Preoperative misdiagnosis of pancreatic and periampullary cancer in patients undergoing pancreatoduodenectomy: A multicentre retrospective cohort study

Introduction: Whereas neoadjuvant chemo(radio)therapy is increasingly used in pancreatic cancer, it is currently not recommended for other periampullary (non-pancreatic) cancers. This has important implications for the relevance of the preoperative diagnosis for pancreatoduodenectomy. This retrospective multicentre cohort study aimed to determine the frequency of clinically relevant misdiagnoses in patients undergoing pancreatoduodenectomy for pancreatic or other periampullary cancer. Methods: Data from all consecutive patients who underwent a pancreatoduodenectomy between 2014 and 2018 were obtained from the prospective Dutch Pancreatic Cancer Audit. The preoperative diagnosis as concluded by the multidisciplinary team (MDT) meeting was compared with the final postoperative diagnosis at pathology to determine the rate of clinically relevant misdiagnosis (defined as missed pancreatic cancer or incorrect diagnosis of pancreatic cancer). Results: In total, 1244 patients underwent pancreatoduodenectomy of whom 203 (16%) had a clinically relevant misdiagnosis preoperatively. Of all patients with a final diagnosis of pancreatic cancer, 13% (87/ 679) were preoperatively misdiagnosed as distal cholangiocarcinoma (n 1⁄4 41, 6.0%), ampullary cancer (n 1⁄4 27, 4.0%) duodenal cancer (n 1⁄4 16, 2.4%), or other (n 1⁄4 3, 0.4%). Of all patients with a final diagnosis of periampullary (non-pancreatic) cancer, 21% (116/565) were preoperatively incorrectly diagnosed as pancreatic cancer. Accuracy of preoperative diagnosis was 84% for pancreatic cancer, 71% for distal cholangiocarcinoma, 73% for ampullary cancer and 73% for duodenal cancer. A prediction model for the preoperative likelihood of pancreatic cancer (versus other periampullary cancer) prior to pancreatoduodenectomy demonstrated an AUC of 0.88. Cancer Center Amsterdam, reef 9, 1105 AZ, Amsterdam,


a b s t r a c t
Introduction: Whereas neoadjuvant chemo(radio)therapy is increasingly used in pancreatic cancer, it is currently not recommended for other periampullary (non-pancreatic) cancers. This has important implications for the relevance of the preoperative diagnosis for pancreatoduodenectomy. This retrospective multicentre cohort study aimed to determine the frequency of clinically relevant misdiagnoses in patients undergoing pancreatoduodenectomy for pancreatic or other periampullary cancer. Methods: Data from all consecutive patients who underwent a pancreatoduodenectomy between 2014 and 2018 were obtained from the prospective Dutch Pancreatic Cancer Audit. The preoperative diagnosis as concluded by the multidisciplinary team (MDT) meeting was compared with the final postoperative diagnosis at pathology to determine the rate of clinically relevant misdiagnosis (defined as missed pancreatic cancer or incorrect diagnosis of pancreatic cancer). Results: In total, 1244 patients underwent pancreatoduodenectomy of whom 203 (16%) had a clinically relevant misdiagnosis preoperatively. Of all patients with a final diagnosis of pancreatic cancer, 13% (87/ 679) were preoperatively misdiagnosed as distal cholangiocarcinoma (n ¼ 41, 6.0%), ampullary cancer (n ¼ 27, 4.0%) duodenal cancer (n ¼ 16, 2.4%), or other (n ¼ 3, 0.4%). Of all patients with a final diagnosis of periampullary (non-pancreatic) cancer, 21% (116/565) were preoperatively incorrectly diagnosed as pancreatic cancer. Accuracy of preoperative diagnosis was 84% for pancreatic cancer, 71% for distal cholangiocarcinoma, 73% for ampullary cancer and 73% for duodenal cancer. A prediction model for the preoperative likelihood of pancreatic cancer (versus other periampullary cancer) prior to pancreatoduodenectomy demonstrated an AUC of 0.88.

Introduction
Pancreatoduodenectomy is performed for both pancreatic cancer and periampullary non-pancreatic cancer (i.e. distal cholangiocarcinoma, ampullary cancer, duodenal cancer). Each of these cancers differ in terms of prognosis and (neo)adjuvant treatment [1,2]. Based on recent randomized trials [3,4], preoperative chemo(radio)therapy is increasingly used in the treatment of pancreatic cancer, but not recommended for periampullary cancers according to the current guidelines [5,6]. Therefore, certainty about the diagnosis prior to pancreatoduodenectomy is important, as it determines the pretreatment strategy and, as such, has direct clinical consequences.
The most likely diagnosis is typically agreed on at multidisciplinary team (MDT) meeting by incorporating clinical presentation, laboratory tests, radiological characteristics and, if available, preoperative cytology/histology. Some studies have previously reported on the accuracy of preoperative diagnostics in pancreatic and periampullary tumours [7,8]. However, most studies only report on the significance of the various diagnostic modalities, such as cross-sectional imaging (CT/MRI) and endoscopic ultrasound (EUS), or investigate staging accuracy rather than distinguishing pancreatic from other periampullary cancers preoperatively [9e11]. No studies have evaluated the accuracy of the diagnosis prior to pancreatoduodenectomy based on all available information presented at the MDT meeting.
This multicentre cohort study aims to evaluate the accuracy and rate of clinically relevant misdiagnosis during the preoperative MDT in a large cohort of patients who underwent pancreatoduodenectomy for pancreatic or periampullary cancer.

Study design and data collection
All consecutive patients who underwent pancreatoduo denectomy in the Netherlands between January 2014 and December 2018 for cancer on final pathology were eligible for this study. Data were retrieved from the mandatory prospectively maintained nationwide Dutch Pancreatic Cancer Audit (DPCA). This audit covers all pancreatic resections and demonstrates a high degree of accuracy and case ascertainment [12]. Centres of the Dutch Pancreatic Cancer Group (DPCG) were asked to participate in the current study [13]. Additional variables, not available in the DPCA, were collected locally by each participating centre. Collected additional data were subsequently matched with the DPCA data based on an encrypted key ID, resulting in an anonymized dataset. Study proposals and data requests utilizing the DPCA are reviewed by the scientific board of the DPCG and need unanimous approval for research purposes [13]. Institutional Review Boards (IRB) of the participating institutions either gave ethical approval or waived the need for ethical approval due to the retrospective and anonymized nature of this study.

Preoperative work-up
The diagnostic workup for pancreatic and periampullary tumours is protocolized within the participating centres, yet minor inter-hospital differences might exist due to the multicentre nature of this study. A multiphase computed tomography (CT) scan of the abdomen, including arterial, venous, and portal contrast phase axial scans performed, using 3-mm slice thickness (i.e. pancreas protocol). As advised in the Dutch treatment guideline online in cases where the CT scan does not reveal a pancreatic or periampullary mass a EUS was performed. EUS is performed by or under the direct supervision of senior gastroenterologists. Contrast-enhanced EUS and elastography was rarely used. MRI is only performed in case of suspect lesions in the liver which cannot be characterized by CT. A weekly multidisciplinary tumour board is held each week and attended by at least one pancreatic surgeon, gastroenterologist, (interventional) radiologist, radiotherapist and pathologist. Each surgical case is discussed. No significant changes were made in the preoperative workup during the study period.

Definitions
Throughout this manuscript, pancreatic ductal adenocarcinoma is referred to as pancreatic cancer and all other non-pancreatic periampullary cancers (distal cholangiocarcinoma, ampullary, and duodenal cancer) are referred to as periampullary cancer. The preoperative diagnosis was collected from the report of the preoperative MDT meeting. When multiple diagnoses were mentioned (for example a differential diagnosis), only the most probable diagnosis was noted if explicitly mentioned or otherwise the first one mentioned. Final pathology was considered as reference standard in the current study. A clinically relevant misdiagnosis was defined as either pancreatic cancer preoperatively misdiagnosed as periampullary cancer, or periampullary cancer preoperatively misdiagnosed as pancreatic cancer. Also, the performed grossing technique during pathology assessment of the pancreatoduodenectomy specimen was collected, defined as either axial slicing as described by Verbeke et al., bivalving as described by Adsay et al., or as other or unknown [14,15]. Vascular involvement on imaging was defined as 90 or more on cross-sectional imaging. Survival was calculated as the time in months between date of diagnosis (if available, otherwise date of surgery) and date of death, or censored at the date of last time of follow-up. Neoadjuvant therapy for pancreatic cancer was not standard treatment during the study period and mostly administered in the setting of the DPCG PREOPANC trial and consisted of 3 courses of Gemcitabine with concurrent radiotherapy (15 Â 2.4 Gy) [4]. Inclusion criteria for the PREOPANC trial (i.e. neoadjuvant therapy) were a WHO performance status of 0 or 1, and adequate hematologic, renal, and hepatic function. Exclusion criteria were cT1 tumour (<2 cm, without vascular involvement), history of malignancy within 5 years, and previous radiotherapy or chemotherapy that precluded treatment.

Statistical analysis
Baseline characteristics were presented with median and interquartile range (IQR) for continuous variables and frequencies with proportions for categorical variables. Chi-square or Fisher's exact test were used for comparing categorical variables as appropriate and Wilcoxon rank-sum test for comparing continuous variables between groups. Preoperative and postoperative diagnoses were cross tabulated to assess accuracy and misdiagnosis rates. Accuracy was defined as the proportion of patients with a preoperative diagnosis that was concordant with final pathology (i.e. the predictive value of the preoperative diagnosis). Misdiagnosis rates were calculated as missed pancreatic cancer preoperatively (1-sensitivity) and incorrectly diagnosed pancreatic cancer preoperatively (1-specificity). A subgroup analysis was performed in all patients in whom preoperatively cytology or histology was performed. Post-hoc survival analyses (per-protocol) were performed using the Kaplan-Meier method and statistical differences between groups were tested using the log-rank test. To evaluate the potentially missed benefit from neoadjuvant therapy, survival of all pancreatic cancer patients was also evaluated in three groups; 1) pancreatic cancer patients who did not receive neoadjuvant therapy, 2) pancreatic cancer patients who did receive neoadjuvant therapy, and 3) pancreatic cancer patients who were preoperatively misdiagnosed as periampullary cancer and did not receive neoadjuvant therapy in accordance with national guidelines.
Predictive preoperative factors associated with correct diagnosis of pancreatic cancer (versus periampullary cancer) were assessed on univariable and multivariable logistic regression, with only statistically significant variables (i.e. P value below 0.05) on univariable analysis selected for the final model. Cut-off values for continuous variables were determined using the receiver operating characteristics (ROC-) curve with highest combination of sensitivity and specificity. Predictive accuracy of the final model was assessed with the area under the curve (AUC) of the ROC-curve. Two-sided P values of lower than 0.05 were considered statistically significant. All analyses were performed using R version 3.6.2 (cran.r-project. org).

Discussion
This multicentre cohort study of 1244 patients who underwent pancreatoduodenectomy demonstrated that 16% of the patients had a preoperative clinically relevant misdiagnosis. In view of the growing role of neoadjuvant treatment for (potentially) resectable pancreatic cancer this may lead to either missed treatment options or incorrect neoadjuvant treatment. 16% of patients with presumed pancreatic cancer, but did not have pancreatic cancer on final pathology, would receive unintended neoadjuvant treatment (116 out of 708 patients). Conversely, 16% of patients with presumed periampullary cancer, who in fact had pancreatic cancer would miss the opportunity of neoadjuvant therapy (87 out of 536 patients). Lastly, a prediction model including clinical, radiological and preoperative pathological parameters demonstrated an AUC of 0.88, which might be of help to systematically determine eligibility for clinical trials on neoadjuvant therapy in pancreatic cancer.
Assessing eligibility for clinical trials on neoadjuvant therapy in pancreatic cancer is typically performed at the preoperative MDT meeting but remains challenging in some cases [16]. The recently published DPCG PREOPANC trial on neoadjuvant therapy in (borderline) resectable pancreatic cancer also included 9 patients (4%) with distal cholangiocarcinoma on final pathology [4]. Another neoadjuvant trial from South Korea did not report on misdiagnosed patients [3]. The misdiagnosis rate of 4% in the PREOPANC trial is lower than found in this study (ranging from 5 to 14% per institution). Presumably, patients with an ambivalent preoperative tumour origin were not included to avoid contamination with nonpancreatic cancer patients in this trial. As mentioned before, during the study period neoadjuvant therapy was only administered in trial setting. This explains the low rate of neoadjuvant therapy for patients with pancreatic cancer (12%). Based on the of the Dutch randomized PREOPANC trial, neoadjuvant chemo-radiotherapy has now (February 2021) become the standard approach to borderline resectable pancreatic cancer in the Netherlands [4]. For all other periampullary cancers, the current Dutch guideline does not advice neoadjuvant treatment as is probably the case in most countries worldwide. Identifying patients who qualify for neoadjuvant therapy should be patient-tailored and based on patient-and tumour characteristics (including patient preference). The current workup for neoadjuvant therapy requires extra invasive tests, yet when these are overcome, neoadjuvant therapy might offer oncological benefits as suggested by several studies [3,4]. The trade-off between risk and benefits is up to the particular treating physician to discuss with the patient. Pancreatic cancer and distal cholangiocarcinoma are from an oncological perspective very comparable entities in terms of tumour biology and prognosis [17]. These cancers are also not histologically distinguishable, neither morphologically nor immunohistochemically. Final diagnosis is based on macroscopic assessment by the pathologist of the most likely origin (e.g., pancreas or bile duct) of the cancer in the resected specimen. Since (randomized) studies on neoadjuvant therapy for (borderline) resectable distal cholangiocarcinoma are still lacking, one cannot exclude the possibility that these patients might also benefit from comparable preoperative chemotherapy as in pancreatic cancer. Some non-randomized series indeed suggest a survival benefit for neoadjuvant therapy in distal cholangiocarcinoma [18]. If this were to be proven correct, than the definition of clinically relevant misdiagnosis would change and shift towards a distinction between pancreatobiliary cancer (i.e. preoperative chemotherapy recommended) and ampullary-duodenal cancer (i.e. not requiring preoperative chemotherapy). In theory, the clinically relevant misdiagnosis rate according to such a definition would decrease to 9.9% (124 out of 1244 patients) in this study. Although the rationale of neoadjuvant therapy also being effective for distal cholangiocarcinoma is still speculative, distinction based on histologic subtype (pancreatobiliary versus intestinal) might be more sound The first percentage is a row percentage and the second percentage a column percentage. Cumulative percentages may not exactly equal 100% due to rounding. than distinction based on anatomical site of tumour origin [19]. Further studies on the efficacy of neoadjuvant regimens for periampullary cancers are to be awaited, as is conclusive evidence on the effect of neoadjuvant therapy for (resectable) pancreatic cancer.
Our study also demonstrates that survival after resection of pancreatic cancer and distal cholangiocarcinoma is comparable. Prognoses after resection of ampullary cancer and duodenal cancer are also comparable but considerably more favourable than pancreatic cancer and distal cholangiocarcinoma. When looking further at the survival curves of patients who underwent pancreatoduodenectomy for pancreatic cancer, neoadjuvant therapy was associated with improved survival (26 months with neoadjuvant therapy versus 19 months without neoadjuvant therapy, p ¼ 0.021). Interestingly, patients whose pancreatic cancer diagnosis was missed preoperatively had a median overall survival of 22 months, which did not significantly differ from those with a correct pancreatic cancer diagnosis whether or not treated with neoadjuvant therapy. It is unclear whether this is due to a type IIerror (underpowered) or that patients with a missed pancreatic cancer diagnosis actually had more favourable disease (more similarities with periampullary cancers).
While final pathology was considered the reference standard in this study, it should be kept in mind that even during pathology assessment distinction between pancreatic and periampullary can be very challenging [20]. The frequent reclassification of tumour origin following slide review, and the wide variation in published incidence of pancreatic (33e89%), ampullary (5e42%) and distal   bile duct (5e38%) cancers indicate that the histopathological distinction between these cancer groups is less accurate than generally believed [21]. The different types of cancers emerge from a continuum of duodenum-ampulla-distal bile duct-pancreatic duct junction, basically hampering to distinguish truly different entities as they are biologically, histologically and embryologically ambiguous [17,22]. Tumour markers and other molecular techniques might further help to distinguish these different types of cancer, preferably prior to surgical/oncological treatment [23e25]. At the same time, more potent chemotherapeutic regimens become available, such as FOLFIRINOX, which shows to be an effective treatment against a broad spectrum of tumours [26]. It will be interesting to watch the outcome of future randomized trials assessing the impact of neoadjuvant FOLFIRINOX in the other periampullary (non-pancreatic) cancers. Currently, the exact diagnosis on final pathology remains of paramount importance, since it determines the indication for adjuvant treatment. Adjuvant therapy after resection of distal cholangiocarcinoma remains controversial [5,6,27]. In some countries, adjuvant therapy is standard of care, whereas in the Netherlands it is only administered in trial setting [28]. Distinguishing pancreatic cancer and distal cholangiocarcinoma also depends on many pathological parameters, including grossing technique, level of expertise, tumour size, et cetera. In the present study, the incidence of pancreatic and periampullary cancers did not differ considerably between performed grossing technique ( Table S1). The exact cause of the wide variability in reported pathology outcomes by different centres (Table 3) remains unknown. It might be partly explained by the unclear classification of (peri) ampullary tumours and the lack of consensus on what is the basis of the classification (macroscopy vs. microscopy vs. immunohistochemistry). This study did not include data on patients with ultimately benign disease since we studied this topic several years ago. In 1629 consecutive pancreatoduodenectomies in the Netherlands (2003e2010) we found a 6.6% rate of unexpected design disease [7].
This study has some limitations. First, when multiple preoperative diagnoses were considered (i.e. a differential diagnosis) this was not taken into account in the analyses. For each patient one preoperative diagnosis was extracted from the MDT report, either the first one listed in the differential diagnosis or the most probable one if explicitly mentioned. Therefore, clinical (un)certainty of the preoperative diagnosis is not reflected. Second, the diagnosis on final pathology is not always entirely independent, as the pathologist might also incorporate preoperative diagnostic findings in his/ her judgment. Misdiagnosis rates might even be higher in case of completely independent assessments. Third, the retrospective nature of additional data collection might have led to information bias. Additional data collection might have been influenced by information already known, either from the prospective database or from other clinical information in the medical chart. Also, the neoadjuvantly treated patients are mostly trial participants and are therefore carefully screened for eligibility criteria. Besides, the survival curves are not based on an intention-to-treat principle and should therefore not be interpreted as such. Hence, these figures are solely for illustrative purposes and one should not draw hard conclusions from these survival estimates.
In conclusion, this is the first study to evaluate the accuracy of the preoperative diagnosis as established during MDT meeting before pancreatoduodenectomy. One in every six patients planned for oncological resection is misdiagnosed either as pancreatic cancer or periampullary cancer. With the increasing use of neoadjuvant therapy for pancreatic cancer (and no indication for neoadjuvant therapy for periampullary cancers), this finding warrants further research to reduce the rate of clinically relevant misdiagnoses.

Financial disclosures
None to declare.