Incidence data
Incidence data derive from population-based cancer registries. These may cover entire national populations but more often cover smaller, subnational areas, and, particularly in developing countries, only major cities. The most important source of information on cancer incidence is the successive volumes of Cancer Incidence in Five Continents CI5 [1,2]. Incidence data are generally associated with some delay as they require time to be compiled and published, but recent information can often be found in routine reports from the registries themselves, commonly available via their websites. While the quality of information from most of the developing countries might not be of sufficient quality, this information is still of unique importance as it often remains the only relatively unbiaised source of information available on the profile of cancer.
Population-based cancer registries can also produce survival statistics by following up their vital status of cancer patients. Survival probabilities [3] can be used to estimate mortality from incidence in the absence of mortality data, and to estimate cancer prevalence.

1. Parkin, D.M., Whelan, S.L., Ferlay, J., and Storm, H. Cancer Incidence in Five Continents, Vol. I to VIII. IARC CancerBase No. 7, Lyon, 2005.
2. Curado. M. P., Edwards, B., Shin. H.R., Storm. H., Ferlay. J., Heanue. M. and Boyle. P., eds (2007) Cancer Incidence in Five Continents, Vol. IX. IARC Scientific Publications No. 160, Lyon, IARC.
3. Sankaranarayanan R. Swaminathan R, Brenner H, Chen K, Chia KS, Chen JG, Law SC, Ahn YO, Xiang YB, Yeole BB, Shin HR, Shanta V, Woo ZH, Martin N, Sumitsawan Y, Sriplung H, Barboza AO, Eser S, Nene BM, Suwanrungruang K, Jayalekshmi P, Dikshit R, Wabinga H, Esteban DB, Laudico A, Bhurgri Y, Bah E, Al-Hamdan N. Cancer survival in Africa, Asia, and Central America: a population-based study. Lancet Oncol. 2010 Feb;11(2):165-73.
4. World Health Organisation (WHO) Databank, Geneva, Switzerland. WHO Statistical Information System http://www.who.int/whosis
5. United Nations, Population division. World Population Prospects, the 2008 revision. (http://www.un.org/, last accessed on 08/11/2009).

The methods used to estimate the country specific burden of cancer are similar to those used in the GLOBOCAN 2002 study [1] and have been described in detail elsewhere [2]. In summary, the most recent disease rates available were applied to the corresponding population of the country in 2008. For GLOBOCAN 2008, the degree of delay in the available data was taken into account by computing predictions of the national incidence and mortality rates to the year 2008, wherever possible. Although historical trends will not always hold in the future, predictions based on relatively linear trend patterns have been shown empirically to be reasonably accurate, particularly in the short-term. Where the availability of annual data was minimal - commonly between 5 and 10 years - simple time-linear models were fitted to these data to predict incidence and mortality for 2008 [3]. Where data series spanning at least 15 years were available, predictions based on age-period-cohort modeling were utilised [4].
Sex- and cancer-specific predictions of the national incidence and mortality rates were performed when at least 50 cancer cases or cancer deaths (all ages) were recorded per year for short-term predictions, and when at least 100 cancer cases or deaths (all ages) were recorded per 5-year period for NORDPRED [4]. Otherwise, the rates for 2008 were estimated as the annual average for the most recent 5-year period available.

The methods to estimate the sex- and age-specific incidence rates of cancer for a specific country are dependant on the availability and the accuracy of data, and fall into one of the following categories, in priority order:

1. National Incidence data (62 countries)
For 4 of these countries, reliable estimates of the national incidence in 2008 were available from local sources and these have been used (1A). When historical data and a sufficient numbers of recorded cases were available, incidence rates were projected to 2008 (1B). Otherwise, the incidence rates from the most recent period were applied to the 2008 population (1C).

2. Local incidence data and national mortality data (52 countries)
The method used for estimating the national incidence has been described in detail elsewhere [5, 6]. Estimates of the incidence to mortality ratios (I_R/M_R) were obtained from log-linear models of the numbers of incident cases based on the aggregation of local/regional registries offset by the numbers of deaths in the same registries, adjusted for sex and age. National incidence (I_N) was obtained on applying these fitted ratios to the corresponding national mortality estimates for 2008 (M_N ):

(1) I_N = M_N * I_R/M_R

Before aggregation, each registry dataset was weighted according to the square root of its population to take into account of the relative size of the population covered. Depending on the accuracy and on the availability of local data, one of two variants of the method was used:

  2A. A country-specific model was used for 11 countries for which several local cancer registries were in operation with the I_R/M_R ratios obtained from the most recent country-specific data (generally around year 2000). A correction [5] was applied to the estimation for breast and prostate cancers to take into account the possible screening-related increase in the incidence rates of these two cancers.

  2B. Regional models were used in the absence of country-specific national or local incidence data, or where they were considered to be of insufficient quality: the I_R/M_R ratios were obtained by the aggregation of cancer registry data in neighbouring countries in the same region. Several models were established, based upon the incidence and mortality data from cancer registries in Cancer Incidence in Five Continents Vol. IX or upon more recent data published on the Internet or provided by external collaborators. These comprised four models for the Americas: Temperate South America, Tropical South America, Black Caribbean and Central America and Latin Caribbean; three for Asia: Eastern Asia, South-Central Asia and South-Central and Western Asia; and two for Europe: Central Europe and the Balkans.

3. Local incidence data. No mortality data (23 countries)
National incidence estimates were derived from the data of one or more cancer registries covering a part of the country (city, state, province etc), wth the approach divided into two categories:
3A. A single cancer registry covering part of a country: the cancer registry data will be used as representative of the country profile.
3B. when more than one local source was available, national incidence rates were built up by some weighted average of the local rates.
In both instances, the cancer incidence was scaled using the ratio of the estimated all cancer mortality (see below, mortality data) to the WHO all cancer mortality envelope for 2008. This correction was performed only if the implied final incidence rates were plausible in relation to the neighbouring countries and the local registry was considered sufficiently complete.

4. Frequency data (13 countries)
For these countries, neither cancer incidence nor mortality statistics are available, or they are considered to lack sufficient accuracy. In these circumstances, we used a set of age- and sex-specific incidence rates for all cancers combined and partitioned these using data on the relative frequency of different cancers (by age and sex). We used data only from sources likely to provide a relatively unbiased picture of population-based relative frequency of different cancers (and corrected pathology-based series using appropriate biopsy proportions (percentage microscopically verified, %MV) for different cancers). For 8 African countries, three sets of age-sex specific incidence rates for all sites combined were used: Eastern, Northern and Western Africa. These rates were produced from the unweighted averages of the observed rates (by sex and age) in registries from Kenya, Malawi, Tanzania, Uganda and Zimbabwe; Algeria, Egypt, Libya and Tunisia; and Cote d’Ivoire, Guinea, Mali, Niger, Nigeria and The Gambia respectively. For the 5 remaining countries, the all cancers rates were computed as the population weighted averages of: observed rates in registries in neighbouring countries (Bangladesh and Cambodia), or estimated national rates in neighbouring countries (Irak, Yemen and Papua New Guinea).

5. No data (34 countries)
No useable data could be identified. The country-specific rates therefore represent simply those of neighbouring countries in the same region.

1. We estimated first the number of endemic (pre-AIDS) KS cases using the percentage frequency of the disease, by sex and age, based on data from Uganda, Kampala (1961-1980) and Nigeria, Ibadan (1971-1990). These percentages were applied to countries in Eastern and Western Africa respectively. For countries in Middle Africa, we applied a simple average of these frequencies.
2. We calculated the number of epidemic (AIDS-related) KS cases, both sexes, for the year 2008, using estimates of AIDS deaths by country in 2007 (source UNAIDS, http://www.unaids.org/), and an estimate of the ratio of deaths from AIDS to incident cases of KS. This ratio was based on observed KS rates in several countries (from the sentinel registries listed below, minus the endemic KS), and was specific by region (varying from 0.4% - 3%). This total number of AIDS-related KS was partitioned by sex and age using sex- and age-specific proportions in the sentinel registries:
   1. Eastern Africa: Kenya (Eldoret), Malawi (Blantyre), Uganda (Kampala), Zimbabwe (Harare) and Tanzania (Kilimanjaro).
   2. Middle Africa: Congo (Brazzaville).
   3. Southern Africa: Botswana, Namibia and Swaziland.
   4. Western Africa: Guinea (Conakry), Mali (Bamako) and Niger (Niamey).

National statistics are collated and made available by the WHO for countries with vital registration, but not all are of the same quality, and some corrections were made before they can be used for estimation purpose:

1. Where necessary, the overall number of deaths was corrected for under-reporting or incompleteness using percentages provided by the WHO.
2. Where the category “ill-defined cause of deaths” (ICD-9 codes 780–799 and ICD-10 codes R00–R99) exceeded 3% of the total causes of deaths, the excess (over 3% which is considered to be the limit as observed in good quality datasets) was partitioned, by sex and age into “cancer deaths” and other specific causes of death. The corrected “cancer deaths” category was partitioned into cancer-specific categories using proportions from the “non-corrected” data. No attempt was made to re-allocate ill-defined cancers (ICD-10 C76-C80, C97) into cancer-specific categories.
3. There are large variations in the accuracy of death certificates related to cancer of the uterus, with many deaths recorded as ‘uterus cancer, not otherwise specified’ (ICD-10 C55). By default, the number of cancer deaths coded as “uterus unspecified” was reallocated to either cervix (C53) or corpus (C54) uterine cancer according to age-specific proportions [7]. For the countries for which country-specific incidence and survival data were available , mortality for cervix uteri (C53) and corpus uteri (C54) cancers was estimated from incidence and 5-year relative survival probabilities.

1. National Mortality data (65 countries).
National mortality data was considered complete for every the cancer site. If no historical data was available, the most recent rates were simply applied to the corresponding 2008 population (1A), otherwise the mortality rates were projected to 2008 using either short-term or long-term prediction method (1B). When the data was incomplete (available only for major cancer sites), the residual group “other sites” was divided according to age- and sex-specific relative frequencies obtained from local mortality in the country or in the region (1C).

2. Sample mortality data (31 countries).
Mortality data are available for a representative sample of the population (or for specific strata within). The age and sex-specific all cancer mortality envelopes for the country in 2008 as provided by the WHO for 2008 were partitioned by site using the sample mortality data.

3. No vital statistic available (88 countries).
For the countries in developing regions without vital registration, mortality was estimated from incidence by a four step process:

  3.1. We applied sets of cancer-, sex- and age-specific incidence to mortality ratios provided by the national cancer registries in three Nordic countries (Denmark, Finland and Norway) for the period 1953-1957 [8] to the estimated national incidence for 2008 (as a reciprocal of formula (1)).

  3.2. We estimated the total number of deaths, by cancer (all ages, both sexes combined) using the estimated country-specific numbers of new cases for 2008 and country-specific survival data. For a given cancer site, mortality (M) is the product of incidence (I) and the probability of dying from the disease:

(2) M = I [k-S_j ]

Where S_j is the relative survival at year j of follow-up and k is a constant depending on j. When 5-year relative survival probabilities are used, the constant k tends to be very close to unity.
For GLOBOCAN 2008, we applied the country-specific survival data to that country and its immediate neighbors: India, Bangladesh, Nepal and Sri Lanka using Indian survival probabilities, and Thailand, Cambodia, Laos and Myanmar using Thai and/or Chinese survival data. For the other countries, we estimated country-specific survival, by cancer, using national macroeconomic data. A study conducted [9] in Europe in elderly patients demonstrated that cancer survival was reasonably correlated with level of Gross Domestic Product (GDP). Based on this assumption we estimated the relationship between cancer-specific 5-year relative survival (both sexes, and all ages) and country-specific GDP per capita [10]. Models predicting site-specific 5-year relative survival from per capita GDP were established using historical survival probabilities from Denmark and Finland, plus 5-year relative survival reported by cancer registries in rural and urban India, and Africa (The Gambia, Uganda (Kampala), and Zimbabwe (Harare, black population) [11, 12]). For each cancer site, we established a log-linear relation between 5-year survival and GDP omitting the data points with high GDP (>10000) from Denmark and Finland for cervical cancer because of the establishment of screening programmes in the recent periods.

  3.3 For each cancer, the sex- and age-specific mortality obtained in 3.1 was scaled using the ratios of the all-ages, both sexes, all cancers combined mortality obtained in 3.2 to that computed in 3.1, in order to take into account differences in survival.

  3.4 Finally, the cancer mortality and the corresponding incidence were scaled to the 2008 all cancer mortality envelope provided by the WHO (see incidence data, method 3) when possible, to maximize comparability between both estimates.

Estimate of mortality from Kaposi sarcoma (KS) in sub-Saharan Africa

Mortality from Kaposi sarcoma in Africa was estimated using incidence and 5-year relative survival.

1. For Uganda and Zimbabwe, we used the KS survival from Kampala and Harare (Black population), respectively.
2. For all others countries, we estimated mortality using survival based on a pooled average of 5-year relative survival probabilities from Uganda and Zimbabwe (both sexes combined, by age).

The methods to estimate the sex age-specific prevalent cases by cancer for a country have been described in detail [13]. Partial prevalence (1-,3- and 5-year prevalent cases) were obtained by combining the estimated annual number of new cases and the corresponding probability of survival by time:
n = n -years prevalent cases of age j years =  where n is the number of incident years cumulated in the partial prevalence (estimates for n equal 1, 3 or 5 have been compiled), is the annual number of new cases of age (j-i+0.5 ) when diagnosed and is the proportion of cases of age (j-i+0.5) surviving (i-0.5 ) years after diagnosis.

1. Ferlay J., Bray F., Pisani P. and Parkin D.M. GLOBOCAN 2002: Cancer Incidence, Mortality and Prevalence Worldwide IARC CancerBase No. 5. version 2.0, IARCPress, Lyon, 2004.
2. Ferlay J, Shin HR, Bray F, Forman D, Mathers C, Parkin DM. Estimates of worldwide burden of cancer in 2008: GLOBOCAN 2008. Int J Cancer. 2010; 127(12):2893-2917.
3. Dyba T, Hakulinen T. Comparison of different approaches to incidence prediction based on simple interpolation techniques. Stat Med. 2000 Jul 15;19(13):1741-52
4. NORDPRED. A software for predicting trends in cancer incidence (http://www.kreftregisteret.no/en/Research/Projects/, last accessed on 17/12/2008).
5. Ferlay J, Parkin DM, Steliarova-Foucher E. Estimates of cancer incidence and mortality in Europe in 2008. European Journal of Cancer 2010;46:765-781.
6. Black RJ, Bray F, Ferlay J, Parkin DM. Cancer incidence and mortality in the European Union: cancer registry data and estimates of national incidence for 1990. Eur J Cancer 1997;33:1075–107.
7. Loos AH, Bray F, McCarron P, Weiderpass E, Hakama M, Parkin D.M. Sheep and goats: separating cervix and corpus uteri from imprecisely coded uterine cancer deaths, for studies of geographical and temporal variations in mortality. Eur J Cancer 2004;40: 2794-803.
8. Gerda Engholm, Jacques Ferlay, Niels Christensen, Freddie Bray, Marianne L. Gjerstorff, Åsa Klint, Jóanis E. Køtlum, Elínborg Ólafsdóttir, Eero Pukkala and Hans H. Storm (2010).
   NORDCAN: Cancer Incidence, Mortality, Prevalence and Prediction in the Nordic Countries, Version 3.7. Association of the Nordic Cancer Registries. Danish Cancer Society (http://www.ancr.nu).
9. Alberto Quaglia , Marina Vercelli, Roberto Lillini, Eugenio Mugnoc,Jan Willem Coebergh, Mike Quinn, Carmen Martinez-Garcia,Riccardo Capocaccia, Andrea Micheli, on behalf of the ELDCARE Working Group. Socio-economic factors and health care system characteristics related to cancer survival in the elderly. A population-based analysis in 16 European countries (ELDCARE project). Critical Reviews in Oncology/Hematology 54 (2005) 117-128
10. A. Maddison web page (http://www.ggdc.net/maddison/, accessed 11-18-2009)
11. Sankaranarayanan R. Swaminathan R, Brenner H, Chen K, Chia KS, Chen JG, Law SC, Ahn YO, Xiang YB, Yeole BB, Shin HR, Shanta V, Woo ZH, Martin N, Sumitsawan Y, Sriplung H, Barboza AO, Eser S, Nene BM, Suwanrungruang K, Jayalekshmi P, Dikshit R, Wabinga H, Esteban DB, Laudico A, Bhurgri Y, Bah E, Al-Hamdan N. Cancer survival in Africa, Asia, and Central America: a population-based study. Lancet Oncol. 2010 Feb;11(2):165-73.
12. Gondos, A, Brenner, H., Chokunonga, E., Borok, M.Z., Chirenje, Z.M., Nyakabau, A.M., Parkin D.M. and Sankila, R. Cancer survival in a southern African urban population – Harare, Zimbabwe. Int J Cancer, 112, 860-864 (2004).
13. Pisani, P., Bray, F., Parkin, D.M. Estimates of the worldwide prevalence of cancer for twenty-five sites in the adult population. Int. J. Cancer: 97,72-81 (2002).
14. Bray, F., Ren, J.S., Masuyer, E., Ferlay, J. Estimates of global cancer prevalence in 2008 for 27 sites in the adult population, submitted.