G, Fiberoptic ductoscopy for breast cancer patients with nipple Wilson BC. Feasibility study of autofluorescence mammary discharge. Surg Endosc 2001; 15: 1340–1345. ductoscopy. J Biomed Opt 2009; 14: 044036.

Supporting information

Additional supporting information may be found in the online version of this article: Appendix S1 Search syntax for ductoscopy in the PubMed, Embase and Cochrane Library databases (Word document) Appendix S2 Quality assessment of the included studies (Word document) Table S1 Quality assessment: risk of bias and level of concern regarding the applicability of the included articles (Word document) Table S2 Categories of index test (ductoscopy) and reference standard (histology), and reclassification used for

analysis of sensitivity and specificity of DSany and DSsusp (Word document)