Dr J.Y. Kang, Department of Gastroenterology, St George's Hospital, Blackshaw Road, London SW17 0QT, UK. E-mail: firstname.lastname@example.org
Background Rates and time trends in mortality from pancreatic cancer vary considerably between countries.
Aim To examine trends and patterns in the incidence of, and the survival and mortality from, pancreatic cancer in England and Wales from 1975 to 2000; in particular, whether incidence and survival rates are related to socio-economic deprivation.
Methods We calculated annual age-specific and overall age-standardized incidence and mortality rates by sex for pancreatic cancer in total, and by subsite. We also estimated survival by sex and age group and by subsite.
Results In males, the age-standardized rate fluctuated in the late 1970s, to peak at 13.0 per 100 000 in 1979, declined steadily by an average of 1.3% per year to around 10.3 per 100 000 in the mid-1990s and then levelled off. For females, the rate peaked at 8.4 per 100 000 in the late 1980s before declining and fluctuating around 7.7 per 100 000 in the late 1990s. Patterns and trends in mortality rates were closely similar to those in incidence, due to the very low survival rates: only 2–3% at 5 years from diagnosis. Survival rates improved only minimally over the period 1971–99. Incidence and mortality rates were slightly higher in both males and females living in the most deprived areas, but survival was not consistently related to socio-economic deprivation.
Conclusions The incidence of, and mortality from, pancreatic cancer in England and Wales have fallen from peak levels observed in the 1970s and 1980s, and levelled off in the 1990s for both sexes; survival rates remain very low.
Pancreatic cancer accounted for just under 3% of diagnosed cases of cancer and just under 5% of deaths from cancer in 2000 in England and Wales1 and for similar proportions across the countries of the European Union.2 Worldwide, there were estimated to be 216 000 new cases of pancreatic cancer and 213 000 deaths from it in 2000. Pancreatic cancer was the 12th most commonly diagnosed cancer and the eighth most common cause of death from cancer.3 It is more common in men than women and is predominantly a disease of elderly people.4
Pancreatic cancer has among the lowest survival rates of any cancer, with <20% of patients surviving for 1 year and <5% for 5 years after diagnosis. Cigarette smoking is the only established risk factor and has been estimated to explain 25–30% of cases.5 Inherited factors account for 5–10% of cases.6 Other possible risk factors include chronic pancreatitis,7 obesity,8 type II diabetes9 and occupation,5 although evidence for specific occupational risk factors is weak. Pancreatic cancer has mainly been reported to be unaffected by socio-economic status,4, 10–13 although a previous report of trends in England and Wales found a small positive gradient with deprivation.14 The symptoms of pancreatic cancer are non-specific and appear relatively late; there is currently, no effective method of early detection. The only potentially curative treatment is surgery, although only 15–20% of patients are eligible15 and in the majority of these, pancreatic cancer will recur within 1–2 years after surgery.16–18
There is wide variation in the incidence of pancreatic cancer around the world, with particularly high rates in Westernized countries, including the US, Japan, and northern Europe, including Great Britain,4 International trends for pancreatic cancer have varied considerably in the second half of the 20th century. While mortality rates for men in North America and Oceania increased until about 1975 and then declined or remained stable, they have continued to increase steadily in many European countries and in Asia. Among women, pancreatic cancer mortality has generally increased worldwide19 and as a consequence, the male to female ratio of the rates has tended to decrease, especially in the United States.19–22
The aims of our study were to: (i) investigate time trends in the incidence of, and mortality from, pancreatic cancer in England and Wales in the last quarter of the 20th century; (ii) determine whether survival rates have changed over this period; and (iii) determine whether incidence and survival rates were related to socio-economic deprivation.
Incidence and mortality
In England, nine regional cancer registries, and in Wales the Welsh Cancer Intelligence and Surveillance Unit, collate information from various sources to make registrations of cancers for their respective catchment areas. These data are carefully checked by the registries to avoid duplication of records. A subset of the collected data for each patient (the minimum dataset) is sent to the National Cancer Intelligence Centre at the Office for National Statistics (ONS). The data are loaded onto a person-based database and validated.1 Validation includes checking the compatibility of the cancer site and the associated histology. Detailed tables on the incidence of all types of cancer are published for a given incidence year once all expected records for that year have been received and validated. The cancer registries are regularly notified of deaths where cancer is mentioned on the death certificate. In addition, cancer patients are followed up to death (or emigration) through the process of ‘flagging’ in the records kept by the National Health Service Central Register. When a ‘cancer-registered’ patient dies of a non-cancer-related death, this is notified back to the cancer registries and the date of death is added to the patient's record.14
Registration of deaths in England and Wales is carried out by the Local Registration Service in partnership with the General Register Office, which is part of ONS. Information about a death is loaded onto the deaths database at ONS and the underlying cause of death is coded by an automatic process for the vast majority of records.23, 24 After validation checks, detailed tables on deaths occurring in a given year are published.1
Cancer registration statistics have been published for incidence years from 1962 onwards25 but there was noticeable under-ascertainment of cases in the 1960s; ascertainment improved after the re-organization of the system in 1971.14 We used validated registrations of cancer, and deaths from cancer, from 1975 to 2000 for analysis. Cases of pancreatic cancer were identified using the eighth or ninth Revisions of the International Classification of Diseases (ICD8 or ICD9) site code 157 from 1975 to 1994 and ICD10 site code C25 thereafter.26–28 Cases were excluded from the analysis if the behaviour of the tumour was classified as benign, uncertain or in situ (<1% of the cases registered between 1975 and 2000), leaving malignant primary and a very small number of metastatic tumours (<0.5% of cases). For consistency across the three revisions of ICD used from 1975 to 2000, cancers located in the body, neck, tail or other parts of the pancreas were grouped together. In ICD8 there were no separate codes for cancers of the pancreatic duct or islets of Langerhans, but each of these accounted for <0.5% of pancreatic cancers between 1979 and 2000. In addition to ICD site code, cases of cancer were classified by histology code: those diagnosed from 1975 to 1978 using MOTNAC29 those from 1979 to 1994 using ICD-O30 and those from 1995 onwards using ICD-O2.31 Pancreatic cancer deaths were identified using ICD8 or ICD9 site code 157. ICD10 was introduced for coding the cause of death in 2001.32, 33
Age-specific incidence and mortality rates were calculated for males and females, and age-standardized rates were calculated by weighting the age-specific rates according to the European standard population. From 1984 to 1992 the Office of Population and Census Statistics (now the Office for National Statistics) changed its interpretation of the rule governing the selection of the underlying cause of death, as defined in ICD9. This resulted in an abrupt decrease in 1984 in the mortality rate from several ‘terminal conditions’, including pneumonia and liver failure, and small increases in mortality from many causes, including most cancers;24, 34 there were corresponding changes in reverse directions in 1993.24 The results presented here are based on data which have been adjusted to account for these and other changes in coding procedures24 and are therefore comparable over the period 1975–2000.
Records for some patients were excluded from the survival analyses,35 for example those whose record at the National Health Service Central Register could not be traced or who had ‘zero survival’. Of the latter cases, most were registered solely from the information provided on the death certificate, but some would have died on the day of diagnosis (‘true zero survival’). In total, 17 919 patients diagnosed in 1996–99 were included in the analysis (8837 men and 9082 women). Data were compared with earlier results for patients diagnosed in 1971–9035 and 1991–95.36 For the analysis by subsite, some subsites for which there were very few cases were excluded (255 patients); data for the body and tail of the pancreas were analysed separately, but the numbers of cases were small, and so the confidence intervals were correspondingly wide. In addition, there were too few cases and deaths in men for 5-year survival to be calculated for these two subsites.
Crude and relative survival rates by age group were calculated for men and women. Crude survival is the probability of survival irrespective of the cause of death. Relative survival allows for the fact that not all cancer patients die from their cancer, by comparing the crude survival among cancer patients with the expected survival in the general population of the same sex and age (the background mortality). The expected survival is based on a life table. The relative survival rate can be interpreted as survival from the cancer concerned in the absence of other causes of death.35
For cancer registrations from 1996 to 2000, each patient was assigned to one of the five categories of socio-economic deprivation based on their electoral ward of residence at diagnosis. These were labelled from most affluent to most deprived. Deprivation categories were based on quintiles of the ward income domain score,36 a subcomponent of the indices of multiple deprivation 2000 for England,37 combined with the comparable but not precisely equivalent index available for Wales.38 The very small proportions (<0.5%) of patients whose records could not be reliably assigned to a deprivation category were excluded. Incidence rates by deprivation category were then calculated using population denominators based on 1998 electoral ward boundaries.39 Relative survival rates by deprivation category were estimated.
Statistical analysis was carried out using the STATA statistical software package (StataCorp LP, College Station, TX, USA). Statistical significance was taken as P < 0.05.
There were around 2900 registered cases of pancreatic cancer in males and 2600 in females in 1975. By 2000, the population of England and Wales had increased from 49.5 to 52.1 million, and the age distribution had shifted towards the elderly: in 2000, 1.1% of males and 2.7% of females were aged 85 and over, compared with only 0.5% and 1.4%, respectively, in 1975. By 2000, the numbers of cases of pancreatic cancer had increased to 3100 in males and 3300 in females.
In males, the age-standardized rate fluctuated in the late 1970s, to peak at 13.0 per 100 000 in 1979, declined steadily by an average of 1.3% per year to around 10.3 per 100 000 in the mid-1990s and then levelled off (Figure 1). Age-standardized incidence in females rose slightly through the late 1970s and 1980s to peak at 8.4 per 100 000, declined to around 7.7 in the mid-1990s and fluctuated in the late 1990s. In 1975, the incidence rate in males (12.3 per 100 000) was around 60% higher than that in females (7.4 per 100 000): a highly significant difference (t-test). By 2000, the difference between the rates (10.5 and 7.9 per 100 000 in males and females respectively) had fallen to about one-third; still a highly significant difference.
Pancreatic cancer is predominantly a disease affecting elderly people: age-specific incidence increased steeply with age from the mid-1950s, and was higher at all ages in men than women (data not shown); in 2000 over 90% of cases occurred in those aged over 55 (Figure 2). From 1975 to 2000, age-specific incidence decreased significantly in men for all age groups below 85 years, although the decreases were slight for those aged under 65 (Figure 3a). In men aged 85 and over, there was a decreasing trend in incidence over the period, but this was of borderline significance (P = 0.07). In women, there was little change in age-specific incidence in those aged under 75, and a slight but significant increase in those aged 75 and over (Figure 3b).
Incidence by subsite
In 2000, just over half (55%) of cases were not coded to a specific area of the pancreas. Of those tumours that were specifically coded, most were located in the head of the pancreas (38% of all cases); around 7% were located in the body, neck, tail or other part of the pancreas. The proportions were very similar for males and females.
Between the late 1970s and late 1990s there were significant decreases in the age-standardized incidence rates of cancers of the head of the pancreas and of the body, neck, tail and other part of the pancreas, while there was a significant increase in the rate of cancers of unspecified part of the pancreas (Figure 4). In the mid-1980s there was an apparent sharp decrease in the rates of cancers of the head of the pancreas and body, neck, tail and other part of the pancreas, accompanied by a simultaneous sharp increase in the rates of unspecified pancreatic cancer. This sudden change was mainly due to changes in coding practices in one of the English cancer registries.
Incidence by morphology
In 2000, only around 30% of cases of pancreatic cancers were diagnosed on the basis of histological verification from a biopsy. A similar proportion (34%) was diagnosed clinically. The majority of malignant pancreatic tumours were recorded as epithelial neoplasms or adenocarcinomas: these two categories accounted for just over 80% of the total. Cystic, mucinous and serous neoplasms accounted for 2% of cases, while 16% of tumours were classed as ‘neoplasms, not otherwise specified’ (NOS). The proportion of tumours classified as neoplasms NOS decreased gradually to a very low level during the period that ICD9 was in use (1979–94), while the proportions of epithelial neoplasms and adenocarcinomas both increased. Following the introduction of ICD10, the proportion of neoplasms NOS has risen.
Since survival from pancreatic cancer was very low throughout the period of study, mortality rates closely followed those for incidence (Figure 1). Age-standardized mortality in males fluctuated in the late 1970s, reached a peak at 12.8 per 100 000 and then declined by an average of 1.3% per year to around 9.7 per 100 000 in the late 1990s. In females, mortality changed very little, increasing slightly to a peak of 7.9 per 100 000 in the mid-1980s and then declining again to around 7.4 per 100 000 in the late 1990s.
In line with incidence, age-specific mortality increased steeply above age 55 and was higher for males than females at all ages. The age-distribution of deaths from pancreatic cancer was therefore closely similar to that for cases. For males, the trends in age-specific mortality were very similar to those for incidence, with significant decreases over the study period for all age groups. For women, however, the pattern in mortality differed slightly from that for incidence, with no significant trends for those aged 65 and over, and small but significant decreasing trends for those aged 20–64.
One-year relative survival for men and women diagnosed with pancreatic cancer in 1996–99 was around 13% and 5-year survival was only 2–3%. For both sexes, relative survival rates decreased with age. In all age groups, both 1- and 5-year relative survival rates for men and women were very similar, but there was a suggestion (based on small numbers of patients) that 5-year survival was higher for women aged 15–39 than for men of the same age (Table 1).40
Table 1. One- and 5-year crude and relative survival (%) from pancreatic cancer in adult* patients diagnosed from 1996 to 1999, by sex, England and Wales
Relative (95% CI)
Relative (95% CI)
n, number of cases included in analysis; 95% CI, 95% confidence interval.
Survival from pancreatic cancer improved only very slightly over the last 30 years of the 20th century. From 1971 to 1990, 1-year survival increased by an average of 1% point every 5 years, and 5-year survival by less than half a percentage point.35 Similar small changes were seen between survival for patients diagnosed in 1991–95 and those diagnosed in 1996–99.40
Survival by subsite
For patients diagnosed in 1996–99, there were quite large differences in 1-year survival rates between subsites (Table 2). In both men and women, 1-year survival from cancer of the head of the pancreas was 16%, compared with only 8% for the body of the pancreas and around 10% for unspecified site within the pancreas. Rates for 5-year survival were more similar between subsites, ranging from 2% to 4%. Over time, 1-year survival has consistently been slightly higher for head of the pancreas and slightly lower for body of the pancreas and for pancreatic cancer unspecified, compared with the rate for all subsites combined. Rates for tail of the pancreas should be viewed with some caution, as the numbers of cases were very small.
Table 2. One- and 5-year relative survival (%) from pancreatic cancer in adult* patients diagnosed from 1986 to 1999, by subsite and sex, men and England and Wales
% (95% CI)
% (95% CI)
n, number of cases included in analysis; 95% CI, 95% confidence interval; –, insufficient number of deaths in period to enable the calculation of a survival rate.
* Aged 15–99.
0.6 (−0.3 to 1.5)
Combining data for 1996–2000, the incidence rate of pancreatic cancer was 11.1 per 100 000 in males living in the most deprived areas and 9.7 per 100 000 in those living in the most affluent areas, a difference of around 14%. The rate in the most deprived category was significantly higher than the rates in each of categories 1–4 (t-test). In females, there was a slight but significant increasing trend in incidence with increasing level of deprivation; the rate was around 10% higher in the most deprived category (8.0 per 100 000) than in the most affluent (7.3 per 100 000).
The difference in 5-year relative survival between the most deprived and most affluent patients (diagnosed in 1996–99) was small, but in opposing directions for men and women. In men, survival was worse in the most deprived, but in women it was worse in the most affluent, although only the difference for women was statistically significant.36
Mortality rates of pancreatic cancer increased gradually in both males and females in England and Wales during the 1950s and 1960s.14 Our results show that, over the last quarter of the 20th century, incidence peaked for both sexes; the male peak occurring in the late 1970s and the female peak in the late 1980s. In males, the incidence rate declined up to the mid-1990s and has since stabilized, while in females incidence declined more gradually up to the late 1990s. Over the study period, the male to female ratio decreased. Our findings are comparable with those reported for North America.19, 22 Mortality rates closely followed those for incidence and the low survival rates improved only marginally over the study period. Our results suggested a small but significant association between the incidence of pancreatic cancer and socio-economic deprivation in both sexes, similar to that previously reported.14 Several other reports have found no such association.10, 11, 13 This difference may possibly be explained by the different measures of socio-economic status employed: this study and the previous one which found a positive association14 used the Carstairs Index of deprivation based on area, whereas others have used social class based on occupation.10, 11, 13 There was no consistent association between survival and deprivation.36
Changes in coding for cancer registration, in diagnostic modalities and in medical procedures can potentially affect time trends in the incidence of pancreatic cancer. Over the study period, three different revisions of the ICD have been used to code cancer registrations. The changeover from ICD9 to ICD10 appears to have had some effects on the classification of pancreatic cancers by histology, but very little effect on classification by subsite and no effect on the overall rates of pancreatic cancer. Mortality data have been adjusted for the relatively small effects of the changes in coding the underlying cause of death, and other procedures, between 1984 and 1992. Until the advent of ultrasonography, computed tomography and endoscopic cholangiopancreatography, pancreatic cancer had been difficult to diagnose without surgery or autopsy. The introduction of these modalities in the 1970s, and their widespread uses in the 1980s and 1990s would have facilitated diagnosis, but would not have accounted for the different time trends between men and women.
The only well-established aetiological factor for pancreatic cancer is cigarette smoking.5 The risk is dose dependent,41 and appears particularly high in individuals with a family history of pancreatic cancer.42, 43 Cigarette smoking became common in men in the early 20th century, about 20 years earlier than in women, and was considerably more prevalent in men by the late 1940s (65% in men and 41% in women in 1948). The proportion of smokers fell among males but remained fairly constant among females for the next 20 years.44 Only after 1970 did the prevalence among women begin to fall, and as levels continued to fall among men, by 2000 the prevalence of smoking was the same in men and women (26%).45
Time trends in the incidence of, and mortality from, lung cancer and gender differences in these trends can be explained by changes in smoking behaviour.14 As time trends for pancreatic cancer were similar to those for lung cancer14 it is likely that changes in smoking behaviour may also explain a large part of the pattern in the trends for pancreatic cancer, particularly the peak and subsequent decline in rates in males. As for lung cancer, there is likely to be a long latency period for pancreatic cancers attributable to smoking, which can be seen from the time differences in the peaks of smoking prevalence and the peaks in incidence and mortality rates. Rates of lung cancer are much higher than those of pancreatic cancer because the risk conferred by smoking is much greater for lung cancer.46
There has been considerable debate about whether obesity increases the risk of pancreatic cancer. A meta-analysis of 14 studies published between 1990 and 2003 concluded that there is a weak association between obesity and pancreatic cancer, with an increased risk of 19% in obese people compared to those with a normal body weight.8 Several studies have reported a slightly higher risk in obese men compared with obese women.47–50 Obesity is associated with abnormal glucose metabolism and type II diabetes, themselves implicated in the development of pancreatic cancer9, 51–53.
The prevalence of obesity has increased over recent years, especially in the UK54 and in the US.55 In England the proportion of the population of both sexes classified as obese increased from under 10% in 198056 to just under a quarter in 2003.57 The prevalence of type II diabetes also increased during the 1990s, particularly in men.56 Given the long latency period for pancreatic cancer, the fairly recent increases in the prevalence of obesity and diabetes are unlikely to have contributed to the observed trends in incidence, but may affect future trends.
The positive association between deprivation and pancreatic cancer incidence may be explained by the higher prevalence of smoking in manual vs. non-manual socio-economic groups.58 As the association is weak, smoking can only explain a fairly small proportion of pancreatic cancer cases. Obesity (and diabetes) is also related to deprivation,57 and hence if obesity affects future trends in pancreatic cancer it is possible that pancreatic cancer may become more strongly associated with socio-economic deprivation.
Pancreatic cancer is likely to become a bigger public health problem in future decades as the proportion of elderly people in the population increases and the prevalence of obesity rises. Public health campaigns to encourage healthy eating may become as important as those aimed at reducing cigarette smoking in potentially reducing the incidence of pancreatic cancer. Without an effective method of early detection, or effective curative treatments suitable for the majority of patients, survival rates are likely to remain very low. However, a major government initiative was recently introduced to improve the management of cancer and a ‘two-week rule’ introduced to ensure that patients suspected of having cancer are seen, investigated and treated promptly.59 Further studies are required to see whether this will have an effect on survival from pancreatic cancer.