Diagnostic accuracy of transvaginal ultrasound for non-invasive diagnosis of bowel endometriosis: systematic review and meta-analysis

Authors


Abstract

Objective

To critically analyze the diagnostic value of transvaginal sonography (TVS) for non-invasive, presurgical detection of bowel endometriosis.

Methods

MEDLINE (1966–2010) and EMBASE (1980–2010) databases were searched for relevant studies investigating the diagnostic accuracy of TVS for diagnosing deep infiltrating endometriosis involving the bowel. Diagnosis was established by laparoscopy and/or histopathological analysis. Likelihood ratios (LRs) were recalculated in addition to traditional measures of effectiveness.

Results

Out of 188 papers, a total of 10 studies fulfilled predefined inclusion criteria involving 1106 patients with suspected endometriosis. The prevalence of bowel endometriosis varied from 24 to 73.3%. LR+ ranged from 4.8 to 48.56 and LR− ranged from 0.02 to 0.36, with wide confidence intervals. Pooled estimates of sensitivities and specificities were 91 and 98%; LR+ and LR− were 30.36 and 0.09; and positive and negative predictive values were 98 and 95%, respectively. Three of the studies used bowel preparations to enhance the visibility of the rectal wall; one study directly compared the use of water contrast vs. no prior bowel enema, for which the LR− was 0.04 and 0.47, respectively.

Conclusions

TVS with or without the use of prior bowel preparation is an accurate test for non-invasive, presurgical detection of deep infiltrating endometriosis of the rectosigmoid. Copyright © 2011 ISUOG. Published by John Wiley & Sons, Ltd.

Introduction

Over the past decade, the use of transvaginal sonography (TVS) has improved the quality of non-invasive assessment of patients with suspected pelvic pathologies. With respect to endometriosis TVS has been shown to be a highly sensitive tool for the detection of ovarian endometriomas1 and is far superior to routine clinical examination alone2, 3. Moore et al.1 systematically reviewed 67 papers on the validity of TVS for the detection of pelvic endometriosis, out of which seven fulfilled the inclusion criteria and focused on TVS imaging of ovarian endometriomas. The prevalence of the condition ranged between 13 and 38%. Sensitivities, specificities and positive (LR+) and negative likelihood ratios (LR−) in six studies using gray-scale ultrasonography ranged between 64 and 89%, 89 and 100%, 7.6 and 29.8 and 0.1 and 0.4, respectively. The authors therefore concluded that TVS should be regarded as a useful test for identifying cystic ovarian endometriosis presurgically.

Recent studies also suggest that TVS could be an accurate method for the detection of endometriosis in extra-ovarian locations, i.e. uterosacral ligament involvement, endometriosis of the rectovaginal space, the pouch of Douglas, the vagina, the urinary bladder and deep infiltrating endometriosis (DIE) of the rectosigmoid3–8. Since TVS is a readily available, cost- and time-effective diagnostic instrument when compared to other radiological procedures such as computed tomography and magnetic resonance imaging (MRI)9, 10, several investigators have further examined the diagnostic value of TVS for the non-invasive detection of DIE infiltrating the bowel. The aim of this work was to systematically analyze the published literature evaluating the role of TVS for the detection of DIE involving the rectosigmoid.

Methods

The MEDLINE (1966–2010) and EMBASE (1980–2010) databases were searched using the following search strategy:

  • 1.(pelvic or ovarian or deep infiltrating) near2 (mass or cyst* or tumo*r)
  • 2.‘ENDOMETRIOSIS’ in MeSH or 1 or ‘BOWEL ENDOMETRIOSIS’
  • 3.2, not case reports, not review articles
  • 4.with checktags ‘female’ and ‘human’
  • 5.with ‘ULTRASOUND’/all subheadings or ‘TRANSVAGINAL’/all subheadings or ‘SONOGRAPHY’

Abstracts of all studies identified were read and manuscripts were then fully reviewed. In addition, reference lists of all reviewed manuscripts were searched for additional data. Study selection and assessment of quality were performed independently by two reviewers (G. H. and J. E).

Selection criteria

All studies included in the present review had to be prospective and were required to involve both TVS examination and surgical exploration of the pelvis either by laparoscopy or by laparotomy (as stated by Moore et al.1). Scientific publications including case reports, studies on adenomyosis or extrapelvic endometriotic disease as well as retrospective case series and review articles were excluded. Studies reporting on pregnant women, rectal ultrasound as the only examination and endoscopic sonography were also excluded from this review. Patients included in the studies presented with either subfertility or symptoms suggestive of endometriosis.

According to the criteria of Moore et al.1, studies were considered to be of good quality when information on recruitment of patients, blinding of ultrasound operators and surgeons and data on the technical equipment were provided. In order to define the stage and severity of disease (i.e. the final endpoint of diagnosis), studies had to describe the anatomical location of deep infiltrating disease combined with histological confirmation of endometriosis. Moore et al.1 considered that studies missing one or two of these criteria were of moderate/poor quality.

Data extraction and presentation of results

As described by Moore et al.1, 2 × 2 tables were created to validate test results against surgical and histopathological findings aimed at defining whether DIE involving the rectosigmoid can be detected by TVS. In addition, QUADAS (quality assessment of diagnostic accuracy of studies) was used to assess the studies11. As described previously, study quality is defined as high when ≥ 9 items out of 14 are met, moderate when 6 items are met and low when < 6 criteria are met12. LR+, LR− and test accuracy were calculated in studies lacking these data. Confidence intervals (CIs) were calculated as described previously1 using CATMaker® statistical software (Centre for Evidence-Based Medicine, Oxford, UK). In two cases raw data (true and false positive and negative rates) were obtained from the authors. Since heterogeneity is a common finding in diagnostic meta-analyses we calculated Cochran's Q and I2 for all measures to assess the significance and magnitude of study heterogeneity13. A forest plot was used to assess eventual outlier studies (data not shown). Potential sources of heterogeneity were explored by random-effects meta-regression. Number of subjects, year of publication, QUADAS scores and prevalence were used as study-level covariates that predicted the logarithm of diagnostic odds ratios (DORs). The studies were weighted by the inverse of variance of the DOR to consider the precision with which each study measured it. Relative DORs were calculated to compare the overall DOR with the adjusted DOR. When heterogeneity between studies is present, a random-effects model can be used to obtain pooled estimates14.

We applied random-effects models with the DerSirmonian–Laird estimator in order to determine overall estimates of sensitivity, specificity, LR+, LR− and DOR. For data with zero counts a continuity correction of 0.5 was added to every value in that study, thereby allowing the calculation of all LRs15. For the assessment of publication bias Egger's test and Begg's test were conducted and funnel plots were investigated (data not shown). The tests for publication bias and funnel plots were performed with R statistical software version 2.11.1 metafor package (R Development Core Team, Vienna, Austria)16. All other analyses were performed with MetaDiSc statistical software version 1.417 (Hospital Universitario, Madrid, Spain).

Results

The initial implementation of the research strategy revealed 188 studies relating to endometriosis and/or adenomyosis and/or ovarian endometriosis diagnosed by laparoscopy or laparotomy and/or ultrasonography. Out of these only 51 papers specifically used TVS and surgical exploration to diagnose DIE. Of these 51 papers, seven were excluded because they were case reports or descriptive in nature. A further 18 papers did not meet the inclusion criteria due to the fact that they were review articles, despite the exclusion of this article type in the primary search process. Finally, three other publications included comments on publications and were also excluded from the final analysis, leaving 23 manuscripts for review3–5, 7, 8, 18–35. Out of these 23 papers, 13 publications were excluded due to methodological problems; three papers were purely retrospective in nature18, 27, 28; four manuscripts were excluded because they involved a group of patients already described in a previous publication by the same group of authors5, 7, 25, 29 and another four papers were not included in the analysis because insufficient information was provided on the presence or absence of bowel endometriosis within the group of women with endometriosis22, 30, 32, 34. Two papers were excluded due to the low number of cases with proven bowel endometriosis24 or insufficient information on the anatomical localization of DIE affecting the bowel35. Thus 10 papers fulfilled the inclusion criteria and were included in the final review. In addition to the quality criteria described by Moore et al.1, the QUADAS scores were calculated for each of the studies. These ranged from 7 to 13, reflecting high methodological quality in eight cases and moderate quality in two cases.

Out of these, four papers evaluated the diagnostic value of TVS for the detection of DIE infiltrating the rectosigmoid excluding other possible locations of DIE in the analysis8, 23, 31, 33. All other works also evaluated the potential of TVS for the visualization of other affected sites of deep infiltrating disease such as the vagina, rectovaginal space, uterosacral ligaments, bladder and pouch of Douglas. Two papers addressed the use of TVS to diagnose DIE involving the rectum, pouch of Douglas, vagina, rectovaginal space, uterosacral ligaments and bladder3, 26 using laparoscopic exploration and histological confirmation of endometriosis as the gold standard test. One work compared TVS with MRI to detect DIE preoperatively compared to the gold standard laparoscopy without histological confirmation21. One paper assessed the diagnostic value of TVS vs. rectal endosonography19; two papers compared TVS with MRI, digital examination4 and rectal endosonography20 to detect DIE presurgically. Data on the quality of these studies including number of patients and cases of bowel endometriosis, recruitment criteria, blinding, information and presence of diagnostic criteria and technical details, data on reference standards and assessment of the grade of study quality are provided in Table 1. Table 2 depicts prevalence rates, sensitivities, specificities, test accuracies, positive predictive values (PPVs), negative predictive values (NPVs) and positive and negative LRs with CIs of all studies included in the final analysis. The number of patients with proven DIE infiltrating the bowel undergoing preoperative TVS ranged from 1721 to 8123. Sensitivity and specificity varied from 67 to 98% and 92 to 100%, respectively.

Table 1. Characteristics of the studies included in the analysis
StudyDiagnostic criteria stated; ultrasound criteria for diagnosisStudy designBlindingRecruitment; cause for referralPatients with suspected endometriosis (n)Cases of rectal/sigmoidal endometriosis (n)Test methodReference standardStudy quality according to Moore criteria (QUADAS score)
  1. MRI, magnetic resonance imaging; PV, per vaginam clinical examination; QUADAS, quality assessment of diagnostic accuracy of studies; RES, rectal endosonography; RWC, rectal water contrast; TVS, transvaginal sonography.

Bazot et al. (2003)19Yes; thickening of the muscularis propria > 3 mm, which is hypoechoic and thin (< 3 mm) under normal circumstancesProspectiveSonographers blindedConsecutive; pain and infertility3022TVS, RESHistologyModerate (9)
Bazot et al. (2004)3Yes; as aboveProspectiveSonographer blindedConsecutive; pain and infertility14247TVSHistology in 29Moderate (8)
Carbognin et al. (2006)21Not describedProspectiveNot statedConsecutive; pain and infertility3217TVS, MRILaparoscopic visualizationPoor (7)
Abrao et al. (2007)4Yes; nodular, predominantly solid, hypoechogenic lesion adhered to the wall of the intestinal loopProspectiveSonographer blindedConsecutive; pain and infertility10454TVS, MRI, PVHistologyModerate (10)
Valenzano Menada et al. (2008)33Yes; rectovaginal hypoechoic mass adherent and/or penetrated into the intestinal wall thickening the muscularis mucosaProspectiveSonographers blindedConsecutive; pain and infertility9023TVS, RWC-TVSHistologyGood (12)
Guerriero et al. (2008)26Yes; presence of some thin band-like echoes departing from the center of the mass as an ‘Indian head dress’.ProspectiveNo information givenConsecutive; pain and infertility8839TVSHistologyGood (11)
Piketty et al. (2009)31Yes; irregular hypoechoic mass, with or without hypo/hyperechoic foci involving colon muscularisProspectiveSonographer blindedConsecutive; pain and infertility13475TVSHistologyGood (12)
Bazot et al. (2009)20Yes; as aboveProspectiveRadiologists blindedConsecutive; pain and infertility9263TVS, MRI, PV, RESHistology in 54Good (12)
Hudelist et al. (2009)8Yes; as aboveProspectiveSonographer not blinded to PVConsecutive; pain and infertility20048TVS, PVHistologyModerate (13)
Goncalves et al. (2010)23Yes; long, nodular, predominantly solid, hypoechogenic lesion adhered to the wall of the intestinal loopProspectiveSonographer blindedConsecutive; pain and infertility19481TVSHistologyGood (12)
Table 2. Analysis of data in the included studies
StudyPrevalence of rectal/sigmoidal endometriosis (n (%))Sensitivity (n (%))Specificity (n (%))PPV (%)NPV (%)Accuracy (%)LR+ (95% CI)LR− (95% CI)
  1. LR+, positive likelihood ratio; LR−, negative likelihood ratio; NPV, negative predictive value; PPV, positive predictive value.

Bazot et al. (2003)1922/30 (73)21/22 (95)8/8 (100)10089970.05 (0.01–0.31)
Bazot et al. (2004)347/142 (33)41/47 (87)92/95 (97)93949427.62 (9.02–84.58)0.13 (0.06–0.28)
Carbognin et al. (2006)2117/32 (53)12/17 (71)15/15 (100)10075844.8 (1.26–18.31)0.29 (0.14–0.61)
Abrao et al. (2007)454/104 (52)53/54 (98)50/50 (100)10098990.02 (0.00–0.13)
Valenzano Menada et al. (2008)3323/90 (26)22/23 (96)67/67 (100)10099990.04 (0.01–0.3)
Guerriero et al. (2008)2639/88 (44)26/39 (67)45/49 (92)8778818.17 (3.11–21.44)0.36 (0.23–0.57)
Piketty et al. (2009)3175/133 (56)68/75 (91)56/58 (97)97899326.29 (6.72–102.83)0.10 (0.05–0.20)
Bazot et al. (2009)2063/92 (68)59/63 (94)29/29 (100)10088960.06 (0.02–0.16)
Hudelist et al. (2009)848/200 (24)46/48 (96)149/152 (98)94999848.56 (15.81–149.10)0.04 (0.01–0.17)
Goncalves et al. (2010)2381/194 (42)79/81 (98)113/113 (100)10098990.02 (0.01–0.10)

As further demonstrated in Table 2, PPVs and NPVs ranged from 87 to 100% and 75 to 99%, respectively. LR+ goes up to infinity in cases lacking false-positive findings and thus are not shown. LR+ of the remaining studies varied from 4.8 to 48.56 and LR− from 0.02 to 0.36.

Three studies used TVS combined with a bowel enema to provide a better visualization of rectal wall anatomy, specifically muscularis propria4, 23, 33. Valenzano Menada et al.33 directly compared whether adding water-contrast in the rectosigmoid (RWC-TVS) during TVS improves presurgical diagnosis of rectal DIE with TVS. The sensitivity of RWC-TVS vs. TVS in identifying rectal DIE was 97 vs. 56%, the specificity 100 vs. 92.5%, the PPV 100 vs. 72% and the NPV 99 vs. 86%. Due to the absence of false-positive cases in the RWC-TVS cohort, LR+ could not be calculated, and LR− was 0.04 vs. 0.47 for TVS.

Meta-analysis of all studies included in the final review yielded significant results of Cochran's Q for all measures except LR+ (sensitivity: P < 0.001, Q = 238.64, df = 9, I2 = 76.7%; specificity: P = 0.037, Q = 17.8, df = 9, I2 = 49.5%; LR+: P = 0.144, Q = 13.4, df = 9, I2 = 33.0%; LR−: P < 0.001, Q = 52.6, df = 9, I2 = 82.9%; DOR: P = 0.001, Q = 27.6, df = 9, I2 = 67.4%), indicating considerable heterogeneity between studies. Significant values for study heterogeneity were mainly caused by two studies21, 26 that were aberrant in respect to sensitivity, LR− and DOR. However, meta-regression including sample size, prevalence, QUADAS and year of publication did not yield any significant results: number of subjects (P = 0.197, relative DOR = 1.02 (95% CI, 0.99–1.04)), prevalence (P = 0.568, relative DOR = 0.97 (95% CI, 0.88–1.08)), QUADAS score (P = 0.319, relative DOR = 1.44 (95% CI, 0.65–3.20)) and year of publication (P = 0.298, relative DOR = 1.35 (95% CI, 0.67–2.71)). This suggests that although there was relevant heterogeneity between the studies, the influence of the covariates was not systematic. To account for heterogeneity we used a random-effects model to perform pooled estimates and estimate the respective CIs (Table 3). As all studies were conducted with women who suffered from pain or infertility, the pooled prevalence of 47% refers to women with specific symptoms.

Table 3. Overall analysis of all studies included in the final analysis using a random-effects model to perform pooled estimates of variables
VariableEstimate (95% CI)
  1. DOR, diagnostic odds ratio (values not adjusted); LR+, positive likelihood ratio (with continuity correction for studies with null-cells); LR−, negative likelihood ratio; NPV, negative predictive value; PPV, positive predictive value.

Sensitivity (%)91 (88.1–93.5)
Specificity (%)98 (96.7–99.0)
LR+30.36 (15.457–59.626)
LR−0.09 (0.046–0.188)
DOR394.3 (116.3–1336.0)
Prevalence (%)47 (36.7–57.3)
PPV (%)98 (96.7–99.6)
NPV (%)95 (92.1–97.7)

Evaluation by both Egger's test (P = 0.221) and Begg's test (P = 0.293) did not show evidence of publication bias for logDOR. This result was confirmed by inspection of the funnel plots, which were all symmetrical for the investigated diagnostic measures (sensitivity, specificity, PPV, NPV, LR+, LR− and accuracy; data not shown).

Discussion

Endometriosis infiltrating the rectosigmoid can be suspected in up to 9–22% of all women with proven endometriosis36, 37. Symptoms of DIE involving the bowel vary greatly, ranging from asymptomatic women with extensive rectal involvement to patients with severe dysmenorrhea and dyschezia38. Treatment strategies include hormonal preparations or surgical excision of endometriotic nodules37, 39, but presurgical staging of DIE is crucial for planning surgical treatment options. The findings of our systematic review clearly suggest that TVS is a highly valuable tool for the non-invasive detection of DIE affecting the rectosigmoid. In addition to sensitivities, specificities, PPVs, NPVs and test accuracies, we recalculated all positive and negative LRs since these reflect the diagnostic accuracy and the clinical usefulness of a test independently of the prevalence of the study condition in the study population1. The prevalence of DIE involving the bowel ranged from 24 to 73%, which may be attributable to different referral patterns and the availability of tertiary referral centers providing sufficient expertise in the presurgical diagnosis and treatment of rectal DIE. LR+ ranged from 4.821 to 27.623 and LR− from 0.0223 to 0.3626. The combined use of TVS with vaginal examination may further increase the diagnostic accuracy, with reported LR+ and LR− 48.56 and 0.048, respectively. In addition, it should be noted that LR+ went up to infinity in five studies (Table 2), suggesting highly accurate presurgical test results.

Performance of a meta-analysis of all included studies by using a random-effects model to calculate pooled estimates revealed LR+ and LR− of 30.36 and 0.09, respectively. According to Altman40, a very helpful test for establishing or excluding a condition is characterized by an LR + > 10 and an LR − < 0.1, while a test is regarded as moderately helpful with an LR+ between 5 and 10 and an LR− between 0.1 and 0.2. As depicted in Table 3, the results of our meta-analysis clearly suggest that TVS is indeed very useful for sonographic diagnosis but also presurgical exclusion of bowel endometriosis. Two studies21, 26 were aberrant with respect to sensitivity, LR− and DOR. This may be attributable to low sensitivity values in association with small patient numbers21 or the use of ‘tenderness guided’ transvaginal sonography26, which may be less sensitive in cases of endometriotic involvement of the upper rectum/lower sigmoid, since these locations may not appear as painful sites when using this technique.

The conclusions of this review are weakened by the fact that blinding of the surgeon was missing in all studies included in the final analysis, which might have a potential influence on the results of the gold standard test, i.e. surgery. On the other hand, information on the preoperative findings is essential for guiding the surgical technique and surgical diagnosis of endometriosis in everyday clinical practice.

An additional variable in the accuracy of the gold standard test, i.e. laparoscopy, is the fact that several studies did not provide sufficient information about the surgical exploration of the patients with DIE3, 20, 21, 26 since complete cul-de-sac obliteration secondary to endometriosis was reported but could not necessarily be surgically cleared. This may, in cases of DIE affecting the non-visible, occluded lower rectum, lead to misinterpretation and observation bias of the gold standard test and consequently negatively affect the validation of the index test, i.e. TVS. Hence, full surgical exploration of patients with an occluded pouch of Douglas is important for the final and accurate diagnosis of bowel endometriosis since DIE may be missed at the time of laparoscopy, thereby questioning the quality of laparoscopic visualization of the pelvis as the gold standard test for the diagnosis of endometriosis.

Another limiting factor to the results of this systematic review is the heterogeneity of populations included in the studies. Some authors failed to provide sufficient data on the selection criteria and referral patterns of centers where studies were conducted. Due to the fact that most patients were treated in tertiary referral centers, it is conceivable that the study populations represent an already preselected patient cohort with a high prevalence of bowel endometriosis. As a consequence, the conclusions of this review may therefore only be applicable to a tertiary referral setting and patients with a high risk for DIE.

In clinical everyday practice, exclusion of DIE is of critical importance since extensive bowel involvement warrants an interdisciplinary approach and referral to a tertiary center. In conclusion, the majority of the published evidence suggests that TVS is a highly useful and easy accessible test for the preoperative detection of DIE infiltrating the rectosigmoid. Whether the widespread use of TVS and the inclusion of TVS in specialist training programs lead to a reduction in the diagnostic delay of patients with endometriosis remains to be seen.

Acknowledgements

The authors want to thank Dr Simone Ferrero and MMag Nadja Fritzer for their support and helpful advice. This work was supported by the OEGEO, Österreichische Gesellschaft für Endokrinologische Onkologie.

Ancillary