Diagnostic value of magnetic resonance versus computed tomography colonography for colorectal cancer: A PRISMA-compliant systematic review and meta-analysis.

BACKGROUND: Advanced colorectal cancers were associated with poor prognosis, and early diagnosis was important for high-risk patients. Colonography is commonly used for diagnosing colorectal cancer. However, a few studies reported the diagnostic value of magnetic resonance colonography (MRC) versus computed tomography colonography (CTC). This study aimed to compare the diagnostic value of MRC versus CTC for colorectal cancer.
METHODS: Twenty-three studies on the diagnosis of colorectal cancer using MRC or CTC were obtained from PubMed, Embase, and the Cochrane Library databases until July 2017. The ratios of sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and receiver operating characteristic (ROC) curve were calculated to compare the diagnostic value of MRC versus CTC.
RESULTS: The summary sensitivity, specificity, PLR, NLR, and area under the ROC for MRC were 0.97 (0.81-1.00), 0.92 (0.80-0.97), 11.71 (4.46-30.73), 0.03 (0.00-0.24), and 0.98 (0.97-0.99), respectively, for diagnosing colorectal cancer. The pooled estimates for CTC in diagnosing colorectal cancer were as follows: sensitivity, 0.96 (0.90-0.98); specificity, 1.00 (0.99-1.00); PLR, 197.32 (73.21-531.85); NLR, 0.04 (0.02-0.11); and area under the ROC, 1.00 (0.99-1.00). No significant differences were found between MRC and CTC for sensitivity, specificity, and NLR. MRC was associated with lower PLR and area under the ROC for diagnosing colorectal cancer compared with CTC.
CONCLUSION: This study demonstrated MRC and CTC as potential diagnostic approaches for colorectal cancer. CTC had a higher diagnostic value of PLR and area under the ROC for colorectal cancer.

Introduction

Colorectal cancer is the most common gastrointestinal malignancy and the second cause of cancer-related deaths, accounting for around 40,340 new cases in the United States in 2013.[ Early detection and removal of polyps before malignant transformation could reduce the incidence of colorectal cancer, determine the treatment strategy, and improve the quality of life.[ Colonography is widely recognized as a reliable method for colorectal cancer diagnosis to distinguish colorectal lesions. However, it is still associated with several limitations. First, observer error and polyp in a blind area might cause inaccuracy in detecting diminutive lesions. Second, the application of colonoscopy is not suitable for obstructing colonic lesions. Finally, a traditional colonoscopy is invasive and uncomfortable for diagnosing colorectal cancer.

Currently, magnetic resonance (MR) and computed tomography (CT) colonography, noninvasive methods based on cross-sectional imaging, are recommended for diagnosing colorectal cancer.[ Although the findings of MR colonography (MRC) and CT colonography (CTC) were associated with a higher diagnostic value for colorectal cancer, the impact of ionizing radiation burden was not neglected.[ However, the ratio between MRC and CTC for diagnosing colorectal cancer was not calculated. Further, the diagnostic value for colorectal cancer was not compared within groups of patients categorized by potential confounders. Numerous recent studies have explored the diagnostic value of MRC versus CTC for colorectal cancer. In this study, a meta-analysis was performed to evaluate the diagnostic value of MRC and CTC for colorectal cancer and compare these diagnostic values among studies or patients with different characteristics.

Methods

Data sources, search strategy, and selection criteria

The present study was performed in accordance with the guidelines for the Preferred Reporting Items for Systematic reviews and Meta-Analyses.[ Ethics approval was not necessary for this study because only de-identified pooled data from individual studies were analyzed. Electronic databases PubMed, Embase, and the Cochrane Library were searched through July 2017 to identify studies on MRC or CTC in diagnosing colorectal cancer. The following core search terms were used: CTC, MRC, virtual colonoscopy, and colorectal cancer. The additional publications in reference lists and citation sections of recovered articles were also searched. Letters, abstracts, and conference proceedings were excluded due to discrepancies common between meeting abstract results and final publication results. Publication languages were limited to English.

The inclusion criteria were as follows: the study had to have a prospective design; the study used MRC or CTC; the study should have regarded conventional colonoscopy with biopsy as the golden standard; participants included in the study had a high risk of colorectal cancer; and the study provided true-positive, false-positive, false-negative, and true-negative results. The literature search and study selection were independently undertaken by 2 authors, and any inconsistencies were resolved by group discussion.

Data collection and quality assessment

Two authors first reviewed the abstract independently and then summarized the full selected studies. Any inconsistencies were settled by group discussion until a consensus was reached. The relevant data abstracted were as follows: first author, publication years, country, sample size, mean age, percentage male, imaging modality, and true and false positive and negative. Quality Assessment of Diagnostic Accuracy Studies (QUADAS)[ was used to evaluate the quality of the studies included in this meta-analysis independently by the 2 authors. Each of the assessment had 7 items and response as “yes,” “no,” or “unclear.” The answer of “yes” meant that a study's risk bias could be judged as low, whereas “no” and “unclear” meant that the risk of bias could be judged as high.

Statistical analysis

The sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and corresponding 95% confidence intervals (95% CIs) were calculated from true-positive, false-positive, false-negative, and true-negative results, which were extracted from each study before data pooling. The bivariate random-effects model [ was used to summarize the sensitivity, specificity, PLR, and NLR, and the hierarchical regression model was used to summarize receiver operating characteristic (ROC) curve and the area under the curve (AUC).[ Q statistic was used to estimate the heterogeneity of individual studies contributing to the pooled estimate. A P value >.10 indicated no significant heterogeneity, whereas a P value ≤ .10 indicated significant heterogeneity for the Q statistic.[ Subgroup analyses were conducted for the sensitivity, specificity, PLR, NLR, and AUC on the basis of sample size, mean age, and percentage of males. The ratios of summary sensitivity, specificity, PLR, NLR, and AUC and the corresponding 95% CIs were estimated according to specific summary effect estimates and 95% CIs after considering the sample size, mean age, and percentage of males.[ Visual inspections of funnel plots were performed using Deeks’ asymmetry test for MRC and CTC.[ All reported P values were 2-sided, and P values <.05 were considered statistically significant for all included studies. Statistical analyses were performed using Stata software (version 10.0; Stata Corporation, TX).

Results

The results of the study selection process are shown in Fig. 1. The initial electronic search identified 562 studies; of these, 519 studies that were duplicate and irrelevant were excluded. A total of 43 potentially eligible studies were selected. After reviewing the full text of each study and browsing the results, 20 studies that did not meet the inclusion criteria were eliminated. Finally, 23 prospective studies with a total of 2628 patients were included in this systematic review and meta-analysis.[ The characteristics of each study are presented in Table 1. A total of 7 studies reported the diagnostic value of MRC,[ and 16 studies reported the diagnostic value of CTC.[ The QUADAS quality assessment of the individual study is presented in Table 2.

Figure 1

Flow chart of study selection process.

Table 1

Characteristics of the studies included in the meta-analysis.

Table 2

Quality evaluation of the included studies using the QUADAS tool.

A total of 7 studies evaluated the diagnostic accuracy of MRC for colorectal cancer. The summary results showed that sensitivity was 0.97 (0.81–1.00; Fig. 2A), specificity was 0.92 (0.80–0.97; Fig. 2A), PLR was 11.71 (4.46–30.73; Fig. 3A), and NLR was 0.03 (0.00–0.24; Fig. 3A). The summary ROC curve was also calculated, and the AUC was 0.98 (0.97–0.99; Fig. 4A). Similarly, 16 studies reported the diagnostic value of CTC for colorectal cancer. The pooled sensitivity was 0.96 (0.90–0.98; Fig. 2B), specificity was 1.00 (0.99–1.00; Fig. 2B), PLR was 197.32 (73.21–531.85; Fig. 3B), NLR was 0.04 (0.02–0.11; Fig. 3B), and AUC was 1.00 (0.99–1.00; Fig. 4B). Further, no significant differences were found between MRC and CTC for the outcomes of sensitivity (ratio of sensitivity: 1.01; 95% CI: 0.90–1.13; P = .858), specificity (ratio of specificity: 0.92; 95% CI: 0.84–1.01; P = .094), and NLR (ratio of NLR: 0.75; 95% CI: 0.04–13.23; P = .844). The ratio of PLR for diagnosing colorectal cancer when comparing MRC and CTC significantly reduced (ratio of PLR: 0.06; 95% CI: 0.02–0.24; P < .001). Also, the ratio of area under the ROC for diagnosing colorectal cancer (MRC vs CTC) significantly reduced (ratio of area under the ROC: 0.98; 95% CI: 0.97–0.99; P = .001).

Figure 2

Forest plots showing sensitivity and specificity for the diagnosis of colorectal cancer. (A) Magnetic resonance colonography (MRC); (B) computed tomography colonography (CTC). Each solid square represents an individual study. Horizontal solid lines passing through solid squares represent 95% CI. Diamond indicates the pooled sensitivity and specificity for all of the studies.

Figure 3

Forest plots showing positive likelihood ratio (PLR) and negative likelihood ratio (NLR) for the diagnosis of colorectal cancer. (A) Magnetic resonance colonography (MRC); (B) computed tomography colonography (CTC). Each solid square represents an individual study. Horizontal solid lines passing through solid squares represent 95% CI. Diamond indicates the pooled sensitivity and specificity for all of the studies.

Figure 4

SROC curve for the diagnosis of colorectal cancer. (A) Magnetic resonance colonography (MRC); (B) computed tomography colonography (CTC). Each ellipse represents individual study estimates. The diamond is the summary point representing the average sensitivity and specificity estimates. The ellipses around this summary point are the 95% confidence region (dashed line) and the 95% prediction region (dotted line).

Four potential confounders, including sample size, mean age, and percentage of males, were stratified in the subgroup analysis of sensitivity, specificity, PLR, NLR, and AUC for colorectal cancer. The results are presented in Table 3. The difference in PLR for diagnosing colorectal cancer between MRC and CTC was prominent if the sample size was less than 100 [0.13 (0.03–0.61)], mean age less than 65.0 years [0.03 (0.00–0.35)], and percentage of males greater than 50% [0.04 (0.00–0.36)]. No other significant differences were found in sensitivity, specificity, PLR, NLR, and area under the ROC.

Table 3

Subgroup analysis.

Publication bias was also calculated according to Deeks’ asymmetry test. The findings of Deeks’ asymmetry test suggested no evidence of publication bias for MRC or CTC for diagnosing colorectal cancer (P value for MRC: .170; P value for CTC: .130; Fig. 5).

Figure 5

Deeks’ funnel plot with a regression line. (A) Magnetic resonance colonography (MRC); (B) computed tomography colonography (CTC).

Discussion

This study evaluated the diagnostic value of MRC and CTC for diagnosing colorectal cancer. Most of the included studies had high methodological quality. The findings indicated that both MRC and CTC had high pooled sensitivity, specificity, PLR, NLR, and area under the ROC for diagnosing colorectal cancer. Further, CTC was found to be superior to MRC for detecting colorectal cancer in terms of PLR and area under the ROC. This higher diagnostic value was mainly related to study sample size less than 100, mean age less than 65.0 years, and percentage of males greater than 50%.

A previous meta-analysis indicated that the pooled sensitivity of MRC in diagnosing colorectal cancer was 0.91 (95% CI: 0.79–0.97), the pooled specificity was 0.98 (95% CI: 0.96–0.99), the pooled diagnostic odds ratio was 576.41 (95% CI: 135.00–2448.56), and the area under the ROC was 0.98. Further, the summary sensitivity of CTC was 0.96 (0.92–0.99), the specificity was 1.00 (0.99–1.00), the diagnostic odds ratio was 1461.90 (544.89–3922.30), and the area under the ROC was 0.99. These findings suggested that the diagnostic accuracy of MRC and CTC was high.[ The inherent limitation of the previous meta-analysis was that the diagnostic value of MRC and CTC in a specific subpopulation was not available, and hence the ratio of effect estimate between MRC and CTC according to potential confounders remained controversial. Therefore, the present systematic review and meta-analysis was performed to evaluate the diagnostic value of MRC and CTC in diagnosing colorectal cancer and the diagnostic accuracy in a specific subpopulation.

Most of the included studies also suggested that the diagnostic value of MRC was appropriate. Ajaj et al[ suggested that MRC had a higher accuracy for detecting colonic lesions greater than 5 mm, and was associated with 93% sensitivity and 100% specificity. Pappalardo et al[ showed that MRC was associated with similar diagnostic accuracy for diagnosing colonic endoluminal lesions compared with conventional colonoscopy. Lauenstein et al[ indicated that barium-tagged MRC was promising for detecting all lesions more than 8 mm in diameter. Luboldt et al[ found that MRC was promising for filtering out individuals with obvious colorectal mass lesions. Leung et al[ recruited 165 patients and correctly identified 3 patients with 75% sensitivity and 99.3% specificity. This moderate diagnostic value of MRC could be because the images of MRC were unsatisfactory for interpretation irrespective of the bowel preparation and distension status.[ Ajaj et al[ included 50 patients and suggested that MRC was permitted using water or air for colonic distension. Furthermore, all included studies suggested that CTC was associated with a higher accuracy for diagnosing colorectal cancer.[ The present study also indicated that MRC and CTC had a high accuracy for diagnosing colorectal cancer. Future studies are still needed to directly compare the diagnostic value of MRC with CTC in a specific population.

The subgroup analysis suggested that the difference in PLR between MRC and CTC for diagnosing colorectal cancer was mainly associated with a study sample size less than 100, mean age less than 65.0 years, and percentage of males greater than 50%. This was probably because that most included studies had a smaller number of patients and higher statistical power. Furthermore, the difference in the ratio of effect estimate was found to be marginal for the area under the ROC in several subsets. Hence, the conclusions might be variable and need verification through large-scale studies in the future. Therefore, the present study was conducted to provide just relative results and a synthetic and comprehensive review.

This study had several strengths. First, a standard protocol and a comprehensive search strategy were used. Furthermore, bivariate random-effects model and hierarchical ROC analyses were also employed in the study. In addition, the large sample size was pooled, and the findings of this study were more robust than those of any individual study. Finally, the ratios of effect estimate were used to compare the diagnostic value of MRC and CTC for diagnosing colorectal cancer in groups of studies or patients categorized by potential confounders.

The study also had some limitations. First, the sources of heterogeneity could not be explored by meta-regression because detailed baseline characteristics were reported only in few studies. Second, the analysis used summarized data, which restricted more detailed analysis. Third, publication bias is always an inevitable problem in a meta-analysis of published studies.

The findings of this study indicated that MRC and CTC for diagnosing colorectal cancer were associated with higher sensitivity, specificity, PLR, NLR, and area under the ROC. When indirectly comparing MRC and CTC, CTC was found to be associated with higher PLR and area under the ROC for diagnosing colorectal cancer compared with MRC. Subgroup analyses suggested that the sample size, mean age, and percentage of males might be important for PLR of MRC and CTC in diagnosing colorectal cancer. Future studies should focus on specific characteristics of individuals to directly compare the diagnostic value of MRC and CTC for colorectal cancer.

Author contributions

Conceptualization: Hongfen Wu, Jihong Zhao.

Data curation: Shuangyan Sun, Changliang Yang.

Formal analysis: Shuangyan Sun, Changliang Yang.

Investigation: Zhicheng Huang, Wen Jiang, Yan Liu.

Methodology: Zhicheng Huang, Wen Jiang, Yan Liu.

Writing – original draft: Shuangyan Sun, Changliang Yang.

Writing – review & editing: Shuangyan Sun, Changliang Yang.