CURRENT STATUS

Restaging of Locally Advanced Rectal Cancer With Magnetic Resonance Imaging and Endoluminal Ultrasound After Preoperative Chemoradiotherapy: A Systemic Review and Meta-analysis Ri-Sheng Zhao, M.D.1 • Hui Wang, M.D., Ph.D.1 • Zhi-Yang Zhou, M.D., Ph.D.2 Qian Zhou, M.P.H.3 • Michael W. Mulholland, M.D., Ph.D.4 1 Department of Colorectal Surgery, The Sixth Affiliated Hospital, Sun Yat-sen University, Guangzhou, Guangdong, China 2 Department of Radiology, The Sixth Affiliated Hospital, Sun Yat-sen University, Guangzhou, Guangdong, China 3 School of Public Health, Sun Yat-sen University, Guangzhou, Guangdong, China 4 Department of Surgery, University of Michigan, Ann Arbor, Michigan

BACKGROUND:  Magnetic resonance imaging and endoluminal ultrasound play an important role in the restaging of locally advanced rectal cancer after preoperative chemoradiotherapy, yet their diagnostic accuracy is still controversial. OBJECTIVE:  Meta-analysis was performed to estimate the diagnostic performance of MRI and endoluminal ultrasound. DATA SOURCES:  Electronic databases from 1996 to

March 2012 were searched. STUDY SELECTION AND INTERVENTIONS:  Either MRI or endoluminal ultrasound was used to restage rectal cancer after chemoradiotherapy or radiation. MAIN OUTCOME MEASURES:  T category, lymph node, and circumferential resection involvement were measured. RESULTS:  The sensitivity estimate for rectal cancer diagnosis (T0) by endoluminal ultrasound (37.0%; 95% CI, 24.0%–52.1%) was higher (p = 0.04) than the sensitivity estimate for MRI (15.3%; 95% CI, 6.5%–32.0%). For T3-4 category, sensitivity estimates of MRI and endoluminal ultrasound were comparable, Financial Disclosure: None reported. Correspondence: Hui Wang, M.D., Ph.D., Department of Colorectal Surgery, The Sixth Affiliated Hospital, Sun Yat-sen University, 26 Yuancun Erheng Rd, Guangzhou, Guangdong 510655, China. E-mail: [email protected]. Dis Colon Rectum 2014; 57: 388–395 DOI: 10.1097/DCR.0000000000000022 © The ASCRS 2014

388

82.1% and 87.6%, whereas specificity estimates were poor (53.5% and 66.4%). For lymph node involvement, there was no significant difference between the sensitivity estimates for MRI (61.8%) and endoluminal ultrasound (49.8%). Specificity estimates for MRI and endoluminal ultrasound were 72.0% and 78.7%. For circumferential resection margin involvement, MRI sensitivity and specificity were 85.4% and 80.0%. LIMITATIONS:  To identify the heterogeneity, metaregression was performed on covariates. However, few of the covariates were identified to be statistically significant because of the lack of adequate original data. CONCLUSION:  Accurate restaging of locally advanced rectal cancer by MRI and endoluminal ultrasound is still a challenge. Identifying T0 rectal cancer by imaging is not reliable. Before performing surgery, restaging is important, but some of the T0-2 patients are likely overestimated as T3-4. Both modalities for lymph node involvement are not very good. Magnetic resonance imaging may be a good method to reassess circumferential resection margin.

KEY WORDS:  Rectal cancer; Chemoradiotherapy; Magnetic resonance imaging; Endoluminal ultrasound; Restaging; Diagnostic accuracy.

P

erformance of preoperative chemoradiation is purported to cause tumor downsizing and thus promote a less radical surgical strategy.1–5 Accurate restaging is of great importance in assigning patients with locally advanced rectal cancer (LARC) after preoperative chemoradiotherapy (CRT) to appropriate treatment. Local excision Diseases of the Colon & Rectum Volume 57: 3 (2014)

389

Diseases of the Colon & Rectum Volume 57: 3 (2014)

has been recommended by some surgeons for patients with T2-3 rectal cancers who achieve complete response after CRT, with the claim that this approach has less morbidity and mortality but similar recurrence rates and overall survival.6–8 Furthermore, some authors have suggested that, combined with salvage surgery, nonoperative management appears to achieve similar local and distant disease control in comparison with patients with a pathological complete response treated by rectal resection.9 Most surgeons restaged patients following CRT to predict complete pathological response, to identify mesorectal margins and thus inform operative approach, to document local response to therapy and improve the prediction of long-term survival outcomes,10–12 and to identify interval development of metastatic disease for patients who might forgo surgery and opt for palliative chemotherapy. A number of studies have focused on the accuracy of restaging LARC after CRT by MRI and endoluminal ultrasound (EUS). However, the conclusions from studies varied. In addition, comparative evaluations of the 2 imaging modalities are limited. The aim of this study is to perform a meta-analysis of comparative diagnostic performance of the 2 imaging modalities when used for restaging LARC after CRT.

were presented to construct a 2 × 2 contingency table containing true-positive, false-positive, false-negative, and true-negative absolute numbers for each study. Studies were excluded if data were published more than once; the study with the most details or the most patients was included. Data Extraction

Data were independently extracted by 2 reviewers who used a standardized data extraction form. Discrepancies were resolved by mutual agreement, and a third reviewer was asked to assess if necessary. For studies in which separate results were provided for multiple independent readers, results of each reader for each outcome were included as a separate data set. Two-by-two contingency tables were extracted or reconstructed for each outcome as follows: T category (T3-4 vs T0-2 or T0 vs T1-4), lymph nodes (N+ vs N−), or circumferential resection margin (CRM) (involvement vs clear). The Quality Assessment of Diagnostic Accuracy Studies tool was used to assess the methodological quality of studies included in this meta-analysis.13 Each of the total of 14 quality items was scored as “yes,” “no,” or “unclear.” Statistical Analysis

MATERIALS AND METHODS Literature Search

A comprehensive computerized systematic literature search was performed to identify the abstracts of articles from studies concerning the diagnostic performance of MRI or EUS in the restaging of LARC after preoperative CRT or radiation alone. The PubMed, Embase, OVID, and Web of Knowledge databases (all from January 1996 to March 2012) were searched with a search strategy similar to PubMed: 1) rectal cancer; 2) staging; 3) chemoradiotherapy; 4) MRI, EUS, and their corresponding related terms; and 5) diagnostic search filter (sensitivity, specificity, accuracy, false negative, false positive, likelihood functions). Reference lists of articles retrieved were also searched manually to identify relevant studies. Reviews, letters, comments, case reports, and articles that did not present raw data were excluded. Study Selection

Studies were selected if they fulfilled all of the following inclusion criteria: 1) More than 20 patients had biopsy-proven LARC with preoperative CRT or ra­ diotherapy (RT). 2) The study involved 1 or 2 of MRI (nonenhanced and phased-array coils) and EUS. 3) Histopathological findings (specimens obtained at surgery) were used as the reference standard, which was compared with the corresponding imaging evaluation. 4) The study was reported in the English language. 5) Sufficient data

Sensitivity, specificity, and diagnostic odds ratios (DORs) with their 95% CIs were estimated by a bivariate r­ andom-effects model in this meta-analysis. The DOR is a summary estimate of diagnostic accuracy, incorporating both sensitivity and specificity. The DOR does not depend on the prevalence of the disease. The DOR of a diagnostic test is the ratio of the odds of positivity in patients with disease relative to the odds of positivity in patients without disease. The value of a DOR ranges from 0 to infinity, with higher values indicating better discriminatory test performance. Bivariate random-effects meta-analysis on sensitivities and specificities are currently one of the main methods recommended synthesizing diagnostic test accuracy studies.14 This model allows for the correlation that can exist between pairs of sensitivity and specificity within a study and also takes into account the heterogeneity beyond chance between studies. Moreover, this model can elicit any significant differences that exist between 2 modalities for diagnosing each outcome. The summary receiver-operating characteristic (ROC) curves were fitted to calculate the area under the ROC curve (AUC), because they summarize the diagnostic performance as a single number: 0.9–1.0 suggested an excellent diagnostic accuracy; 0.8–0.9 suggested a very good diagnostic accuracy; 0.7–0.8 suggested a good diagnostic accuracy; 0.6–0.7 suggested a sufficient diagnostic accuracy; 0.5–0.6 suggested the diagnostic accuracy was bad; and 0.5 (or lower) suggested the diagnostic test was not useful.15

390

Receiver-operating characteristic plane and the Spearman correlation coefficient between sensitivity and specificity can account for the threshold effect in diagnostic studies, which occurs when studies use different cutoff points or thresholds to define a positive or negative test result. To account for heterogeneity, covariate analysis and metaregression were performed. If heterogeneity did exist, the DOR and AUC were estimated within the subgroups of studies according to the metaregression method proposed by Moses et al.16 All the statistical analyses were performed with the use of SAS statistical software (version 9.1, SAS Institute, Inc, Cary, NC) and Meta-DiSc (version 1.4) software packages.

Zhao et al: Accuracy of Imaging in Restaging Rectal Cancer

Initial search gave 1116 titles and abstracts 333 were duplicated; 663 did not examine restaging; 63 were reviewed 57 potentially eligible full texts were assessed 8 potential eligible full texts were identified from reference lists

RESULTS

24 studies were included in this meta-analysis

Search Results and Study Selection

After data extraction, 24 articles met the inclusion criteria and were used for this meta-analysis. Study identification and inclusion were shown in a flowchart as Figure 1. Data Extraction

From the 24 included articles, 1201 patients and 59 data sets were included. Among these, 739 patients and 34 data sets were for MRI assessment, and 651 patients and 25 data sets were for EUS assessment. Grouping the articles according to investigated imaging modality, 15 articles studied the performance of MRI, and 11 articles studied the performance of EUS. All studies used histopathology after surgery as a reference standard. The characteristics of included studies are summarized in Table 1. According to the Quality Assessment of Diagnostic Accuracy Studies tool, the methodological quality of each study eligible for this meta-analysis is summarized in ­Table 2. All studies satisfied 6 to 13 of 14 terms, averaging 9. All the included patients had LARC with a long course of CRT or RT alone and afterward received histological validation. All withdrawals from the studies were explained. However, poor reporting of details concerning several quality items happened in most studies, such as blindness between imaging and histopathology or clinical data available to the radiologist.

Other language (6); not relevant (8); less than 20 patients (4); abstract only (4); not TNM staging (2); not enough data to construct 2×2 table (17)

FIGURE 1.  Flowchart describing the identification and inclusion of studies.

95% CI, 6.5%–32.0%), but the difference of specificities (93.9% by EUS and 94.6% by MRI) were not significant, which suggested that most of T0 category rectal cancer could not be correctly identified, but few were misdiagnosed. From the DOR value, we knew that the odds of confirmed T0 rectal cancer for the patients diagnosed T0 by EUS were 9.09 times higher than the odds for the patients diagnosed T1-4 by EUS. The odds were higher than by MRI (3.18), but did not achieve significance. T3-4 vs T0-2

For each outcome, summary estimates of sensitivity, specificity, and DOR with corresponding 95% CIs are presented in Table 3 with the use of the bivariate random-effects model. None was found to exhibit threshold effects by ROC plane and calculation of the Spearman correlation coefficient.

The sensitivity estimates of EUS and MRI were comparable: 87.6% and 82.1%, suggesting that the probability of correctly diagnosing T3-4 was high. Endoluminal ultrasound and MRI specificities were poor: 66.4% and 53.5%. The corresponding misdiagnosis rates by EUS and MRI were 33.6% and 46.5%. In the overall consideration of sensitivity and specificity, the T2 category (or lower) tumors were easier to be overstaged to T3-4 compared with the T3-4 category to be downstaged to T0-2 when both modalities were applied in clinical restaging. The DOR for EUS showed that the odds of histopathologically confirmed T3-4 category for the T3-4 diagnosed by EUS were 17.81 times higher than for the T0-2 diagnosed by EUS, which was not significantly higher than that for MRI (5.34) because of wide 95% CIs that partially overlapped. According these results, EUS was slightly superior to MRI in restaging T category of LARC after preoperative CRT.

T0 vs T1-4

Lymph Node Involvement

The sensitivity estimate for rectal cancer diagnosis (T0) by EUS (37.0%; 95% CI, 24.0%–52.1%) was higher (p = 0.043) than the sensitivity estimate for MRI (15.3%;

The sensitivity estimates for MRI (61.8%) were not of significant difference from EUS (49.8%). The specificity estimates for MRI and EUS were comparable: 72.0% and

Data Analysis

R

R P

R

R R P

R

R

2011 Netherlands

2010 Spain 2009 Italy

2009 Korea

2009 Netherlands 2009 Korea 2009 Italy

2009 Israel

2009 UK

Engelen20

Arbea21 Mezzi22

Cho23

Dresen24 Kim25 Barbaro26

Beer-Gabel27

Suppiah28

R

P

R

R

R

R R R

2007 UK

2007 Italy

2007 Sweden

2006 Israel

2005 Taiwan

2005 Taiwan 1999 Germany 1996 USA

Allen33

Maretto34

Torkzad35

Maor36

Chen37

Kuo38 Rau39 Bernini40

MRI EUS EUS

MRI

EUS

MRI EUS MRI

MRI

MRI MRI

EUS EUS

MRI

EUS

MRI MRI MRI

MRI

EUS

EUS MRI

MRI

EUS Independently

Independently

Independently (2) Independently

Assessment

1.5T Flexible 7.0 MHz

1.5T

Flexible

1.0T Flexible 1.5T

1.5T

1.5T 1.0T

Flexible 7.5 MHz

1.5T

10 MHz

1.5T 1.5T 1.5T

3.0T

Flexible

Independently Not available Independently

Independently

Independently

Not available Independently Consultation

Consultation

Independently (2) Not available

Independently Independently

Consultation

Independently

Independently (3) Independently Consultation

Independently

Independently

Not available Not available 1.5T Independently

1.0T/1.5T

7.5 MHz

1.5T Flexible

Protocol

N

T N T N N N T

N N T

N CRM T

N CRM T

N T N N T

CRM T

N T N T N T N N T N T N T N T CRM T

No. Treatment time

Time interval

42 Not available 17 (1–45) Size≥5 mm, border, signal AJCC 235 Not available Not available 235 Signal, border, shape, size ≥10 mm AJCC 82 5 wk Immediately Signal, round or oval, adjacent to rectal cancer 82 T3/T4: penetrating through the bowel wall 79 6–8 wk 22 (1–109) Size, border, signal 53 Signal, border, shape, size ≥10 mm, adjacent to tumor 50 Not available Just before surgery AJCC 39 6 wk Within 1 wk Not available 39 AJCC 39 Not available 39 AJCC 30 5 wk 1–3 wk Border, signal 30 AJCC 67 Not available 26 (1–109) 62 5.5–6 wk 4 (1–20) The distance ≤2 mm between tumor and MRF T3/T4: extension into perirectal fat or adjacent 53 Not available Within 1 wk structure or viscus The distance 5 wk 18 (14–24) Not available 49 51 Not available Not available Circle or oval, size ≥5 mm T3/T4: extension into perirectal fat or adjacent 44 Not available 1–2 wk structure or viscus Round, border, size >5 mm, signal 44 6 wk 24 (1–72) Subjective criteria 64 T3/T4: threatening the MRF or definite involvement 80 5 wk Not available of the fascia 80 Size ≥10 mm, signal, round shape 80 The distance≤1 mm between tumor and MRF T3/T4: extension into perirectal fat, adjacent structure 30 Not available 43 (11–89) or viscus, CRM ≤2 mm 43 Not available 1–3 wk Size ≥5 mm 41 Size ≥5 mm T3/T4: disruption of muscularis propria or extension 25 Not available 40 (26–65) into adjacent organ Not available 25 Not available 7–14 days All detected lymph nodes, irrespective of their size 25 Not available 50 Not available Before 8 wk Size >5 mm 50 Not available 36 Not available Before 8 wk Round shape, signal, border 33 Not available 4–6 wk T3/T4: extension into perirectal fat or adjacent 21 5–6 wk After 4 wk structure or viscus Signal, oval or round shapes; adjacent to tumor, not 21 rectal wall

Outcome Cutoff

CRT CRT CRT

CRT

CRT or RT

RT

CRT

CRT

CRT

CRT or RT

RT CRT

CRT

RT

CRT or RT CRT CRT

RT

CRT CRT

CRT

CRT

CRT CRT

Treatment

No. = number of patients; P = prospective; R = retrospective; CRM = circumferential resection margin; AJCC = TNM stage classification developed by the American Joint Committee on Cancer; MRF = mesorectal fascia; CRT = chemoradiotherapy; RT = radiotherapy; EUS = endoluminal ultrasound.

R P

2008 Netherlands 2008 UK

Vliegen31 Kulkarni32

2008 Korea Huh29 Radovanovic30 2008 Serbia

R R

P

2011 Italy

Lambregts Pastor18

MRI EUS

Design Imaging

Marone19

Year Country

P P

17

2011 Netherlands 2011 Spain

Author

TABLE 1.   Characteristic of studies included for assessment Diseases of the Colon & Rectum Volume 57: 3 (2014)

391

392

Zhao et al: Accuracy of Imaging in Restaging Rectal Cancer

TABLE 2.   Methodological quality of studies in this meta-analysis Study ID Lambregts Pastor18 Marone19 Engelen20 Arbea21 Mezzi22 Cho23 Dresen24 Kim25 Barbaro26 Beer-Gabel27 Suppiah28 Huh29 Radovanovic30 Vliegen31 Kulkarni32 Allen33 Maretto34 Torkzad35 Maor36 Chen37 Kuo38 Rau39 Bernini40 17

Aa

Bb

Cc

Dd

Ee

Ff

Gg

Hh

Ii

Jj

Kk

Ll

Mm

Nn

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Unclear Unclear Yes Yes Yes Yes Yes Yes Yes Unclear Yes Yes Yes Yes

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Yes Unclear Yes No Unclear Yes Yes No Yes Yes No Yes Unclear Yes No Unclear No Yes No Yes Unclear Unclear Unclear Unclear

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Yes Yes Yes Yes Yes Unclear Unclear Yes Yes Yes Unclear Unclear Yes Unclear Yes Yes Yes Yes Yes Unclear Unclear Unclear Yes Unclear

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Unclear

Yes Yes Yes Yes No Yes Yes Yes Yes Yes Yes No Yes No Yes Yes Yes No Yes No Yes Yes No Yes

No Yes No Yes No No No Yes No Yes No No No No Yes Yes Yes Yes Yes No No No Unclear Unclear

Yes Unclear Unclear Unclear Unclear Unclear Unclear Yes Yes Unclear Unclear Unclear Yes Unclear Yes Unclear Yes Unclear Unclear Yes Unclear Unclear Unclear Unclear

Yes Unclear Yes Unclear Unclear Unclear Unclear Unclear Yes Unclear Unclear Unclear Unclear Unclear Unclear Unclear Unclear Unclear Unclear Unclear Unclear Yes Unclear Unclear

Unclear Unclear Yes Unclear Unclear Unclear Unclear Unclear Yes Yes Yes Unclear Yes Unclear Unclear Unclear Unclear Unclear Yes Unclear Yes Unclear Unclear Unclear

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

LARC = locally advanced rectal cancer; CRT = chemoradiotherapy; RT = radiotherapy. a All the patients had LARC rectal cancer with a long course of CRT or RT alone. b Selection criteria were clearly described. c Histopathological validation was appropriate. d The time period between restaging and histopathological confirmation did not change the actual condition of tumor. e All of the patients received histological validation. f The same histopathological protocol was performed regardless of imaging results. g Histopathology was independent of imaging. h Imaging protocol was sufficient to permit replication. i Histopathological protocol was sufficient to permit replication. j The imaging results were interpreted without knowledge of the results of the histopathology. k The histopathological results were blinded to the imaging results. l The same clinical data were available when imaging results were interpreted as would be available when diagnostic imaging was used in practice. m Uninterpretable results were reported. n Withdrawals from the study were explained.

78.7%. The DORs for both modalities were also similar: 4.33 and 3.96. The diagnostic accuracy of both modalities in restaging lymph node involvement was not very good. Magnetic resonance imaging seemed to be a little superior to EUS, but did not achieve statistical difference. Circumferential Resection Margin Involvement

No data were available for EUS concerning this outcome. Magnetic resonance imaging sensitivity and specificity for circumferential resection margin involvement were 85.4% and 80.0%. Both sensitivity and specificity were high. In Table 3, the estimated DOR for MRI in detecting histopathologically confirmed positive CRM involvement was 27.6. This means that, for MRI, the odds of positivity among patients with CRM involvement are 27.6 times higher than the odds of positivity among patients without CRM involvement. From the data available, the diagnostic accuracy of MRI in reassessment CRM after preoperative CRT was excellent according to the AUC value.

Covariate Analysis and Metaregression

With respect to the factors that can potentially affect diagnostic performance, backward stepwise metaregression was performed to find the effects of these covariates: 1) design (prospective vs retrospective); 2) sample size (≥50 vs