Ductal Dilatation of ≥5 mm in Intraductal Papillary Mucinous Neoplasm Should Trigger the Consideration for Pancreatectomy: A Meta-Analysis and Systematic Review of Resected Cases

Simple Summary Intraductal papillary mucinous neoplasms (IPMN) are common but difficult to manage since accurate tools for diagnosing malignancy are unavailable. This study evaluates the diagnostic value of main pancreatic duct (MPD) diameter for detecting IPMN malignancy, using a meta-analysis of published data. The result suggests that malignancy is highly prevalent in IPMN with ductal dilatation of >5 mm. Abstract Intraductal papillary mucinous neoplasms (IPMN) are common but difficult to manage since accurate tools for diagnosing malignancy are unavailable. This study tests the diagnostic value of the main pancreatic duct (MPD) diameter for detecting IPMN malignancy using a meta-analysis of published data of resected IPMNs. Collected from a comprehensive literature search, the articles included in this analysis must report malignancy cases (high-grade dysplasia (HGD) and invasive carcinoma (IC)) and MPD diameter so that two MPD cut-offs could be created. The sensitivity, specificity, and odds ratios of the two cutoffs for predicting malignancy were calculated. A review of 1493 articles yielded 20 retrospective studies with 3982 resected cases. A cutoff of ≥5 mm is more sensitive than the ≥10 mm cutoff and has pooled sensitivity of 72.20% and 75.60% for classification of HGD and IC, respectively. Both MPD cutoffs of ≥5 mm and ≥10 mm were associated with malignancy (OR = 4.36 (95% CI: 2.82, 6.75) vs. OR = 3.18 (95% CI: 2.25, 4.49), respectively). The odds of HGD and IC for patients with MPD ≥5 mm were 5.66 (95% CI: 3.02, 10.62) and 7.40 (95% CI: 4.95, 11.06), respectively. OR of HGD and IC for MPD ≥10 mm cutoff were 4.36 (95% CI: 3.20, 5.93) and 4.75 (95% CI: 2.39, 9.45), respectively. IPMN with MPD of >5 mm could very likely be malignant. In selected IPMN patients, pancreatectomy should be considered when MPD is >5 mm.


Introduction
Pancreatic cystic lesions are common. The prevalence of these lesions is around 50% in the general population and increases with age [1]. Of these cystic lesions, pancreatic cystic neoplasm (PCN) management is challenging and important in modern pancreatology. The challenge lies in the difficulty in accurately discerning completely benign PCNs from ones with potential for malignancy progression [2][3][4][5].
Intraductal papillary mucinous neoplasm (IPMN) is a pre-cancerous lesion that accounts for at least half of all PCNs [6]. Morphologically, IPMNs can be divided into two major categories: (1) the branch-duct IPMN (BD-IPMN) that only involves the peripheral pancreatic ducts, and (2) main-duct IPMN (MD-IPMN) and mixed-type IPMN that involve the main pancreatic duct (MPD) and/or the branch ducts [7]. The morphologic and radiologic classification for identifying BD-IPMN, MD-IPMN, or mixed-type IPMN is extremely important. It is well known that BD-IPMNs have a very low risk for cancer progression; in contrast, MD-and mixed-type IPMNs are more prone to becoming cancerous [2,8]. Studies have shown that it is safe to surveil small BD-IPMNs (below 3-4 cm) that do not have any radiological features suggestive of malignancy [8][9][10]. In contrast, IPMNs involving the MPDs are more aggressive and generally need to be treated surgically [6,11].
The International Consensus Guidelines and the European Guidelines for managing IPMNs concordantly suggest an aggressive approach for managing MD-and mixed-type IPMNs. However, the role of MPD dilatation in deciding the indication for surgical resection of IPMN is moderately different in the two guidelines. The International Consensus Guidelines for IPMN management suggests that MD-or mixed-type IPMN with MPD dilatation of 10 mm or more should be referred for surgical treatment (high-risk stigmata) [11]. However, the same guideline suggests that surgery could still be considered for patients with 5-9.9 mm MPD (a worrisome feature), if there is the presence of mural nodule(s) ≥5 mm, cytology positivity for malignancy, or main-duct features suspicious for malignancy involvement [11]. The first European Guideline, published in 2013, recommended lowering the cutoff of MPD dilatation for surgery indication from 10 mm to 6 mm [2]. This approach was later supported by retrospective analysis that showed the implementation of MPD dilatation of 10 mm as a cutoff for surgery indication risks the possibility of IPMN undertreatment, as patients could have already developed invasive carcinoma (IC) or high-grade dysplasia (HGD) [12][13][14]. For better management of IPMNs, the latest evidence-based European Guidelines, published in 2018, recommends surgical resection of IPMN with MPD dilatation of 5-9.9 mm, if the patient is fit and has a long life expectancy (relative indications for surgery) [6]. This approach has also been recently supported by large retrospective surgical analyses [15].
Meanwhile, some studies have demonstrated a safe conservative approach in managing patients with suspected MD-or mixed-type IPMNs that have MPD dilatation of 5-9.9 mm [16,17]. Despite the compelling data, those studies need to be reassessed since high-volume centers have demonstrated that the accuracy of the pre-operative diagnosis of PCNs is approximately 60-80% [18][19][20]. Therefore, a high percentage of the suspected MD-or mixed-type IPMNs in those studies could very likely represent other more benign pancreatic diseases, such as chronic pancreatitis [18][19][20]. Hence, the optimal study design to analyze the disease status of PCNs is to evaluate a cohort with available histology data [21].
The role of MPD dilatation is important for establishing surgical indication. However, currently there is no consensus across existing guidelines regarding the degree of ductal dilatation that warrants pancreatectomy. This study is designed to evaluate the association of HGD and IC with ≥5 mm and ≥10 mm MPD. Sensitivity and specificity for detecting HGD and IC were calculated using the two MPD cutoffs.

Article Search/Selection and Outcome Assessment
This systematic review and meta-analysis was performed by following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Guideline [22]. Literature search was performed on 28 February 2020 by a professional librarian at the Strauss Health Sciences Library. The following databases were queried: Ovid MEDLINE(R) ALL 1946 to 27 February 2020; Embase; Web of Science; Google Scholar. The search strategy focused on obtaining all existing literature related to MPD dilatation in IPMN and its association to malignancy. The terms including but not limited to: intraductal papillary mucinous or intraductal mucinous papillary or ipmn or ipmt; dilat or size or diameter or cut off or cutoff or mm or millimeter or milli meter were used for the systematic search. Full search strategy is provided in online Figure S1. The articles obtained from the search were uploaded to the Endnote version X9 citation management application, in which duplicated search results were eliminated. The final list of articles for initial review was uploaded to Covidence, a software designed for reviewing and selecting articles for systematic review and meta-analysis.
In the initial phase of reviewing articles, Y.H.A.W. and A.O. independently studied all records uploaded to Covidence. If the title or abstract of the articles were relevant to the topic of this study, the independent reviewers (Y.H.A.W. and A.O.) would thoroughly perform a screening of the entire article to assess for eligibility of inclusion. Non-English articles were excluded at the title/abstract review phase. Should there be a disagreement in eligibility of the included articles, D.N. would resolve the discrepancy.
Studies evaluating the association between MPD, IPMN (including BD-IPMN, MD-IPMN, or mixed-type IPMN), and malignancy were included in this study. We specifically only included studies that reported MPD ranges that could be used to create two cut-offs: ≥5 mm and ≥10 mm. Studies also had to include the counts or rates of malignancy and non-malignancy for each MPD category. We also required that included articles show histological diagnosis of lesions, specifically whether they were non-malignant or malignant. In this study, histology classification and definition are in accordance with the 2015 Baltimore Consensus Meeting Guideline [23] or the current World Health Organization guidelines [24] for IPMN. In short, HGD or IC were defined as malignancy, whereas lowgrade dysplasia or moderate-grade dysplasia were defined as non-malignancy. In the event of encountering two or more publications that had identical cohorts, only the publication with the most complete dataset was included in this study.
The primary outcomes of this study were the proportion of patients with HGD, IC, or malignancy.

Data Extraction
Study characteristics and data were independently extracted by two investigators (Y.H.A.W. and A.O.) and recorded on a standardized data extraction form. Any discrepancies were resolved by other reviewers (S.F., D.N., L.B., and K.L.C.). Data extracted and used for analysis are listed in Table 1. The final extracted data were reviewed by Y.H.A.W., A.O., S.F., L.B., and K.L.C.

Risk of Bias Analysis
Y.H.A.W., S.F., and A.O. assessed the quality of the included articles by conducting an evaluation using the Risk of Bias in Non-randomized Studies of Interventions (ROBINS-I) [41]. ROBINS-1 is designed to assess an article by evaluating 7 domains of risk of bias (Table S1). The assessment will grade the 7 domains of an article with low, moderate, serious, critical, or not assessable risk. In the end, the 7 domains were collectively analyzed for every article included in this study. Additionally, Funnel plots and Egger's regression tests were used to assess publication bias ( Figure S2).

Data Synthesis and Statistical Analysis
Tests of association: Random effects models were used to estimate adjusted diagnostic odds ratios (OR) and 95% confidence intervals for the pooled data. These were estimated using the meta package in RStudio (RStudio, Boston, Massachusetts), which provides functions for diagnostic meta-analysis [42]. A continuity correction was applied to all cells in a 2 × 2 table when necessary. We evaluated the association between cutoffs of ≥5 mm and ≥10 mm and classification of HGD, IC, and malignancy. There were three primary comparisons made for the two cutoffs: (1) classification of non-malignancy and malignancy; (2) classification of non-malignancy and HGD, and (3) classification of non-malignancy and IC. Therefore, we estimated six ORs using two cutoffs and three disease classifications. Forest Plots were used to compare individual-study ORs and the pooled OR. Forest plots include OR estimates from both the fixed effects model and the random effects model. Heterogeneity amongst the included publications was assessed using Cochrane's Q test and I 2 . Based on the results of these tests, this paper utilized only the random effects estimates for interpretations.
Diagnostic tests: Pooled sensitivities and specificities and 95% confidence intervals for the same six comparisons were also estimated using the mada package (R Foundation for Statistical Computing, Vienna, Austria). Using a random effects model, the mada package implements a bivariate estimation of sensitivity and specificity described by Reitsma et al. [43]. This bivariate approach is necessary because the sensitivity and specificity of a test are interrelated; therefore, univariate approaches to estimation are inappropriate. From these pooled values for sensitivity and specificity, we then estimated AUC for each comparison. The summary receiver operating characteristic (SROC) curves were plotted to assess the spread of diagnostic measures for each comparison and cutoff.
We conducted an additional analysis to evaluate the impact of including the large study by Del Chiaro et al. (senior author of this study) [15]. This was done by repeating the analyses described above excluding that study.

Tests of Association
Both the ≥5 mm and ≥10 mm cutoffs in resected cases were significantly associated with higher risk of malignancy compared to <5 mm and <10 mm, respectively (≥5 mm: OR = 4. 36 Figure 2. Summary of the odds of HGD and IC for different MPD cutoffs are in Table S2.

Diagnostic Tests
For all six comparisons, the tests for equality of sensitivities and specificities were all found to be significant (p < 0.001). This indicated heterogeneity between studies, and therefore we estimated pooled sensitivities and specificities using random effects models ( Table 3). The six SROC curves from the random effects models are included in Figure 3.   Using a cutoff point of 5 mm, specificity was 58.6% and sensitivity was 74.8% for classification of malignancy. The AUC was 0.716. For the cutoff point of 10 mm, specificity was 86.4% and sensitivity were 33.8% for classification of malignancy. The AUC was 0.586. A cutoff of 5 mm had 70.1% specificity and 72.2% sensitivity for classification of HGD. The AUC was 0.769. A cutoff of 10 mm had 88.7% specificity and 35.7% sensitivity for classification of HGD. The AUC was 0.587. A cutoff point of 5 mm had 69.7% specificity and 75.6% sensitivity for classification of IC. The AUC was 0.786. A cutoff of 10 mm had 88.2% specificity and 36.6% sensitivity for classification of IC. The AUC was 0.587 ( Figure 3).

Subset Analysis Excluding Del Chiaro et al.
The analysis excluding the large study by Del Chiaro et al. [15] yielded very similar results to the overall results presented in this study. The results of this analysis are provided in Table S2.

Discussion
The correct clinical management of IPMNs is crucial for the prevention of pancreatic cancer [46]. An overtreatment of low-grade dysplasia lesions could result in unnecessary morbidity and mortality related to pancreatic surgery. Surgery is indicated for IPMNs in an attempt to remove IC and HGD. The latter is the optimal pre-invasive histology form and time-point for surgical intervention. Unfortunately, there is no available method that can effectively discriminate HGD from IC, except a few experimental approaches that are not yet implemented in clinical practice [15,47,48]. Recent studies reported that pancreatectomy-related mortality has decreased from 7.3% since 2000 [49], and the benchmark for postoperative mortality after pancreaticoduodenectomy published in 2019 was ≤1.6% [50], which is drastically lower than the 17-42% probability of five-year survival of resected IC [50][51][52][53][54]. Therefore, surgery performed on IC could be too late for extending the survival of patients, and it could be more beneficial to resect MD-IPMN while in HGD form.
According to the European evidence-based guideline for PCN management, IPMN MPD dilatation of 5-9.9 mm and ≥10 mm are relative and absolute indication for surgery, respectively [6]. Recent studies have demonstrated IPMN with MPD dilatation of ≥5 mm have a malignancy rate of 30-90% [12,15,28,33,[55][56][57][58][59]. In this meta-analysis of 20 retrospective studies, malignancy was detected in nearly 46.8% of resected IPMN patients with 5-9.9 mm MPD and 63.5% of resected IPMN patients with ≥10 mm MPD. In total, 53.1% of resected IPMN patients with ≥5 mm MPD had malignancy. On the contrary, only 18.4% of resected cases with MPD < 5 mm were malignant IPMN. Pooled OR for malignancy calculated in this meta-analysis showed that MPD cutoff set at ≥5 mm was higher than ≥10 mm (OR = 4.4 vs. 3.2) in the resected cases. To improve the overall survival rate of malignant IPMN, HGD should be surgically removed before allowing it to progress to IC. Pooled OR from this meta-analysis showed that the odds of HGD were higher in ≥5 mm MPD than ≥10 mm MPD (OR = 5.7 vs. 4.4) in resected cases. Similarly, pooled OR for IC was also higher in ≥5 mm MPD than in ≥10 mm MPD (OR = 7.4 vs. 4.8). These data suggest that ductal dilatation of ≥5 mm should trigger the consideration for pancreatectomy.
The pooled sensitivities of ≥5 mm MPD in predicting IPMN HGD and IC were 72.2% and 75.6%, respectively. When using ≥5 mm MPD as cutoff, SROC AUC was 0.769 and 0.786 for HGD and IC, respectively, which were higher than those of ≥10 mm cutoff (AUC = 0.657 and 0. 587 for HGD and IC, respectively). The predictive role of ≥10 mm MPD for malignancy is unquestionable, but ≥5 mm MPD cutoff should also be considered as a highly sensitive factor for detecting HGD and/or IC. This lower cutoff could potentially identify malignancy in advance and improve survival of IPMN patients. However, Marchegiani et al. suggested that utilization of MPD dilatation as the sole indicator for pancreatectomy could encourage unnecessary surgical procedure and that most individuals with 5-9 mm MPD should be managed expectantly, performing conversion surgery when tumor progression is identified during close, regular follow-ups [17]. In the observation arm of the same study, 3 out of 46 (6.5%) IPMN patients with 5-9 mm MPD eventually underwent surgery [17]. However, Salvia et al. reported that the diagnostic accuracy for MD-IPMNs is only approximately 80% [19], implying some of those IPMN patients who underwent conservative treatment did not actually have IPMN [21]. If a PCN were to present with mural nodule(s) or cyst(s), fine-needle aspiration (FNA) with real-time endoscopic ultrasound (EUS) could be utilized to obtain histology proof and to make a diagnosis. It is important to note that EUS can accurately identify morphologic features of PCNs but is only 51% accurate in discerning the difference between mucinous and non-mucinous lesions [60]. The accuracy would increase up to 79% if CEA measurements were obtained from cystic fluid using EUS-FNA [60]. However, mural nodules and cysts ≥40 mm have only been reported in 1.2% to 21.3% and 23.2% to 27.5% of malignant IPMNs, respectively [13,15], and FNA would not have been a possible approach if ductal dilatation were the only cross-sectional imagining finding. Studies that include resected cases disproportionately favor inclusion of more serious cases that require surgery, but at least the diagnosis could be confirmed with surgical pathology. Including only pathologically confirmed IPMN cases for analysis, this study shows that many HGD and IC patients would be missed if the decision to operate were made at MPD ≥ 10 mm associated with low sensitivity.
The results in this study underline the need for a surgical evaluation for MPD ≥5 mm and are not conclusive evidence that suggest the risk for developing cancer is higher in IPMN with 5-9.9 mm MPD. On the other hand, it could also be possible that IPMN with MPD dilatation over 10 mm are slow progressive diseases that would turn invasive when MPD reaches larger dilatation. Using ≥5 mm MPD to select surgical candidates might also identify low-grade dysplasia. Considering that the highest incidence of IC occurs with MD-IPMN, we could assume that the surgical treatment of MD-IPMN with low-grade dysplasia is perhaps not an overtreatment when compared to surgical treatment of BD-IPMN with low-grade dysplasia [21]. Nevertheless, it is important to know that the currently available information for IPMN management is not perfect, and it is important to diagnose IPMN with existing sensitive diagnostic tools to increase the overall survival rates of the patients, as the mean frequency of malignancy in MPD dilatation of ≥5 mm was 61.6% and that of IC IPMN was 43.1% [11].
Recognizing the potential for the Del Chiaro et al. study [15] of heavily influencing the results presented in this study given its overall size, we completed a full analysis of the data excluding this study. The results and conclusions were unchanged.
Our study has limitations related to the use of retrospective studies that lack conservatively managed cohorts as a control group. Cases could not be stratified based on the imaging modalities used. The most accurate cutoff line for MPD dilatation could not be determined in this study since data for MPD dilatation are presented in ranges. In addition, other features associated with IPMN malignancy, such as mural nodule, cytology, elevated CA 19-9, or cyst diameter, could not be included in this analysis [11]. In addition, this study includes only resected IPMNs. Therefore, the results of this study are not representative of the unresected IPMN population. However, this study was performed based on recently published study that showed MPD dilatation is currently the best predictor of HGD or IC in IPMN [15]. Although estimates of heterogeneity above 50% were observed for some of our analyses, we found no reason for publication bias based on visual inspection of the funnel plot and the results of the Egger's test (Egger Regression Model). Additionally, and as stated in the methodology, all of our statistical models were estimated using random model effects, which have previously been used successfully to account for this limitation [42]. Even with some limitations related to the retrospective nature of this study, this analysis contains the largest volume of IPMN cases analyzed and our methodological approach properly accounted for study-level variation.

Conclusions
MPD dilatation is an important predictive factor of IPMN malignancy and 5 mm is a highly sensitive cutoff that detects high-risk pre-cancerous or cancerous lesions in resected cases. It is important to note that this study cannot draw a conclusion for non-surgical cases. However, the need for pancreatectomy should be thoroughly evaluated in patients with ductal dilatation of ≥5 mm. The result of this study implies that MPD dilatation over 5 mm should trigger the referral of a patient to a high-volume center for further consultation. The decision to perform resection should only be considered after careful evaluation of multiple aspects related to the general conditions and the expectancy of life of a patient. In conjunction with new biomarkers or diagnostic modalities such as pancreatoscopy [61,62], MPD dilatation detection could improve surgical patient selection and reduce overall IPMN malignancy mortality.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/cancers13092031/s1. Figure S1: Detailed Search Strategy for Existing Online Literature, Figure  S2: Funnel plot with Diagnostic Odds Ratio (DOR) and Egger's Test for publication bias assessment, Table S1: ROBINS-I Risk of Bias Assessment of included Articles, Table S2: Odds ratio, Sensitive, and Specificity for different comparisons and cutoffs.

Conflicts of Interest:
None of the authors have any personal conflict of interest to declare, except for Del Chiaro who is a co-principal investigator on a Boston Scientific study investigating the role of intraoperative pancreatoscopy in IPMN patients. Del Chiaro has also been awarded an industry grant from Haemonetics Inc. to conduct a multicenter study that evaluates the prognostic implications and changes of thromboelastographic (TEG) in pancreas cancer patients.