Global surveillance of cancer survival 1995–2009: analysis of individual data for 25 676 887 patients from 279 population-based registries in 67 countries (CONCORD-2)

Background Worldwide data for cancer survival are scarce. We aimed to initiate worldwide surveillance of cancer survival by central analysis of population-based registry data, as a metric of the eﬀ ectiveness of health systems, and to inform global policy on cancer control. Methods Individual tumour records were submitted by 279 population-based cancer registries in 67 countries for 25·7 million adults (age 15–99 years) and 75 000 children (age 0–14 years) diagnosed with cancer during 1995–2009 and followed up to Dec 31, 2009, or later. We looked at cancers of the stomach, colon, rectum, liver, lung, breast (women), cervix, ovary, and prostate in adults, and adult and childhood leukaemia. Standardised quality control procedures were applied; errors were corrected by the registry concerned. We estimated 5-year net survival, adjusted for background mortality in every country or region by age (single year), sex, and calendar year, and by race or ethnic origin in some countries. Estimates were age-standardised with the International Cancer Survival Standard weights. Findings 5-year survival from colon, rectal, and breast cancers has increased steadily in most developed countries. For patients diagnosed during 2005–09, survival for colon and rectal cancer reached 60% or more in 22 countries around the world; for breast cancer, 5-year survival rose to 85% or higher in 17 countries worldwide. Liver and lung cancer remain lethal in all nations: for both cancers, 5-year survival


Introduction
The global burden of cancer is growing, particularly in countries of low and middle income. The need to implement eff ective strategies of primary prevention is urgent. 1,2 Prevention is crucial but long term. If WHO's global target of a 25% reduction in deaths from cancer and other non-communicable diseases in people aged 30-69 years is to be achieved by 2025 (referred to as 25 × 25), 3 we will need not only more eff ective prevention (to reduce incidence) but also more eff ective health systems (to improve survival). 4 In the fi rst international comparison of cancer survival, a transatlantic study of patients diagnosed during 1945-54, survival for 12 cancers in three US states was typically higher than in six European countries. 5 In 2008, a global comparison of population-based cancer survival (CONCORD) showed very wide variations in survival from cancers of the breast (women), colon, rectum, and prostate. 6 That analysis included 1·9 million adults (age 15-99 years) diagnosed with cancer during 1990-94 and followed up until 1999 from 31 countries (16 with 100% population coverage) on fi ve continents.
Three large international comparisons of cancer survival have been published since 2008. The European cancer registry study on survival (EUROCARE)-5 provided survival estimates for all cancers for patients diagnosed during 2000-07 in 29 countries in Europe. 7 In SurvCan (cancer survival in Africa, Asia, the Caribbean, and Central America), relative survival estimates were reported for patients diagnosed during 1990-2001 in 12 low-income and middle-income countries. 8 The International Cancer Benchmarking Partnership published survival estimates for four common cancers for patients diagnosed during 1995-2007 in six highincome countries. 9 These three studies diff er with respect to geographic and population coverage, calendar period, and analytical methods and do not enable worldwide comparison of cancer survival.
Surveillance of cancer survival is seen as important by national and international agencies, cancer patient advocacy groups, departments of health, politicians, and research agencies. Cancer survival research is being used to formulate cancer control strategies, 9 to prioritise cancer control measures, 10 and to assess both the eff ectiveness 11,12 and cost-eff ectiveness 13 of those strategies.
We designed CONCORD-2 to initiate long-term worldwide surveillance of cancer survival on the broadest possible basis. Our aim is to analyse progress toward the overarching goal in the Union for International Cancer Control's World Cancer Declaration 2013: "there will be major reductions in premature deaths from cancer and improvements in quality of life and cancer survival". 14

Cancer registries
We identifi ed population-based cancer registries that were operational in 2009 and had either published reports on survival or were known to follow up registered cancer patients to establish their vital status. Many registries had met quality criteria for inclusion in either the quinquennial compendium Cancer Incidence in Five Continents, 15,16 published by the International Association of Cancer Registries (IACR) and the International Agency for Research on Cancer (IARC), or similar compendia; other registries were established more recently.
We invited all these registries to contribute data for patients diagnosed during all or part of the 15-year period 1995-2009, including data on their vital status at least 5 years after diagnosis, or at Dec 31, 2009, or a later year. Of 395 registries invited, 306 (77%) agreed to participate: of these, 24 (8%) did not submit data, either because of resource constraints (n=4), legal constraints (1) or reversal of the original decision (3), or because they could not provide complete follow-up data (6) or did not respond to further communication (10). We excluded three registries because they provided data that did not adhere to the protocol and could not be rectifi ed, leaving 279 participating registries (71% of those invited).
Among the cancers suggested by participating registries, the ten we prioritised for study (referred to as index sites) accounted collectively for almost two-thirds of the estimated global cancer burden in 2008, both in developed and developing countries. 4 They comprised cancers of the stomach, colon, rectum, liver, lung, breast (women), cervix, ovary, and prostate in adults (age 15-99 years), and leukaemia in adults, and precursor-cell acute lymphoblastic leukaemia in children (age 0-14 years).

Ethics approval
We obtained approval for CONCORD-2 from the Ethics and Confi dentiality Committee of the UK's statutory National Information Governance Board (now the Health Research Authority; ECC 3-04(i)/2011) and the National Health Service (NHS) research ethics service (southeast; 11/LO/0331). We obtained separate statutory or ethics approval (or both) in more than 40 other jurisdictions to secure the release of data. Registries in all other jurisdictions obtained their own ethics approval locally.
We applied strict security constraints to the transmission of data fi les. We gave every registry a set of unique numeric codes for the name of every fi le; these codes have no meaning outside the CONCORD-2 study. All data fi elds were numeric or coded. We developed a fi le transmission utility deploying 256-bit advanced encryption security, with random, strong, one-time passwords that were generated automatically at the point of data transmission but sent separately, thus eliminating the need for email or telephone exchanges to confi rm passwords. We also provided free access to a similar commercial utility (HyperSend; Covisint, Detroit, MI, USA) that complies with US federal law on the secure transmission of sensitive health data.

Protocol
We fi nalised the protocol (in which we defi ned the data structure, fi le transmission procedures, and statistical analyses) after a 2-day meeting in Cork, Ireland, in September, 2012, with 90 members of the CONCORD Working Group from 48 countries (the protocol was revised by October, 2012). English poses a communication barrier in many countries; therefore, native speakers translated the protocol into Chinese (Mandarin), Portuguese, and Spanish, and other native speakers did back-translation to check the translation against the English original. We made the protocol available in all four languages. We held protocol workshops in Argentina (for Spanish-speaking South American researchers), Brazil, China, India, Japan, Puerto Rico, Russia, and the USA (for North America), which we followed up with conference calls and online seminars. We responded to telephone or email queries in Chinese, English, French, Italian, Portuguese, and Spanish.
We defi ned countries, states, and world regions by their UN names and codes (as of 2007). 17 Only Cuba and Puerto Rico provided data from the Caribbean and Central America so we grouped them with South America as America (Central and South). We wrote this Article and prepared the maps without prejudice to the status, boundaries, or name of any country, territory, or region. We have shortened some names for convenience (eg, Korea for South Korea), which does not have any political signifi cance. We created world maps and 27 regional maps in ArcGIS version 10, using digital boundaries (shapefi les) of countries and subnational regions from the Database of Global Administrative Areas (GADM 2.0). 18 We obtained national populations for 2009 from the UN Population Database 17 or national authorities (Canada, Portugal, and the UK) and subnational populations from the relevant registries.
We defi ned solid tumours by anatomical site (topography) and leukaemia by morphology (table 1). We coded topography and morphology according to the International Classifi cation of Diseases for Oncology (3rd edn; ICD-O-3). 19 For ovarian cancer, we included the fallopian tube, uterine ligaments, and adnexa, and the peritoneum and retroperitoneum, where high-grade serous ovarian carcinomas are often detected. We excluded Kaposi's sarcoma and solid tumours with lymphoma morphology.
The classifi cation of leukaemias and lymphomas has changed since the mid-1990s. To minimise diff erences in the range of leukaemia subtypes included in our analyses, we asked registries to provide data for all haemopoietic malignant diseases in adults and children, as defi ned by the ICD-O-3 morphology code range 9590-9989. In consultation with specialists in the cancer registrybased project on haematologic malignancies (HAEMACARE) group, 20 we selected subtypes of adult leukaemia from nine morphology groups, 21 excluding myelodysplastic and myeloproliferative neoplasms such as chronic myeloid leukaemia (appendix p 2). Precursorcell acute lymphoblastic leukaemia is the most common form of leukaemia in children; we included HAEMACARE group 15-a relatively homogeneous group comprising precursor-cell lymphoblastic lymphoma and precursor-cell lymphoblastic leukaemia (B-cell, T-cell, and not otherwise specifi ed), and we refer to these six entities as acute lymphoblastic leukaemia. 22 For survival analyses, we included only invasive primary malignant diseases (ICD-O-3 behaviour code 3). To facilitate quality control and comparisons of the intensity of early diagnostic and screening activity, however, we asked registries to submit data for all solid tumours at each index site, including those that were benign (behaviour code 0), of uncertain or borderline malignancy (1), or in situ (2).
We asked registries to submit full dates (day, month, year) for birth, diagnosis, and death or last known vital status, both for quality control and to enable comparable estimation of survival. 23 When the day of diagnosis or the day or month of birth or last known vital status were missing, we developed an algorithm to standardise the imputation of missing dates for all populations (details available on request). Participating registries completed a detailed question naire on their methods of operation, including data defi nitions, data collection procedures, coding of anatomical site, morphology and behaviour, the tracing of registered cancer patients to ascertain their vital status, and how tumour records are linked with data on vital status.

Quality control
The quality and completeness of cancer registration data can aff ect both incidence and survival estimates and, thus, the reliability of international comparisons. 26 We developed a suite of quality control programs, 27 extending the checks used in the fi rst CONCORD study, 6 cross-checked with those used in the EUROCARE study, 28 IARC/IACR tools for cancer registries, 29 and WHO's classifi cation of tumours. 22,[30][31][32] We applied these checks systematically in three phases and sent registries a detailed report on how to revise and resubmit their data, if needed, after every phase.
First, we sent registries a protocol adherence report that showed, for every cancer, the proportion of tumour records that were coded in compliance with the protocol. Second, we checked the data in every tumour record for logical coherence against 20 sets of criteria, including eligibility (eg, age, tumour behaviour), defi nite errors (eg, sex-site errors and invalid dates or date sequence), and possible errors including a wide range of inconsistencies between age, tumour site, and morphology. 27 We sent registries exclusion reports that showed, for every index cancer and calendar period, the number of tumour records in each category of defi nite or possible error, the number of tumours registered from a death certifi cate only or detected at autopsy, and the number of patients whose data could be included in survival analyses. When we identifi ed errors in classifi cation, coding, or pathological assignment, we asked registries to correct and resubmit their data. Finally, we analysed: the proportion of tumour records with morphological verifi cation or non-specifi c morphology; distributions of the day and month of birth, diagnosis, and last known vital status; and proportions of patients who died within 30 days, were reported as lost to follow-up, or were censored within 5 years of diagnosis.

Follow-up for vital status
Cancer registries use various methods to ascertain the vital status (alive, dead, emigrated, lost to follow-up) of registered cancer patients. In countries with limited administrative infrastructure, so-called active follow-up can be used to establish vital status via direct contact with the patient, the family, or a local authority (eg, a village headman), or by home visit. Many registries in both high-income and low-income countries also seek information from the hospital or the treating clinician in hospital or primary care.
Most registries link their database with a regional or national index of deaths, using identifi ers such as name, sex, date of birth, and identity number. Tumour records that match to a death record are updated with the date of death. Many registries also use other offi cial databases (eg, hospital and primary care databases, social insurance, health insurance, drivers' licences, and electoral registers) to establish the date on which a patient was last known or believed to have been alive, to have migrated within the country, or to have emigrated to another country. Cancer registrations are updated with the vital status and the date of last known vital status. These methods are typically summarised as passive follow-up.
Some registries receive information on the vital status of all registered patients on an almost continuous basis, or at least every month or every 3 months. Other registries seek to trace the vital status of patients registered in a particular calendar year only, 1 year or even 5 years after the end of that year: this approach can increase the proportion of patients lost to follow-up. It also means that 5-year survival estimates for more recently diagnosed patients cannot be obtained, even with the period approach.
We asked all 279 participating registries how they ascertained the vital status of registered cancer patients. Of 243 registries that responded to the question, 147 (60%) stated that they used only passive follow-up, 92 (38%) that they used both passive and active follow-up, and four (2%) only active follow-up.

Statistical analysis
Most registries submitted data for patients diagnosed from 1995 to 2009, with follow-up to 2009 or later; some registries only began operation after 1995 or provided data for less than 15 years. We were able to estimate 5-year survival using the cohort approach for patients diagnosed in 1995-99 and 2000-04, because in most datasets, all patients had been followed up for at least 5 years. We used the period approach 33 to estimate 5-year survival for patients diagnosed during 2005-09, because 5 years of follow-up data were not available for all patients (appendix p 174).
We estimated net survival up to 5 years after diagnosis for both adults and children. Net survival represents the cumulative probability that the cancer patients would have survived a given time, say 5 years or more after diagnosis, in the hypothetical situation that the cancer was the only possible cause of death. Net survival can be interpreted as the proportion of cancer patients who survive up to that time, after eliminating other causes of death (background mortality). We used the recently developed Pohar Perme estimator 34 of net survival implemented with the program stns 35 in Stata version 13. 36 This estimator takes unbiased account of the fact that older patients are more likely than younger patients to die from causes other than cancer-ie, that the competing risks of death are higher for elderly cancer patients.
To control for the wide diff erences in background mortality between participating jurisdictions and over time, we constructed 6514 life tables of all-cause mortality in the general population of each country or the territory covered by each participating registry, by age (single year), sex, and calendar year of death, and by race or ethnic origin in Israel (Arab, Jewish), Malaysia (Chinese, Malay, Indian), New Zealand (Māori, non-Māori), and the USA (Black, White). The method of life table construction depended on whether we received raw data (numbers of deaths and populations) or mortality rates, and on whether the raw data or the mortality rates were by single year of age (so-called complete) or by 5-year or 10-year age group (abridged). We checked the life tables by examination of age-sex-mortality rates, life expectancy at birth (appendix p 175), the probability of death in the age bands 15-59 years, 60-84 years, and 85-99 years and, where necessary, the model residuals.
Of the 279 participating registries, 21 provided complete life tables that did not need interpolation or smoothing, for each calendar year. For 172 registries, we obtained raw data from either the registry, the relevant national statistical authority, or the Human Mortality Database. 37 We derived life tables for 1996 and 2010 if possible, each centred on three calendar years of data (eg, 1995-97, 2009-11) to increase the robustness of the rates. We modelled raw mortality rates with Poisson regression and fl exible functions to obtain smoothed complete life tables extended up to age 99 years. We then created life tables for every calendar year from 1997 to 2009 by linear interpolation between the 1996 and 2010 life tables. 38 Rather than extrapolate, we used the 1996 life table for 1995.
62 of 279 registries provided abridged mortality rates, or complete mortality rates that were not smoothed. We used the Ewbank relational model 39 with three or four parameters to interpolate (if abridged) and smooth the mortality rates for the registry territory against a high-quality smooth life table for a country with a similar pattern of mortality by age. We could not obtain reliable data on all-cause mortality for 24 registries. We took national life tables published by the UN Population Division 40 and interpolated and extended them to age 99 years with the Elandt-Johnson method. 41 For each country and registry, we present estimates of age-standardised net survival for each cancer at 5 years after diagnosis. We report cumulative survival probabilities National registries in smaller countries are shown in boxes at diff erent scales. 28 regional maps and a world map for childhood acute lymphoblastic leukaemia are in the appendix (pp . National coverage Regional coverage Regional territory (no data) No coverage 5000 10 000 km 0  42 For children, we estimated survival for the age groups 0-4 years, 5-9 years, and 10-14 years; we obtained age-standardised estimates by assigning equal weights to the three age-specifi c estimates. 43 We derived CIs for both unstandardised and age-standardised survival estimates assuming a normal distribution, truncated to the range 0-100. We derived SEs with the Greenwood method 44 to construct the CIs We did not estimate survival if fewer than ten patients were available for analysis. If between ten and 49 patients were available for analysis in a given calendar period (1995-99, 2000-04, 2005-09), we merged data for two consecutive periods. For less common cancers in the smallest populations, we sometimes needed to merge data for all three periods. When between ten and 49 patients in total were available, we only estimated survival for all ages combined. If 50 or more patients were available, we attempted survival estimation for each age group. If an age-specifi c estimate could not be obtained, we merged data for adjacent age groups and assigned the combined estimate to both age groups. If two or more age-specifi c estimates could not be obtained, we present only the unstandardised estimate for all ages combined.

Role of the funding sources
The funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all data in the study and had fi nal responsibility for the decision to submit for publication.

Results
279 cancer registries from 67 countries provided data for this study (fi gure 1; appendix pp . Nine African countries took part (ten registries), eight countries were in Central and South America (27 registries), Canada and the USA comprised North America (57 registries), 16 countries were in Asia (50 registries), 30 European countries participated (128 registries), and New Zealand and Australia represented Oceania (seven registries). For countries with less than 100% coverage of the population, the country name is used for brevity in the text (eg, Libya, the USA), but a more accurate term is used in the tables (eg, Libya [Benghazi], US registries). Some registries provided data for only part of their territory.
We examined records for 28 685 445 patients diagnosed with cancer of the stomach, colon, rectum, liver, lung, breast (women), cervix, ovary, and prostate in adults (age 15-99 years), leukaemia in adults, and precursor-cell acute lymphoblastic leukaemia in children (age 0-14 years) during the period 1995-2009 (table 2). Of these, 1 682 081 (5·9%) records were for an in situ cancer, mostly of the cervix, breast, colon, or prostate. The proportions of in situ cancer are not comparable directly because some registries do not record in situ cancer, others did not submit data for index sites in which in situ malignant disease is common, and screening programmes in which in situ cancers are frequently detected were introduced in some countries during 1995-2009. The variation between continents is still of interest: for example, a little over 1% of cervical cancers in African registries were in situ, compared with 20% in Central and South American registries and 81% in Oceania. For breast cancer in situ, the variation was from 0·1% in African registries to 16% in North American registries and about 4-5% in other regions of the world (appendix pp . Patients with in situ cancer were not included in survival analyses. We excluded a further 360 773 (1·3%) patients either because their year of birth, month or year of diagnosis, or year of last vital status were unknown, or because the tumour was not primary invasive malignant disease (behaviour code 3) or the morphology was that of Kaposi's sarcoma or lymphoma in a solid organ, or for other reasons (table 2). The proportion of patients with an unknown date of last vital status ranged from 0% to 40% or more for some cancers in some African registries. Proportions are presented in the appendix (pp 3-63) for each registry, for all cancers combined, and for each cancer separately.
Of 26 642 591 patients eligible for inclusion in the survival analyses, 905 841 (3·4%) were excluded because their cancer was registered from a death certifi cate only or discovered at autopsy (table 2), and 59 863 (0·2%) were excluded for other reasons, including defi nite errors (eg, unknown vital status or sex, sex-site error, or invalid dates or sequence of dates) or possible errors (eg, apparent inconsistencies between age, cancer site, and morphology) for which the record was not later confi rmed as correct by the relevant registry.
Of 25 676 887 patients available for survival analyses (96·4% of those eligible), pathological evidence of malignant disease (histological, cytological, or haematological fi ndings) was available for 23 338 015 patients for all cancers combined (91·1%; table 2), ranging from 83·1% in Asian registries, 85·5% in African registries, and 87·4% in Central and South American registries to 90-95% in Europe, Oceania, and North America. The range of pathological evidence at a national level was very wide, from 15% in The Gambia, 36% in Mongolia, and 66% in Chinese registries, up to 99% or more in Belgium, Mauritius, and Sweden. For 938 703 (3·7%) patients, morphological features were poorly specifi ed (eg, malignant neoplasm or tumour, ICD-O-3 codes 8000-8005): this proportion also varied widely, from around 1% in North American registries to 17% for all African registries combined and as high as 59% in The Gambia. Data for every registry are shown in the appendix (pp 3-63). NA=not available. *100% coverage of the national population. †100% coverage of the national population for childhood leukaemia only. ‡South Korea. ¶In situ malignant disease (ICD-O-3 behaviour code 2): some registries do not register in situ cancers, other registries did not submit them. Other: records with incomplete data; or tumours that are benign (behaviour code 0), of uncertain behaviour (1), metastatic from another organ (6), or unknown if primary or metastatic (9); or patients falling outside the age range 0-14 years (children) or 15-99 years (adults); or other conditions. ||DCO=tumours registered from a death certifi cate only or detected solely at autopsy. Other: vital status or sex unknown; or invalid sequence of dates; or inconsistency of sex-site, site-morphology, age-site, age-morphology, or age-site-morphology. † † MV=microscopically verifi ed. Non-specifi c morphology (solid tumours only): ICD-O-3 morphology code in the range 8000-8005. Censored: patients diagnosed during 1995-2004, with last known vital status "alive" but less than 5 years of follow-up. in most populations, but in some countries it changed substantially between the earliest and latest years for which data were available, from a decline of 6-9 years in South Africa and Lesotho (attributable largely to HIV/AIDS), 45 to an increase of 6 years or more in Estonia, Latvia (for males), and South Korea, and in some regions of Brazil (males), China, and Germany (males; data not shown). Whenever possible, fi ndings are presented for patients diagnosed during 1995-99, 2000-04, and 2005-09, by continent, country, and registry (fi gures 2 to 4; appendix pp 3-173). When data were available for more than one registry in a given country, survival estimates were derived by pooling data for that country, excluding data from registries for which estimates were judged less reliable (fi gures 2 and 3). Survival estimates were fl agged as less reliable if a higher than usual proportion of patients was excluded from analyses because their cancer was registered from the death certifi cate only, or had an unknown date for last vital status, or because not all deaths were ascertained. Less reliable estimates are not always outliers in the global distribution, but when they are, they have been omitted from this discussion. Less reliable estimates are also excluded from the distribution of survival among registries in each continent (fi gure 4  (18-19%). Survival was less than 10% in Gibraltar and Libya, but those two estimates are based on fewer than 100 cases (table 4; appendix pp [64][65][66][67][68][69][70][71][72][73][74][75][76][77][78][79][80]. In most countries, survival from stomach cancer remained in the narrow range of 25-30% from 1995-99 to 2005-09. Very large increases were seen in South Korea (from 33% to 58%) and China (from 15% to 31%), but survival rose by less than 10% in some countries on all continents (appendix p 153). Survival from stomach cancer fell by 6-17% in Brazil, Cyprus, Malaysia, Thailand, and Turkey, declines that were not seen for most other cancers in these registries. We could not assess survival trends for stomach cancer in African countries. The range of 5-year survival estimates for stomach cancer in 2005-09 varied widely between registries in Africa, Asia, and Central and South America (appendix p 164).
Data for colon cancer are available for 3 613 067 patients (   . Survival estimates for every country are ranked from highest to lowest within every continent; for ease of reference, the ranking for 2005-09 is used for 1995-99 and 2000-04. Error bars represent 95% CIs. Grey bars represent African countries; red bars represent America (Central and South); light green bars represent America (North); purple bars represent Asian countries; blue bars represent European countries; and dark green bars represent Oceania. *100% coverage of the national population. †National estimate not age-standardised. §National estimate fl agged as less reliable because the only estimate or estimates available are from a registry or registries in this category.
Data for prostate cancer are available for 4 999 267 men (table 3). 189 registries in 48 countries contributed data for 1995-99, 241 registries in 57 countries provided data for 2000-04, and 240 registries in 60 countries had data for 2005-09 (appendix pp [64][65][66][67][68][69][70][71][72][73][74][75][76][77][78][79][80]. Among the 61 countries that provided data on prostate cancer, the range in age-standardised 5-year net survival is very wide, from less than 40% to greater than 95%. For men diagnosed during 2005-09, survival was 90% or higher in Austria, Belgium, Brazil, Canada, Cyprus, Ecuador, Finland, France, Germany, Israel, Italy, Lithuania, Puerto Rico, and the USA (table 4; appendix p 149). In the USA, where widespread prostate-specifi c antigen (PSA) testing was introduced around 1990, 5-year survival has been higher than 90% since 1995-99. Prostate cancer survival was 80-89% in 19 countries in Central and South America, Asia, Europe, and Oceania. In 18 other countries, survival ranged widely (50-79%), but in Libya and Mongolia it was 40-41%. Striking and persistent Data are net survival estimates (%) with 95% CI. Italics denote survival estimates that are not age-standardised. When too few patients were available for analysis in any calendar period, data were merged and the survival estimates are underlined. Follow-up was shorter than 5 years for six registries: Libya (Benghazi); The Gambia; Argentina (Mendoza); China (Lianyungang); Indonesia (Jakarta); and Colombia (Manizales: stomach, colon, breast, cervix, and prostate). ALL=acute lymphoblastic leukaemia. *100% coverage of the national population. †100% coverage of the national population for childhood leukaemia only. ‡South Korea. §Survival estimate considered less reliable.  . With a few exceptions, survival seems to be low in east Asia (eg,from 19% in Japan to 23% in South Korea and Taiwan), high in west Asia (eg, from 33% in Turkey to 53% in Qatar), with a mixed picture in other Asian countries (eg, from 7% in Jordan to 40% in Indonesia). Survival estimates for adult leukaemia from Jordan, India, and Saudi Arabia might be less reliable for international comparison, but the overall pattern of leukaemia survival in Asia is still informative. Survival increases of 10-16% for adult leukaemia were seen in China, Denmark, Germany, Iceland, Latvia, Sweden, and New Zealand. Smaller rises of 5-9% were noted in North America, Israel, Japan, South Korea, andten European countries. In Malta, 5-year survival fell from 39% in 1995-99 (based on 142 adults) to 19% for 2005-09 (128 adults;appendix p 161). This pattern is surprising, because data quality is very high (appendix pp [54][55][56][57][58] and survival trends for all solid tumours seem to be normal. Smaller declines were seen in several countries, such as Slovakia (from 41% to 37%) and Slovenia (from 44% to 38%).
Data for acute lymphoblastic leukaemia in children are available for 74 343 patients (table 3). 173 registries in 42 countries contributed data for 1995-99, 215 registries in 50 countries provided data for 2000-04, and 213 registries in 53 countries provided data for 2005-09. In Romania (Cluj), data were only available for eight children and survival was not estimated. Of 53 countries, 32 provided data with 100% national population coverage. The geographic range in survival for acute lymphoblastic leukaemia in children was very wide. For patients diagnosed during 2005-09, age-standardised 5-year net survival was 90% or higher in Austria, Belgium, Canada, Germany, and Norway and 80-89% in 21 countries on various continents (table 4; appendix p 151). In many countries, however, 5-year net survival is still lower than 60%, even after adjustment for the very high background mortality in childhood. Survival was less than 50% in Indonesia, Mongolia, and Lesotho, although these estimates are based on very small numbers. The range of survival estimates for childhood acute lymphoblastic leukaemia in Central and South America (16 registries) and Asia (23 registries) is much lower than the range in North America (48 registries), Europe (83 registries), and Oceania (seven registries; appendix p 173). 5-year survival for childhood acute lymphoblastic leukaemia rose by 10% or more between 1995-99 and2005-09 in Belarus, Belgium, Bulgaria, China, Colombia, Lithuania, Norway, Portugal, South Korea, Spain, Taiwan, and 83-86% in 1995-2004 to 76% for 2005-09). Survival trends could not be assessed in Africa.

Discussion
With CONCORD-2, we have initiated worldwide surveillance of trends in cancer survival. In the fi rst CONCORD study, 6 comparable estimates of cancer survival worldwide were provided: the study included 1·9 million patients diagnosed with breast, colorectal, or prostate cancer during 1990-94 and followed up to 1999 in 31 countries (panel). CONCORD-2 extends coverage to 25·7 million patients diagnosed with an invasive primary cancer during the 15-year period 1995-2009 in 67 countries. The ten index cancers represent about two-thirds of the overall cancer burden in both low-income and highincome countries. 4 Individual patient data provided by 279 population-based cancer registries were prepared with standardised quality-control procedures and subjected to centralised analysis with the latest statistical methods.
The fi ndings do not cover all countries, but they provide at least some population-based cancer survival estimates for 67 countries (26 of low or middle income) that are home to two-thirds of the world's population, including national data for 40 countries. The estimates are derived from analysis of raw data on the survival of individual cancer patients up to 5 years after diagnosis. Until now, for comparison of global or continental survival, researchers generally needed to interpret scattered reports produced with diverse cancer defi ni tions, quality-control criteria, and survival estimators, for diff erent calendar periods, and age-standardised to diff erent sets of weights. 46 More speculative comparisons have been based on modelling of mortality-incidence ratios, sometimes with data from neighbouring regions or countries, 47 with all the attendant assumptions. 48 Even after adjustment for the wide international variation in levels of mortality from other causes, and with due allowance for variation in quality of data, the global range in 5-year survival from ten cancers in adults and acute lymphoblastic leukaemia in children is very wide. For most cancers, survival in Africa, Asia, and Central and South America is lower, and the range in survival much wider, than in Europe, North America, and Oceania. The wider range is only partly attributable to the fact that not all cancer registries could provide data covering the 15 years from 1995 to 2009; for example, many of the Chinese registries contributed data for 2000-04 but not 2005-09. In North America and Oceania, population coverage was higher than 80% and the same registries generally provided data for the entire period 1995-2009 (fi gure 4; appendix pp 163-73): survival for most cancers was high on a global scale, with a fairly narrow range in estimates between registries.
5-year net survival from stomach cancer is generally in the range 25-30%, but it is very high (50-60%) in Japan, South Korea, and, to a lesser extent, Taiwan. High survival from stomach cancer in Japan, 49 South Korea,50 and Taiwan 51 is well known, and is likely to be attributable to intensive diagnostic activity, early stage at diagnosis, and radical surgery. Survival varies according to sub-site, morphological type, and stage. Types of cancer with better prognosis might also be more common in Japan and South Korea, but the striking worldwide diff erences in survival suggest important lessons could be learnt from these countries about diagnosis and treatment.
5-year survival has risen for colon and rectal cancers in most developed countries and regions, including North America, Europe, Oceania, and parts of east Asia (South Korea and urban areas in China); increases in breast cancer survival have also been noted in these regions and in parts of Central and South America. These trends are likely to be attributable to earlier diagnosis, reduction in postoperative mortality, 52 and more eff ective treatment. 53,54 For rectal cancer, preoperative radiotherapy and total mesorectal excision reduce local recurrence and extend survival, [55][56][57] which could account for improvements noted in Canada, Finland, the Netherlands, Norway, Sweden, and the USA, where survival was already high (55-60%) for patients diagnosed in 1995-99 and rose further for those diagnosed during 2005-09 (62-65%). These trends accord with those reported from the Netherlands, 58 Scotland, the Nordic countries, 59 and elsewhere in Europe. 60 Liver and lung cancer remain lethal in both developing and developed countries, with 5-year survival generally lower than 20%, indicating that most patients are still diagnosed when they are inoperable. Primary prevention aimed at reducing tobacco and alcohol consumption, and prevention of chronic hepatitis, will be especially important for these cancers. The very low survival estimate for liver cancer in The Gambia (5%) is based on a sample of only 85 patients diagnosed during 1995-97 who were followed up for less than 5 years, to the end of 1998; it is not age-standardised, but it is unlikely to be far wrong: patients in The Gambia tend to present with very advanced disease and cirrhosis and are not amenable to surgery. 61 Overall completeness of registration is low, but the incidence of liver cancer is comparable with that of other west African populations. 62 Data from the national cancer registry for The Gambia, set up in 1986 to support the IARC's Gambia Hepatitis Intervention Study, 63 have been analysed previously, 64 but more recent data were unavailable, so we cleaned and analysed them here alongside all other datasets, with permission from IARC.
The global range in 5-year survival from cervical cancer is very wide, from less than 40% to more than 70%. The overall decline in survival from 66% to 61% in France between 1995-99 and 2000-04 was seen in all nine registries (appendix p 105). The decrease might be attributable to removal of less aggressive tumours by more intensive cervical screening for preinvasive lesions. 65,66 Survival from cervical cancer in the Nordic countries was stable or rose slightly over the same period. 67 By comparison, lower survival in low-income and middle-income countries is striking, since invasive cervical cancer is potentially curable with early detection by screening and appropriate surgery. 68 5-year survival from ovarian cancer is generally in the range 30-40% in most parts of the world, but the overall range is much wider. Diversity in international survival might be attributable partly to variations in the proportion of tumours classifi ed as type I (typically early-stage and slow-growing) and type II (typically late-stage and aggressive). 69 Diff erences in stage at diagnosis and treatment are also likely to be important. 70 Diff erential

Systematic review
In the fi rst global comparison of population-based cancer survival (CONCORD), 6 wide variations in survival from cancers of the breast (women), colon, rectum, and prostate were reported among 1·9 million adults diagnosed during 1990-94 and followed up to 1999 in 31 countries (16 countries had national coverage). More recent studies have diff ered with respect to geographic and population coverage, calendar period, and analytical methods, and they do not enable worldwide comparison of survival trends. [7][8][9] With CONCORD-2, we have extended coverage to 25·7 million cancer patients diagnosed during the 15-year period 1995-2009 in one of 67 countries (26 of low or middle income), of which 40 countries had national coverage.

Interpretation
The ten index cancers we selected for analysis represent two-thirds of the overall cancer burden in both low-income and high-income countries. 5-year survival from colon, rectal, and breast cancers has increased in most developed countries. Liver and lung cancer remain lethal in both developing and developed countries. Striking increases in prostate cancer survival have occurred in many countries, but trends vary widely. The range in cervical and ovarian cancer survival is very wide, but improvements have been slight. In east Asia, stomach cancer survival is very high, suggesting lessons could be learnt, whereas survival for adult and childhood leukaemia is remarkably low. The global range in survival from precursor-cell acute lymphoblastic leukaemia in children is very wide, suggesting major defi ciencies in the management of what is now a largely curable disease. The fi ndings of our study can be used to assess the extent to which investment in health-care systems is improving their eff ectiveness.
classifi cation of borderline and invasive tumours might also contribute. Overall, however, worldwide survival trends show very little improvement between 1995-99 and 2005-09 (appendix p 158). This fi nding accords with the absence of improvement reported from many developed countries. 7,9 Striking increases in 5-year survival from prostate cancer have occurred in many countries, but global trends varied widely. Examples include three northern European countries, all with nationwide cancer registration. 5-year survival in Lithuania jumped from 52% for men diagnosed during 1995-99 to 92% for those diagnosed during 2005-09. The rise in Latvia was from 52% to 74%: access to health care in these countries has improved, and opportunistic PSA screening began in 2000. 71 In Denmark, survival rose from 46% to 77% over the same period, having been stable at 40% throughout the period 1982-94, 72 during which time survival increased rapidly in the other four Nordic countries. 73 The Danish Urology Society advised against PSA testing in asymptomatic men in the early 1990s, 74 but this advice is now followed less widely. By contrast, survival in North America and Oceania was already very high in the late 1990s, and increases since then have been much smaller. In Africa, we were unable to assess a trend.
Survival from both adult and childhood leukaemia in east Asia is surprisingly low. The low survival for adult leukaemia in Japan, South Korea, and Taiwan is especially surprising, because survival from solid tumours is generally high. Could ethnic or genetic factors play a part? This possibility has been suggested in a recent comparison of survival from chronic lymphocytic leukaemia between Taiwan and the USA. 75 Leukaemia survival is also low in China, but haematological malignant diseases have received low priority in cancer control there, with limited access to health insurance and chemotherapy, 76 and medical resources in rural areas are poor. 77 The global range in 5-year survival from acute lymphoblastic leukaemia in children is very wide, from less than 60% in several countries to 90% or higher in Austria, Belgium, Canada, Germany, and Norway. This fi nding confi rms that major defi ciencies are present in the management of what is now a largely curable disease. 78 Failure to start or complete treatment, usually for fi nancial reasons, is an important contributor to the survival defi cit in developing countries. 79 Standardised quality controls were applied systematically to all datasets. Detailed discussions were held with every registry to identify and correct any errors or artifacts in the data. Many registries resubmitted their data after correction, which greatly improved data quality and comparability. The overall proportion of eligible tumours excluded from analysis was low (3·6%), but it was much higher for some registries and varied widely between cancers. For some populations, mostly in low-income and middle-income countries, these exclusions will have biased survival estimates upwards. Thus, the proportion of cancer registrations from a death certifi cate only was typically higher in countries where survival is low. This leads to exclusion from analysis of a group of patients who tend to have low survival, 80 leading to overestimation of the level of survival in that population. This bias would tend to reduce international diff erences.
Various indications suggest that the data submitted by some registries were not exhaustive, either because there were fewer cancer patients than expected or because the full range of haemopoietic malignant diseases was not represented in some of the leukaemia datasets. The smaller number of cancer registrations in Poland for 1995-99 refl ects a national strike of doctors in 1997, but we have little reason to suppose this type of incompleteness would bias survival estimates.
Pathological confi rmation of diagnosis was available for more than 90% of cancers included in the analyses (98·5% for childhood acute lymphoblastic leukaemia), and less than 4% of malignant diseases were assigned to a non-specifi c morphology code. Nevertheless, considerable variation was noted, and pathological evidence was much less complete for some populations in low-income and middle-income countries (table 2; appendix pp . Several registries reported high proportions of intestinaltype adenocarcinoma in the colon and rectum: this morphological type was originally described (in 1965) for carcinoma of the stomach 81 and is included in ICD-O-3 (M8144). A similar issue arose with cholangiocarcinoma (M8160) coded as arising in the liver (ICD-O-3 site code C22.0) rather than the intrahepatic bile duct (C22.1). If we were told that pathologists frequently use these terms for malignant disease of the large bowel or liver, respectively, we included the patients in our analyses.
The distribution of cancers within an organ by anatomic sub-site or morphological type can diff er between populations, so any diff erences in survival by sub-site or morphological features could aff ect comparisons of overall survival. We will address the eff ect on survival of these diff erences in biology with more detailed analyses, particularly for cancers of the stomach, lung, and ovary. Leukaemia comprises a broad and heterogeneous group of diseases. We excluded chronic myeloid leukaemia; survival for other major groups will be investigated in more detail.
Premalignant and small malignant lesions can be detected more frequently in countries with mass screening programmes or intensive early diagnostic activity, particularly for cancers of the breast, cervix, colon, rectum, and prostate. Diff erences in tumour stage at diagnosis can contribute to international variations in overall survival between low-income countries. 8 Wide diff erences in tumour stage at diagnosis and stage-specifi c survival have also been recorded among high-income countries. 59,70,[82][83][84] High-resolution studies of tumour stage at diagnosis, treatment, and adherence to guidelines have helped account for international diff erences in survival. 55,[85][86][87] The comparability of data gathered routinely on cancer stage remains poor in developed countries, 88 even though the TNM classifi cation 89 has been available for more than 60 years. We will examine in more detail the extent to which available data on tumour stage can explain the very wide global diff erences in survival reported by us here.
We imputed the day of diagnosis in data from registries that only record (or were only allowed to submit) the month and year of diagnosis. A few of those registries also submitted survival time in days; our imputation achieved similar results. The eff ect on short-term survival of minor variations in the date of diagnosis is generally small 90 and cannot account for the very wide international diff erences in 5-year survival. 91 Loss to follow-up of cancer patients in registries using active follow-up varied widely, but most registries also used several passive follow-up techniques. Diff erences between the databases used for passive follow-up can aff ect survival estimates. 92,93 When information for all deaths is incomplete or inaccessible from administrative systems, active follow-up by the registry augments completeness of ascertainment of vital status, particularly in low-income and middle-income countries. 94 Some registries did not have the resources to follow up all their patients for vital status. Others could not provide follow-up data for at least 5 years after diagnosis for all their patients; for those registries, we have presented survival at 3 or 4 years if possible.
If age-specifi c (and thus age-standardised) survival estimates could not be produced, non-standardised estimates for all ages combined were presented. In some analyses, data had to be pooled across two or three calendar periods, restricting presentation of survival trends. For some countries or regions with very small populations, no survival estimate could be made at all for less common cancers, because very few patients were available for analysis.
We used a rigorously enforced protocol, with centralised data evaluation and analysis to enhance comparability, but international survival comparisons should still be interpreted with caution. Data quality varies widely: 95,96 we provided detailed indices of data quality at country and registry level (table 2; appendix pp 3-63), which should be taken into account. Not all countries could provide data for 2005-09. Also, the range in size between the smallest and largest populations included in this report is greater than 1000-fold, both for registries with national coverage (eg, Gibraltar includes 29 000 people and the UK covers 61·8 million people) and those with regional coverage (eg, Nunavut in Canada represents 33 000 people whereas California in the USA includes 37·0 million people). These diff erences are refl ected in the numbers of patients and the width of CIs around survival estimates. However, lack of precision because of small numbers does not necessarily imply that the survival estimates are incorrect or unreliable: high quality and completeness of data and follow-up can be easier to achieve in small or island populations than in large urban populations.
For robust international comparison of cancer survival, diff erences and trends in background mortality according to age, sex, region, and ethnic origin must be taken into account. In the populations covered by these data, the range in background mortality was very wide, measured by life expectancy at birth (46-87 years in females and 45-81 years in males), and by the change in life expectancy between 1995 and 2009 (appendix p 175), and in other metrics such as the probability of death in middle age (data not shown). We created more than 6500 complete life tables of background mortality to capture these diff erences.
For children with cancer, usual practice is to present the observed probability of survival, including all causes of death, 97 rather than net survival, because mortality from other causes is typically very low, at least in developed countries. Here, however, we have estimated net survival for children with acute lymphoblastic leukaemia because, among the 53 countries for which data could be analysed, mortality from other causes in childhood varied very widely. In 2002, infant mortality ranged from less than one death per 1000 livebirths to more than 120 deaths per 1000 livebirths (in some African populations); under-5 mortality ranged from less than one death per 1000 livebirths to more than 200 deaths per 1000 livebirths; and the overall probability of death before age 15 years ranged from one death per 1000 livebirths to more than 250 per 1000 livebirths (data not shown). For a worldwide comparison of survival from childhood acute lymphoblastic leukaemia, it seemed especially important to eliminate the eff ect of this wide variation in background mortality between countries and over time.
Net survival was age-standardised in most estimates for both adults and children. Age standardisation minimises the risk of reporting international diff erences or trends in cancer survival that are attributable solely to international diff erences or changes over time in the age distribution of cancer patients. 42 We included both fi rst and higher order cancers in our analyses. The eff ect of multiple primary cancers on overall survival is typically only 1-2%, 98 but the proportion of such cancers in a given population is aff ected by the set of rules used to defi ne them 99 and by the longevity of the registry. 100 Some participating registries began operation in the 1950s whereas others only started after 2000. In long-established registries, 10% or more of patients might be registered with more than one cancer. 101 This proportion is lower in newer registries, because a second cancer will typically be registered as the patient's fi rst. Restriction to fi rst primaries can also aff ect international comparison of survival trends, because the number of long-term survivors at high risk of another cancer is increasing, particularly in high-income countries. 102 Exclusion of second cancers would, therefore, tend to bias international survival comparisons in favour of wealthier countries. 103 The rules for defi ning multiple primary cancers diff er between North America and the rest of the world, 24,25 but in a novel step, data from registries in North America were fi rst converted to IACR defi nitions used elsewhere, before being submitted for analysis. This alteration will have minimised any eff ect on international survival comparisons presented here.
To maintain the breadth of global surveillance of survival, we retained some datasets that seemed less suitable for international comparison than all other estimates, but we fl agged these survival estimates to inform interpretation. The number of fl agged estimates is larger than in the fi rst CONCORD study 6 because more registries are from low-income countries and the data cover a much longer period. Residual errors and artifacts in data undoubtedly exist, but they are unlikely to account for global patterns and trends in cancer survival.
We used an unbiased estimator of net survival. 104 To our knowledge, this is the fi rst time this estimator has been used for an international comparison. We used the period approach 33 to estimate survival up to 5 years after diagnosis for patients diagnosed during 2005-09 (appendix p 174). This approach off ers reliable prediction of the eventual survival of recently diagnosed patients who have not all been followed up for 5 years. 105 A small part of the global range in survival could be attributable to diff erences in the intensity of diagnostic activity. The introduction of new diagnostic techniques in wealthier countries, such as PSA testing for prostate cancer, has led to more patients being diagnosed at an early stage of disease, typically with a good prognosis, thus infl ating both incidence and survival. We were not able to use the proportion of in situ cancers for international comparison of the intensity of diagnostic activity for cancers of the colon, rectum, breast, cervix, or prostate. Some registries do not collect data for in situ tumours, whereas some registries that do collect this information did not include these data in their submissions. In poorer countries, by contrast, many patients still die undiagnosed or untreated. 68 For some cancers, both incidence and survival in countries with the most intensive diagnostic activity could be infl ated slightly by overdiagnosis, but the eff ect on the global range of survival estimates is probably small. Equally, in the poorest countries, under-registration of cancer patients with the worst prognosis might lead to underestimation of incidence and overestimation of survival. Even though some survival estimates in low-income and middle-income countries might be too high for this reason, it is striking that for cancers of the colon, rectum, lung, and breast, and particularly for leukaemia in adults and children, the range of estimates in Africa and Central and South America for patients diagnosed during 2005-09 is still much lower than in North America and Oceania during 1995-99, 10 years earlier (fi gure 4; appendix pp 163-73). As reported elsewhere, 68 these patterns strongly suggest inadequate access to early diagnosis and optimum treatment.
National health-care systems must manage all cancer patients, however they are diagnosed, even if some patients might not have been diagnosed before widespread adoption of new diagnostic techniques or screening programmes. In a given country, incidence and survival estimates refl ect current approaches to prevention, diagnosis, and treatment. 6 Coherent assessment of preventive and health-care strategies, therefore, requires that all cancer patients are included, no matter how they are diagnosed, in both incidence and survival estimates. Projections of the future burden of cancer 106 are based on the same cancer incidence data.
Some cancer registries followed up their patients for the fi rst time so they could participate in CONCORD-2. Other registries, not all of them in low-income countries, were prevented from participating by scant resources either to follow up registered patients for vital status or to prepare data for submission. This defi cit underscores the continued fragility, low coverage, and scarcity of resources for cancer registries. 4,107,108 In many countries, even the basic infrastructure of a civil registration system and vital statistics is defi cient. 109 This absence is especially severe in Africa, where several participating countries have also been subject to civil or military confl ict within the past 10-15 years and where, with few exceptions, assessment of recent survival trends from available data was almost impossible.
Cancer registries are crucial to our understanding of the global cancer burden, 107 and they need to be funded and equipped to gather, analyse, and publish incidence and survival data at national or regional level. Worldwide monitoring of cancer incidence has been done since the 1960s, with centralised data collection and standardised methods in Cancer Incidence in Five Continents. 16 IARC's Global Initiative for Cancer Registry Development is an important stimulus to promote high-quality data collection and cancer registration in low-income and middle-income countries. 108 Both WHO 3 and the UN 110 have recognised cancer as a worldwide public health issue of growing concern. However, if cancer registration is to develop further in support of the 25 × 25 goals and in the evaluation of clinical care, 111 WHO and the UN will need to address the growing legal and procedural diffi culties in obtaining primary health data and in accessing them for research. For example, legislation now at the fi nal stage of discussion in the European Union would make cancer registration and most forms of public health research either impossible or illegal in 28 European countries. 112,113 The CONCORD programme at the London School of Hygiene & Tropical Medicine (LSHTM) represents the establishment of worldwide surveillance of cancer survival by centralised quality control and analysis of population-based registry data, as a comparative metric of the eff ectiveness of health systems. It will provide part of the evidence base for global policy on cancer control and should contribute to the overarching goal of the World Cancer Declaration 2013 14 and, more broadly, to the "revolution in metrics for global health". 114 At a national level, cancer outcomes are aff ected by the organisation and funding of access to health services. 115 Improvements in cancer survival have been reported after major political and economic changes in Estonia, 116 Lithuania, 117 and Germany. 118 In turn, low survival has aff ected the development of cancer strategy in countries such as Algeria, 119 Brazil,120,121 Mexico,122 China,India,and Russia,123 and in many wealthier countries. 4 Some of the conclusions drawn from these analyses are similar to those for patients diagnosed 20-25 years ago. 6 The fi ndings of this study can be used to assess the extent to which investment in health-care systems is improving their eff ectiveness. We will examine survival trends and diff erentials in relation to health economic indicators to assess why improvements in survival are so slow and unequal.
Most of the wide global range in cancer survival is probably attributable to inequity in access to optimum diagnostic and treatment services, 6 both in rich [124][125][126] and poor 127,128 countries. Availability of linear accelerators varies more than ten-fold worldwide, from one machine per 500 000 population to less than one per fi ve million people, and more than 30 countries in Africa and Asia have no radiotherapy service at all. 129 Cancer survival in Europe has been associated with gross national product, total national expenditure on health and investment in health technology (eg, CT scanners, radiotherapy units), 130 and with suboptimum allocation of available resources. 86 The global economic cost of cancer from premature death and lost productivity was estimated at US$895 billion in 2008, excluding direct treatment costs estimated at $300 billion. 131 Even in wealthy countries, the rapidly growing costs of cancer treatment have raised concerns about the growing use of tests, imaging, and treatments that are expensive but have marginal value. 132 At the same time, closing the rich-poor divide in access to cancer treatment has been described as "an equity imperative". 133,134 The fi ndings reported here confi rm the global divide in outcomes.
The fi rst international study of cancer survival was published 50 years ago. 5 In the same year, Alexander Langmuir, founder of the US Centers for Disease Control and Prevention's epidemic intelligence service, commented on national outbreaks of infectious disease: "good surveillance does not necessarily ensure the making of the right decisions, but it reduces the chances of wrong ones". 135 His view applies today to non-communicable diseases such as cancer, for which long-term surveillance of incidence, mortality, and survival is increasingly important. Survival is a key metric of overall progress in cancer control. 4 Continuous worldwide surveillance of cancer survival should become both an indispensable source of information for cancer patients and researchers and a stimulus for politicians to improve health policy and health-care systems.

Contributors
CA, HKW, RM-G, CS, GAS, BR, HS, TCT and MPC drafted the protocol; CA, HKW, GAS, W-QC, OJO, MJS, HY, TM, MB-L, TCT, and MPC obtained statutory and ethics approvals; HKW, FB, CJJ, RM-G, CS, GAS, W-QC, OJO, MJS, HY, TM, MB-L, HS, and TCT contributed to data acquisition; CA, DS, FB, MPC and BR prepared the life tables; CA, HC, RH, DS, X-SW, FB, JVA, AB, BR, and MPC had access to all raw data; CA, HC, RH, DS, X-SW, FB, JVA, CJJ, AB, and MPC did the data preparation, quality control and analyses, and checked the results; CA and MPC drafted the report. All authors contributed to writing the fi nal report and approved the version to be published. All members of the CONCORD Working Group had access to the results at all steps of data preparation, quality control, and analyses, and contributed to interpretation of the fi ndings.