Carbohydrate antigen (CA) 19-9 is the most widely used tumour marker in pancreaticobiliary malignancy. Overall sensitivity (78–89%) and specificity (67–87%) are low, and in around 7% of the population who lack the Lewis (a) antigen, CA19-9 will remain negative ³³ . In small tumours, sensitivity decreases further. The marker can also be elevated in a number of other malignant diseases (for example, gastric adenocarcinoma) and benign diseases, particularly those causing jaundice (for example, primary biliary cirrhosis, cholestasis and cholangitis), and in smokers ³⁴ . In addition, variation has been reported among commercially available assays, which may impact on interpretation ³⁵ . Therefore, to improve the sensitivity of the marker in current clinical practice, it is always interpreted in the context of cross-sectional imaging findings ³³ .

Other commercially available tumour markers that have a role in diagnosing pancreaticobiliary cancer include carcinoembryonic antigen (CEA) and CA125. CEA is a glycosyl phosphatidyl inositol cell surface-anchored glycoprotein that is involved in cell adhesion. When elevated, it is highly suggestive of colorectal cancer, but it is also increased in about a third of patients with BTC ^{36– 38} . CA125 is a protein encoded by the MUC16 gene and is a large membrane-associated glycoprotein with a single transmembrane domain. When elevated, it is suggestive of ovarian cancer, but it is also increased in about 40–50% of patients with pancreaticobiliary malignancy, particularly when there is peritoneal involvement ³⁸ .

Owing to the limitations of existing biomarkers, over the last few years several studies have evaluated various combinations of biomarkers to supplement or ultimately replace existing biomarkers. Biomarker panels using combinations of markers, often including CA19-9, have been particularly successful in detecting small tumours and early disease. Validation studies have also shown that these markers can differentiate PDAC from relevant benign conditions and in some cases detect tumours up to 1 year prior to diagnosis with a specificity of 95% and a sensitivity of 68% ⁷ ( Table 1 and Table 2).

In pancreaticobiliary malignancy and PDAC in particular, metastatic disease occurs at a very early stage in tumour development. This is demonstrated by the fact that patients who underwent resection of small primary tumours (<2 cm) with no clinical evidence of metastatic disease had a 5-year survival after pancreatectomy of less than 18% owing to recurrent metastatic disease ³⁹ . Tumour development is driven by a series of cumulative genetic abnormalities; therefore, genetic and epigenetic changes have been explored as diagnostic targets in circulating tumour cells, cell-free DNA (cfDNA) and non-coding RNA ( Table 3– Table 5). Owing to the position and composition of pancreaticobiliary tumours, tissue samples are frequently acellular, making diagnostics challenging. Recently, the utility of next-generation sequencing was explored as a technique that allows the detection of low-abundance mutations and abnormalities in small amounts of material ⁴⁰ . Changes in the metabalome are also being explored as a potential diagnostic tool in pancreaticobiliary malignancy ⁴¹ .

Patients with an indeterminate stricture on cross-sectional imaging are typically referred for an ERCP and biliary brushing with or without endobilary biopsy to obtain tissue for diagnosis, with or without therapeutic stenting ²⁸ . Although these techniques do not compromise resection margins in potentially resectable cases, sensitivity remains low (9–57%) and patients frequently have to undergo multiple procedures to obtain a diagnosis ^{28– 30} . Bile can be easily obtained at the time of ERCP and, owing to its proximity to the tumour, is a potentially important source of diagnostic biomarkers in these cancers ( Table 6). Unfortunately, owing to the invasiveness of ERCP, the role of these biomarkers is limited to diagnosis rather than screening or surveillance in these tumours.

Urine provides a very easy and acceptable source for biomarker analysis. In BTC, a 42-peptide panel (consisting mostly of fragments of interstitial collagens) correctly identified 35 of 42 BTC patients with a sensitivity of 83% and a specificity of 79% ⁴² . In PDAC, the three-biomarker panel (LYVE-1, REG1A and TFF1) has been validated in a multi-centre cohort of 371 samples. When comparing PDAC stage I–IIA (resectable disease) with healthy urines, the panel achieved AUCs of 0.97 (95% confidence interval of 0.93–1.00). The performance of the urine biomarker panel in discriminating PDAC stage I–IIA was superior to the performance of serum CA19-9 ( P=0.006) ⁴³ ( Table 7).

Recently, pre-diagnostic symptom profiles have been investigated as an alternative way of detecting hepato-pancreato-biliary (HPB) cancers at an early stage ^{8, 9, 16, 44} . It is now recognised that the onset of PDAC and BTC is heralded by a collection of gastrointestinal and constitutional symptoms ⁴⁵ . Although overlap occurs with other benign and malignant conditions, certain symptoms such as back pain, lethargy and new-onset diabetes have been identified as particularly suggestive of PDAC. Commonly performed blood tests such as liver function tests, glucose and haemoglobin also typically become abnormal in the months preceding diagnosis ⁴⁶ . Therefore, cancer decision support tools have been produced from combinations of symptoms and risk factors. In the UK, they have been introduced into general practices in 15 cancer networks to date ⁸ , and their utility is currently being audited ⁴⁷ . Modification to existing tools to enhance their diagnostic accuracy can be expected in the future.

If there is a mass lesion on cross-sectional imaging, endoscopic ultrasonography with fine-needle aspiration (EUS-FNA) provides an alternative method for visualising and sampling the extra-hepatic biliary tree, pancreas, gallbladder or peri-hilar lymph nodes. EUS-FNA has a diagnostic accuracy for PDAC of between 65% and 96% ^{48, 49} . In BTC, a single-centre study reported a sensitivity of 73%, which was significantly better in distal compared with proximal tumours (81% versus 59% respectively, P=0.04) ⁵⁰ . Recently, developed fine core biopsy needles appear to have improved diagnostic accuracy over traditional FNA needles, but randomised trials are awaited ^{49, 51, 52} . Rapid onsite examination by a cytopathologist is used in some centres, particularly in North America, and has been shown to improve the yield of EUS-FNA in individual centres ^{53, 54} but this trend has not been borne out in recent randomised controlled trials ⁵⁵ .

To improve the diagnostic accuracy of EUS, it can also be combined with novel adjuncts such as contrast agents (SonoVue ^®), transient elastography (TE) or confocal laser endomicroscopy (CLE). TE allows the measurement of the tissue firmness, which tends to be increased in malignant tissue. In a recent single-centre study from the UK, quantitative strain measurements were found to have high sensitivity but low specificity for the detection of PDAC ⁵⁶ . The technology to perform the techniques is available on most modern EUS machines and adds little time to the overall procedure time. The technique can be performed equally well by endosonographers with limited experience ^{57, 58} and is particularly advantageous in cases where the diagnosis remains uncertain after standard EUS has been performed ⁵⁹ . Contrast-enhanced EUS is performed with agents such as SonoVue ^® and allows visualisation of the early arterial phase and late parenchymal phase enhancement of the pancreas. Pancreatic tumours are generally hypovascular compared with the surrounding parenchyma ^{60, 61} . Dynamic contrast EUS is a relatively novel method that allows the non-invasive quantification of the tumour perfusion compared with the pancreatic parenchyma by using software that is now built into a number of EUS scanners. The use of this technology is evolving but is expected to be most applicable when predicting tumour response to chemotherapeutic agents, particularly new drugs against vascular angioneogenesis ^{62, 63} .

Recently, a needle-based confocal endomicroscope has also been developed which can be passed through a 19G FNA needle to assess indeterminate masses, cysts or lymph nodes. Malignancy in the hepatobilary tract is identified by the presence of irregular vessels, vascular leakage and large dark clumps ( Figure 1) ⁶⁴ . In a recent study of 25 patients with indeterminate pancreatic masses referred for EUS-FNA, needle-based CLE was shown to be a safe and feasible technique ⁶⁵ .

ERCP is typically undertaken when imaging demonstrates an indeterminate biliary stricture and tissue acquisition is required for cytological or histological assessment. Biliary brush cytology and endobiliary biopsy have a sensitivity for malignancy of 9–57% ^{29, 30, 66, 67} . Most HPB tumours exhibit chromosomal aneuploidy ⁶⁸ ; therefore, in some centres, fluorescence in situ hybridisation and digital image analysis are used to assess for the presence of DNA abnormalities in brush cytology ^{30, 69} . Although these techniques have been adopted by only a few centres, the presence of polysomy is highly suggestive of BTC ^{30, 69} .

Poor diagnostic accuracy in biliary brush and endobiliary samples has been attributed to their being non-targeted samples obtained with only fluoroscopic guidance ⁷⁰ . The single-operator cholangioscopy system (SpyGlass, Boston Scientific Corporation, Natick, MA, USA) introduced in 2006 and now superseded by the SpyGlass DS system enables intrabiliary biopsies under direct vision via small disposable forceps ( Figure 1). In a recent systematic review, the sensitivity and specificity of cholangioscopy-guided biopsies in the diagnosis of malignant biliary strictures were 60.1% and 98.0%, respectively ⁷¹ . Higher sensitivities are observed for intrinsic biliary malignancy compared with extrinsic compressing tumours ⁷² . Several techniques have been employed to augment the visualised mucosa during cholangioscopy, including chromendoscopy with methylene blue ^{73– 75} , narrow-band imaging ^{76, 77} and autofluorescence ⁷⁸ .

During ERCP, a “CholangioFlex” confocal probe (Mauna Kea Technologies, Paris, France) can be placed down the working channel of a cholangioscope or duodenoscope to obtain real-time CLE images, which are akin to standard histology ( Figure 1). If the images obtained from a point on the biliary mucosa contain dark areas, this is highly suggestive of malignancy ^{79, 80} . The diagnostic accuracy of probe-based CLE was recently validated in a prospective multi-centre international study with 112 patients (71 with malignant lesions). Tissue sampling alone had a sensitivity, specificity and diagnostic accuracy of 56%, 100% and 72%, respectively. In comparison, ERCP with probe-based CLE had a sensitivity, specificity and diagnostic accuracy of 89%, 71% and 82%, respectively. Diagnostic accuracy increased to 88% when probe-based CLE and tissue sampling results were combined ⁸¹ . CLE is also feasible in the pancreatic duct during pancreaticoscopy but, owing to concerns over pancreatitis, is rarely used. In a case report by Meining et al., the presence of a main duct-intraductal papillary mucinous neoplasia was confirmed by clear views of typical finger-like projections ⁸² . Intraductal ultrasound in small studies has also been shown to have a diagnostic accuracy of up to 90% ⁸³ .