Biomarker Risk Score Algorithm and Preoperative Stratification of Patients with Pancreatic Cystic Lesions

Abstract

Introduction:

Pancreatic cysts are incidentally detected in up to 13% of patients undergoing radiographic imaging. Of the most frequently encountered types, mucin-producing (mucinous) pancreatic cystic lesions may develop into pancreatic cancer, while non-mucinous ones have little or no malignant potential. Accurate preoperative diagnosis is critical for optimal management but has been difficult to achieve, resulting in unnecessary major surgery. Here, we aim to develop an algorithm based upon biomarker risk scores to improve risk stratification.

Methods:

Patients undergoing surgery and/or surveillance for a pancreatic cystic lesion with diagnostic imaging and banked pancreatic cyst fluid were enrolled in the study following informed consent (n=163 surgical, 67 surveillance). Cyst fluid biomarkers with high specificity for distinguishing non-mucinous from mucinous pancreatic cysts (vascular endothelial growth factor [VEGF], glucose, carcinoembryonic antigen [CEA], amylase, cytology, and DNA mutation) were selected. Biomarker risk scores were used to design an algorithm to predict preoperative diagnosis. Performance was tested using surgical (retrospective) and surveillance (prospective) cohorts.

Results:

In the surgical cohort, the biomarker algorithm outperformed the preoperative clinical diagnosis in correctly predicting the final pathological diagnosis (91% vs. 73%; P<0.000001). Specifically, non-mucinous serous cystic neoplasms (SCN) and mucinous cystic neoplasms (MCN) were more frequently correctly classified by the algorithm than clinical diagnosis (96% vs 30%; P<0.000008 and 92% vs. 69%; P=0.04, respectively). In the surveillance cohort, the algorithm predicted a preoperative diagnosis with high confidence based upon a high biomarker score and/or consistency with imaging from ≥1 follow-up visits.

Conclusion:

A biomarker risk score-based algorithm was able to correctly classify pancreatic cysts preoperatively. Importantly, this tool may improve initial and dynamic risk stratification, thus reducing overdiagnosis and underdiagnosis.

Precis

We developed a novel biomarker scoring system and risk algorithm to provide a more objective and accurate means of determining preoperative pancreatic cyst diagnosis. This type of tool may improve the initial risk stratification of new patients or existing patients who are currently under surveillance.

INTRODUCTION

Pancreatic cancer remains one of the deadliest cancers due to its often advanced stage at presentation and the lack of therapeutic options. The 5-year relative survival rate by stage of diagnosis is 39.4% for localized (confined to pancreas), 13.3% for regional (spread to regional lymph nodes), and only 2.9% for distant (metastasis); the combined rate is 9% (1). To improve the poor prognosis, early detection and prevention are key. Three precursors have been identified for pancreatic cancer – pancreatic intraepithelial neoplasia (PanIN), intraductal papillary mucinous neoplasm (IPMN), and mucinous cystic neoplasm (MCN) (2). While PanINs are typically microscopic, IPMN and MCN are macroscopic pancreatic cystic lesions detectable on radiographic imaging so provide a clinical window of opportunity for early detection.

Pancreatic cysts are being identified more frequently in the general population due to increased awareness and advances in imaging technology. Up to 13% of patients undergoing radiographic imaging for an unrelated reason are diagnosed with an incidental pancreatic cyst (3–6). Pancreatic cystic lesions can be broadly classified into two groups, non-neoplastic (i.e. pseudocysts) or neoplastic (7). The latter group can be further categorized into mucinous or non-mucinous cystic neoplasms. Importantly, mucinous pancreatic cystic lesions such as IPMN and MCN may develop into pancreatic cancer, but certain non-mucinous ones such as serous cystic neoplasms (SCN) are largely benign with little to no potential for malignant transformation. Other less common non-mucinous cyst types such as cystic neuroendocrine tumors (PaNET) and solid pseudopapillary neoplasms (SPN) may, however, have malignant potential. Within the mucinous category, pancreatic cysts have variable malignant potential such that the majority with low malignant potential may be safely managed without surgery. For the management of presumed IPMN and MCN, International Consensus Guidelines based upon imaging and clinical features were created in 2006 and subsequently revised but have low specificity despite high sensitivity for identifying cysts that should be resected (8–11). Even with the implementation of these criteria, low-risk or benign lesions may be misclassified and resected, emphasizing the need for more accurate early stratification.

It is challenging to determine optimal management (resection versus surveillance) of pancreatic cysts (12). Optimal management may change for an individual pancreatic cyst over time as a function of evolving clinical, radiographic, cytologic, biochemical and DNA data. An accurate preoperative diagnosis is the most important first step to inform optimal pancreatic cyst management. Diagnostically, the most common pancreatic cystic lesions encountered clinically are IPMN, MCN, SCN and pseudocysts. Although these pancreatic cyst types possess characteristic radiographic features, imaging modalities such as computed tomography (CT), magnetic resonance imaging (MRI) and endoscopic ultrasound (EUS) are often unable to accurately diagnose them or differentiate benign from premalignant/malignant ones (12, 13). The preoperative clinical diagnosis has been reported to differ from the final pathological diagnosis following surgery in up to one-third of pancreatic cyst cases (14, 15). The stakes are high, as cases of overdiagnosis may result in “overtreatment” or potentially unnecessary resection of low-risk pancreatic cysts. Conversely, “undertreatment” or monitoring cystic lesions that are at high risk of malignancy may miss the opportunity to prevent or cure early pancreatic cancer.

Recently, tissue/fluid biomarker discovery efforts have identified candidates that can distinguish between pancreatic cyst types and supplement imaging in clinical decision-making (16–18). Of particular interest is pancreatic cyst fluid obtained by EUS-guided fine needle aspiration (EUS-FNA), a minimally invasive procedure, for biochemical, cytological and DNA analysis. We hypothesized that integrating selected high specificity cyst fluid biomarkers would improve diagnostic accuracy and thus initial risk stratification of patients. In the present study, we designed an algorithm based upon biomarker risk scores and evaluated its performance in both surgical and surveillance cohorts.

METHODS

Patient population and data collection.

Patients undergoing resection (retrospective cohort, 2002–2020) or surveillance (prospective cohort, 2016–2020) for pancreatic cystic lesions were identified in an institutional database. Those with diagnostic imaging (at least one CT, MRI or EUS) and banked pancreatic cyst fluid (obtained prior to/during surgery or during surveillance for the respective cohorts) were consented and enrolled in the study. Demographic, clinical (e.g., incidental versus symptomatic presentation), radiographic (presumed diagnosis provided by radiologist), cytology and biologic marker information prior to surgery or during surveillance were gathered from our institutional database supplemented by electronic medical records in accordance with the Indiana University Institutional Review Board. The preoperative differential diagnosis was obtained from the indication for surgery as stated in the surgeon’s clinical or operative notes. Final pathological diagnosis was obtained from the pathology report. Cyst size (largest diameter of index lesion) was obtained from the pathology report or if not available, from the preoperative imaging report.

Cyst fluid biomarkers and scoring.

Pancreatic cyst fluid was collected at the time of EUS-FNA (either cohort) or surgery (surgical cohort) and aliquoted for storage at −80. Samples were thawed on ice prior to assay. A majority of the surgical cohort was previously assayed for vascular endothelial growth factor (VEGF), carcinoembryonic antigen (CEA) and glucose (19–21). Each marker listed below was scored according to the criteria or threshold for mucinous/non-mucinous cysts (0 or 1 point each). At a minimum, each sample was assayed for VEGF; other markers were scored if fluid and/or data was available.

Vascular endothelial growth factor (VEGF) –

VEGF levels were assayed by enzyme-linked immunosorbent assay (ELISA, R&D) and scored based upon the published threshold of >5000 pg/ml to identify non-mucinous SCN with 100% sensitivity and 87.3% specificity (19).

Glucose –

Cyst fluid glucose was measured using a patient glucometer and scored according to the threshold of 50 mg/dl to differentiate non-mucinous from mucinous pancreatic cysts. We previously published that cyst fluid glucose (≤ 50 mg/dl) is 92% sensitive and 87% specific for mucinous pancreatic cysts (20).

Carcinoembryonic antigen (CEA) –

Cyst fluid CEA levels were determined during cyst fluid analysis as noted in patient medical records or by laboratory assay (20) and scored using thresholds of <10 ng/ml for non-mucinous SCN or >192 ng/ml for mucinous cysts (22). CEA >800 ng/ml is 98% specific for mucinous cysts (23). We previously reported that the VEGF-A/CEA combination is 95.5% sensitive and 100% specific for SCN (19).

Amylase –

Cyst fluid amylase levels were determined during cyst fluid analysis as noted in patient records and scored based on the published threshold of <250 units/L to exclude pseudocysts. Amylase <250 has 98% specificity for cysts without pancreatic ductal connectivity (i.e., SCN and MCN) (23).

Cytology –

If the cytology report included terms consistent with the presence of a mucinous cyst such as “mucin”, “mucinous epithelial cells”, “viscous fluid”, or “IPMN”, one point was given to rule in mucinous cysts (18).

Mutation –

KRAS/GNAS gene alterations or mutations detected in cyst fluid as noted in patient medical records or in a prior study using Safe-SeqS sequencing (24) were assigned one point to rule in mucinous cysts (25). KRAS mutation is very specific (96%) but only 45% sensitive for mucinous cysts (26); GNAS mutation identifies IPMN with 100% specificity (25). Mutations/alterations (i.e. loss of heterozygosity [LOH] on 3p) in the von Hippel-Lindau (VHL) gene or cadherin-associated protein beta 1 (CTNNB1) gene mutations are predictive of SCN or SPN respectively (24, 27, 28).

Algorithm.

The algorithm was developed and trained using the surgical cohort. Testing was performed by using the algorithm to predict preoperative diagnosis when blinded to the sample’s final diagnosis, simulating intended use of the algorithm in a surveillance cohort. The algorithm’s performance was determined by comparing the diagnosis predicted in this blinded manner to the final pathological diagnosis.

Statistics.

Descriptive data were expressed as mean ± SEM or median (interquartile range). For two-group comparisons, the Mann-Whitney (continuous variables) or Fisher’s exact (categorical) test was performed. To assess diagnostic performance, the preoperative diagnosis (either clinical or from algorithm) was “correct” only if it exactly matched the final pathological diagnosis of pseudocyst, SCN, mucinous cyst (IPMN/MCN), PaNET or SPN. If multiple differential diagnoses were given or possible, the preoperative diagnosis was “incorrect”. McNemar’s test was performed to compare diagnostic performance of the two methods (algorithm or clinical diagnosis for the same sample). Statistical significance was defined as a P value < 0.05. Statistical analysis was performed using GraphPad Prism (version 8).

RESULTS

Patients undergoing resection for pancreatic cystic lesions were identified, and those with diagnostic imaging prior to surgery as well as banked pancreatic cyst fluid samples from EUS or surgery were enrolled in the study as the surgical cohort (n=163). Final pathological diagnosis included IPMN (n=73; 40 low-grade, 22 high-grade, 11 invasive), MCN (n=36), SCN (n=27), PaNET, (n=8), SPN (n=3), and pseudocyst (n=16). Of the neoplastic cysts, 38 were classified as non-mucinous (SCN, PaNET, SPN) and 109 as mucinous (IPMN, MCN). Non-neoplastic pseudocysts were included with the non-mucinous cysts for analysis in the present study.

The non-mucinous group was significantly younger than the mucinous group (median age [interquartile range, IQR] 58 [44–69] vs. 65 [52–74] years, respectively; P=0.024) (Table 1). There was no significant difference in proportion of males/females between the non-mucinous and mucinous groups (31.5% vs. 27.5% males, respectively; P=0.71), but a difference in race was evident (87% vs. 96.3% white, respectively; P=0.042). Incidentally discovered pancreatic cysts comprised approximately 31% of each group (P>0.999). Median cyst size (IQR) was significantly larger in non-mucinous cysts at 3.6 (2.2–5.8) cm compared to 2.9 (2.0–4.2) cm in mucinous cysts (P=0.02), but interestingly, did not differ significantly between incidentally discovered and symptomatic cysts (P=0.6, not shown).

Table 1.

Characteristics of Surgical and Surveillance Pancreatic Cyst Cohorts

Characteristic	Nonmucinous (n = 54)	Mucinous (n = 109)	p Value (nonmucin ous vs mucinous)	Total surgical cohort (n = 163)	Surveillance cohort (n = 67)	p Value (surgical vs surveillance)
Age, y, median (IQR)	58 (44–69)	65 (52–74)	0.024	63 (49–72)	72 (58–77)	0.0003
Sex, m, n (%)	17 (31.5)	30 (27.5)	0.71	47 (28.8)	19 (28.4)	>0.999
Race, white, n (%)	47 (87.0)	105 (96.3)	0.042	152 (93.3)	64 (95.5)	0.76
Incidental discovery, n (%)	17 (31.5)	34 (31.2)	>0.999	51 (31.3)	51 (76.1)	<0.0001
Cyst size, cm, median (IQR)	3.6 (2.2–5.8)	2.9 (2.0–4.2)	0.02	3.0 (2.0–4.6)	2.0 (1.5–3.0)	<0.0001
Surgery, n (%)	54 (100)	109 (100)	>0.999	163 (100)	4 (6.0)	-
No follow-up past initial EUS, n (%)*	n/a	n/a	-	n/a	7 (11.1)	-
At least 1 follow-up visit with imaging, n (%)*	n/a	n/a	-	n/a	56 (88.9)	-
≥2 follow-up visits with imaging, n (%)*	n/a	n/a	-	n/a	30 (47.6)	-
Total follow-up, mos, mean ± SEM (range)	n/a	n/a	-	n/a	19.7 ± 1.6 (4 – 67)	-

^*At Indiana University Health

IQR, interquartile range; n/a, not applicable; SEM, standard error of the mean

Cyst fluid biomarkers with reported evidence of high specificity (>85%) at distinguishing pancreatic cystic lesions were chosen for inclusion (see Methods). Combinations of these markers were employed to reinforce confidence in a preoperative diagnosis, critical for patients undergoing surveillance. To classify non-mucinous SCN, Vascular endothelial growth factor (VEGF) >5000 pg/ml, Glucose >50 mg/dl, Carcinoembryonic antigen (CEA) <10 ng/ml, and Amylase <250 units/l were selected. For mucinous pancreatic cystic lesions, Glucose <50 mg/dl, CEA >192 ng/ml, Cytology (mucinous), and Mutation (KRAS/GNAS) were used. Biomarkers scores were calculated by earning 1 point for meeting the criteria or threshold for each of the 4 non-mucinous or mucinous biomarkers; the respective scores are referred to by the biomarker initials VGCA and GCCM (underlined above). A high VGCA or GCCM score “rules in” either a non-mucinous SCN or a mucinous (IPMN/MCN) cyst respectively. In this surgical cohort, a VGCA score of 3 or 4 was 96% specific (5 false positives were PaNET), 63% sensitive, and 91% accurate for SCN (likelihood ratio of 15.75); a GCCM score ≥2 was 98% specific, 83% sensitive, and 88% accurate for mucinous cysts (likelihood ratio of 41.5). Furthermore, positive Glucose<50 and CEA>192 (score of 2) demonstrated 100% specificity for mucinous cysts; conversely, a score of “0” was 96% specific for a non-mucinous cyst (both biomarkers required).

The biomarker risk score-based algorithm was developed and trained using the surgical cohort (Figure 1). The first decision point is VEGF (> or < 5000 pg/ml), followed by calculation of the VGCA and GCCM scores. Based upon the risk scores alone meeting certain criteria, a diagnosis of “SCN” or “mucinous cyst” can be obtained with high confidence. Other possible diagnoses may be assigned by exclusion such as “not SCN”, “not SCN/not muc (mucinous)” or “other”, which may be further refined by confirmatory imaging, cytology or mutation status. For example, of the 16 known pseudocysts, 15 were categorized as “not SCN” or “not SCN/not muc” by the algorithm, and 12 were subsequently correctly classified by at least one confirmatory imaging. Only a small number of PaNET and SPN were included in this study; these entities are very rare and can be diagnosed accurately by positive cytology. Of 8 resected PaNET, 5 scored as possible SCN (VGCA score 3 or 4) but 4 were corrected by cytology/EUS to PaNET (5^th one had outside cytology which was negative), demonstrating the importance of cytology/imaging in discriminating between SCN and NET.

Figure 1.

Biomarker Risk Score Algorithm for Initial Stratification. The diagram depicts the decision-making process which begins with a patient who presents with a pancreatic cyst on imaging and has cyst fluid available for analysis. The vascular endothelial growth factor (VEGF) test is the entry point followed by VGCA and GCCM score calculations, as indicated, to assist with predicting preoperative diagnosis. If scores alone are not predictive, then imaging and/or cytology may be informative. Possible preoperative diagnoses are shown in the shaded boxes. CEA, carcinoembryonic antigen; EUS, endoscopic ultrasound, muc, mucinous; PaNET, cystic pancreatic neuroendocrine tumor; pseudo, pseudocyst; pt, point; SCN, serous cystic neoplasm; SPN, solid pseudopapillary neoplasm; VHL, von Hippel-Lindau

The performance of the algorithm was compared with that of the preoperative clinical diagnosis (indication for surgery) at predicting the final pathological diagnosis, the gold standard (Table 2). For non-mucinous SCN, the biomarker algorithm significantly outperformed the preoperative clinical diagnosis at correctly predicting the final diagnosis (96% [26/27] vs. 30% [8/27] respectively; P=0.000008); the 1 incorrect case predicted by the algorithm to be “SCN or mucinous cyst” had unusual characteristics of SCN and PaNET on final pathology. Of the 26 correct algorithm-predicted SCN, 20 (77%) were not predicted with certainty by imaging alone (either multiple differential diagnoses including SCN or incorrect diagnosis); characteristic microcystic/honeycomb features were detected on imaging in only 6 cases. Furthermore, in the cases missed by clinical diagnosis, the diagnosis was “cystic lesion” or “IPMN”. Considering all non-mucinous cysts, the algorithm correctly diagnosed 87% compared to 46% by preoperative clinical diagnosis (P<0.000001) and similarly for the entire surgical cohort (91% vs. 73% respectively; P<0.000001). The algorithm was also superior in the entire younger surgical cohort (<50 years old), correct in 96% of cases versus 62% by clinical diagnosis (P=0.0006). Although there was no significant difference in performance for the mucinous group as a whole (94% algorithm vs. 86% clinical; P=0.08), MCNs in particular were correctly classified as “mucinous cyst” more frequently by the algorithm versus a clinical diagnosis of an undefined cystic lesion (92% vs. 69%; P=0.04). For pseudocysts, 75% were correctly predicted by the algorithm versus 50% by clinical diagnosis (P=0.125). Performance of the two methods was the same for IPMN, PaNET, and SPN at 95, 75, and 100% correct for each, respectively (P=1.0). For IPMN, neither biomarker risk scores nor predictions correlated with dysplastic grade. Average GCCM score was 2.5, 2.4 and 2.5 for low-grade, high-grade and invasive groups, respectively. Percent correct prediction by the algorithm did not differ significantly between the three grades (93% correct for low, 95% high, 100% invasive; P=0.62). Taken together, the algorithm significantly improves the preoperative classification of SCN and MCN (mucinous).

Table 2.

Preoperative Predictive Performance of the Biomarker Algorithm Compared with Clinical Diagnosis

Variable	Biomarker algorithm, correct diagnosis*, n (%)	Clinical diagnosis, correct*, n (%)	p Value†
SCN (n = 27)	26 (96)	8 (30)	0.000008
MCN (n = 36)	33 (92)	25 (69)	0.04
Pseudocyst (n = 16)	12 (75)	8 (50)	0.125
IPMN (n = 73)	69 (95)	69 (95)	1
PaNET (n = 8)	6 (75)	6 (75)	1
SPN (n = 3)	3 (100)	3 (100)	1
Nonmucinous (n = 54)	47 (87)	25 (46)	<0.000001
Mucinous (n = 109)	102 (94)	94 (86)	0.08
Total surgical cohort (n = 163)	149 (91)	119 (73)	<0.000001
Surgical cohort, <50 y old (n = 45)	43 (96)	28 (62)	0.00006

^*Correct: preoperative diagnosis matches the final pathological diagnosis (“mucinous cyst” for IPMN/MCN); if >1 preoperative diagnosis or unclear diagnosis, then not “correct”

^†McNemar’s test

IPMN, intraductal papillary mucinous neoplasm; MCN, mucinous cystic neoplasm; PaNET, cystic pancreatic neuroendocrine tumor; SCN, serous cystic neoplasm; SPN, solid pseudopapillary neoplasm

Finally, the algorithm was tested on patients being monitored for a pancreatic cystic lesion with cyst fluid obtained during EUS-FNA (prospective cohort, n=67). Characteristics of the total surveillance and surgical cohorts were compared (Table 1). Median age (IQR) of the surveillance group was 72 (58–77) years, significantly older than the surgical cohort at 63 (49–72) years (P=0.0003). Male/female status (28% male) and race (>90% white) were similar for both cohorts. Pancreatic cysts were incidentally discovered more frequently in the surveillance than surgical group (76.1 vs 31.3%, P<0.0001). Cyst size (IQR) was significantly smaller in the surveillance group (2.0 [1.5–3.0] vs 3.0 [2.0–4.6] cm, P<0.0001). Within the surveillance group, 4 have undergone surgery to date. Of the remainder, 7 (11.1%) had no follow-up at least at our institution beyond the initial EUS, 56 (88.9%) had at least 1 follow-up visit with imaging (CT/MRI), and 30 (47.6%) had 2 or more follow-ups with imaging. The mean total follow-up time was 19.7 months (range: 4–67 months).

Patient VCCA and GCCM risk scores were calculated, and the predicted diagnosis was determined using the biomarker algorithm (Table 3 and details in eTable 1). The differential diagnosis from EUS or CT/MRI imaging done prior to EUS as well as from any follow-up visits was noted and compared to the algorithm prediction. Of 4 that have been resected (patients #3, 11, 13, 44), the algorithm-predicted diagnosis was confirmed in 3 (2 IPMN, 1 NET); the remaining case was an extremely rare benign-acting cyst - namely, acinar cystadenoma. The VGCA and GCCM risk scores predictive of SCN or mucinous cysts were high (≥3 or ≥2) in 80 or 100% of cases respectively, providing high confidence in the diagnoses independent of imaging. In 80% of presumed SCN, the algorithm-predicted diagnosis narrowed down the differential diagnoses from imaging which included SCN; the algorithm confirmed an EUS or needle-based confocal laser endomicroscopy (nCLE) diagnosis of SCN in 73%. In 91% of presumed mucinous cysts (IPMN/MCN), the algorithm confirmed the single diagnosis from imaging (same diagnosis from 1 or more imaging exams) and similarly for 80% identified as mucinous by exclusion. This demonstrates stability of the cyst and increases the level of confidence in the accuracy of the algorithm-derived preoperative diagnosis. The algorithm also predicted 1 PaNET (patient #6), 1 pseudocyst (#25), and 2 “other” (#8, 18).

Table 3.

Surveillance Cohort: Algorithm-Predicted Diagnosis and Comparison with Imaging Differential Diagnosis

Diagnosis predicted by biomarker algorithm	% of total under surveillance (total n = 63)	Predict SCN: VGCA≥3 or mucinous cyst: GCCM≥2, n/N (%)	Narrows differential diagnosis†, n/N (%)	Confirms imaging diagnosis†, n/N (%)	Disagrees with diagnosis†, n/N (%)
SCN (N = 15)	24	12/15 (80)	12/15 (80)	11/15 (73)	1/15 (7)
Mucinous cyst (ie IPMN/MCN, N = 34)	54	34/34 (100)	3/34 (9)	31/34 (91)	0
PaNET (N = 1)	2	0	0	1/1 (100)	0
Mucinous cyst, exclusion* (“not SCN”, N = 10)	16	0	2/10 (20)	8/10 (80)	0
Pseudocyst, exclusion* (“not SCN”, N = 1)	2	0	0	1/1 (100)	0
Other, exclusion* (“not SCN, not mucinous”, N = 2)	3	0	0	0	2/2 (100)

VGCA score for nonmucinous serous cystic neoplasm (SCN): Vascular endothelial growth factor >5000, Glucose>50, Carcinoembryonic antigen (CEA) <10, Amylase<250 (1 point each); GCCM score for mucinous cysts: Glucose<50, CEA>192, Cytology (mucin), Mutation (KRAS/GNAS) (1 pt each)

^*Diagnosis predicted by exclusion (ie “not SCN” or “not SCN/not mucinous”)

^†Algorithm “narrows” down differential diagnoses obtained from imaging (CT/MRI/endoscopic ultrasound [EUS]) or confirms single diagnosis from EUS/multiple imaging exams or “disagrees” with imaging diagnosis

IPMN, intraductal papillary mucinous neoplasm; MCN, mucinous cystic neoplasm; PaNET, pancreatic neuroendocrine tumor

eTable 1.

Surveillance Cohort: Biomarker Risk Score, Differential Diagnosis, and Predicted Diagnosis

No.	VEGF (pg/mL)	Glucose(mg/dL)	CEA (ng/mL)	Amylase (units/L)	VGCA score	Cytology result	DNA mutate on/LOH	GCCM score	Differential diagnosis						Predicted diagnosis from algorithm (biomarker/imaging)	Algorithm vs imaging‡
									From EUS	Earlier MRI	Earlier CT	From follow-up MRI†	From follow-up CT†	Surgery, final pathological diagnosis
1	18630	93	<0.2	-	3	Hypocellular	LOH 3p	0	SCN*	SPN/MCN/SCN	-	Cystic mass	-	-	SCN	Confirms EUS/nCLE, narrows
2	10000	88	-	-	2	-	None	0	SCN	sb IPMN>SCN	-	sb IPMN; SCN; SCN/sb IPMN	-	-	SCN	Confirms EUS, narrows
3	90667	<20	38730	57	2	Mucin, Pseudo cyst	KRAS/GNAS	4	pseudo/MCN	-	MCN/SCN/IPMN	-	-	PDAC, likely IPMN invasi ve	Muc cyst (IPMN)	Confirms final dx
4	119604	99	1.5	15	4	Benign	LOH 3p	0	SCN	SCN	-	sb IPMN and SCN (2)	-	-	SCN	Confirms
5	52	<20	595	4991	0	Mucin	LOH	4	sb IPMN	Sb IPMN	-	-	-	-	Muc cyst	Confirms
6	2001	133	-	-	1	PaNET	-	0	PaNET	PaNET	-	PaNET; PaNET	-	-	PaNET	Confirms
7	64677	128	<0.2	-	3	Benign	None	0	Communicates with MPD,	IPMN	-	Main duct IPMN; SCN/IPMN	-	-	SCN	Narrows
8	125	133	0.1	300	2	Benign	None	0	SCN, dilated MPD	-	-	sb IPMN; dilated main pancrea tic duct; sb IPMN	-	-	Not SCN, not muc; other?	Disagrees
9	1968	<20	7.1	59097	1	Benign	None	1	Mural nodule	sb IPMN	-	-	-	-	Not SCN; muc cyst?	Confirms
10	6611	119	<0.2	178	4	Benign	None	0	sb IPMN	sb IPMN/SCN	-	-	-	-	SCN	narrows
11	2365	155	<0.2	-	2	PaNET	None	0	PaNET	PaNET	-	-	-	PaNET	Not SCN, not muc; PaNET	Confirms final dx
12	53769	107	0.7	156	4	Mucin	None	1	Multiple SCN	IPMN	-	IPMNs; IPMN	-	-	SCN	Confirms EUS, narrows
13	548	<20	4.6	2412	1	Macroph age	None	1	SCN; MCN	-	-	-	-	Acinar cystadeoma	Not SCN; muc cyst?	Disagrees with final dx
14	210147	118	7.2	-	3	Hypocellular	None	0	SCN	SCN/MCN/IPMN	-	SCN/MCN/IPMN; SCN	-	-	SCN	Confirms EUS, narrows
15	175846	127	<0.2	-	3	Mucinous	None	1	SCN	SCN	-	Stable; stable	-	-	SCN	Confirms
16	216	<20	426	-	0	Benign	None	2	No communicati On with PD	sb IPMN	-	sb IPMN	-	-	Muc cyst	Confirms
17	15416	82	1.4	308	3	Hypocellular	VHL history	0	SCN	Panccyst, not PaNET	-	Stable cystic pancreatic lesions; stable cysts	-	-	SCN	Confirms EUS, narrows
18	790	86	78	-	1	Benign	None	0	sb IPMN	IPMN	-	sb IPMN	-	-	Not SCN, not muc; other?	Disagrees
19	106935	112	5.7	-	3	Hypocellular	None	0	SCN	sb IPMN, 2 cysts (IPMN, index SCN)	-	sb IPMN; sb IPMN > SCN	-	-	SCN	Confirms EUS, narrows
20	11215	139	2.3	-	3	Benign	None	0	No communicati on with MPD	-	sb IPMN	Small pancrea tic cyst; tiny cyst	-	-	SCN	Disagrees
21	5749	<20	77900	-	1	Hypocellular	KRAS	3	Communicati on with MPD	sb IPMN	-	-	-	-	Muc cyst	Confirms
22	17083	112	0.4	-	3	Benign	None	0	Mucin ous vs serous	-	sb IPMN	sb IPMN; sb IPMN	-	-	SCN	Narrows
23	1613	42	627	-	0	Benign	None	2	No communicati on with PD	-	Pancrea tic cyst	Stable panc cyst	Stable panc cysts	-	Muc cyst	Narrows
24	911	<20	-	-	0	Mucin	None	2	sb IPMN	sb IPMN	-	sb IPMN; sb IPMN	-	-	Muc cyst	Confirms
25	866	85	25	34814	1	Pseudocyst	None	0	Pseudocyst	Pseudocyst/IPMN	-	Enlarging IPMN/pseudo cyst	-	-	Not SCN/muc cyst; pseudo?	Confirms
26	7808	<20	5218	-	1	Mucinous	KRAS	4	Mucinous cyst	IPMN	-	Stable sb IPMN	Stable	-	Muc cyst	Confirms
27	901	<20	-	-	0	Acellular	None	1	sb IPMN,	sb IPMN	Cystic neo plasm	-	Stabl e cyst	-	Not SCN; muc cyst?	Confirms
28	755	<20	94	24648	0	Hypocellular	KRAS	2	sb IPMN	sb IPMN/SCN	-	Stable IPMN	-	-	Muc cyst	Confirms
29	790	<20	24	31539	0	Acellular	KRAS	2	sb IPMN	PaNET	-	IPMN/cystic PaNET; sb IPMN	-	-	Muc cyst	Narrows
30	223	<20	14	2941	0	Hypocellular	None	1	sb IPMN	sb IPMN	-	Stable sb IPMN	-	-	Not SCN; muc cyst?	Confirms
31	2875	<20	2.6	133325	1	Mucinous	KRAS	2	Communicati on with PD	-	-	-	IPMN/MCN, stable cysts	-	Muc cyst	Confirms
32	1141	<20	-	-	0	Hypocellular	KRAS/GNAS	2	sb IPMN	sb IPMN	-	Stable sb IPMN; stable sb IPMN	-	-	Muc cyst (IPMN)	Confirms
33	0	<20	395	-	0	Mucin	KRAS	4	sb IPMN	sb IPMN	-	Stable sb IPMN	-	-	Muc cyst	Confirms
34	935	<20	177	-	0	Hypocellular	None	1	sb IPMN	-	Cyst	sb IPMN	-	-	Not SCN; muc cyst?	Confirms
35	5724	<20	37700	-	1	Mucinous	None	3	sb IPMN	sb IPMN	-	sb IPMN	Stable sb IPMN	-	Muc cyst	Confirms
36	1862	<20	96	-	0	Hypocellular	None	1	sb IPMN	sb IPMN	-	sb IPMN stable; stable IPMN	-	-	Not SCN; muc cyst?	Confirms
37	17	<20	72	-	0	Hypocellular	KRAS/GNAS	2	sb IPMN	sb IPMN	SCN/IPMN	Stable cyst; stable sb IPMN	-	-	Muc cyst (IPMN)	Confirms
38	275	<20	328	-	0	IPMN	KRAS	4	sb IPMN	IPMN	-	Stable cyst; stable sb IPMN	-	-	Muc cyst	Confirms
39	1811	<20	46	-	0	Benign	KRAS/GNAS	2	No communicati on with PD	IPMN	-	Stable sb IPMN; stable sb IPMN	-	-	Muc cyst (IPMN)	Confirms
40	1167	<20	115	-	0	Benign	None	1	sb IPMN	-	IPMN	-	Stable sb IPMN; stable sb IPMN	-	Not SCN; muc cyst?	Confirms
41	3484	<20	-	-	0	-	KRAS/GNAS	2	Done but no diagno sis	-	-	sb IPMN; sb IPMN	-	-	Muc cyst (IPMN)	Confirms
42	352	<20	209	<10	1	Mucin	KRAS/GNAS	4	sb IPMN	sb IPMN	-	Stable sb IPMN	-	-	Muc cyst (IPMN)	Confirms
43	112	<20	70	19800	0	Benign	None	1	Seroufluid	sb IPMN	-	IPMN	-	-	Not SCN; muc cyst?	Narrows
44	866	<20	137	560	0	Mucin	KRAS/GNAS/LOH+	3	Cystic lesion	-	Pseudo cyst/cysic	-	-	IPMN, low grade	Muc cyst (IPMN)	Confirms final dx
45	824	<20	-	-	0	Benign	KRAS	2	sb IPMN	sb IPMN	-	Stable sb IPMN	-	-	Muc cyst	Confirms
46	2952	<20	33880	-	0	Benign	None	2	sb IPMN	sb IPMN	-	-	-	-	Muc cyst	Confirms
47	1958	<20	-	-	0	Mucin	-	2	sb IPMN	-	sb IPMN	-	-	-	Muc cyst	Confirms
48	1759	<20	111	-	0	Hypocellular	GNAS	2	sb IPMN	IPMNs	-	Stable sb IPMN	Stable	-	Muc cyst (IPMN)	Confirms
49	6058	<20	6643	7.2	2	-	KRAS	3	Multiple cysts	IPMN	-	Stable IPMNs	-	-	Muc cyst	Confirms
50	213	<20	207	-	0	Mucinous	None	3	sb IPMN	sb IPMN	-	Stable sb IPMN	-	-	Muc cyst	Confirms
51	1509	<20	74	-	0	Mucin	KRAS	3	sb IPMN	-	IPMN	IPMN	-	-	Muc cyst	Confirms
52	169	<20	134	-	0	Benign	KRAS	2	sb IPMN	sb IPMN	-	Stable IPMN	-	-	Muc cyst	Confirms
53	118212	53	<0.2	-	3	Benign	None	0	SCN	sb IPMN	-	sb IPMN; stable sb IPMN	-	-	SCN	Confirms EUS, narrows
54	1524	<20	-	-	0	Hypocellular	KRAS/GNAS	2	sb IPMN	sb IPMN	-	Stable sb IPMN	-	-	Muc cyst (IPMN)	Confirms
55	1583	No result	109	23367	0	Mucin	KRAS	2	sb IPMN	cysts	-	Stable sb IPMN	-	-	Muc cyst	Confirms
56	74	<20	58	80	1	Mucinous	KRAS	3	sb IPMN multiple	sb IPMN	-	sb IPMN stable	-	-	Muc cyst	Confirms
57	81	<20	153	18849	0	Hypocellular	None	1	sb IPMN	sb IPMNs	-	Stable sb IPMN	-	-	Not SCN; muc cyst?	Confirms
58	8193	Too viscous	28	11490	1	Pseudocyst	None	1	Cystic neoplasm	-	Pseudo cyst/MCN	-	-	-	Not SCN; muc cyst?	Narrows
59	1487	<20	2.2	112580	1	Benign	None	1	sb IPMN	PD communi cation	-	sb IPMN	-	-	Not SCN; muc cyst?	Confirms
60	162	<20	1188	-	0	Mucin	KRAS	4	sb IPMN	pseudo/cyst/SCN/muc cyst	-	sb IPMN	-	-	Muc cyst	Narrows
61	545	<20	-	-	0	Mucin	KRAS/GNAS	3	sb IPMN	sb IPMN	-	sb IPMN	-	-	Muc cyst (IPMN)	Confirms
62	913	<20	142	140536	0	Mucin	None	2	sb IPMN	IPMN	-	IPMNs	-	-	Muc cyst	Confirms
63	302	<20	365	-	0	-	None	2	sb IPMN	Mixed IPMN	-	stable mixed IPMN	-	-	Muc cyst	Confirms
64	92597	110	24	-	2	Hypocellular	None	0	SCN	IPMN	-	sb IPMN	-	-	SCN	Confirms EUS, narrows
65	258	<20	298	-	0	Benign	None	2	sb IPMN	sb IPMN	-	sb IPMN	-	-	Muc cyst	Confirms
66	2392	<20	3	138106	1	Mucin	None	2	sb IPMN	sb IPMNs	-	sb IPMN	-	-	Muc cyst	Confirms
67	49684	89	-	-	2	SCN	None	0	SCN*	Mixed IPMN	Cyst/pseudo/cystic	-	-	-	SCN	Confirms EUS/nCLE, narrows

VGCA score for nonmucinous serous cystic neoplasm (SCN): Vascular endothelial growth factor (VEGF) >5000, Glucose>50, Carcinoembryonic antigen (CEA) <10, Amylase <250 (1 pt each); GCCM score for mucinous cysts: Glucose <50, CEA >192, Cytology (mucin), Mutation (KRAS/GNAS) (1 pt each)

^*EUS/nCLE

^†Each CT/MRI imaging diagnosis listed is from a separate follow-up visit (separated by semi-colons)

^‡“Narrows”: biomarker score narrows down differential diagnosis (dx) from imaging to the predicted diagnosis; “Confirms”: Confirms single EUS or imaging or pathological diagnosis; “disagrees”: score predicts different diagnosis than provided by imaging or final pathological diagnosis

IPMN, intraductal papillary mucinous neoplasm; LOH, loss of heterozygosity; MPD, main pancreatic duct; muc, mucinous; MCN, mucinous cystic neoplasm; nCLE, needle-based confocal laser endomicroscopy; PaNET, pancreatic neuroendocrine tumor; PD, pancreatic duct; sb, side-branch; VHL, von Hippel-Lindau

DISCUSSION

Pancreatic cystic lesions are detected due to symptoms or incidentally by diagnostic imaging such as CT/MRI. However, these imaging modalities rely on interpretation and are often imperfect at distinguishing between mucinous premalignant/malignant and non-mucinous or benign pancreatic cysts. High-resolution EUS and FNA with cyst fluid analysis may be performed in cysts >1cm or those with worrisome/high-risk features to assist with diagnosis. Despite this, several studies highlight the need for more accurate preoperative diagnosis to spare patients from potentially morbid pancreatic surgery. In a large multinational study of SCN patients, the indication for surgery was unclear diagnosis in 60% of cases (950 of 1590 total) (27). Another study reported that SCN and other benign cysts were resected in 11% (15/136) of incidentally discovered cysts due to preoperative misclassification as mucinous cysts or unclear diagnosis; furthermore, of 23 cases under surveillance and ultimately resected, 6 (26%) diagnosed preoperatively as mucinous or unclear diagnosis were benign on final pathology (14). Similarly, Cho et al reported that 20% of their cases had a premalignant/malignant preoperative diagnosis but a benign pathological diagnosis (15).

In the present study, we integrated protein, metabolite, cytological and molecular markers chosen for their high specificity in order to confidently rule in non-mucinous SCN or mucinous pancreatic cysts. High VGCA and GCCM biomarker risk scores alone were able to predict SCN or mucinous pancreatic cysts (IPMN/MCN) with 91% and 88% accuracy, respectively. In cases, where a potential diagnosis was predicted by exclusion, the algorithm incorporated imaging to assist in providing a more definitive preoperative diagnosis. Overall in the surgical cohort, we demonstrated that the algorithm significantly outperformed the clinical diagnosis (indication for surgery) in correctly predicting the final pathological diagnosis (91% vs. 73% correct). Specifically, this was driven by the algorithm’s superior performance in the SCN (96% vs. 30% clinical diagnosis) and mucinous MCN (92% vs. 69%) subsets. Only 6 of 27 SCN in our study exhibited the characteristic microcystic pattern on imaging; the algorithm correctly predicted SCN in the remaining 20 of 21 cases (95%), demonstrating its utility in this group lacking typical imaging features. Use of such a biomarker-based algorithm may therefore prevent the misclassification and subsequent overdiagnosis/overtreatment of SCN as mucinous/other and underdiagnosis/undertreatment of MCN as an undefined cystic lesion.

Algorithm performance was also assessed in the younger surgical cohort (<50 years old). Pancreatic cysts are less common in younger patients, and surgery is more often considered to avoid prolonged surveillance and associated costs. The algorithm was superior to clinical diagnosis, correctly predicting final pathological diagnosis in 96% versus 62% of cases, respectively. This suggests that performing EUS-FNA and biomarker analysis may be particularly useful in guiding the management of this younger cohort. For such patients, the algorithm test (likelihood ratios of 15.75 and 41.5 for serous or mucinous cysts, respectively, based on scores alone) could potentially be used in a Fagan nomogram to calculate post-test probability from pre-test probability determined by patient demographics (i.e. age, sex) and cyst features (i.e. location, duct communication). Performance of the biomarker algorithm was also evaluated in a surveillance cohort where a predicted diagnosis of SCN or mucinous cyst was provided with high confidence in a majority of cases, based upon high biomarker risk scores and consistency with imaging on extended follow-up. Importantly, this supports the clinical utility of the algorithm for its intended patient population.

The algorithm increases the likelihood that low-risk or benign pancreatic cystic lesions will be accurately identified for monitoring – the initial level of stratification. This would address the relatively low specificity of the International Consensus Guidelines discussed earlier. Pancreatic cysts correctly categorized as mucinous (IPMN/MCN) by our algorithm would then undergo further stratification with the International Consensus Guidelines to determine whether the cyst develops worrisome or high-risk features. In addition, biomarkers of dysplasia may supplement these guidelines to assess whether a pre-malignant/malignant mucinous cyst should be considered for resection or can be safely monitored (29).

Limitations of this study include being performed at a single, high volume institution. Validation in a larger multi-institutional study would be more optimal. Although the surveillance cohort allows us to assess the algorithm’s performance relative to its intended use, a definitive diagnosis is not available for the majority of this group. Continued follow-up and ultimately resection may be required for validation. Sample viscosity precluded glucose measurement in some cases but also serves as indicator of a mucinous cyst in our study. Faias et al similarly noted that 11% (9/82) of cyst fluid samples, all from mucinous cysts (24%, 9/38), were unable to be read by the glucometer due to sample viscosity (30). In our surgical cohort, cyst fluids were obtained either during EUS-FNA or surgery. Testing cyst fluid obtained at the time of resection rather than during preoperative EUS in the surgical cohort could be a limitation of this study. However, in a subset for which both fluids from the same patient were available, similar levels of the markers VEGF or glucose were detected in the two fluids. This suggests that cyst fluids obtained preoperatively may be as predictive as those obtained during surgery but should be further validated.

CONCLUSIONS

We developed a novel biomarker scoring system and risk algorithm to provide a more objective and accurate means of determining preoperative pancreatic cyst diagnosis. This type of tool may improve the initial stratification of new patients or existing patients who are currently under surveillance. Although EUS-FNA is currently performed for symptomatic or worrisome/high-risk cysts, our findings support wider use of EUS-FNA followed by biomarker analysis to increase preoperative diagnostic accuracy. In turn, clinical care and its associated costs may be optimized.

ABBREVIATIONS:

CEA

carcinoembryonic antigen

CTNNB1

cadherin-associated protein beta 1

EUS-FNA

endoscopic ultrasound-guided fine needle aspiration

IPMN

intraductal papillary mucinous neoplasm

LOH

loss of heterozygosity

MCN

mucinous cystic neoplasm

nCLE

needle-based confocal laser endomicroscopy

PanIN

pancreatic intraepithelial neoplasia

PaNET

cystic pancreatic neuroendocrine tumor

SCN

serous cystic neoplasm

SPN

solid pseudopapillary neoplasm

VEGF

vascular endothelial growth factor

VHL

von Hippel-Lindau