Participants

Ninety-one experts met selection criteria and were invited to vote. Eighty-two completed the first round, 76 completed the second round (response rate 84%). The 76 final responders included 37 gastroenterologists, 16 surgeons, 7 pathologists, 6 radiologists, 5 geneticists, 3 oncologists, and two epidemiologists, from 11 countries and four continents; 70 (92%) worked in a university hospital setting. They had practised their profession for a median of 22 (IQR 15) years, and had been involved in a pancreatic cancer surveillance programme for a median of 10 (IQR 12) years.

Recommendation statements

A summary of the statements that reached consensus is provided in tables 1 and 2. All voting statements and results are provided in online supplementary table S2. A summary of the main consensus recommendations is provided in table 3.

Who should be screened?

Age, family history and germline mutation status are the major criteria for determining eligibility for pancreatic surveillance. The number of first- and second-degree relatives with pancreatic cancer can be used to quantify pancreatic cancer risk.15 For example, the estimated lifetime risk of developing pancreatic cancer for an individual with two first-degree relatives with pancreatic cancer is ~8%.15 16 Family history of pancreatic cancer is also a risk factor for patients identified as having incidentally detected pancreatic cysts.17 Current surveillance recommendations for a family history (generally in one blood relative) are the same as for those without a family history.18 19

Consensus on family history recommendations for pancreatic surveillance (ie, having at least one first-degree relative and one second-degree relative with pancreatic cancer) were the same as in the 2013 guidelines. Obtaining a comprehensive cancer family history from newly diagnosed patients with pancreatic cancer can help to identify family members who may benefit from surveillance. The average lifetime risk of developing pancreatic cancer (~1 in 64 in the USA) is too low for population-based screening.20 21

Germline mutation carriers

Pancreatic surveillance is recommended for carriers of germline deleterious variants in cancer susceptibility gene,22–26: BRCA2, ATM, BRCA1, PALB2, CDKN2A, STK11, MLH1 and MSH2. Recommendations for age and family history vary by gene. Surveillance for CDKN2A and STK11 (Peutz-Jegher syndrome) mutation carriers is recommended irrespective of patients' family history of pancreatic cancer, because of their high lifetime risk. Since the previous consensus, ATM mutation carriers have been added to the list recommended for surveillance. For carriers of mutations in ATM, BRCA2 and PALB2, the consensus among experts was to recommended surveillance for mutation carriers who have a blood relative with pancreatic cancer. Consensus on family history criteria for BRCA1 mutation carriers was not reached (table 1), but consensus was reached for recommending that BRCA1 mutation carriers undergo surveillance (online supplementary table S2).

Surveillance is recommended for patients with hereditary pancreatitis, with most experts recommending age 40 or 20 years after the first pancreatitis attack (online supplementary table S2), irrespective of gene status. The pancreatitis susceptibility genes, PRSS1, CPA1 and CTRC, are associated with significantly increased risk of developing pancreatitis.27–30 Deleterious variants in CPA1 and CPB1 associated with pancreatic cancer risk may not always progress through a clinical syndrome of pancreatitis.31

Deleterious variants in the known pancreatic cancer susceptibility genes account for ~10–20% of the familial clustering of pancreatic cancer.26 32 33 Deleterious variants have also been reported in ~5–10% of patients with apparently sporadic pancreatic cancer.34–39 These variants also confer risk for other cancers.40 Therefore, germline testing should be considered for individuals eligible for pancreatic cancer surveillance.41 42 Recent National Comprehensive Cancer Network (NCCN) guidelines recommend offering gene testing for patients with newly diagnosed pancreatic ductal adenocarcinoma; this recommendation did not reach consensus among CAPS experts (online supplementary table S2). American Society of Clinical Oncology guidelines also consider the role of gene testing for patients with pancreatic cancer.43 The average lifetime risk of developing pancreatic cancer has been estimated in prospective studies for carriers of deleterious variants in BRCA2, BRCA1, CDK2NA, PRSS1, MLH1 and MSH2, 12 25 29 44–46 and odds of developing pancreatic cancer from case/control analysis for ATM, PALB2 and TP53.35 Many individuals undergo gene testing because a germline mutation was identified in a blood relative who developed cancer. The risk of pancreatic cancer in ATM/BRCA/PALB2 mutation carriers without a pancreatic cancer family history is not well defined. A study of BRCA mutation carriers found no difference in pancreatic cyst prevalence by family history; although the authors suggested pancreatic surveillance is appropriate for BRCA mutation carriers irrespective of family history, more evidence on this question is needed.47 Although family history remains an important risk assessment tool,48 pedigrees are often small and family histories incomplete. Further research is needed to better define how family history of pancreatic cancer in mutation carriers influences their risk of developing the disease.

At what age should pancreatic surveillance begin?

For individuals who meet familial risk criteria (without a defined genetic mutation) the consensus was that surveillance should begin at age 50 or later; but some experts thought surveillance should begin at 50, others at age 55. Most pancreatic surveillance programmes lower the age at which surveillance is initiated for individuals with a first-degree relative with young-onset pancreatic cancer (age <50).49 For mutation carriers, with a deleterious germline variant, the recommended age to initiate surveillance is generally age 50 (BRCA2, ATM, PALB2), though some groups would start surveillance at age 45 and earlier still for the higher-risk genes; surveillance from age 40 is recommended for CDKN2A mutation carriers; 16% of p16-Leiden mutation carriers with pancreatic cancer were diagnosed at age <4550 and at age 30–40 for those with Peutz-Jegher syndrome.3 When to initiate pancreatic surveillance for these mutation carriers did not reach consensus (online supplementary table S2). To date, most pancreatic cancers in high-risk individuals with familial risk only (ie, no known susceptibility gene mutation) under surveillance are diagnosed after age 55.10 The average age of pancreatic cancer diagnosis among individuals with a family history of pancreatic cancer is younger than for those with no family history, and mutation carriers who develop pancreatic cancer are diagnosed at a slightly younger age (~3–5 years) than those with sporadic forms of the disease.34–36 The average age for diagnosis of pancreatic cancer is slightly younger in smokers (by several years) than in non-smokers.51 52 The relative risk of pancreatic cancer among smokers is not considered sufficiently high to recommend surveillance in the absence of other risk factors. There was no consensus that the age at which surveillance begins should be modified for high-risk individuals who smoke. As pancreatic cancer risk factors become better defined, it should become possible to provide more accurate individualised risk assessment that can be used to provide personalised recommendations for pancreatic surveillance.53

What tests should be used for pancreatic surveillance?

Most pancreatic surveillance protocols for high-risk individuals use pancreatic imaging with MRI/magneticretrograde cholangiopancreatography (MRCP) and/or endoscopic ultrasound (EUS), with pancreatic-protocol CT reserved for individuals unable to have MRI or EUS. The preference for EUS and MRI/MRCP rather than CT is based on their superiority at detecting subcentimetre pancreatic cysts, and avoidance of ionising radiation.1 Subcentimetre pancreatic cysts are detected in up to 50% of high-risk individuals, depending on the age of the cohort.10 54 However, pancreatic cysts have low malignant potential and although their detection can help risk stratification, the primary responsibility of pancreatic imaging tests is to detect pancreatic cancers. In this respect, EUS may be better for detecting small pancreatic ductal adenocarcinomas, although this evidence is based on studies of small numbers of individuals55; some centres use EUS as the primary test for pancreatic surveillance. EUS diagnostic yield is highly operator dependent.56 EUS has previously been shown to be better at detecting small neuroendocrine tumours.57 EUS also identifies subtle non-specific parenchymal abnormalities, which in a high-risk setting may represent the effects of pancreatic intraepithelial neoplasia (PanIN) with associated lobulocentric atrophy.58–60

Pancreatic-protocol CT can accurately detect early-stage pancreatic cancers and performs similarly to MRI in detecting evidence of cancer/high-grade dysplasia associated with intraductal papillary mucinous neoplasms (IPMN).61 In addition, CT can, in principle, also be used to quantify changes in abdominal fat and lumbar muscle mass with the emergence of pancreatic cancer.62 63 Observation of this early wasting could help to detect early pancreatic cancers, although this approach has not been tested in prospective studies. The availability of deep learning and advanced radiomics protocols may help radiologists to identify and quantify subtle abnormalities in the pancreas by CT.64 65

Based on current evidence, the consensus among experts is that MRI/MRCP and EUS should be the first-line tests for pancreatic surveillance, in part because of the cumulative radiation exposure with frequent CT, but developments in low-dose CT imaging may necessitate re-evaluation of the role of CT in surveillance. One European study examined the diagnostic yield of performing MRI/MRCP annually with EUS or limiting EUS to every third year unless there are significant changes in MRI scans. The authors found no significant difference in the diagnostic yield among the different surveillance protocols,66 although a larger sample size and longer follow-up would be required to definitively answer this question. Less expensive, short-protocol MRI has been evaluated for cyst surveillance.67 Some Japanese centres have evaluated abdominal ultrasound as a screening modality,68 as detailed pancreas sonographic images are feasible in thin individuals, but there was consensus that abdominal ultrasound should not be a first-line test for pancreatic surveillance.

The experts also considered the role of CA19-9 testing. Although the role of CA19-9 testing has not been studied in high-risk individuals with familial/genetic risk, its diagnostic performance characteristics have been extensively studied.69–71 CA19-9 could have diagnostic value in individuals in whom the pre-test probability of pancreatic cancer is significant, although this question requires further investigation. For this reason, there was consensus that CA19-9 testing should be performed when there is concern about the possibility of pancreatic cancer, such as when worrisome features are found on pancreatic imaging.

Experts also reached consensus that glucose testing (fasting glucose or HbA1C) to detect new-onset diabetes was reasonable for high-risk individuals. There was also consensus that the emergence of new-onset diabetes in a high-risk individual should prompt additional investigation. Population guidelines recommend fasting glucose or HbA1c testing for individuals with risk factors for diabetes such as overweight or obesity,72 although there are concerns about the potential to overtreat individuals with pre-diabetes. Epidemiological studies show that 0.4% to 0.8% of patients with new-onset diabetes aged ≥50 will be diagnosed with pancreatic cancer within 3 years.62 73–75 A model incorporating weight loss, age and trend in glucose level can help to identify patients with new-onset diabetes more likely to have pancreatic cancer76; other models incorporating additional parameters are being evaluated.77 One study estimated the average glucose level for a given tumour size, predicting that when glucose levels reach diabetic levels (126 mg/dL), pancreatic tumour volume is ~2–8 mL (~diameter of 1.6–2.5 cm).76 There is no direct evidence that glucose monitoring is of additional value for improving detection of pancreatic cancer for individuals undergoing regular pancreatic imaging. Nonetheless, given the higher risk of pancreatic cancer in high-risk individuals compared with the general population, the consensus was that new-onset diabetes in a high-risk individual should prompt further testing for the presence of pancreatic cancer.

Experts discussed circulating tumour DNA (ctDNA) and its potential to contribute to pancreatic surveillance.78–80 CtDNA testing is beginning to emerge as a clinical test81; further studies are needed to define its role for patients under pancreatic surveillance. Other biomarker tests are also undergoing evaluation for their potential for early detection,63 82–84 but more study is needed to determine their diagnostic performance.

The experts recognised that many high-risk individuals meeting criteria for pancreatic surveillance (particularly mutation carriers) are at increased risk of developing other cancers; these individuals should undergo surveillance for other cancers tailored to their germline mutation status and cancer family history.

Surveillance questions

There was consensus that patients with normal pancreata, or without concerning lesions, should undergo annual pancreatic imaging surveillance. Surveillance of high-risk individuals occasionally identifies small (<1 cm) pancreatic neuroendocrine tumours (PanNETs)1–11 although it is not certain if these lesions are more common in this population. Most incidentally detected PanNETs have low malignant potential.85 There was consensus that patients with small (<1 cm diameter) PanNETs can also undergo annual surveillance detection and that treatment of PanNETs (>1 cm) could be considered a success of surveillance. However, recent studies report small (<2 cm) PanNETs with low-risk characteristics on biopsy (eg, low Ki-67) can be safely followed up.86–88 Current neuroendocrine neoplasms guidelines recommend surveillance of asymptomatic non-functional low-risk (by grade, Ki-67 by EUS-fine-needle aspiration) PanNETs (2 cm).88

Although patients in a high-risk programme commonly have pancreatic abnormalities (depending on age and other risk-factors, up to 50% will have pancreatic cysts; many also have subtle non-specific EUS parenchymal abnormalities), only a minority will develop concerning lesions. There was consensus that annual surveillance is appropriate for those with these abnormalities (figure 2). Furthermore, there was consensus that CDKN2A mutation carriers with concerning pancreatic abnormalities that do not lead to immediate surgery (eg, mild main pancreatic duct dilation, stricture without mass) should undergo additional testing such as EUS/fine-needle aspiration, and if they do not proceed to surgery after multidisciplinary review, should undergo close follow-up imaging in 3–6 months.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Decision flow-chart for the management of pancreatic abnormalities found during surveillance. EUS, endoscopic ultrasound; FNA, fine-needle aspiration; MPD, main pancreatic duct; MRCP, magnetic retrograde cholangiopancreatography.

The predicted progression rate suggests that stage I pancreatic cancers can progress to stage IV disease within 1 year,89 which may explain why interval pancreatic cancers are occasionally diagnosed despite annual surveillance, even in the absence of concerning lesions (worrisome features or solid lesions)18 19 on prior scans.

Few studies have evaluated factors that influence compliance with long-term pancreatic surveillance; one such study found that many high-risk individuals drop out of regular surveillance.90 Factors affecting long-term compliance with surveillance require further study.

When should surgery be performed?

Many factors are considered when deciding if surgical resection is appropriate for patients with concerning imaging findings, including a patient’s estimated risk of pancreatic cancer based on their gene mutation status, family history, operative risk, comorbidities, life expectancy and compliance with surveillance. Decision-making is best undertaken by an experienced, expert multidisciplinary team. There was consensus that high-risk individuals should undergo pancreatic resection for broadly similar indications to individuals without known familial/genetic risk, based on established guidelines—for example, those with worrisome features.18 19 91 92 Thus, generally, surgical resection in a patient with multifocal pancreatic cysts should manage the dominant, worrisome lesion. There was consensus that patients with solid lesions of indeterminate pathology and >5 mm should undergo pancreatic resection if additional evaluation does not yield a definitive preoperative diagnosis.

Some high-risk individuals develop multiple precursors throughout their pancreas, and those who undergo pancreatic resection for IPMN can have concomitant high-grade PanIN.59 This raises the question of whether resection criteria for high-risk individuals should include less concerning lesions than for those with sporadic disease, or if total pancreatectomy should be considered. There is no evidence to support this approach unless there are concerning lesions affecting multiple regions of the gland. Total pancreatectomy is a major operation, although studies have reported that morbidity and mortality are similar to those of Whipple operations, and diabetes-related mortality, is quite rare.93 94 There was also no consensus that surgical resection was indicated for less worrying lesions, such as suspected IPMN of 2 cm or with mild main pancreatic duct dilatation.

Consensus was also reached that the operative approach to a resectable pancreatic cancer should be the same for high-risk individuals and those with sporadic pancreatic cancer. Patients with sporadic IPMN who have had partial pancreatectomy have a 5–10% risk of developing pancreatic cancer95–98; ongoing surveillance of these individuals is needed. Studies have identified factors associated with metachronous disease.97 98 In some cases, metachronous disease represents re-emergence of a previously resected IPMN, raising the possibility that precancerous cells might spread through the main pancreatic duct.98–100 In germline mutation carriers, particularly those at highest risk of pancreatic cancer, the possibility of multiple primary cancers should be considered.101

What are the goals of surveillance?

The primary goal of pancreatic surveillance is to prevent death from pancreatic cancer and prevent its emergence by identifying and treating precursor lesions. As with the first CAPS consensus guideline, there was consensus that the main pathological _targets of surveillance are stage I pancreatic cancers and precursors with high-grade dysplasia either in PanIN or IPMN (online supplementary table S3). Since the last consensus meeting, the classification of pancreatic precursors has been slightly revised.102

Published studies of surveillance programmes reveal evidence of downstaging of pancreatic cancers, with most pancreatic cancers diagnosed as stage IIB or stage I.7 10 The detection and management of high-grade dysplasia in PanIN and IPMN remains an important goal of surveillance; these lesions are not only more commonly detected by pancreatic imaging, they are more likely to have pancreatic precursor lesions than in patients without such a family history.103

The imaging characteristics of IPMN can be useful in identifying evidence of high-grade dysplasia within IPMN, but this is not the case for PanIN, most of which are microscopic lesions that cannot be identified with available technologies. Most pancreatic ductal adenocarcinomas are thought to arise from PanIN. This is thought to be true for patients with sporadic pancreatic cancers and also for those with a familial/inherited susceptibility to develop pancreatic cancer. Most pancreatic cancers from such individuals have genetic signatures consistent with PanIN origin,26 and pancreatic cancers detected during surveillance often arise in areas of the gland separate from pancreatic cysts.95 104 Indeed, many pancreatic cancers associated with IPMN are genetically distinct from the IPMN.105 Since PanIN generally do not cause specific imaging abnormalities, high-grade PanIN (previously known as PanIN-3) is diagnosed only by surgical pathology review of pancreatic resections undertaken for other concerning imaging findings. In some cases, evidence of pancreatic neoplasia can be inferred by the presence of mutations detected in secretin-stimulated pancreatic fluid samples,104 106 and multifocal PanIN lesions by imaging findings of lobulocentric atrophy,59 but further investigation is needed to determine the value of these tests for patients under pancreatic surveillance.

Areas for future research

Emerging technologies such as CT detection of muscle and fat wasting as well as subtle changes in the pancreas using deep learning have yet to be applied to the high-risk setting. Similarly, the value of glucose monitoring in detecting new-onset diabetes for patients already undergoing routine pancreatic imaging is not known. The development of non-invasive blood tests, such as tests for ctDNA, provides hope that these will eventually improve the early detection of pancreatic cancer. The emergence of interval (ie, presenting before their annual surveillance) advanced-stage pancreatic cancers in some patients under regular pancreatic imaging suggests that biological characteristics, such as early lymph node metastases and venous invasion even with small cancers,107 make early detection efforts particularly challenging. The main factors used in clinical practice to assess the risk of pancreatic cancer in high-risk individuals remain family history, gene mutation status, age and pancreatic imaging abnormalities. Other known factors, such as diabetes and metabolic syndrome markers, smoking status, other cancer family history, gene variants identified through genome-wide meta-analysis,108 and circulating biomarkers that predict future risk,109 could help to improve risk stratification, particularly if models could be developed and validated.

Cost-effectiveness models of pancreatic surveillance have been reported. One recent paper estimated that MRI is more cost-effective in the USA overall, with EUS more cost-effective for highest-risk individuals (relative risk >20)110; cost-effective model results depend on cost estimates and MRI and EUS costs vary considerably. Ultimately, pancreatic surveillance programmes need to demonstrate better evidence that survival from pancreatic cancer can be improved by surveillance or even that the detection and treatment of high-grade dysplasia can lower the incidence of pancreatic cancer.111 Indeed, the US Preventive Services Task Force recommended that pancreatic screening should not be carried out,21 112 113 although it excluded the study by Vasen et al,7 because it focused on mutation carriers and completed its literature review before the recent CAPS study was published.10 These two studies show that pancreatic surveillance of high-risk individuals can lead to downstaging of pancreatic cancers diagnosed. Such downstaging is associated with better survival compared with historical controls, particularly when surveillance detects stage I cancers. The need to evaluate long-term outcomes necessitates pancreatic surveillance be undertaken in academic settings. Efforts to implement a pancreatic surveillance programme need to be balanced with its costs. This is a difficult balance to achieve since the harms of overdiagnosis can take many years to become evident, as is the case for other cancers.114–116 The evaluation of long-term outcomes of high-risk individuals participating in pancreatic surveillance programmes,7 95 96 117 including the potential for harm, should continue.