The Accuracies of Diagnosing Pancreas Divisum by Magnetic Resonance Cholangiopancreatography and Endoscopic Ultrasound: A Systematic Review and Meta-analysis

Magnetic resonance cholangiopancreatography (MRCP), MRCP after secretin stimulation (S-MRCP) and endoscopic ultrasonography (EUS) are all selected to diagnose pancreas divisum. However, the accuracies of three diagnosis remain unclear. The aim is to address the diagnostic accuracies of MRCP, S-MRCP and EUS on pancreas divisum. We searched PubMed, MEDLINE and EMBASE databases from inception to January, 2015. Of the 536 citations retrieved, 16 studies were included. For MRCP diagnosis on pancreas divisum, the area under the hierarchical summary receiver-operating characteristic (HSROC) curve was 0.90 (95% confidence interval [CI] 0.87 to 0.92), and for S-MRCP and EUS, 0.99 (95% CI 0.97 to 0.99) and 0.97 (95% CI 0.96 to 0.98). Sensitivity and specificity for MRCP were 0.59 (95% CI 0.45 to 0.71) and 0.99 (95% CI 0.96 to 1.00); for S-MRCP, 0.83 (95% CI 0.66 to 0.92) and 0.99 (95% CI 0.96 to 1.00); for EUS, 0.85 (95% CI 0.67 to 0.94) and 0.97 (95% CI 0.90 to 0.99). Comprehensive comparison of three diagnostic techniques to pancreas divisum, S-MRCP was more reliable than MRCP and EUS on the effect of the diagnostic test.

The main results are listed in Fig. 2 and Table 3. In ten studies on MRCP, the sensitivity and specificity for MRCP diagnosis were 0.59 (95% CI 0.45 to 0.71) and 0.99 (95% CI 0.96 to 1.00), respectively; positive likelihood ratio (LR+ ) and negative likelihood ratio (LR− ) were 87.83 (95% CI 15.25 to 505.81) and 0.42 (95% CI 0.30 to 0.58); the diagnostic odds ratio (OR) was 211.33 (95% CI 32.14, 1389.76). In five studies on S-MRCP, the sensitivity and specificity for S-MRCP diagnosis were 0.83 (95% CI 0.66 to 0.92) and 0.99 (95% CI 0.96 to 1.00), respectively; LR+ and LR− were 65.48 (95% CI 20.85 to 205.71) and 0.17 (95% CI 0.08 to 0.37); the diagnostic OR was 376.89 (95% CI 75.02, 1893.37). In seven studies on EUS, the sensitivity and specificity for EUS diagnosis were 0.85 (95% CI 0.67 to 0.94) and 0.97 (95% CI 0.90 to 0.99), respectively; -LR+ and LR− were 26 Sensitivity analysis. We systematically removed one portion of data randomly and recalculated the Log OR for the remaining studies of MRCP, S-MRCP and EUS, respectively ( Table 4). The three results were similar, indicating that no single data could alone significantly influence the combined analysis of MRCP, S-MRCP and EUS.
Bias Diagnostics. No significant publication bias for the three techniques on MRCP, S-MRCP and EUS was detected through the Begg-Mazumdar test and the Harbord-Egger test (all P > 0.05). No visual publication bias was found in the funnel plot for MRCP, S-MRCP and EUS, respectively (Fig. 3).

Discussion
The finding of the meta-analysis suggested that the diagnosis of pancreas divisum using whichever technique such as MRCP, S-MRCP or EUS, has a high diagnostic performance. The lowest area under HSROC curve in the three techniques reached 0.90. Pancreas divisum is the congenital anomaly in which the dorsal and ventral pancreatic glands drain separately into the duodenum 31 . Although ERCP has been proved to have high diagnostic accuracy in the excellent description of the biliary and pancreatic ductal system and has been accepted as the gold standard for diagnosing pancreas divisum 31,32 , it is an invasive procedure and represents a certain degree of complications 33 . MRCP is non-invasive diagnostic technique and EUS is a minimally invasive test, which both give negligible complications. The two techniques have been widely used to investigate the pancreatic and ductal system in clinical practice. However, the specific accuracies of diagnosing pancreas divisum by MRCP, S-MRCP and EUS remain unclear. The number of related studies on the diagnosis of pancreas divisum with MRCP or EUS is increasing year by year, so it is the right time to combine these studies together and to compare the accuracies of different techniques on diagnosing pancreas divisum.
MRCP is one of the first choices for examining biliary and pancreatic ductal system in a non-invasive way. Secretin is used to strengthen the visualization of the pancreatic duct at MRCP, and its working mechanism is mainly to stimulate pancreas to secrete fluid and bicarbonate and to improve the visualization of the pancreatic duct 10,21 . Actually, this meta-analysis showed the area under HSROC curve for S-MRCP (0.99, 95% CI, 0.97-0.99) was larger than that for MRCP (0.90, 95% CI, 0.87-0.92); the diagnostic OR for S-MRCP (376.89, 95% CI, 75.02-1893.37) was larger than that for MRCP (211.33, 95% CI, 32.14-1389.76). This suggested that S-MRCP was superior to MRCP in terms of the effect of the diagnostic test. If the likelihood ratios are greater than 10 but less than 0.1, the results will suggest strong evidence for meeting a diagnosis in or out, respectively. The meta-analysis showed the pooled likelihood ratios LR+ and LR− for S-MRCP were 65.48 (95% CI, 20.85-205.71) and 0.17 (95% CI, 0.08-0.37); the pooled likelihood ratios LR+ and LR− for MRCP were 87.83 (95% CI, 15.25-505.81) and 0.42 (95% CI, 0.30-0.58). Combining these results together, we concluded that S-MRCP was more reliable than MRCP on diagnosing pancreas divisum.
EUS has definite advantages over other options in evaluating biliary and pancreatic ductal system, and it is less invasive than ERCP. EUS could achieve detailed imaging of the biliary and pancreatic ductal system without injecting contrast into these ducts. Therefore, pancreas divisum could also be detected by minimally invasive techniques like EUS, but EUS could obviate the associated risks like ERCP 34 . This meta-analysis showed the area under HSROC curve for EUS was 0.97 (95% CI, 0.96-0.98), suggesting that EUS was superior to MRCP but slightly inferior to S-MRCP in terms of the effect of the diagnostic test. Additionally, the meta-analysis showed the pooled likelihood ratios LR+ and LR− for EUS were 26.80 (95% CI, 8.05-89.27) and 0.16 (95% CI, 0.07-0.38), respectively. However, the diagnostic OR for S-MRCP (376.89, 95% CI, 75.02-1893.37) was larger than that for EUS (167.89, 95% CI, 36.06-762.69). This suggested that EUS was more reliable than MRCP in diagnosing pancreas divisum, but it was inferior to S-MRCP.
There are several limitations about the present meta-analysis of literature data. First, there are 10 studies of totally 856 patients for MRCP diagnosis of pancreas divisum, so the MRCP diagnosis involves twice as many studies as S-MRCP has, and nearly twice as many patients as EUS has. So the number of studies and patients are different among the three methods. Second, ERCP is considered as the gold standard for pancreas divisum, but fails to cannulate a certain proportion. We didn't account the part of ERCP failure rates in this meta-analysis. Additionally, therapeutic potential of ERCP could add perspective, it's pity that there were no further explanations in these studies. Third, recurrent pancreatitis or chronic pancreatitis may destroy or alter pancreatic ducts, there

Table 1. Study design and statistical characteristics of included studies on MRCP, S-MRCP and EUS.
were no clear explanations in these studies, which could result in a certain bias in the meta-analysis. Fourth, a large portion of studies are observational and nonrandomized controlled studies. These suggest that there are some confounding factors influencing the meta-analysis. Fifth, S-MRCP and EUS are relatively newer technology than MRCP, publications may favor positive reports, so each of the included studies needed to be made quality assessment to avoid the potential bias as much as possible. The results are limited by the quality and quantity of the present data. Certainly, no significant publication bias for the three techniques was detected through the Begg-Mazumdar test and the Harbord-Egger test, and the results were further confirmed by the funnel plot. A large part of studies and patients were performed in tertiary centers, but it was not clear whether gastrointestinal radiologists or general radiologists reviewed the results. Therefore, there is expertise bias in this study because of differnent hospitals and different experienced doctors.
In conclusion, comprehensive comparison of three diagnostic techniques to pancreas divisum, S-MRCP was more reliable than MRCP and EUS on the effect of the diagnostic test.

Methods
Literature retrieval strategy and selection. We conducted an online literature retrieval or search from PubMed, MEDLINE and EMBASE databases from inception to January, 2015. The key words of the heading used in the search included "pancreas divisum" AND ("cholangiopancreatography, magnetic resonance" OR "magnetic Resonance Imaging" OR "magnetic resonance cholangiopancreatography" OR MRCP); "pancreas divisum" AND Study C1 C2 C3 C4 C5 C6 C7 C8 C9 C10 C11 C12 C13 C14 Kushnir VM et al. 16     ("endoscopic ultrasound" OR EUS OR ultrasound OR endosonography). No restrictions were imposed on the choice of languages. The titles and abstracts that included these terms were in detail for potential inclusion. The full text of the remaining articles including the references was ascertained to have contained related information.
We included studies evaluating the detection of pancreas divisum with MRCP and/or S-MRCP and/or EUS as the reference standard. The studies could be represented by 2 × 2 tables with true-positive, false-negative, false-positive and true-negative values. We excluded those studies that did not involve the detection of pancreas divisum, those with insufficient data, as well as those primarily designed as the reviews, editorials, case reports or meta-analysis. We discussed and resolved disagreement between the investigators who evaluated the detection of pancreas divisum.
Data extraction. Data were seperately extracted and printed in standardized paper forms. The following data were collected for all studies: study design, period of study/year of publication, country, number of patients, mean age, male to female ratio, criteria for pancreas divisum, tertiary center and main outcomes reported.
Assessment of study quality. The Quality Assessment of Diagnostic Accuracy Studies tool was used to evaluate the quality of the studies. We rated the quality of key study design characteristics of a total of 14 items 35 . The 14 items were rated to evaluate the quality of key study design characteristics in this analysis.
Statistical analysis. Meta-analysis for diagnosis of pancreas divisum was performed under a linear mixed model approach to calculate summary estimates of sensitivity, specificity, LR+ , LR− , and diagnostic OR all of which were then fitted into a HSROC curve 36,37 . We assessed the publication bias qualitatively and quantitatively by funnel plots and bias indicators, including by the Begg-Mazumdar test and the Harbord-Egger test 38,39 . We performed sensitivity analysis to calculate whether any single study was contributing undue weighting to the analysis. We removed one portion of study data and checked the pooled results to see whether there was any significant change in test performance. We used Stata V.12 to perform the calculations.