World-wide trends in net survival from pancreatic cancer by morphological sub-type: An analysis of 1,258,329 adults diagnosed in 58 countries during 2000 – 2014 (CONCORD-3)

Background: Ductal adenocarcinomas and neuroendocrine tumours are the two main morphological sub-types of pancreatic cancer. Using data from CONCORD-3, we examined whether the distribution of morphological sub-types could help explain international variations in pancreatic cancer survival


Introduction
Pancreatic ductal adenocarcinomas and pancreatic neuroendocrine tumours are the two main morphological sub-types of pancreatic cancer. Arising from exocrine cells in the pancreas, pancreatic ductal adenocarcinomas are the commonest sub-type, comprising around 85% of all pancreatic cancers. Pancreatic neuroendocrine tumours, which originate in islet cells of the pancreas, comprise only 3-4% of all pancreatic cancers [1][2][3].
The majority of exocrine pancreatic tumours are diagnosed at advanced stages due to lack of obvious symptoms during early-stage disease, effective population-based screening programmes, aggressive behaviour of tumours [4]. Symptoms for exocrine tumours, such as jaundice, weight loss, abdominal pain, loss of appetite and fatigue, are usually only present in advanced-stage disease, and are often attributed to other illnesses, such as pancreatitis or ulcers. Thus, five-year survival from these tumours is very poor (less than 5%) [2,4].
Pancreatic neuroendocrine tumours are either functional or nonfunctional, depending on whether the tumour produces hormones (e. g., insulin). Functional tumours may cause symptoms before the tumour has grown or spread due to the increase in hormones, while patients with non-functional tumours generally only have symptoms after the tumour has grown or spread to other areas of the body [3]. The majority of neuroendocrine tumours are functional (60-70%) [1]. Five-year survival from neuroendocrine tumours is relatively high (50-60%) [1,5].
Given the high proportion of aggressive pancreatic ductal adenocarcinomas, pancreatic cancer survival for all morphological sub-types combined is poor world-wide. The CONCORD programme established global surveillance of cancer survival in 2015 with CONCORD-2 [6], which has since been updated with the publication of CONCORD-3 [7]. CONCORD-3 included data on over 37.5 million cancer patients diagnosed with one of 18 common cancers, contributed by 322 population-based cancer registries in 71 countries. CONCORD-3 reported some variation in five-year survival from pancreatic cancer between countries, ranging from 5% to 15% world-wide: 5-year survival was highest in Asia, while survival was less than 15% in all other countries. Trends in pancreatic cancer survival from 2000-2004 to 2010-2014 were generally flat, though survival did increase by 3-5% for some countries [7].
Previous studies examining the distribution of morphological subtypes have been limited to one country [5,8]. Additionally, studies examining survival have been limited to either one morphological sub-type or to one country or region, if including more than one morphological sub-type [5,[9][10][11]. Using data from all 295 population-based registries participating in CONCORD-3 that submitted data on pancreatic cancer, we examined whether the distribution of morphological sub-types could help explain international variations in survival. Additionally, we provided estimates of survival from pancreatic cancer, by morphological sub-type, by country and over time.

Methods
We estimated age-standardised one-and five-year net survival by country, calendar period of diagnosis (2000-2004, 2005-2009, 2010-2014) and morphological sub-type, using data from 295 population-based cancer registries in 58 countries for 1,258,329 adults (aged 15-99 years) diagnosed with pancreatic cancer during 2000-2014. We included only primary, invasive tumours (ICD-O-3 behaviour code 3) in the analysis. If a patient was diagnosed with two or more primary, invasive tumours of the pancreas, then only the first record was included. Patients whose cancer registration was from a death certificate or autopsy only were excluded, because their true survival time was unknown. Follow-up data on the patient's vital status were available until 31 December 2014.
We defined three morphological sub-types based on the WHO classification of tumours of the digestive system [1] and on the International Classification Diseases of Oncology, 3rd edition [12] (ICD-O-3) morphological codes: carcinomas, neuroendocrine tumours and other specified tumours (Table 1). Tumours with non-specific morphology (ICD-O-3 codes 8000-8005) were included in carcinomas. A separate category for tumours with unknown morphology was also defined. We included all tumours regardless of whether the cancer registry reported the tumour as morphologically verified. If a tumour was reported by the registry as not morphologically verified but had a specific ICD-O-3 code (i.e., any code except 8000-8005), then we recoded this tumour as morphologically verified, because a specific morphology code implies that morphological verification had been completed.
Net survival was estimated using the Pohar Perme estimator [13]. Net survival is the probability of a cancer patient to survive their cancer up to a given time since diagnosis, e.g. one year, after controlling for competing risks of death (background mortality), which are higher in the elderly. To account for the differences in background mortality between regions and over time, we used life tables of all-cause mortality that were specific to each region, single year of age, sex, calendar year and, where possible, race or ethnic group. The Pohar Perme estimator was implemented using stns [14] in Stata version 15. We used the cohort approach [15] to estimate net survival for patients diagnosed during 2000-2004 and 2005-2009, because at least one year of follow-up data were available for all patients by the end of 2014 for the estimation of one-year survival, and at least five years of follow-up data were available for the estimation of five-year survival. We used the period approach [16] to estimate survival for patients diagnosed during 2010-2014, because five years of follow-up data were not available for all patients by 31 December 2014.
We produced survival estimates for five age groups (15-44, 45-54, 55-64, 65-74 and 75-99 years) and obtained age-standardised estimates for all ages combined, using the International Cancer Survival Standard (ICSS) weights [17]. If an age-specific estimate could not be produced, or fewer than 10 patients were available for analysis in an age group, data for adjacent age groups were pooled and the re-estimated survival was used for both of the original age groups. If two or more age-specific estimates could not be produced, or fewer than 10 patients were available for analysis in two or more age groups, only the Table 1 Morphological sub-types.
The CONCORD-3 protocol, the ethical approvals and the data quality control procedures have been described elsewhere [7].

Results
The proportion of patients registered only from a death certificate or at autopsy varied over time and between countries ( Table 2). For most countries, the proportion of death certificate or autopsy only registrations decreased over time. Of the 1,258,329 adults included in the analysis, 75.7% had tumours that were reported by the registry as morphologically verified or had a specific morphology code. The proportion of tumours morphologically verified, however, varied between countries and over time (Table 3): during 2010-2014, it was only 24.2% in Chile but 100% in Kuwait. The proportion of tumours that were morphologically verified generally increased over time. In 20 of the 58 countries, the proportion of morphologically verified tumours increased by 10% or more from 2000-04 to 2010-14. In 17 countries, this proportion decreased, but the decrease was less than 10% in all but 4 countries. Given that a specific morphology code was required for a tumour to be categorised as either a neuroendocrine tumour or a tumour with other specified morphologies, 100% of these tumours were morphologically verified. For carcinomas, however, tumours with nonspecific morphology were included in this sub-type. Thus, the proportion of tumours morphologically verified differed between countries, but generally it increased over time. All carcinomas were morphologically verified in 10 countries in 2010-2014, while morphological verification was less than 50% in seven countries.
The distribution of the known morphological sub-types varied between countries and over time (Fig. 1). Carcinomas were by far the most common morphological sub-type, comprising around 90% or more of all pancreatic tumours in all countries. Neuroendocrine tumours were rare, generally 0-10% of all tumours. Tumours with other specified morphologies were extremely rare, only comprising of less than 1% of all tumours.
The highest proportion of tumours reported as carcinoma was 100% in South Africa in 2010-2014 (Table 4), though the majority of these carcinomas were non-specific (Supplementary Table 1), and none were morphologically verified (Table 3).
In 2010-2014, the proportion of carcinomas was the lowest in Norway (89.9%). The proportion of carcinomas decreased over time in all but eight countries (South Africa, Argentina, Martinique, Uruguay, Malaysia, France, Romania and Sweden). The proportion of neuroendocrine tumours was highest in Norway (10.0%) and lowest in Uruguay (0.6%) ( Table 4). This proportion increased over time in all but eight countries (Argentina, Martinique, Uruguay, India, Malaysia, France, Romania and Sweden). The proportion of neuroendocrine tumours increased more than 5% in 29 countries (Algeria; Guadeloupe and Martinique; Canada and the United States; seven counties in Asia; 16 countries in Europe; and Australia), with consequent decreases in the proportion of carcinomas in all but two of these countries (Martinique and France).
The proportion of tumours with other specified morphologies remained extremely small (1% or less) throughout 2000-2014 (Table 4): it was relatively stable over time in 18 countries and it decreased slightly in 16 countries. Eight countries had slight increases in the proportion of other specified morphologies. The highest proportion of other specified morphologies was in Algeria and Argentina (1.1%).
The proportion of tumours with unknown morphology decreased over time in 21 countries, while it increased in five countries (Supplementary Table 2). Unknown morphology was the second commonest morphological group in countries that reported tumours of this sub-type, though its proportion varied widely between countries. In 2010-2014, unknown morphology was highest in Croatia (49.8%).

Carcinomas
Age-standardised one-year net survival from carcinomas was generally poor, ranging from 10% to 30% (Table 5, Fig. 2). One-year survival for adults diagnosed during 2010-2014 was highest in Kuwait (40.3%, 95% CI: 29.6-51.1%) and lowest in the Russian Federation (14.9%, 13.5-16.4%). Improvements in one-year survival were modest, with increases of less than 5% in 10 countries (Canada; Malaysia, Taiwan, Turkey; five countries in Europe; and New Zealand), 5-10% in 12 countries (Martinique; United States; Korea; eight countries in Europe; and Australia) and more than 10% in eight countries (Kuwait and Singapore; and six countries in Europe). One-year survival decreased over time in Chile; China and Thailand; Croatia, Iceland, Russian Federation and Slovakia.
Five-year survival decreased over time in seven countries (Chile; China and Thailand; Croatia, Iceland, Poland and the Russian Federation). There was some variation world-wide in age-standardised five-year survival from carcinomas (Fig. 4).

Neuroendocrine tumours
Age-standardised one-year net survival from neuroendocrine tumours was much higher than that of carcinomas, ranging from 40-80% (Table 5, Fig. 5). During 2010-2014, one-year survival was highest in Switzerland (83.7%, 95% CI: 77.2-90.2%), and lowest in China (29.8%, 24.4-35.2%). Improvements in one-year survival were also greater than those of carcinomas, increasing less than 5% in six countries (Korea and Turkey; and France, Germany, Italy and Portugal), 5-10% Ireland and Taiwan, and more than 10% in 12 countries (Canada and the United States; nine countries in Europe; and Australia). Four countries (China; Belgium, Finland and Poland) showed small decreases in one-year survival over time.
Age-standardised five-year survival from neuroendocrine tumours was also much higher than that of carcinomas, in the range 20-50% (Table 6, Fig. 6). The highest five-year survival for adults diagnosed during 2010-2014 was in Slovakia (75.0%, 95% CI: 58.4-91.5%) and the lowest survival was in China (18.4%, 13.6-23.2%). There were large improvements in five-year survival from neuroendocrine tumours in most countries: survival increased 10-15% in eight countries (Canada; Taiwan; and six countries in Europe) and more than 15% in six countries (United States; Finland, Netherlands, Slovakia and the United Kingdom; and Australia), while it decreased only in Belgium, China and Denmark. There was wide variation world-wide in age-standardised five-year survival from neuroendocrine tumours (Fig. 4).

Other specified morphologies
Age-standardised one-year net survival from tumours of other specified morphologies could only be estimated for the United States (Table 5) due to small numbers. In 2010-2014, one-year survival from these tumours (60.7%, 95% CI: 53.9-67.5%) was slightly lower than  Table 4 Number and proportion of tumours with known morphological sub-type, by country and calendar period of diagnosis (2000-2004, 2005-2009, 2010-2014): adults (15-     that of neuroendocrine tumours but much higher than survival for carcinomas.

AFRICA
As for one-year survival, age-standardised five-year survival from tumours of other specified morphologies could only be estimated for the United States (Table 6). Five-year survival (41.5%, 95% CI: 34.0-49.1%) followed a similar pattern as one-year survival, with survival lower than that of neuroendocrine tumours but much higher than that of carcinomas. Improvements in one-and five-year survival were similar, showing a 15% increase over time.

Unknown morphology
Age-standardised one-year net survival from tumours with unknown morphology was the lowest of all morphological sub-types, ranging from 10% to 20% (Table 5). One-year survival was highest in Turkey (18.3%, 95% CI: 10.4-26.2%) and lowest in Slovenia(7.8%, 5.4-10.1%). Oneyear survival from tumours with unknown morphology decreased over time in 9 countries and increases in one-year survival were generally less than 5%.
Age-standardised five-year net survival from tumours with unknown morphology followed a similar pattern as one-year survival ( Table 6). Five-year survival was generally similar to or lower than five-year survival for carcinomas, in the range 0-10%. Five-year survival was highest in Turkey (14.1%, 95% Ci: 7.0-21.1%) and lowest in the United Kingdom (2.2%, 0.9-3.5%). Unlike one-year survival, five-year survival for tumours of unknown morphology was slightly higher than that of carcinomas in Poland, Taiwan and Turkey. Five-year survival increased in only 5 countries, with an improvement of less than 5% in each country.

Discussion
Including high-quality data from 295 population-based cancer registries in 58 countries, this is to our knowledge the largest study examining trends in pancreatic cancer survival by morphological sub-type to date. Net survival estimates were produced using the same protocol for data collection, the same stringent three-phase data quality control procedures, the same robust methods for each registry, and life tables of background mortality that were specific to single year of age, sex, race/ ethnicity (where possible), country and the calendar year of death.
The results from these analyses show that survival from carcinomas of the pancreas remains poor in many countries world-wide, despite some improvement during the 15-year period from 2000 to 2014.
Survival from neuroendocrine tumours was much higher and there were larger improvements in survival over time for this sub-type.
The distribution of known morphological sub-types changed over time, with the proportion of carcinomas decreasing and the proportion of neuroendocrine tumours increasing in the majority of countries. This result is in contrast to previous studies that have reported an increase in the incidence of both pancreatic carcinomas and pancreatic neuroendocrine tumours, which would result in the distribution of the morphological sub-types remaining the same over time [5,8,10]. This difference can be explained by the inclusion of all tumours, regardless of morphological verification, in our study. Though morphological verification of pancreatic tumours remains low relative to other solid tumours, the proportion of tumours that are morphologically verified increased by 10% or more in 20 of the 58 countries. Previous studies examining the incidence of carcinomas excluded non-morphologically verified tumours, thus, the increase in the incidence of carcinomas is likely due to the increase over time in the proportion of tumours morphologically verified [5,8,10].
Additionally, the proportion of tumours with unknown morphology decreased in 21 of the 27 countries that registered tumours of this subtype. When tumours of unknown morphology were present in the distribution, a decrease in the proportion of unknown morphology generally corresponded to an increase in the proportion of both carcinomas and neuroendocrine tumours over time (Supplementary Table 2). This is in line with results from other studies showing increases in incidence for both carcinomas and neuroendocrine tumoursas the proportion of tumours morphologically verified increases over time, the proportion of unknown morphology decreases resulting in increases in both carcinomas and neuroendocrine tumours [5,8,10].
There is little international variation in survival from carcinomas, while the international variation from pancreatic neuroendocrine tumours is wider. Furthermore, there is no clear geographical pattern in survival for the different morphological sub-types. Some countries, such as the United States, Korea, Portugal and Italy have higher survival than other countries in the same continent for each morphological sub-type.  Age-standardised one-year net survival (NS, %): adults (15-99 years) diagnosed with pancreatic cancer by morphological sub-type, country and calendar period of diagnosis (2000-2004, 2005-2009, 2010-2014    However, for the majority of the countries, higher survival for one morphological sub-type does not mean that the country also has higher survival for the others. Age-standardised one-and five-year net survival from tumours with unknown morphology was the lowest of all the morphological subtypes. As the proportion of unknown morphology generally decreased over time, so did the survival from these tumours. Therefore, it is likely that these tumours are actually aggressive carcinomas that are not biopsied at the time of diagnosis. Several of the survival estimates were flagged as less reliable, and this was largely due to a high proportion of patients being registered through death certificate only (DCO). Around 7% of all records were registered through DCO, though this proportion varied widely between countries.
Given that all carcinomas were included in the same group, this sub-type is not as homogenous as a group comprising solely of pancreatic ductal adenocarcinomas (ICD-O-3 morphology codes 8020, 8035, 8480, 8490, 8500, 8510, 8560 and 8576) [1]. However, only 8.8% of the patients included in analysis were diagnosed with pancreatic ductal adenocarcinomas. Restricting our analyses only to these tumours would have excluded patients diagnosed with adenocarcinomas (ICD-O-3 morphology code 8140, 45.0% of all patients included in the analyses). We have included some carcinomas that may have better survival than pure adenocarcinomas: serous cystadenocarcinoma (8441), solid pseudopapillary carcinoma (8452), mucinous cystadenocarcinoma (8470), acinar cell carcinoma (8550), acinar cell cystadenocarcinoma (8551) and pancreatoblastoma (8971) [1]. However, the total proportion of carcinomas with these specific morphologies was only 0.6%, thus inclusion of these carcinomas with adenocarcinomas is unlikely to increase survival substantially.
Likewise, within our definition of neuroendocrine tumours, there are some morphologies that have been known to be more aggressive than neuroendocrine tumours in general. Neuroendocrine carcinomas (8246), including large cell (8013) and small cell (8041), have survival similar to other carcinomas [1]. The proportion of neuroendocrine tumours that are neuroendocrine carcinomas was relatively high (63.5%), so the survival estimates for all neuroendocrine tumours may be slightly lower because of the inclusion of neuroendocrine carcinomas.
There are only small variations in the world-wide distribution of pancreatic cancer morphological sub-types, and these variations do not explain fully international differences in five-year survival for all pancreatic tumours combined. However, these analyses highlighted a decreasing trend in the proportion of pancreatic carcinomas that has not been reported previously in the literature.
Stage at diagnosis is an important prognostic factor for pancreatic cancer. The CONCORD-3 database includes data on stage at diagnosis submitted by participating registries that use several different staging classifications. We are finalising a staging algorithm to allow for robust comparisons of stage-specific survival between countries. Future analyses of stage-specific survival for all morphologies combined and by morphology may provide further insight into the international variations in survival.
Survival from pancreatic carcinomas remains poor world-wide and trends for survival from carcinoma showed little improvement over time. Despite slight declines in the proportion of carcinomas over time, carcinomas continue to be the majority of pancreatic tumours. Thus, efforts to develop effective population-based screening techniques, to increase early diagnosis and to decrease resistance to treatment of these tumours, are needed.
Pancreatic neuroendocrine tumours have much higher survival, owing to the fact that the majority of these tumours are functional and are, therefore, symptomatic at earlier stages than carcinomas. Increases in survival from neuroendocrine tumours were also greater than that of carcinomas, indicating that enhancements in diagnostic techniques and treatments have helped improve survival over time. If Survival estimate considered less reliable, because 15% or more of patients were (a) lost to follow-up or censored alive within five years of diagnosis (or if diagnosed in 2010 or later, before 31 December 2014), or (b) registered only from a death certificate or at autopsy, or (c) registered with incomplete dates, i.e., unknown year of birth, unknown month and/or year of diagnosis or unknown year of last vital status. Italics denote survival estimates that are not age-standardised. Survival estimates of 100% that are shown without a confidence interval are those for which the first event (either death or censoring) in that group of patients occurred more than five years after diagnosis.
* Data with 100% coverage of the national population.
the proportion of neuroendocrine tumours continues to increase, efforts are needed to ensure that patients are diagnosed early and treated adequately.

Ethics approval and consent to participate
The Cancer Survival Group maintains approval for processing sensitive personal data for the CONCORD programme from the UK's statutory Health Research Authority (reference ECC 3-04(i)/2011; last update 2 October 2021), the National Health Service Research Ethics Service (11/LO/0331;6 October 2021), and the Ethics Committee of the London School of Hygiene & Tropical Medicine (12171; 6 October 2021).

Funding
This project was supported by the Institut National du Cancer (France), La Ligue Contre le Cancer, US Centers for Disease Control and Prevention, Swiss Re, Swiss Cancer Research Foundation, Swiss Cancer League, Rossy Family Foundation, US National Cancer Institute (USA) and the American Cancer Society.

Role of the funding source
The funding sources played no part in the design, data collection, quality control, analysis, interpretation of the findings, writing of the manuscript or the decision to submit for publication. The corresponding author had full access to all the data and responsibility for submission for publication.
The interpretation of the findings in this report, and the opinions,   (15-99 years) diagnosed with pancreatic cancer, by continent, country and period of diagnosis (2000-2004, 2005-2009, 2010-2014): carcinomas. * Data with 100% coverage of the national population. † National estimate not age-standardised. § National estimate flagged as less reliable.

Table 6
Age-standardised five-year net survival (NS, %): adults (15-99 years) diagnosed with pancreatic cancer by morphological sub-type, country and calendar period of diagnosis (2000-2004, 2005-2009, 2010-2014     -- § Survival estimate considered less reliable, because 15% or more of patients were (a) lost to follow-up or censored alive within five years of diagnosis (or if diagnosed in 2010 or later, before 31 December 2014), or (b) registered only from a death certificate or at autopsy, or (c) registered with incomplete dates, i.e., unknown year of birth, unknown month and/or year of diagnosis or unknown year of last vital status Italics denote survival estimates that are not age-standardised Survival estimates of 100% that are shown without a confidence interval are those for which the first event (either death or censoring) in that group of patients occurred more than five years after diagnosis * Data with 100% coverage of the national population Fig. 3. Distribution of age-standardised five-year net survival (%), adults (15-99 years) diagnosed with pancreatic cancer, by continent, country and period of diagnosis (2000-2004, 2005-2009, 2010-2014): carcinomas. * Data with 100% coverage of the national population. † National estimate not age-standardised. § National estimate flagged as less reliable. Fig. 4. Global range of age-standardised five-year net survival (%) estimates for carcinomas and neuroendocrine tumours of the pancreas, by continent and calendar period of diagnosis (2000-2004, 2005-2009, 2010-2014): adults (15-99 years). Each box-plot shows the range of survival estimates among all cancer registries for which suitable estimates could be obtained for patients diagnosed in each calendar period, in each continent. The number of registries included in each box-plot is shown in parentheses. Survival estimates considered less reliable are not included (see text). The vertical line inside each box represents the median survival estimate among all contributing registries (the central value in the range, or 50th centile). The box covers the inter-quartile range (IQR) between the lower and upper quartiles (25th and 75th centiles). Where there are only a few widely scattered estimates, the median may be close to the lower or upper quartile. The extreme limits of the box-plot are 1.5*IQR below the lower quartile and 1.5*IQR above the upper quartile. Open circles indicate 'outlier' values, outside the range.