Global estimates of cancer prevalence for 27 sites in the adult population in 2008

Authors

  • Freddie Bray,

    Corresponding author
    1. Section of Cancer Information, International Agency for Research on Cancer, Lyon, France
    • Section of Cancer Information, International Agency for Research on Cancer, 150 cours Albert Thomas, F-69372 Lyon, Cedex 08, France
    Search for more papers by this author
    • Tel.: +33-4-72-73-84-53, Fax: +33-4-72-73-86-96

  • Jian-Song Ren,

    1. Section of Cancer Information, International Agency for Research on Cancer, Lyon, France
    Current affiliation:
    1. Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, Maryland, USA
    Search for more papers by this author
  • Eric Masuyer,

    1. Section of Cancer Information, International Agency for Research on Cancer, Lyon, France
    Search for more papers by this author
  • Jacques Ferlay

    1. Section of Cancer Information, International Agency for Research on Cancer, Lyon, France
    Search for more papers by this author

Abstract

Recent estimates of global cancer incidence and survival were used to update previous figures of limited duration prevalence to the year 2008. The number of patients with cancer diagnosed between 2004 and 2008 who were still alive at the end of 2008 in the adult population is described by world region, country and the human development index. The 5-year global cancer prevalence is estimated to be 28.8 million in 2008. Close to half of the prevalence burden is in areas of very high human development that comprise only one-sixth of the world's population. Breast cancer continues to be the most prevalent cancer in the vast majority of countries globally; cervix cancer is the most prevalent cancer in much of Sub-Saharan Africa and Southern Asia and prostate cancer dominates in North America, Oceania and Northern and Western Europe. Stomach cancer is the most prevalent cancer in Eastern Asia (including China); oral cancer ranks as the most prevalent cancer in Indian men and Kaposi sarcoma has the highest 5-year prevalence among men in 11 countries in Sub-Saharan Africa. The methods used to estimate point prevalence appears to give reasonable results at the global level. The figures highlight the need for long-term care targeted at managing patients with certain very frequently diagnosed cancer forms. To be of greater relevance to cancer planning, the estimation of other time-based measures of global prevalence is warranted.

Prevalence measures the absolute number (or relative proportion) of individuals in the population affected by a given disease and requiring some form of medical care. Although incidence and mortality are considered the key measurements of cancer burden and the demands for cancer services, prevalence is an important supplementary indicator for developing strategies for service provision. Unlike incidence and mortality, however, there is no unique definition of cancer prevalence. Total (or complete) prevalence is the number of persons in a defined population alive at a given time who have had cancer diagnosed at some time in the past. This may be estimated directly in population-based cancer registries by counting the number of cases still alive and present at a specified point in time; however, this approach requires registration and follow-up for vital status over many years. Furthermore, the resource requirements for treating newly diagnosed patients are quite different from those for supporting long-term survivors,1 and with increasing numbers of patients considered clinically cured, alternative definitions of prevalence are commonly used. Limited duration prevalence is the number of patients diagnosed with cancer within a fixed time in the past and is likely to be pertinent in estimating the needs for cancer services according to specific phases of cancer care from diagnosis through to palliation and death. A number of country- and cancer-specific approaches to prevalence estimation based on mathematical models have been recently published.2–7

Previously, we had described and validated methods that estimated limited-duration cancer prevalence worldwide (as a product of incidence and survival) by region in 1990,8 and we compiled and presented a set of revised estimates at the country level in GLOBOCAN for 20009 and 2002.10 This article updates the estimates to 2008 and reports 1-, 2- to 3- and 4- to 5-year prevalence for the adult population (aged 15 years or older) by world region and predefined categories of human development.11 The prevalence estimates therefore pertain to the number of cancer cases diagnosed between 2004 and 2008 who were still alive at the end of 2008; partitioning these into varying durations serves as an aid in identifying the resources required during the phases of initial treatment (within 1 year), clinical follow-up (2–3 years) up to the point of cure (4–5 years). Although the computational procedures remain as in the previous exercise,8 the sources of survival used to calculate the global prevalence figures have been updated as a result of the availability of more recent and geographically extensive datasets, particularly in low- and medium-income countries.

In our article, we describe and interpret the patterns of sex-specific prevalence for 27 cancer sites by level of development, across 13 world regions and 184 countries. We also include a validation exercise that compares our estimates against those observed within the five Nordic countries.12 We end by discussing the need and requirements for cancer-specific measures of prevalence that are more relevant in the planning of cancer services.

Methods and Data Sources

Estimation of limited duration prevalence

The n-year prevalence at age k was estimated from year-specific incidence rates and survival probabilities according to the following formula:

equation image

where ICk is the annual number of new cases at age k, S is the absolute survival amongst patients with cancer and Sk(t) is the proportion of cases diagnosed at age k and alive at time t after diagnosis, with n the number of years as prevalent cases following diagnosis. The algorithm requires estimation of the distribution of incident cases by single year of age, obtained by interpolating the cumulative distribution of cases by age using a logistic function on fixing the total number of cases. Person-years at risk were obtained from the United Nations national population estimates for 2008 (the 2008 revision) to derive incidence cases from the incidence rates for the same year. We assume that (i) incidence rates and the corresponding person-years at risk are constant in the preceding n years; (ii) annual incidence cases occur at mid-year and (iii) the curve of absolute survival at time intervals of 6 months follows the Weibull distribution. Prevalent cases were aggregated into the corresponding age groups as used for incidence (see below).

Sources of incidence and survival data

Estimates of the worldwide incidence in 2008 for 27 cancers (see list below) in each of the 184 countries were extracted by age group (15–44, 45–54, 55–64 and >65 years) and sex from the GLOBOCAN 2008 database13 (http://globocan.iarc.fr) as compiled by the International Agency for Research on Cancer. The sources of absolute survival used in the prevalence estimations in the specified regions/countries of the world are given in Appendix Table 1. In brief, three sources of site-specific survival data were available:

  • (i)The IARC study on cancer survival in mainly low- and middle-income countries estimated sex-specific absolute survival of over half a million cases of 1 to 56 cancer types from 27 population-based cancer registries in 14 countries within Eastern and Western Africa, the Caribbean, Central America and four regions of Asia.14, 15 Cases were those registered during 1990–2001, with the diagnostic period varying for individual registries. Based on individual or aggregated registry-specific 1- through to 5-year absolute survival in a single country, we were able to estimate corresponding national cancer prevalence in Uganda, Zimbabwe, China, Republic of Korea, Singapore, Thailand and India. To estimate survival in other low- and medium-income countries or for specific cancer sites for which survival estimates were not available in any one of the above countries, we developed three proxies of age and site-specific survival based on the arithmetic mean of registry-specific survival: Human Development Index I (HDI I) comprised a weighted average of survival from Korean registries and Singapore and represents survival in countries with very high levels of the HDI (a summary measure of human development, see below), and HDI II was the average survival from Chinese, Indian and Thai registries, alongside the registries in Uganda and Zimbabwe (e.g., representing survival in low to medium HDI). Finally, HDI III was an unweighted average of HDI I and HDI II and represented an intermediate survival figure. The level of human development in a given country determined which of the three survival indexes were assigned to a given country to estimate cancer prevalence (Appendix Table 1).
  • (ii)The Surveillance, Epidemiology and End Results (SEER) programme in the United States produced estimates of 1- to 5-year absolute survival based on registry data from nine areas covering ∼10% of the US population for cases diagnosed between 2001 and 2007. We used all-race survival proportions to estimate prevalence in North America and for selected sites in Australia.
  • (iii)The EUROCARE-4 project provided 1- to 5-year absolute figures from several European cancer registries for the period 1995–1999.16 Where possible, country-specific survival estimates were used, based on one or more cancer registries within that country, or elsewhere, in European countries where no local survival data were available, three sets of regional estimates were prepared (Appendix Table 1).

In addition, country-specific sources of survival were used in Japan and Australia. For composite sites such as oral cavity and mouth, we used, where necessary, weighted averages of sub-site survival proportions.

Validation

We compared the 5-year point prevalence based on actual counts of prevalent cancer cases for 25 cancer sites by sex obtained from the Association of the Nordic Cancer Registries (Denmark, Finland, Iceland, Norway and Sweden) to our estimates derived from incidence and survival obtained in the same five countries. The aggregated values are compared in Appendix Table 2 as ratios of 5-year prevalence to incidence, as well as absolute numbers of 5-year prevalence.

Overall, the results suggested that for the majority of cancers, our estimates of prevalence were reasonably consistent with those observed, particularly for highly prevalence cancers in colorectal and breast, as well as melanoma of the skin (relative differences in counts <10%). However, for prostate cancer, the relative difference was 22%, whereas some of the cancer-specific differences were far greater than 25%. The largest errors (in both sexes) were observed for brain cancers (related to classification issues related to inclusion of benign tumours) and for several poor prognosis cancers (oesophagus, liver, gallbladder and pancreas), with the higher prevalence to incidence ratios in the observed sources pointing to an underestimate of absolute survival in our data.

Presentation of results

Tabular and graphical representations of 5-year cancer prevalence in 2008 are presented according to duration from diagnosis (within 1 year, 2–3 years and 4–5 years earlier) for the adult population (aged 15 years and older): they thus constitute the cancer prevalence among patients diagnosed and alive at the end of 2008 who were diagnosed in 2008, 2006–2007 and 2004–2005, respectively. These estimates are calculated for all cancer sites combined excluding non-melanoma skin cancers (ICD-10 C00–43, C45–96) and for the following specific types: lip and oral cavity (C00–08), nasopharynx (C11), other pharynx (C09–10, C12–14), oesophagus (C15), stomach (C16), colorectum (C18–21), liver (C22), gallbladder (C23–24), pancreas (C25), larynx (C32), trachea, bronchus and lung (C33–34), melanoma of skin (C43), Kaposi sarcoma (C46, estimated in GLOBOCAN 2008 only for Sub-Sahara Africa), breast (C50), cervix uteri (C53), corpus uteri (C54), ovary (C56), prostate (C61), testis (C62), kidney (C64–66), bladder (C67), brain, nervous system (C70–72), thyroid (C73), Hodgkin lymphoma (C81), non-Hodgkin lymphoma (C82–85, C96), multiple myeloma (C88, C90) and leukaemia (C91–95). Calculations were based on four age groups: 15–44, 45–54, 55–64 and >65 years and are reported here separately by sex. We present the prevalence as overall numbers and as crude proportions, either per 100000 or as percentages; because prevalence is intended to measure the absolute load on health services, standardisation by age is not applied.

For global and selected country-specific analyses, we present the prevalence estimates in 13 world regions and according to the traditional dichotomy of developed and developing areas. To further examine prevalence patterns by levels of socioeconomic development, we linked the country-specific prevalence estimates to the corresponding HDI, as estimated for 2007 by the United Nations Development Programme (UNDP).11 The HDI (based on the 2007 UNDP estimates) is a composite index of three basic dimensions of human development, namely, a long and healthy life (based on life expectancy at birth), access to knowledge (based on a combination of adult literacy rate and primary to tertiary education enrolment rates) and a decent standard of living (based on GDP per capita adjusted for purchasing power parity in US$). We partitioned the HDI using the UNDP-defined distribution of the countries into four levels (see footnote in Table 1): low HDI (HDI < 0.5); medium HDI (0.5 ≤ HDI < 0.8); high HDI (0.8 ≤ HDI < 0.9) and very high HDI (HDI ≥ 0.9).

Results

Global prevalence by HDI category

Table 1 displays population totals and 5-year prevalence of all cancers (as cases and percentages) by world area, level of development and sex. The 5-year global prevalence for all cancers combined is estimated to be 28.8 million in 2008, and thus there were almost 29 million persons living with cancer in 2008 who were diagnosed within the last 5 years. The percentage of persons affected by cancer in 2008 and diagnosed within the previous 5 years is considerably higher in very high HDI areas relative to those residing in both low HDI (a sevenfold difference) and medium HDI areas (a fivefold difference). Close to two-thirds of the prevalent cases occur in the very high and high HDI settings (62.2%). Countries that constitute very high levels of human development encompass close to half (47.2%) of the 5-year global prevalence (13.6 million cases) despite comprising only 17% of the world's population. In contrast, the 3.2 billion living in medium HDI countries represent almost two-thirds (64.3%) of the population worldwide in 2008, but only one-third (35.8%) of the global cancer prevalence figure is of 10.3 million. In low HDI countries, the 5-year cancer prevalence of 480,000 (1.7% of the global total) is estimated in a population of less than 220 million (4.5% of the world total).

Table 1. Population (in thousands) and all-sites 5-year prevalence of cancer cases (in thousands, %) aged older than 15 years by sex, region and level of development
inline image

The three measures in Tables A3–A5 crudely approximate the number of cases undergoing initial standard treatment (1 year), those under clinical follow-up (2–3 years) and those in remission and entering long-term follow-up (4–5 years) by cancer site, sex, world region and level of development. These statistics when aggregated are equal to the prevalence of all cases diagnosed within 5 years in Table 1. On average, in very high HDI areas, 0.41% (459 and 397 every 100,000 men and women, respectively) of the adult population are undergoing treatment for cancer (Supporting Information Appendix Table 3), 0.67% are under clinical follow-up (Supporting Information Appendix Table 4) and 0.54% are in remission (Supporting Information Appendix Table 5). The corresponding figures for high, medium and low HDI areas are considerably lower than these values in both sexes.

Global prevalence by world region

Figure 1a shows the 5-year prevalence and the proportions of cases surviving for 1 year, 2–3 and 4–5 years after diagnosis by world region. For most cancer sites, 2- to 3-year prevalence (Supporting Information Appendix Table 4) is greater than the prevalence in the first year after diagnosis (Supporting Information Appendix Table 3) and that of cases at 4–5 years after diagnosis (Supporting Information Appendix Table 5). In terms of the absolute load of cancer on health services by world region, the highest 5-year prevalence is seen in Eastern Asia, with over 7 million alive in 2008 and diagnosed with cancer within the previous 5 years (Fig. 1a). North America and Western Europe follow with figures of ∼4.7 and 3.0 million, respectively. South-Central Asia (including India) ranks fourth with 2.7 million affected persons, considerably lower than the Eastern Asian figure despite the former having an adult population size only 7% smaller than the latter (Table 1). However, when the corresponding populations in each region are taken into account (Fig. 1b and Table 1), 5-year prevalence proportions (as a crude percentage) are highest in the most developed areas of the world, with a range of 1.5–2% of these populations affected by cancer: Western Europe (1.9%) and Australia/New Zealand (1.8%) have the highest prevalence proportions, followed by Northern America and the rest of Europe. The corresponding proportions in Africa, the Middle East and South-Central Asia are all less than 0.3%.

Figure 1.

(a) Bar chart of 5-year cancer prevalence counts (millions) by world area; stacked bars denote cancer prevalence among patients alive at the end of 2008 who were diagnosed in 2008, 2006–2007 and 2004–2005, respectively, for example, <1 year, 2–3 and 4–5 years after diagnosis. All cancers combined (except non-melanoma skin cancer), both sexes and aged 15 years and older sorted by magnitude of prevalence counts. (b) Bar chart of 5-year cancer prevalence proportions (%) by world area; stacked bars denote cancer prevalence among patients alive at the end of 2008 who were diagnosed in 2008, 2006–2007 and 2004–2005, respectively, for example, <1 year, 2–3 and 4–5 years after diagnosis. All cancers combined (except non-melanoma skin cancer), both sexes and aged 15 years and older sorted by magnitude of prevalence proportions. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Global prevalence by sex

It is worth noting that the relative magnitude of 5-year prevalence of all cancers combined varies considerably between the sexes across the world regions according to degree of human development. Overall, there is a considerably higher prevalence burden in women (Table 1) in Sub-Saharan Africa, South-Eastern Asia, South-Central Asia and Oceania (excluding Australia/New Zealand) where the male to female ratio is 0.6. In contrast, the prevalent numbers in Southern and Western Europe and Australia/New Zealand indicate a slight excess in men. The sex ratio thus increases with the level of development, with low HDI countries exhibiting 5-year male:female prevalence ratios of around 0.6 compared to estimates of about 1.1 in very high HDI areas.

Global prevalence by cancer site and country

Figure 2 depicts the 5-year cancer-specific prevalence and the proportions of cases surviving for 1 year, 2–3 and 4–5 years, respectively. With 5.2 million cases, female breast cancer is the most prevalent neoplasm worldwide, with one in six cancer survivors in 2008 diagnosed within the previous 5 years. Prostate and colorectal cancers rank second and third globally, with a similar number of 5-year prevalent cases (around 3.2 million cases each). The three neoplasms in combination are responsible for over two-fifths of the prevalence burden. Lung, stomach and cervix cancers rank fourth to sixth.

Figure 2.

Bar chart of global 5-year cancer prevalence counts (millions) by cancer site; stacked bars denote cancer prevalence among patients alive at the end of 2008 who were diagnosed in 2008, 2006–2007 and 2004–2005, respectively, for example, <1 year, 2–3 and 4–5 years after diagnosis. All cancers combined (except non-melanoma skin cancer), both sexes and aged 15 years and older sorted by magnitude of prevalence counts. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Figures 3a3c show global maps of the cancers that lead, in terms of 5-year prevalence, at the national level in 184 countries among males, females and both sexes combined, respectively. Prostate cancer ranks as the most prevalent cancer in men in 111 countries worldwide, including all countries within the Americas, most of Europe and 31 countries of the 46 Sub-Saharan African countries (Fig. 3a). Colorectal cancer ranks highest in men in 25 countries, including 13 Asian countries, and stomach cancer is the most prevalent cancer in Eastern Asia (including China), with oral cancers ranking as the most prevalence cancer in Indian men. Kaposi sarcoma has the highest 5-year prevalence among males in 11 countries in Sub-Saharan Africa. In women, breast cancer ranks as the most prevalent cancer in the vast majority of countries worldwide, that is, in 145 of 184 countries (Fig. 3b). The key geographical areas where there are notable exceptions include Sub-Saharan Africa and parts of Asia (including India), where cervix cancer is the neoplasm most affecting women at 5 years (in 37 countries worldwide).

Figure 3.

Global maps depicting the most prevalent cancer in the adult population at 5 years in 184 countries among (a) men, (b) women and (c) both sexes.

When taking both sexes in combination, a complex pattern emerges: breast cancer continues to be the most prevalent cancer in men and women combined in the overwhelming majority (112) of countries globally. However, cervix cancer remains the most prevalent in a further 33 countries, predominantly in Sub-Saharan Africa and Southern Asia. Thus, female-specific cancers can be observed as the most prevalent form of cancer in both sexes combined in three-quarters of the countries of the world. In a further 28 countries, prostate cancer is the most prevalent, largely in the highest developed areas including North America, Oceania and Northern and Western Europe. The prevalence of breast, cervix or prostate cancer thus ranks as the most prevalent cancer in almost 95% of the 184 countries examined. There are notable exceptions, including the top ranking of colorectal cancer prevalence in Japan, stomach cancer in China and Kaposi sarcoma in Botswana and Malawi in sub-Saharan Africa.

Global prevalence by cancer site and HDI

The five most prevalent cancers diagnosed within 5 years according to four levels of human development (Fig. 4) emphasise the extent to which the cancer profiles among survivors differ both by place of residence as well as corresponding level of resource. The most prevalent cancer site irrespective of the level of development is female breast cancer with around 18% of all cancer cases diagnosed within 5 years and alive at the same time in each of the four HDI regions. After breast cancer, cancers of the prostate, colorectum, lung and bladder are the most prevalent in very high HDI areas, the former three of which constitute 6.7 million prevalent cases, almost half (49.3%) of the entire prevalence burden in these areas. The same three cancers rank one, two and three in high HDI areas; however, cervix ranks as the fourth most prevalent cancer after lung cancer. Prostate cancer does not feature in the five highest ranking neoplasms in medium HDI, with female breast cancer by far the most prevalent cancer in these countries, with 1.8 million cases, and cervix, colorectal and stomach cancer each constituting a further 850,000–900,000 prevalence cases at 5 years. Finally, as with medium HDI areas, the 5-year prevalence of breast and cervix ranks as the two top-most prevalent cancers in low HDI areas, but with the order of ranking reversed; Kaposi sarcoma ranks third in these areas above prostate and colorectal cancer; altogether, the five neoplasms comprise 57% of the total cancer prevalence in low HDI settings.

Figure 4.

Bar charts of incidence and 5-year cancer prevalence counts (in thousands) in the adult population by the level of human development: very high, high, medium and low and both sexes. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Discussion

Our global estimate of 28.8 million people alive in 2008 who had been diagnosed with cancer in the previous 5 years represents a useful supplement to the corresponding figures of incidence (12.7 million new cases of cancer) and mortality (7.6 million cancer deaths) worldwide. Although the methods used to estimate limited duration prevalence appear to give fairly reasonable results for most if not every cancer site according to the validation exercise, the overall global figure is probably on the conservative side given that the survival sources available in this study (other than SEER) largely pertained to the prognosis of patients diagnosed during the 1990s. We might thus expect that access to more recent survival data for certain forms of cancer (particularly in higher resource areas) would capture further improvements in prognosis leading to a greater number of global cancer survivors in 2008. On the other hand, the availability of survival data in low- and middle-income countries14, 15 has increased considerably as the previous exercise,8 and as a result, may more accurately convey prognosis in lower resource settings and consequently the estimates of prevalence.

It is important to assess the major drivers of the observed geographical variations in these estimates of cancer prevalence worldwide and their relation with incidence, mortality and survival. The magnitude of disease incidence and the distribution of frequent cancers are key determinants of crude prevalence so that differences by region or level of societal development partly reflect variations in risk. Close to two-thirds of the 5-year prevalent cases occur in countries that can be considered ‘developed’ (HDI ≥ 0.8), and close to half of the prevalence burden is seen in areas of very high human development (HDI ≥ 0.9) that comprise only one-sixth of the world's population. The overall magnitude of 5-year cancer prevalence is therefore directly related to levels of human development, and the vast differences in the magnitude of prevalence in Northern America and Europe compared to Sub-Saharan Africa and South-Central Asia may not only be partially explained by longer average life expectancy at higher levels of HDI (with risk of cancer greatest at older ages) but will also reflect the higher incidence of cancers associated with higher fatality rates in many lower HDI settings, together with poorer average outcomes following diagnoses of those cancers requiring more sophisticated and resource-intensive treatment options to reduce case fatality.

Breast cancer continues to be the most prevalent cancer in the overwhelming majority of countries globally, with cervix cancer ranking second overall but in first place in terms of 5-year prevalence in much of Sub-Saharan Africa and Southern Asia, with prostate cancer the most prevalent cancer in North America, Oceania and Northern and Western Europe. Thus, one of breast, cervix or prostate cancer ranks as the most prevalent form of the disease in almost all countries of the world, indicating the need for improvements in care and service provision targeted at managing patients diagnosed with these neoplasms. In many settings, screening at the population level may have inflated the prevalence burden for those cancers where precursor lesions are not the target. Instances would include breast and prostate cancer, and one might argue that limiting the extent of PSA testing would be one of the means to reduce the prevalence of the latter neoplasm. We also noted a considerably higher number of female cancer survivors in many low- to medium-resource countries that reflects a sex-specific differential in the distribution of cancers and their prognosis, with a preponderance of neoplasms associated with higher fatality rates occurring in men, including tumours of the liver, oesophagus and stomach.

Cancer prevalence can be directly estimated from population-based cancer registry data by simply counting the number of registered cases still alive and present at a specified point in time. This approach requires not only the long-term registration of new cases but also the complete follow-up of the vital status of the catchment population over many years. Estimation-based methods have been proposed,17 and there are recent examples of country- and cancer-specific approaches to calculate prevalence from mathematical models in greater Europe2 and in country-specific estimations of partial and complete prevalence in France,3 the United Kingdom,4 Italy,5 Japan6 and South Korea.7 Prevalence estimates in the United States have been derived from state-specific registry data18 as well as using SEER data.19 There have also been a number of interesting developments that have focussed on examining trends and projecting forward partial prevalence estimates in France,3 Japan6 and Quebec.20 Approaches orientated toward specific cancers have been applied in estimating the number of breast cancer survivors by state in the United States21 and on accounting for breast cancer data limitations in The Netherlands,22 whereas colorectal cancer prevalence has been examined using data from the SEER program on partitioning the proportion of patients alive into four predefined phases of cancer care. Mariotto et al.23 have recently estimated multiple cancer prevalence using SEER data, calculating the number of patients with cancer alive and diagnosed with more than one primary cancer in the past. Colonna et al.24 had previously studied a representative sample of 5-year prevalent cases of colorectal cancer in France, estimating that about 20% of the prevalent cases experienced metastatic or local relapse after initial treatment.

We estimated limited duration cancer prevalence for the 184 countries of the world based on a function of respective estimates of cancer incidence and absolute survival. As we previously indicated,8 the fitting of advanced statistical modelling techniques would be inappropriately complex for such a systematic exercise that requires a set of prevalence estimates for 27 cancers by sex and country; our estimates are derived from incidence and survival data that themselves are based on estimates with varying levels of associated accuracy and completeness.13, 14 However, we consider that the methods used here yield reasonably satisfactory results for most sites on the basis of validation exercises described in Appendix Table 2. We have shown the 5-year prevalence estimates for 2008 are compatible with those observed in the Nordic countries (http://ancr.nu); however, the use of less recent survival estimates may partially explain our underestimation of Nordic prevalence relative to the observed counts for cancers associated with poorer survival.

We believe that our approach is straightforward enough to be reproducible for a given country and of sufficient accuracy to be of assistance in the planning of cancer health services at the local, regional and global level. The estimates of 5-year prevalence at 1 year, 2–3 and 4–5 years are directly applicable to the evaluation of initial treatment, clinical follow-up and point of cure, respectively, for the majority of cancers. The prevalence of all survivors estimated in terms of total prevalence would drastically overestimate the cost of assistance as the large majority of long-term survivors do not require medical attention. Complete prevalence may however be of specific relevance in estimating the number of breast cancer survivors who require some form of care, given the disease remains a chronic condition that affects prognosis decades after diagnosis, at both pre-menopausal25 and post-menopausal ages.26 One might thus consider estimates of complete prevalence for breast cancer extrapolated from limited duration prevalence portioned into longer lengths of duration (e.g., 10–20 years and >20 years), as has been done recently (for all cancer types) in the United Kingdom.4 An issue would be the lack of long-term follow-up time in most countries, leading to the requirement of model-based assumptions in predicting long-term survival proportions.

There are several weaknesses in our study that relate to the data sources available and to the specifics of the method applied. Prevalence is a rather complex fusion of cancer incidence, case fatality and other influences operating in affected individuals prior to death or ‘cure’. In building up the global estimates of prevalence from incidence and survival at the country level, we have relied on national incidence estimates obtained from GLOBOCAN 2008. The methods devised at IARC in compiling the data have evolved with time, although, the underlying principle remains as one of the reliance on the best available data on cancer incidence. Although national incidence data are available in 62 countries, there are limited or no data in 45 predominantly low-income countries worldwide.13 In terms of survival, we were able to augment available European, US, Australian and Japanese data with estimates provided by the SURVCAN project based on diagnostic and follow-up data obtained from 26 cancer registries in 14 Eastern and Western African, Caribbean, Central American and Asian countries.15 However, for the vast majority of low- and middle-income countries, there are no survival estimates available, and therefore, we created proxies based on SURVCAN, linking assumptions of survival according to a given country's level of human development.

A further limitation may be the restriction of limited duration prevalence to 5 years. We assume that for most cancer sites, cases surviving 5 years after diagnosis experience the same survival as the general population, and thus much of the workload due to medical acts occurs within these first 5 years following diagnosis. A logical addition would be to extend the limited duration estimates to better incorporate the point of cure. There is increasing evidence from cancer-specific estimates of 5-year conditional survival, from cured proportion models and visual inspection of the asymptotes of survival curves against time from diagnosis that the point of statistical cure for a number of cancers is between 5 and 10 years, and thus a logical extension to the method might simply be to additionally estimate 10-year prevalence. This could be easily achieved on extrapolating the fitted survival curves a further 5 years from diagnosis and assuming a suitable distribution for the survival times. The key assumption in the present estimations of constant incidence is likely to be less credible in the long term, and therefore, trend-based predictions of annual incidence would be required.

For prevalence measures to become more relevant in planning needs, they should take into account the natural history of the specific cancers under study.27, 28 Proportion-cured models and estimates of proportion treated have been used to estimate disability-adjusted life years from cancer-specific natural history models,27, 28 and on the basis of such derivations, one might explore the possibility to obtain the prevalent numbers of patients (i) initially diagnosed who are treated with curative intent, (ii) requiring post-diagnostic monitoring and treatment for recurrent disease and (iii) those in terminal phases requiring palliative care. Linked to this concept is the work in the United States focussed on attaching the associated financial costs in delivering care to patients with specific cancer types according to the phases of diagnosis, rehabilitation and palliation, with financial costs shown to be highest during initial treatment and end-of-life care.29, 30

Given increasing future needs, the incorporation of likely scenarios of future incidence and survival trends to predict limited duration prevalence estimates over the next two decades would be a useful addition. A number of simple methodologies can be used in calculating future prevalence, including assumptions that the prevalence:incidence ratio will remain fixed or on assuming constant or trend-based incidence and/or survival in the future.30

Incidence, mortality and survival are the basic cancer indicators providing the necessary foundations to plan, monitor and evaluate cancer control activities, with population-based cancer registries the institutions central to their delivery. In service planning, indicators such as prevalence can aid the development of strategies for service provision by measuring both the absolute number and relative proportion of the population of individuals affected by cancer who will require some form of medical attention. Our global estimates of almost 29 million persons living with cancer in 2008 and diagnosed within the previous 5 years is thus a useful supplement to the indicators currently included in GLOBOCAN, and limited duration prevalence estimates are now available online for the above cancer types by sex for each of the 184 countries worldwide and by UN-defined world region (http://globocan.iarc.fr).

Acknowledgements

The authors thank the Japan Cancer Surveillance Research Group (Miyagi, Yamagata, Niigata, Fukui, Osaka and Nagasaki) for sending their data for use in the calculations of prevalence. J.-S. Ren was an IARC Fellow during the period 2009–2011 and funded by the IARC Fellowship programme during this period.

APPENDIX

Appendix Table 1, Appendix Table 2

Table Appendix Table 1. Sources of absolute survival by area/country
inline image
Table Appendix Table 2. Cancer- and sex-specific (global) comparisons of prevalence:incidence ratios [based on observed 5-year prevalence and incidence 2008 (NORDCAN) on January 1, 200812] and our estimates for 2008 (based on available national incidence and survival sources)
inline image

Ancillary