Comparing accuracy of high-risk features for detecting advanced neoplasia in pancreatic cystic lesions: a systematic review and meta-analysis

Introduction

Pancreatic cystic lesions (PCL) are often detected incidentally in asymptomatic patients who undergo cross-sectional imaging. The prevalence of PCL ranges from 13-45% in patients who undergo contrast-enhanced computed tomography and magnetic resonance imaging (MRI)/magnetic resonance cholangiopancreatography [1-3]. PCL display a wide variety, most of which are benign [4]. Some cysts, such as solid pseudopapillary neoplasms, mucinous cystic neoplasms and intraductal papillary mucinous neoplasms, have malignant potential and can evolve into pancreatic cancer. Pancreatic cancer has a high mortality and typically presents with lymphadenopathy or metastasis when curative interventions are not feasible [5]. This makes early identification of high-risk PCL imperative.

Endoscopic ultrasound (EUS) is an excellent modality for evaluating PCL and defining cyst features, including size, mural nodularity, internal features, relation of the cyst with pancreatic duct, and location within the gland [6]. It also facilitates the aspiration of cyst fluid for analysis for carcinoembryonic antigen, amylase, and cytology. These features make EUS a useful diagnostic tool in risk stratification for PCL [7].

The current literature suggests that specific high-risk features (HRF) on cross-sectional imaging are associated with a cyst’s progression to malignancy [8-10]. These HRF include: (a) dilated pancreatic duct (≥5 mm); b) presence of a solid component or mural nodule; and c) cyst size ≥3 cm. The American Gastroenterological Association (AGA) in 2015 recommended further evaluation with an EUS if 2 or more HRF are identified [11]. The subsequently published American College of Gastroenterology (ACG) guideline, however, recommends EUS for 1 or more HRF [12]. This discrepancy in the recommendations for EUS in PCL can lead to confusion in risk-stratifying patients and raise questions about which guideline should be used in clinical practice [13]. Several studies have assessed the diagnostic accuracy of these guidelines, with variable results. We performed a systematic review and meta-analysis to compare the diagnostic accuracy of using ≥1 or ≥2 HRF for assessing the risk of advanced neoplasia (AN) and performing EUS in PCL.

Materials and methods

Results

The initial literature search resulted in 38 articles. After a review of the abstracts and full texts to exclude redundant and duplicate studies, 20 articles were screened and 8 studies were included in the final analysis, based on the selection criteria above (Fig. 1). All the 8 studies were retrospective cohort studies; they included 2 multicenter studies and 6 single-center studies. Seven studies were from the United States and one study was from Japan [18-25]. The quality of the studies was assessed using the Newcastle-Ottawa scale and is reported in Supplementary Table 2.

Figure 1 Study flow diagram depicting search strategy, screening, and identification of studies for final analysis

HRF, high-risk features; DDW, digestive disease week; ACG, American College of Gastroenterology; UEGW, United European Gastroenterology Week

Overall, the 8 studies included 1882 patients (40.4% male; mean age 64.5 years; Table 1). Among these, Kohli et al [20], Sighinolfi et al [23] and Imbe et al [19] reported the location of the cyst, most commonly in the head of the pancreas (39.5%). The diagnosis of advanced neoplasia was made based on a histological evaluation of resected pancreas [18,21,23], cytological analysis of cells aspirated from pancreatic cysts [20,24], or a combination of the two [22]. Singhi et al also performed molecular testing on aspirated pancreatic cyst fluid to assess mutations in KRAS, GNAS, VHL, TP53, PIK3CA, and PTEN genes [24]. Imbe et al did not provide details of how pancreas cancer was defined [19].

Table 1 Demographic information from individual studies

Diagnostic accuracy of recommendations for EUS for detecting advanced neoplasia

EUS for ≥2 HRF

The AGA recommends performing EUS only for PCL with ≥2 HRF [11]. Seven studies provided information regarding the presence of ≥2 HRF in the PCL, seen in 234 patients [18,20-25]. The sensitivity and specificity of ≥2 HRF in detecting advanced neoplasia were 41.7% (95%CI 19.5-67.8%) and 90.8% (95%CI 81.9-95.5%), respectively. The corresponding positive and negative predictive values were 30.4% (95%CI 19.4-44.2%) and 94.3% (95%CI 89.6-97.0%), respectively (Table 2).

Table 2 Performance characteristics of number of high-risk features in predicting advanced neoplasia in pancreatic cystic lesions

EUS for ≥1 HRF

The ACG recommends performing EUS in the presence of 1 or more HRF seen on cross-sectional imaging [12]. Four studies provided information regarding the presence of ≥1 HRF in the PCL, seen in 329 patients [18-20,22]. The sensitivity and specificity of ≥1 HRF in detecting advanced neoplasia were 77.1% (95%CI 66.1-85.3%) and 72.7% (95%CI 50.4-87.5%), respectively. The corresponding positive and negative predictive values were 17.9% (95%CI 10.3-29.4%) and 98.1% (95%CI 90.8-99.6%), respectively (Table 2).

Missed rates for AN

Different investigators have provided varying amounts of information regarding AN in relation to HRFs. Seven studies [18,20-25] provided cytological or histological assessment to confirm the presence of AN. In the pooled analysis of these studies, 173 patients demonstrated AN. Of these, 77 had at least 2 HRF (44.5%), meeting the threshold for EUS using the AGA criteria. However, 96 (55.5%) patients with definite AN had <2 HRF and hence would not undergo EUS per the AGA recommendations.

Five studies [18,20-22,24] provided the number of total HRFs in the study population. In the pooled analysis of these 5 studies, 88 patients were reported to have AN. Of these 88 patients, 20 had no HRF (22.72%), 33 had 1 HRF (37.5%), and 35 had 2 HRFs (39.7%).

Three studies [18,20,22] described individual HRFs in the PCLs. Among the 201 patients with 1 HRF reported in the studies, 26 harbored AN, with a pooled incidence of 12.93%.

Using pooled data from all studies reporting relevant data, the rate for missed AN was calculated. The missed AN rate when using 2 HRF was 4.7% (95%CI 2.4-8.8; I²=87%) in all patients with PCL and 58.3% (95%CI 32.3-80.5; I²=85%) in patients with confirmed neoplasia. The missed AN rate with 1 HRF was 1.8% (95%CI 0.8-4.2; I²=58%) in all patients with PCL and 23.6% (95%CI 15.8-33.6; I²=0) in patients with confirmed neoplasia (Fig. 2-5). Overall, there was significant heterogeneity among the studies, with an I² value of 96.47%.

Figure 2 Absolute pooled missed advanced neoplasia rates using ≥2 high-risk features in all patients with pancreatic cysts

CI, confidence interval

Figure 3 Missed advanced neoplasia rates using the ≥2 high-risk features criterion in patients with confirmed advanced neoplasia

CI, confidence interval

Figure 4 Absolute pooled missed advanced neoplasia rates using ≥1 high-risk features in all patients with pancreatic cysts

CI, confidence interval

Figure 5 Missed advanced neoplasia rates using the ≥1 high-risk features criterion in patients with confirmed advanced neoplasia

CI, confidence interval

Discussion

Increasing use of high-resolution abdominal imaging has led to more frequent incidental detection of asymptomatic PCL. Two professional gastroenterology societies, the ACG and AGA, have published consensus guidelines on the suggested approach to these lesions [11,12]. Both guidelines have risk-stratified PCL based on their morphology on cross-sectional imaging. This approach is crucial in identifying the minority of PCL that harbor early invasive cancer or high-grade dysplasia. The 2015 AGA guideline recommended the performance of EUS only if 2 or more HRF are present [11]. This conditional recommendation is based on a poor quality of evidence and precludes performing an EUS in patients with 1 HRF, who may still be at higher risk for development of AN compared to patients with no HRF [20]. The ACG, however, recommends performing an EUS in the presence of any HRF [12].

In this study, we demonstrated that performing EUS in patients with at least 1 HRF, as suggested by the ACG, had a sensitivity of 77.1% for diagnosing AN. Restricting EUS to patients with ≥2 HRF, reduced the sensitivity for AN to 41.7%. Indeed, this pooled analysis demonstrated that 37.5% of patients with AN had only 1 HRF.

Other investigators have similarly reported that following the AGA recommendations and performing EUS only in patients with at least 2 HRF reduces sensitivity for the detection of high-risk lesions. Kohli et al demonstrated that using the AGA guidelines reduced the sensitivity for detecting pancreatic malignancy to 50%, and suggested that performing EUS in patients with at least 1 HRF on imaging would increase the sensitivity to 100%. On the other hand, the number of EUS procedures performed would have been reduced by 91% if AGA guidelines were followed, compared to a 67% reduction for EUS procedures performed for 1 HRF [20]. Singhi et al similarly showed that AN was detected with a sensitivity and specificity of 62% and 75% by using AGA guidelines. Moreover, 45% of advanced neoplastic cysts were missed by the AGA guidelines. They proposed an alternative algorithm to manage PCL, which included molecular analysis of cyst fluid aspirate [24]. In a study by Sahar et al, abiding by the AGA guidelines and using 2 HRF as threshold for performing EUS, the diagnosis of malignant and high-risk premalignant lesions (including pancreatic adenocarcinoma, mucinous cystadenoma, neuroendocrine tumors, and intraductal papillary mucinous neoplasm with dysplasia) had a 40% sensitivity and 100% specificity. If EUS was used based on a threshold of 1 HRF on imaging, malignant and high-risk premalignant lesions would have been identified with 80% sensitivity and 95% specificity [22]. However, it is not known if an early diagnosis of AN within a pancreatic cyst necessarily prevents progression to cancer or has a mortality benefit.

It is also notable that the 3 individual HRF have a variable risk of progression to malignancy. The AGA technical review calculated that the risk of malignancy in a pancreatic cyst was the highest with mural nodularity (odds ratio [OR] 7.73, 95%CI 3.38-17.67), followed by a cyst size >3 cm (OR 2.97, 95%CI 1.82-4.85). A dilated pancreas duct did not increase the odds of malignancy (OR 2.38, 95%CI 0.71-8.0) [8]. Hence, a risk stratification algorithm that considers these 3 HRF to be equivalent may not be ideal.

Certain investigators have also questioned the rationale for performing EUS in the presence of ≥2 HRF on MRI since the interobserver agreement between EUS and MRI is poor and the misdiagnosis rate of cross-sectional imaging is high. In view of the limited accuracy of cross-sectional imaging, EUS can improve diagnostic yield with its higher resolution and can also provide aspirated fluid for chemical and cytological analysis [26].

An EUS can also help in the down-stratification of PCL, which can lead to avoiding unnecessary imaging procedures and a consequent reduction in healthcare costs [27,28]. In the study by Sahar et al, EUS was useful in classifying 14 cysts as pseudocysts and serous cystadenomas and consequently down-stratifying this group and removing them from a surveillance program [22]. It is notable that an EUS with cyst aspiration is a relatively safe endoscopic procedure with a low complication rate in the hands of an experienced operator [29].

Different scientific organizations have provided broadly similar recommendations for the management of PCL, but there are some significant variations among them. The European Society of Gastrointestinal Endoscopy recommends EUS as an adjunct to other imaging modalities, especially if the cystic lesion demonstrates clinical or radiological features of concern [4] and if a precise diagnosis changes patient management [30]. It also recommends against performing EUS for lesions ≤10 mm in diameter with no high-risk stigmata [30]. Notably, this guideline uses a cyst size of ≥4 cm as a cutoff, instead of the 3 cm recommended by the ACG and AGA guidelines. The revised Fukuoka guidelines recommend EUS for patients with worrisome features on imaging, including a cyst of ≥3 cm, enhancing mural nodule <5 mm, thickened enhanced cyst walls, main pancreas duct size of 5-9 mm, abrupt change in the duct caliber with distal pancreatic atrophy, lymphadenopathy, an elevated serum level of carbohydrate antigen 19-9 and a rapid rate of cyst growth >5 mm/2 years [9]. The revised guidelines are more aggressive than those in 2012 and the Sendai guidelines in 2006, with the recommendation for initial surveillance to occur at a shorter interval [10,31]. The American College of Radiology, however, stratifies risk based on age at presentation, interval increase in cyst size, and overall cyst size [32]. It recommends EUS even for cysts >1.5 cm in size, especially if main duct communication is visualized or interval growth is detected on serial imaging [32]. It is notable that the ACG and AGA guidelines were authored by gastroenterologists, whereas the other guidelines were authored by an expert multidisciplinary panel consisting of gastroenterologists, surgeons, radiologists, and pathologists [31]. There is no single guideline that has sufficient accuracy to definitively guide clinical decision making [31].

Recently, Wu et al published a meta-analysis comparing the Fukuoka and AGA guidelines in risk-stratifying the malignant potential of pancreatic cysts. The study revealed that the diagnostic accuracy was similar and “unsatisfactory” with either guideline [33]. Our meta-analysis, in contrast, compared 2 American society guidelines and demonstrates that the sensitivity of the AGA guideline is lower, with a higher missed rate for advanced neoplasia compared to the ACG guideline. The ACG guideline, however, has a higher utilization of EUS for assessing PCL.

This systematic review and meta-analysis is the first study to perform a pooled analysis of multiple retrospective studies to assess the diagnostic accuracy of the AGA and the ACG guidelines, based on the number of HRF on cross-sectional imaging for EUS evaluation, and provides an objective assessment of each of these guidelines in risk-stratification. It also provides comparative objective data regarding the reduction in unnecessary EUS procedures based on 2 commonly used guidelines. Given the small number of populations in these retrospective studies, our results could have been underpowered. However, given the uncertain natural history of pancreatic cysts and their slow rate of progression to malignancy, prospective trials are challenging to perform.

This meta-analysis is heavily limited by the significant heterogeneity in the individual source studies. The presence or absence of advanced neoplasia was based on varying criteria, including a combination of cyst fluid analysis and cytology [20,22,24], histopathological review of resected specimens [18,23,25], imaging features [19], or a combination of the above [21]. Similarly, the follow up after EUS was not uniform, as some investigators performed resection of the cyst [18,21,23,25] whereas others opted for surveillance [19,20]. This lack of uniformity impedes a clear understanding of the natural history of the pancreatic cysts. Finally, some studies assessed the HRF and EUS findings as a method to select candidates for surgery, rather than solely to assess the indication for an EUS.

While these studies assessed the primary outcome differently, our aim was to compare the existing literature on diagnostic accuracy and prediction of AN while using either ≥1 or ≥2 HRF. Previous guidelines have also formulated their decision based on such available data [11,12]. In our review, carefully planned prospective randomized trials comparing the 2 strategies are lacking and no single study has conclusively answered the question of how many criteria should be used to refer a patient for EUS.

In conclusion, this systematic review and meta-analysis suggests that performing EUS for PCL with at least 1 HRF could offer a higher sensitivity in detecting advanced neoplasia compared to ≥2 HRF, with a high negative predictive value. Given the safety profile of EUS, it may be reasonable to consider using ≥1 HRF as the criterion for performing an EUS in select patients with PCL. However, prospective multicenter randomized studies are required to validate these findings.