Global burden of cancer attributable to high body-mass index in 2012: a population-based study

Transcription

1 Global burden of cancer attributable to high body-mass index in 2012: a population-based study Melina Arnold*, Nirmala Pandeya*, Graham Byrnes, Andrew G Renehan, Gretchen A Stevens, Majid Ezzati, Jacques Ferlay, J Jaime Miranda, Isabelle Romieu, Rajesh Dikshit, David Forman, Isabelle Soerjomataram Lancet Oncol 2015; 16: Published Online November 26, S (14) See Comment page 3 *Contributed equally Section of Cancer Surveillance (M Arnold PhD, J Ferlay MSc, D Forman PhD, I Soerjomataram PhD), Biostatistics Group (G Byrnes PhD), and Nutrition and Metabolism Section Epidemiology Group (I Romieu PhD), International Agency for Research on Cancer, Lyon, France; School of Population Health, University of Queensland, Brisbane, QLD, Australia (N Pandeya PhD); Faculty Institute of Cancer Sciences, University of Manchester, Manchester Academic Health Science Centre, Manchester, UK (Prof A G Renehan PhD); Department of Health Statistics and Information Systems, WHO, Geneva, Switzerland (G A Stevens DSc); MRC-PHE Centre for Environment and Health and Department of Epidemiology and Biostatistics, School of Public Health, Imperial College London, London, UK (Prof M Ezzati FMedSci); CRONICAS Centre of Excellence in Chronic Diseases and School of Medicine, Cayetano Heredia University, Lima, Peru (J Jaime Miranda PhD); and Department of Epidemiology, Tata Memorial Hospital, Mumbai, India (R Dikshit PhD) Correspondence to: Dr Melina Arnold, Section of Cancer Surveillance, International Agency for Research on Cancer, Lyon, France arnoldm@fellows.iarc.fr Summary Background High body-mass index (BMI; defined as 25 kg/m² or greater) is associated with increased risk of cancer. To inform public health policy and future research, we estimated the global burden of cancer attributable to high BMI in Methods In this population-based study, we derived population attributable fractions (PAFs) using relative risks and BMI estimates in adults by age, sex, and country. Assuming a 10-year lag-period between high BMI and cancer occurrence, we calculated PAFs using BMI estimates from 2002 and used GLOBOCAN2012 data to estimate numbers of new cancer cases attributable to high BMI. We also calculated the proportion of cancers that were potentially avoidable had populations maintained their mean BMIs recorded in We did secondary analyses to test the model and to estimate the effects of hormone replacement therapy (HRT) use and smoking. Findings Worldwide, we estimate that or 3 6% of all new cancer cases in adults (aged 30 years and older after the 10-year lag period) in 2012 were attributable to high BMI. PAFs were greater in women than in men (5 4% vs 1 9%). The burden of attributable cases was higher in countries with very high and high human development indices (HDIs; PAF 5 3% and 4 8%, respectively) than in those with moderate (1 6%) and low HDIs (1 0%). Corpus uteri, postmenopausal breast, and colon cancers accounted for 63 6% of cancers attributable to high BMI. A quarter (about ) of the cancer cases related to high BMI in 2012 could be attributed to the increase in BMI since Interpretation These findings emphasise the need for a global effort to abate the increasing numbers of people with high BMI. Assuming that the association between high BMI and cancer is causal, the continuation of current patterns of population weight gain will lead to continuing increases in the future burden of cancer. Funding World Cancer Research Fund International, an Commission (Marie Curie Intra-an Fellowship), Australian National Health and Medical Research Council, and US National Institutes of Health. Copyright World Health Organization. Published by Elsevier Ltd/Inc/BV. All rights reserved. Introduction High body-mass index (BMI; defined as 25 kg/m² or greater) is a known risk factor for various chronic diseases and mortality. Although prevalence varies widely, overweight and obesity have been increasing worldwide, raising concerns about their effect on health. Recent statistics showed that 35% of the adult population (aged 20 years and older) worldwide is overweight (BMI 25 kg/m²), including 12% that is classified as obese (BMI 30 kg/m²). 1 The prevalence of high BMI ranges from about 10% in many Asian and African countries to more than 90% in Pacific island nations such as the Cook Islands and Nauru. According to recent estimates, 1,2 the global prevalence of excess bodyweight in adults increased by 27 5% between 1980 and 2013, although the increase has slowed in recent years in some an countries and the USA. 3 7 Continuous updates of the scientific literature have confirmed the association between high BMI and risk of oesophageal adenocarcinoma and colon, rectal, kidney, pancreas, gallbladder (women only), postmenopausal breast, ovarian, and endometrial cancers The estimated increase in risk of these cancers due to high BMI ranges from 3% to 10% per unit increase in BMI. 14 A recent estimate from Global Burden of Disease (GBD) study 15 showed that 3 9% of cancer mortality in 2010 could be attributed to high BMI. However, this estimate did not take into account the lag time necessary for high BMI to lead to the development of a new cancer. Additionally, relating risk factors to mortality in the estimation of disease burden can be problematic because of the potential role of reverse causation. 16 Potential confounders and effect modifiers of the association between BMI and cancer, such as the use of hormone replacement therapy (HRT) and smoking, also need to be taken into account. 17,18 In this study, we aimed to estimate the global population attributable fraction (PAF) of cancer incidence in 2012 attributable to high BMI in 2002, acknowledging the time lag between the exposure (high BMI) and outcomes (cancer incidence). We also aimed to test the robustness of the estimates in a series of sensitivity analyses, including assessment of the role of smoking and HRT use as potential effect modifiers or confounders of the association between high BMI and cancer incidence Vol 16 January 2015

2 Methods Study design In this population-based study, we quantified the effect of high BMI on cancer incidence in the adult population worldwide, tested the effect of various model assumptions, and assessed the effect of confounders and effect modifiers. We applied PAFs according to sex, age, cancer site, and country to national estimates of cancer incidence estimates, using data for mean BMI, cancer incidence, and corresponding risk estimates. We used the BMI estimates reported by Global Burden of Metabolic Risk Factors of Chronic Diseases Collaborating Group. 1 The details of the applied model and its assumptions in the estimation of mean BMIs have been reported elsewhere. 19 For this study, we obtained the annual estimates of mean BMI and the corresponding SDs for adults aged 20 years and older for each country by sex and age group (20 34, 35 44, 45 54, 55 64, 65 74, and 75 years) in 1982 and 2002 (appendix pp 2 7). In our primary analysis, we included only cancers reported by the World Cancer Research Fund (WCRF) as having sufficient evidence to be associated with high BMI These include oesophageal adenocarcinoma and colon, rectal, kidney, pancreatic, gallbladder, postmenopausal breast, corpus uteri, and ovarian cancers, collectively defined here as high-bmi-related cancers. In view of the differences in risk of colon and rectal cancer associated with obesity, we estimated PAFs for these two cancer sites separately. Similarly, only adenocarcinoma of the oesophagus was included because of the absence of an association between excess bodyweight and oesophageal squamous-cell carcinoma. The sex-specific relative risks (RRs) for the sites included in the analysis were obtained from the standardised metaanalysis estimates by Renehan and colleagues 14 and the WCRF Continuous Update Project In these metaanalyses, risk estimates were pooled from from cohort studies that mainly used cancer incidence as an outcome (apart from pancreatic cancer, for which studies included mortality as an outcome). In a secondary analysis we included thyroid cancer and non-hodgkin lymphoma as additional cancer sites, which might be associated with high BMI, 14,20 but were not listed by WCRF as having sufficient evidence. The exact sources and sizes of RRs are described in the appendix (pp 8 9). Calculation of PAF We calculated PAFs on the basis of the approach suggested by the Comparative Risk Assessment Collaborative Group, using the formula: 21 PAF= RR(x)P(x)dx RR(x)P* (x)dx RR(x)P(x)dx where P(x) is the population distribution of BMI, P*(x) is the distribution of theoretical minimum BMI, RR(x) is the RR of cancer associated with BMI, and dx indicates that the integration was done with respect to BMI. The theoretical minimum distribution of BMI was defined as a BMI distribution with a mean of 22 kg/m² and an SD of 1 kg/m², at which the disease burden is assumed to be lowest at the population level. 15,22 We used a log-logit function to characterise the shape of the RR across BMI units. Furthermore, we assumed no risk for BMI less than 22 and no risk increase for BMI greater than 40, since estimates of RR beyond these points were scant. A pictorial illustration and a more detailed description of these assumptions of the risk function are presented in the appendix (pp 8 9). We calculated age-specific, sex-specific, and countryspecific PAFs for individual high-bmi-related cancer sites. We then derived the number of cancer cases attributable to high BMI by multiplying age-specific, sexspecific, country-specific, and cancer-specific PAFs by the corresponding numbers of incident cancers in We calculated overall national, regional, and global estimates of the total attributable proportion of cancer related to high BMI by summing the numbers of attributable incident cases and dividing them by the total number of cancer cases in each subgroup. We estimated 90% uncertainty limits for PAFs using Monte Carlo simulation. We also computed a counterfactual scenario (ie, a model of incidence if mean BMIs had remained at their 1982 values) to provide a more realistic view about the preventable proportion of the current burden of cancers caused by high BMI. The analysis was done by replacing the theoretical minimum distribution with the BMI distribution that was reported in in 1982, an attainable value in the past in each country and probably a more realistic goal for prevention than the mean BMI of 22 kg/m² used in our main analysis. Using this approach, we estimated what the PAF would be if population mean BMIs had stayed constant at their 1982 values. A more detailed description of PAF inputs and calculation is presented in the appendix (pp 31 33). Cancer incidence and attributable cancer burden The numbers of incident cancers in 2012 by age (in adults aged 30 years and older after 10-year lag period), sex, and country were obtained from GLOBOCAN Countries were grouped into 12 geographical regions (appendix pp 34 40): sub-saharan Africa (eastern, middle, southern, and western Africa); Middle East and north Africa (western Asia and northern Africa); Latin America and the Caribbean (central and south America and the Caribbean); North America; east Asia (eastern Asia, including China); southeast Asia; south-central Asia (southern Asia, including India); eastern ; northern ; southern ; western ; and Oceania (including Australia and New Zealand). 24 Furthermore, countries also were grouped by 2012 human development index (HDI; very high, high, moderate, or low). 25 Because the separate incidences of See Online for appendix Vol 16 January

3 Oesophageal adenocarcinoma Colon Rectum Pancreas Kidney Total PAF n PAF n PAF n PAF n PAF n n PAF high BMI* PAF all Region Sub- Saharan Africa Middle East and north Africa Latin America and the Caribbean North America 15 5% ( ) 34 3% ( ) 35 6% ( ) 44 4% ( ) East Asia 17 9% ( ) Southeast Asia Southcentral Asia Eastern Northern Southern Western 13 8% ( ) 9 7% ( ) 38 2% ( ) 44 0% ( ) 43 0% ( ) 42 6% ( ) Oceania 44 2% ( ) HDI group Low HDI 7 8% ( ) Medium HDI 16 4% ( ) High HDI 34 2% ( ) Very high HDI World 43 2% ( ) Total 33 3% ( ) 125 (88 169) 391 ( ) 1193 ( ) 4293 ( ) 1390 ( ) 91 (66 128) 389 ( ) 637 ( ) 2262 ( ) 527 ( ) 1911 ( ) 361 ( ) 175 (64 326) 1703 ( ) 1735 ( ) 9955 ( ) ( ) 5 0% ( ) 16 1% ( ) 15 4% ( ) 21 0% ( ) 6 9% ( ) 4 2% ( ) 4 2% ( ) 16 0% ( ) 18 4% ( ) 18 1% ( ) 18 7% ( ) 19 1% ( ) 3 3% ( ) 5 9% ( ) 14 5% ( ) 16 8% ( ) 13 0% ( ) 330 ( ) 2111 ( ) 4058 ( ) ( ) 9120 ( ) 951 ( ) 997 ( ) 6082 ( ) 3756 ( ) 7002 ( ) 8536 ( ) 1212 ( ) 341 ( ) 7399 ( ) 8533 ( ) ( ) ( ) 2 6% ( ) 8 1% ( ) 7 6% ( ) 10 9% ( ) 3 4% ( ) 2 1% ( ) 1 7% ( ) 8 2% ( ) 9 5% ( ) 9 4% ( ) 9 7% ( ) 9 9% ( ) 1 5% ( ) 2 8% ( ) 7 4% ( ) 8 4% ( ) 6 2% ( ) 144 ( ) 631 ( ) 1099 ( ) 2792 ( ) 3386 ( ) 301 ( ) 353 ( ) 2586 ( ) 1337 ( ) 1930 ( ) 2847 ( ) 399 ( ) 127 (61 239) 2778 ( ) 3104 ( ) ( ) ( ) 4 3% ( ) 10 8% ( ) 10 0% ( ) 14 2% ( ) 4 4% ( ) 2 6% ( ) 2 8% ( ) 10 3% ( ) 12 2% ( ) 12 0% ( ) 12 5% ( ) 12 8% ( ) 2 5% ( ) 4 1% ( ) 9 6% ( ) 11 2% ( ) 8 4% ( ) 175 ( ) 697 ( ) 1294 ( ) 3390 ( ) 2625 ( ) 165 ( ) 255 ( ) 1829 ( ) 845 ( ) 1379 ( ) 1959 ( ) 233 ( ) 117 (63 199) 2340 ( ) 2717 ( ) 9672 ( ) ( ) 5 9% ( ) 19 5% ( ) 19 0% ( ) 25 0% ( ) 8 2% ( ) 5 6% ( ) 5 1% ( ) 18 7% ( ) 21 8% ( ) 21 1% ( ) 21 8% ( ) 22 6% ( ) 3 9% ( ) 7 6% ( ) 17 4% ( ) 21 3% ( ) 16 6% ( ) 94 (68 129) 1150 ( ) 2366 ( ) 9801 ( ) 4878 ( ) 248 ( ) 481 ( ) 4382 ( ) 2042 ( ) 3206 ( ) 4984 ( ) 600 ( ) 134 (62 243) 4365 ( ) 5437 ( ) ( ) ( ) 867 ( ) 4980 ( ) ( ) ( ) ( ) 1756 ( ) 2474 ( ) ( ) ( ) ( ) ( ) 2804 ( ) 895 ( ) ( ) ( ) ( ) ( ) 4 6% ( ) 14 5% ( ) 14 4% ( ) 20 8% ( ) 6 0% ( ) 3 6% ( ) 3 7% ( ) 13 8% ( ) 18 3% ( ) 16 1% ( ) 17 2% ( ) 17 9% ( ) 3 1% ( ) 5 3% ( ) 13 0% ( ) 15 9% ( ) 11 9% ( ) 0 4% ( ) 2 0% ( ) 2 0% ( ) 3 5% ( ) 0 9% ( ) 0 5% ( ) 0 4% ( ) 3 1% ( ) 3 8% ( ) 3 3% ( ) 3 3% ( ) 3 4% ( ) 0 3% ( ) 0 7% ( ) 2 1% ( ) 3 2% ( ) 1 9% ( ) Data are population attributable fractions (PAFs) or numbers of cancer cases (n), both reported with 90% uncertainty intervals. BMI=body-mass index. HDI=Human Development Index. *PAF high BMI=proportion of high-bmi-related cancers attributable to high BMI. PAF all=proportion of all cancer (excluding non-melanoma skin cancers) attributable to high BMI. Table 1: Estimated PAFs and numbers of cancer cases associated with high BMI in men in Vol 16 January 2015

4 Oesophageal adenoarcinoma Colon Rectum Gallbladder Pancreas Breast (postmenopausal) Corpus uteri Ovary Kidney Total PAF n PAF n PAF n PAF n PAF n PAF n PAF n PAF n PAF n n PAF high BMI* PAF all Region Sub- Saharan Africa Middle East and north Africa Latin America and the Caribbean North America 27 3% ( ) 43 9% ( ) 41 2% ( ) 48 2% ( ) East Asia 21 5% ( ) Southeast Asia Southcentral Asia Eastern Northern Southern Western 22 5% ( ) 17 6% ( ) 46 4% ( ) 43 8% ( ) 44 3% ( ) 42 1% ( ) Oceania 43 5% ( ) HDI group Low HDI 13 8% ( ) Medium HDI 22 6% ( ) 151 ( ) 172 ( ) 507 ( ) 706 ( ) 582 ( ) 54 (45 64) 289 ( ) 228 ( ) 579 ( ) 111 ( ) 416 ( ) 67 (64 70) 150 (86 221) 850 ( ) 4 8% ( ) 11 1% ( ) 9 3% ( ) 11 1% ( ) 3 9% ( ) 3 5% ( ) 3 4% ( ) 10 7% ( ) 9 4% ( ) 9 6% ( ) 9 3% ( ) 9 8% ( ) 2 9% ( ) 4 0% ( ) 302 ( ) 1254 ( ) 2670 ( ) 6052 ( ) 4229 ( ) 694 ( ) 636 ( ) 4543 ( ) 1829 ( ) 2943 ( ) 3707 ( ) 592 ( ) 253 ( ) 4112 ( ) 2 2% ( ) 5 6% ( ) 4 6% ( ) 5 6% ( ) 2 0% ( ) 1 8% ( ) 1 4% ( ) 5 3% ( ) 4 8% ( ) 4 9% ( ) 4 7% ( ) 5 0% ( ) 1 4% ( ) 1 9% ( ) 128 (96 167) 334 ( ) 603 ( ) 1009 ( ) 1228 ( ) 174 ( ) 227 ( ) 1371 ( ) 441 ( ) 647 ( ) 875 ( ) 122 ( ) 104 (60 158) 1279 ( ) 20 4% ( ) 53 2% ( ) 48 4% ( ) 53 4% ( ) 25 0% ( ) 22 7% ( ) 13 2% ( ) 52 5% ( ) 47 3% ( ) 48 9% ( ) 48 7% ( ) 49 1% ( ) 11 7% ( ) 23 2% ( ) 279 ( ) 1501 ( ) 4930 ( ) 3131 ( ) ( ) 819 ( ) 2249 ( ) 3450 ( ) 656 ( ) 2193 ( ) 2631 ( ) 222 ( ) 882 ( ) ( ) 5 6% ( ) 11 8% ( ) 9 5% ( ) 11 2% ( ) 3 9% ( ) 3 5% ( ) 3 2% ( ) 10 7% ( ) 9 6% ( ) 9 8% ( ) 9 5% ( ) 9 9% ( ) 3 3% ( ) 4 3% ( ) 216 ( ) 514 ( ) 1377 ( ) 2625 ( ) 1796 ( ) 200 ( ) 237 ( ) 1796 ( ) 695 ( ) 1122 ( ) 1533 ( ) 158 ( ) 138 (84 202) 1774 ( ) 6 1% ( ) 14 7% ( ) 12 1% ( ) 14 3% ( ) 5 1% ( ) 4 8% ( ) 4 1% ( ) 13 3% ( ) 11 8% ( ) 11 9% ( ) 11 4% ( ) 12 4% ( ) 4 4% ( ) 5 6% ( ) 2759 ( ) 6760 ( ) ( ) ( ) 8143 ( ) 2959 ( ) 4736 ( ) ( ) 7513 ( ) 9096 ( ) ( ) 1740 ( ) 2980 ( ) ( ) 24 8% ( ) 47 7% ( ) 41 6% ( ) 47 8% ( ) 19 6% ( ) 18 9% ( ) 16 5% ( ) 44 7% ( ) 41 5% ( ) 41 1% ( ) 40 5% ( ) 42 4% ( ) 17 3% ( ) 20 6% ( ) 2142 ( ) 4609 ( ) 8022 ( ) ( ) ( ) 2842 ( ) 3241 ( ) ( ) 5707 ( ) 8013 ( ) 9562 ( ) 1287 ( ) 2005 ( ) ( ) 2 3% ( ) 6 4% ( ) 5 5% ( ) 6 8% ( ) 2 2% ( ) 2 2% ( ) 1 5% ( ) 5 9% ( ) 5 5% ( ) 5 3% ( ) 5 3% ( ) 5 7% ( ) 1 6% ( ) 2 2% ( ) 252 ( ) 617 ( ) 900 ( ) 1554 ( ) 965 ( ) 399 ( ) 540 ( ) 1639 ( ) 544 ( ) 672 ( ) 758 ( ) 108 ( ) 301 ( ) 1814 ( ) 11 1% ( ) 34 7% ( ) 30 3% ( ) 34 5% ( ) 13 8% ( ) 12 9% ( ) 12 4% ( ) 32 8% ( ) 30 0% ( ) 29 8% ( ) 28 7% ( ) 30 7% ( ) 9 3% ( ) 14 3% ( ) 184 ( ) 1256 ( ) 2269 ( ) 8084 ( ) 4045 ( ) 321 ( ) 526 ( ) 5317 ( ) 1728 ( ) 2231 ( ) 3791 ( ) 428 ( ) 234 ( ) 4052 ( ) 6413 ( ) ( ) ( ) ( ) ( ) 8465 ( ) ( ) ( ) ( ) ( ) ( ) 4722 ( ) 7048 ( ) ( ) 7 6% ( ) 18 2% ( ) 15 8% ( ) 19 2% ( ) 8 3% ( ) 6 2% ( ) 5 4% ( ) 18 2% ( ) 14 9% ( ) 15 3% ( ) 14 5% ( ) 15 2% ( ) 5 5% ( ) 8 2% ( ) 2 0% ( ) 7 0% ( ) 6 4% ( ) 9 4% ( ) 3 0% ( ) 2 2% ( ) 1 7% ( ) 9 9% ( ) 7 9% ( ) 8 1% ( ) 7 8% ( ) 7 2% ( ) 1 5% ( ) 2 7% ( ) (Table 2 continues on next page) Vol 16 January

5 Corpus uteri Ovary Kidney Total Colon Rectum Gallbladder Pancreas Breast (postmenopausal) Oesophageal adenocarcinoma PAF all PAF n PAF n PAF n PAF n PAF n PAF n PAF n PAF n PAF n n PAF high BMI* (Continued from previous page) 7 5% ( ) 16 6% ( ) ( ) 6604 ( ) 31 6% ( ) 2424 ( ) 5 7% ( ) ( ) 43 6% ( ) ( ) 12 5% ( ) 2692 ( ) 10 0% ( ) 6285 ( ) 49 0% ( ) 1845 ( ) 5 0% ( ) 6539 ( ) 9 9% ( ) 796 ( ) High HDI 41 5% ( ) 7 8% ( ) 15 5% ( ) ( ) ( ) 29 8% ( ) 4409 ( ) 5 4% ( ) ( ) 41 8% ( ) ( ) 12 2% ( ) 7665 ( ) 9 0% ( ) ( ) 41 5% ( ) 3933 ( ) 4 4% ( ) ( ) 8 9% ( ) 2066 ( ) 44 2% ( ) Very high HDI World 5 4% ( ) 13 1% ( ) ( ) ( ) 25 9% ( ) 8948 ( ) 4 0% ( ) ( ) 34 0% ( ) ( ) 10 2% ( ) ( ) 7 8% ( ) ( ) 32 3% ( ) 7160 ( ) 3 6% ( ) ( ) 7 6% ( ) 3862 ( ) Total 33 8% ( ) Data are population attributable fractions (PAFs) or numbers of cancer cases (n), both reported with 90% uncertainty intervals. BMI=body-mass index. HDI=Human Development Index. *PAF high BMI=proportion of high-bmi-related cancers attributable to high BMI. PAF all=proportion of all cancer (excluding non-melanoma skin cancers) attributable to high BMI. Table 2: Estimated PAFs and numbers of cancer cases associated with high BMI in women in 2012 colon and rectal cancers and the incidence of oesophageal cancer by histological subtypes are not reported in GLOBOCAN, 23 we estimated the numbers of these cancers by subtypes using country-specific and sexspecific proportions of subtypes reported in Cancer Incidence in Five Continents volume X (appendix p 10). 26,27 The precise time lag between development and duration of high BMI and the occurrence of cancer is not well established. However, the general perception is that excess bodyweight does not initiate cancer, but rather promotes of cancer to clinical presentation over several years. Renehan and colleagues 28 assumed a 10-year lag time on the basis of the scientific literature, wherein the average follow-up time of 10 years showed the beneficial effect of weight loss on subsequent cancer risk. 29,30 With no additional information available, we chose to assume a 10-year lag period in this study, mapping high BMI prevalence in 2002 (by sex, age, and country) to cancer incidence in Sensitivity analyses In estimating the PAFs of cancers attributable to high BMI, we made several assumptions about the population BMI distribution and the RRs function. To assess their effect on the results, we repeated the analyses while changing the following assumptions: exposure definition (categorical vs continuous BMI; appendix pp 13 14); BMI distribution (normal vs log-normal; appendix pp 15 16); shape of the RR (linear or log-linear vs log-logit; appendix pp 17 19); and region-specific versus global RR estimates (appendix pp 20 23). Because smoking and use of HRT 8,11,35 are known effect modifiers of the association between bodyweight and cancer, we estimated PAFs stratified by current smoking status (for pancreatic cancer) or HRT use (for postmenopausal breast cancer, ovarian cancer, and endometrial cancer) and assessed the bias that might have occurred when these interactions were ignored (appendix pp 24 27). Furthermore, because studies have shown a protective effect of high BMI on premenopausal breast cancer, 8,14,36 we also assessed the potentially adverse effects of decreasing population BMI on the incidence of premenopausal breast cancer and its effect on the total PAF (appendix pp 28 30). Role of the funding source The funders of the study 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 the data in the study and had final responsibility for the decision to submit for publication. Results Worldwide, our result show that an estimated or 3 6% of all new cancers (or 12 8% of all high-bmirelated cancers) in adults in 2012 were attributable to high BMI. By sex, (1 9%) new cancers in men (table 1) and (5 4%) in women (table 2) were 40 Vol 16 January 2015

6 attributable to high BMI. The attributable burden was larger in countries with very high and high HDIs (PAF 5 3% and 4 8%, respectively) than in those with moderate (1 6%) and low (1 0%) HDIs. Region-specific estimates show that all three Asian regions and sub-saharan Africa had the lowest PAFs, ranging from 0 4% to 0 9% of total cancers (3 6% to 6 0% of total high-bmi-related cancers) in men and 1 7% to 3 0% of total cancers (5 4% to 8 3% of total high-bmi-related cancers) in women. North America had the highest PAFs, at 3 5% of total cancers (20 8% of high-bmi-related cancers) for men and 9 4% of total cancers (19 2% of high-bmi-related cancers) for women. For the remaining regions (Middle East and north Africa, Latin America and the Caribbean, Oceania, and all an regions), the PAF ranged from 2 0% to 9 9% of total cancers (14 4% to 18 2% of high-bmi-related cancers) in both sexes. With respect to the regional contribution to new high- BMI-related cancers in 2012, the North American region contributed the most ( or 23 0% of the total worldwide cases attributable to high BMI), and sub- Saharan Africa the least (7300 or 1 5%; tables 1, 2). Eastern had the greatest share of attributable Men PAF (%) 3 1 < < < < <0 3 No data Women PAF (%) 8 5 < < < < <1 6 No data Figure 1: PAF of new cancer cases in 2012 caused by high BMI in men and women, by country PAF=population attributable fraction. Vol 16 January

7 North America Northern Western Oceania Southern Eastern Latin America and the Caribbean Middle East and north Africa East Asia Southeast Asia Sub-Saharan Africa South-central Asia Men Women Incidence rate Figure 2: Age-standardised incidence rate of high-bmi-related cancers and high-bmi-related cancers attributable to high BMI (per people) in 2012 Incidence data are age-standardised to the world standard population. Light bars show total incidence rates of high-body-mass-index (BMI)-related cancers, and dark bars show those attributable to high BMI. burden among the an regions ( or 33 8% of the total an cases attributable to high BMI). Despite the low PAF (1 8%), the east Asia region had the second largest number of cases attributable to high BMI (70 000) after North America, because of its large population size. Country-specific PAFs for men and women are shown in figure 1 and in the appendix (pp 34 40). In men, the highest PAF of 5 5% was in the Czech Republic, followed by 4 5% in Jordan and Argentina, and 4 4% in the UK and Malta. The greatest between-country differences within a region were in Latin America and the Caribbean, where the PAF ranged from 4 5% in Argentina to 0 7% in Haiti and Jamaica. In women, Barbados had the highest PAF, with 12 7% of cancers attributable to high BMI, followed by the Czech Republic (12 0%) and Puerto Rico (11 6%). As for men, between-country differences were largest in Latin America and the Caribbean, where the PAF ranged from 12 7% in Barbados to 1 6% in Haiti. Countries in sub-saharan Africa had consistently lower overall PAFs than those in other regions, of less than 2% in men and less than 4% (with the exception of Mauritius and South Africa) in women. PAF also varied greatly by cancer site, ranging from 6 2% for rectal cancer to 33 3% for oesophageal adenocarcinoma in men and from 3 6% for rectal cancer to 34 0% for cancer of the corpus uteri and oesophageal adenocarcinoma in women (tables 1, 2). Despite having a large estimated PAF of more than 30%, oesophageal adenocarcinomas accounted for only (or 10 0%) of the total worldwide cancer cases attributable to high BMI in men and 4000 (or 1 1%) in women. Colon cancer in men and postmenopausal breast cancer in women contributed the largest number of cancer cases attributable to high BMI. In men, colon and kidney cancer together contributed about two-thirds of the new cancer cases attributable to high BMI (90 000). In women, postmenopausal breast cancer and cancer of the corpus uteri contributed about two-thirds of the new cancer cases attributable to high BMI ( ). We noted substantial differences between men and women in PAFs for colon cancer (13 0% vs 7 6%). Sex differences in the numbers of attributable cases were largest for colon cancer and oesophageal adenocarcinoma, with and attributable cases, respectively in men and only and 4000 attributable cases in women (tables 1, 2). The incidence of high- BMI-related cancers attributable to high BMI was relatively higher for women than for men in all regions (figure 2). Particularly, in regions with a fairly low incidence of high-bmi-related cancers, such as Asia and sub-saharan Africa, the proportion of new cancer cases attributable to high BMI was two to three times greater for women than for men. In our counterfactual scenario, we calculated that if BMI had remained as recorded in 1982, about a quarter ( cases) of cases of high-bmi-related cancers in 2012 could have been averted. In other words, a quarter of all high-bmi-related cancers could be attributed to the increase in BMI between 1982 and 2002 (appendix pp 31 33). About 0 9% (0 5% in men and 1 3% in women) of all cancers diagnosed in 2012 could therefore be regarded as realistically avoidable by prevention of high BMI. The realistically attributable fraction was greatest in countries with a very high or high HDI, where 83 2% of these potentially avoidable cancers occurred. In a high-burden region such as North America, this proportion translated into more than cases, or 35 6% of all attributable cancer cases that could be linked to the increase in BMI since With respect to specific cancer sites, about 10 7% of all oesophageal adenocarcinomas (5600), 8 5% of all corpus uteri (27 000), 4 9% of all kidney (15 000) and 2 5% of all postmenopausal breast cancers (28 000) could have been avoided if BMI had not increased between 1982 and Vol 16 January 2015

8 When PAF for pancreatic cancer was corrected for smoking status, we estimated that it ranged between 0 6% and 18 3% for both men and women (appendix pp 24 25), dependent on country. Compared with the unadjusted PAFs, this adjustment increased the PAF by 0 5 percentage points in men and 0 9 percentage points in women. This difference was largest in the UK for both men and women. HRT non-users had substantially higher PAFs than HRT users; PAFs for HRT non-users ranged from 50 4% to 65 0% for corpus uteri cancer and from 8 3% to 12 4% for postmenopausal breast cancer and ovarian cancer, dependent on country. For HRT users, we noted PAFs between 8 6% and 20 8% for corpus uteri cancer and PAFs below 0% for postmenopausal breast and ovarian cancers, because of slightly protective RRs (appendix pp 26 27). When comparing the PAFs corrected for HRT use to the unadjusted PAFs, the difference was small for most countries, ranging from 0 to 14 percentage points for corpus uteri cancer and from 0 to 5 percentage points for postmenopausal breast and ovarian cancers. The difference was larger in countries where the prevalence of HRT use was low (<55%) and a large proportion of women had high BMI eg, Germany and Russia. The results for thyroid cancer, non-hodgkin lymphoma, and premenopausal breast cancer are reported in the appendix (pp 28 30). In the sensitivity analyses, the choice of BMI data type and distribution did not substantially affect the results (appendix pp 13 16), although PAFs changed with the use of different RR functions (appendix pp 17 19) and region-specific RRs (appendix pp 20 23). Discussion Our results show that about 3 6% of all new cancers in adults aged 30 years and older (excluding non-melanoma skin cancer) in 2012, or 12 8% of high-bmi-related cancers, could be attributed to high BMI. These figures are equivalent to an estimated new cancers that might have been caused by high BMI. Postmenopausal breast, corpus uteri, and colon cancer accounted for 72 5% of the total attributable cases in women, whereas in men kidney and colon cancers accounted 66 0% of all attributable cases. 63 5% of the global cancer cases related to high BMI were in the North American and an regions, although the PAF was also large in Oceania, Latin America and the Caribbean, and the Middle East and north Africa. Assuming that the association between high BMI and cancer is causal, the continuation of current patterns of population weight gain will lead to continuing increases in the future burden of cancer. Most importantly, about one quarter of the total cases attributable to high BMI ( cancers) could potentially have been avoided if the global population mean BMI had remained the same as was recorded in Our results show that the issue of cancer burden related to high BMI mainly affects higher-resource regions, particularly North America and. Besides the unequal distribution of cancer cases attributable to high BMI worldwide, we noted substantial differences within regions; for example, in Latin America and the Caribbean PAFs for women ranged from 12 7% in Barbados to 1 6% in Haiti. Although high-bmi-related cancers have become a global issue, 37 the transition from increasing, to stabilising, to possibly decreasing obesity prevalence occurs at different rates in different countries and regions. In a few countries such as the UK and the USA, where BMI increased substantially in the 1980s and 1990s, the BMI increase has since slowed, but in most countries the average BMI has continued to increase steadily since the 1980s. 2 The results of our secondary analysis, in which historical BMI was used as an achievable population mean BMI, could be used to measure the changing effect of BMI on the burden of cancer. Taking into account both current population mean BMIs and their changes over time, the increase in PAF has been greatest in the Middle East and north Africa, Latin America and the Caribbean, North America, and Oceania. By contrast, eastern maintained a similar (high) population mean BMI between 1982 and 2002, so despite the large current PAF, only very few cases are attributable to the change in BMI in that period. The varying pattern in BMI distribution and trends across countries emphasises the need for future research into the cumulative effects of overweight and obesity on the burden of cancer and other chronic diseases. Investigators of independent pooled studies have reported an attenuated risk of high BMI in smokers for pancreatic and thyroid cancers. In our study, taking into account the differential effect by smoking status produced different estimates, dependent on a country s smoking prevalence. In high-income countries such as the UK and the USA, because of the high past prevalence of tobacco smoking 38 and high present BMI, 1 the PAF of pancreatic cancer related to high BMI was slightly underestimated in our uncorrected analysis. By contrast, in low-income countries such as Ghana, where smoking prevalence has only started to rise, 39 the effect of high BMI on pancreatic cancer was slightly overestimated or was not large enough to be appreciable. Another important effect modifier in the relation between BMI and cancer is HRT use, wherein the risk of female hormone-driven cancers related to high BMI is largely attenuated or even eliminated among HRT users. 8,11,35 In our sensitivity analysis, we showed that most postmenopausal breast, ovary, and corpus uteri cancers attributable to high BMI occurred among HRT non-users. The falling use of HRT since the early 2000s 40,41 has contributed to a decrease in breast cancer incidence in countries where use was high; this decrease in use will probably translate into a higher proportion of cases being attributable to high BMI and therefore amenable to prevention by weight loss. This study adds important insights to the contribution of lifestyle and exogenous risk factors on cancer risk. Previous studies have quantified the global cancer burden Vol 16 January

9 Panel: Research in context attributable to infections (2 million new cases in 2008, PAF 16 1%) 42 and smoking (1 4 million cancer deaths in 2000, PAF 21%). 43 Ours is the most comprehensive study so far reported to provide worldwide estimates of the burden of cancer due to high BMI (panel). A report from the GBD study 15 provided estimates of the burden of cancer due to high BMI, but those results are not directly comparable to ours because PAF was presented as a proportion of deaths or disability-adjusted life years attributable to high BMI, whereas incidence was the outcome in our study. Furthermore, in the GBD study, information about high BMI prevalence and cancer mortality was obtained for the same year, not allowing for a lag between the exposure and cancer development and mortality. A few other studies have provided estimates of cancer incidence associated with high BMI, 17,28,44 46 but these were limited to an populations. For example, Renehan and colleagues 28 estimated that 2 5% of all cancer cases in men and 4 1% in women were related to high BMI. Our estimates for the an regions ranged between 3 1% and 3 8% in men and between 7 8% and 9 9% in women. Such differences in estimates are to be expected, since our estimates are based on more recent data for both the prevalence of high BMI and the incidence of cancer, both of which have increased greatly over the past decade. 19,47 Another strength of this study is the use of age-specific, sex-specific, and country-specific estimates of BMI and the latest available estimates of cancer incidence. Although these data were estimates and therefore careful interpretation of the results is advised, the best available estimates were used. We made many assumptions when estimating the PAFs; however, our sensitivity analyses Systematic review We searched Medline for articles published in any language up to Jan 1, 2014, using the search terms obesity, body-mass index, cancer risk, cancer incidence, attributable fraction, avoidable, and preventable. We identified several studies that provided estimates of the burden of cancer attributable to high body-mass index (BMI) in specific countries or regions, 17,28,44 46 as well as a report 15 from the Global Burden of Disease study that included estimates of deaths or disability-adjusted life years attributable to high BMI. However, no previous study had provided global estimates of cancer incidence attributable to high BMI. Interpretation Our results show that 3 6% of all new cancers in adults in 2012 (a total of cases) are attributable to excess bodyweight. This finding emphasises the need for a global effort to abate the continuing increases in overweight and obesity worldwide. Assuming a causal link between high BMI and cancer incidence, if the current pattern of population weight gain continues, it will lead to further increases future burden of cancer, especially in regions such as Latin America and the Caribbean and north Africa, where the largest increases in the prevalence of obesity have occurred in the past three decades. 2 Our results should be used to inform health policy in terms of targets for prevention programmes, while emphasising existing gaps in our knowledge about the association between BMI and cancer. showed that changing these assumptions made little difference to the reported PAFs. One of the assumptions that we tested was related to the evidence of non-linear associations between BMI and several cancers eg, oesophageal, colon, breast, and endometrial cancers. 44 We opted for a log-logit RR function for all cancer sites included in this study instead of a linear function, which partly addressed this issue. In the sensitivity analyses, we tested different RR functions, which had only small effects on the final PAF estimates. Alongside point estimates for the PAF, we presented 90% uncertainty intervals to provide a measure of reliability. However, these uncertainties do not take into account uncertainties in the cancer data from the GLOBOCAN 2012 database, which provides a qualitative ranking of data quality for each country-specific estimate. 23 Quantification of this uncertainty and incorporation of additional uncertainties from the modelling and estimation processes remains a major challenge and therefore was not attempted in our analysis. 48,49 Another limitation of this study includes the assumption of constant RRs across very diverse populations. The risks of some cancers associated with excess bodyweight have been reported to vary by ethnic group and geographical location. 14,50 Variation exists in the distribution of body fat between ethnic groups and how this is reflected in the BMI measure. For example, within the USA, African-American and Hispanic women are more likely to be obese than white and Asian-American women, yet white and Asian-American women have higher body fatness at similar BMIs. 51 Other anthropometric measures, such as waist circumference or waist-to-hip ratio, have been suggested as better predictors of obesity-related health outcomes than BMI. 52,53 Furthermore, rural and urban differences in the prevalence of obesity have been reported In our study, some variation in the distribution of BMI between ethnic groups might have been captured by our use of countryspecific BMI estimates. However, residual variation ie, within countries was not accounted for in the models. Because very little information is available for subnational populations such as ethnic groups and because of the absence of comparable global prevalence data for other anthropometric measures and their risk estimates, we could not do additional analyses to address these issues. Another drawback is the assumption of the absence of time-dependent effects of high BMI on cancer risk, which cannot be completely captured by age-specific BMI data and lag time. We assumed a 10-year lag in our modelling, but we recognise that the time-related effects of excess adiposity are likely to vary between cancer types. Results from some studies 57,58 have shown that the risk of cancer from high BMI accumulates with the number of years lived with excess weight, suggesting that the risk can better be predicted from years of life lived with high BMI. Longer duration of obesity has also been linked to other diseases and conditions, such as coronary artery 44 Vol 16 January 2015

10 calcification, a precursor of coronary heart disease. 59 Although this finding is in line with the biological mechanisms underlying the association between obesity and the development cancer, studies examining this aspect of obesity are a recent development and neither risk estimates nor the exposure information are available for every type of high-bmi-related cancer. Lastly, the estimation of the PAF is based on the assumption that the association between high BMI and each cancer type included in our study is causal. 60 We thus assume that reducing BMI will lead to a reduction in the incidence of these cancers. Excess bodyweight has been shown to increase circulating levels of oestrogens and bioactivity of IGF-1, hence promoting the development of cancer. 61 However, epidemiological studies that report risk associations between BMI and cancer are prone to several limitations. Residual confounding might account for the association between obesity and some types of cancer, and this possibility was not accounted for in our analysis. We have tried to overcome this issue by exclusively using risk estimates based on large meta-analyses that included only highquality studies and, whenever possible, only cohort studies. Based on our results, historical and continuing increases in the global prevalence of high BMI, especially in younger cohorts, are expected to translate into further increases in cancer burden in the future. Changes in the prevalence of strong effect modifiers such as HRT use are likely to increase the proportions of cancers attributable to high BMI, particularly among women. The large burden of cancers attributable to high BMI in North America,, Oceania, Latin America and the Caribbean, and the Middle East and north Africa points to the importance of weight-control programmes in these regions. Our results should inform health policy in terms of targets for prevention programmes, while emphasising existing gaps in our knowledge about the association between BMI and cancer. It also emphasises the need for research into effective interventions to control weight gain to avoid further increases in the burden of cancer related to high BMI. Contributors MA, NP, and IS contributed to data collection, study design, analysis, and wrote the first draft of the report. GB and AGR contributed to study design, analysis, and finalising the report. GAS and ME contributed to study design, data collection, analysis, and finalising the report. JF contributed to data collection and finalising the report. DF, RD, IR, and JJM contributed to study design and drafting of the report. All authors read and approved the final report. Declaration of interests We declare no competing interests. Acknowledgments This project was funded by the World Cancer Research Fund International (grant number SG 2012/619). NP was funded by the Australian National Health and Medical Research Council (NHMRC project 31691). IS was funded by a Marie Curie Intra-an Fellowship from the an Commission (project number ). We would like to thank Rachel Thompson and Martin Wiseman from World Cancer Research Fund International for their crucial input at several stages of the project. JJM is with the CRONICAS Centre of Excellence in Chronic Diseases at Cayetano Heredia University (Lima, Peru), supported by federal funds from the US National Heart, Lung And Blood Institute (National Institutes of Health, Department of Health and Human Services), under contract number HHSN C. GAS is a staff member of WHO. Data for body-mass index by hormone replacement therapy status were collected by the WHO MONICA investigators and have been made available for this publication by the WHO MONICA project. The views expressed in this report are those of the authors and do not necessarily represent the decisions, policy, or views of WHO, nor the views of individual investigators from the WHO MONICA project. References 1 Stevens GA, Singh GM, Lu Y, et al. National, regional, and global trends in adult overweight and obesity prevalences. Popul Health Metr 2012; 10: Ng M, Fleming T, Robinson M, et al. Global, regional, and national prevalence of overweight and obesity in children and adults during : a systematic analysis for the Global Burden of Disease Study Lancet 2014; 384: García-Alvarez A, Serra-Majem L, Ribas-Barba L, et al. Obesity and overweight trends in Catalonia, Spain ( ): gender and socio-economic determinants. Public Health Nutr 2007; 10: Sperrin M, Marshall AD, Higgins V, Buchan IE, Renehan AG. Slowing down of adult body mass index trend increases in England: a latent class analysis of cross-sectional surveys ( ). Int J Obes (Lond) 2014; 38: Flegal KM, Carroll MD, Kit BK, Ogden CL. Prevalence of obesity and trends in the distribution of body mass index among US adults, JAMA 2012; 307: Norberg M, Lindvall K, Stenlund H, Lindahl B. The obesity epidemic slows among the middle-aged population in Sweden while the socioeconomic gap widens. Glob Health Action 2010; 3: Faeh D, Bopp M. Excess weight in the canton of Zurich, : harbinger of a trend reversal in Switzerland? Swiss Med Wkly 2010; 140: w WCRF, AICR. Continuous update project report. Food, nutrition, physical activity, and the prevention of breast cancer. World Cancer Research Fund and American Institute for Cancer Research, WCRF, AICR. Continuous update project report. Food, nutrition, physical activity, and the prevention of colorectal cancer. World Cancer Research Fund and American Institute for Cancer Research, WCRF, AICR. Continuous update project report. Food, nutrition, physical activity, and the prevention of pancreatic cancer. World Cancer Research Fund and American Institute for Cancer Research, WCRF, AICR. Continuous update project report. Food, nutrition, physical activity, and the prevention of endometrial cancer. World Cancer Research Fund and American Institute for Cancer Research, WCRF, AICR. Continuous update project report. Food, nutrition, physical activity, and the prevention of ovarian cancer. World Cancer Research Fund and American Institute for Cancer Research, WCRF, AICR. Food, nutrition, physical activity, and the prevention of cancer: a global perspective. Washington, DC: World Cancer Research Fund and American Institute for Cancer Research, Renehan AG, Tyson M, Egger M, Heller RF, Zwahlen M. Body-mass index and incidence of cancer: a systematic review and meta-analysis of prospective observational studies. Lancet 2008; 371: Lim SS, Vos T, Flaxman AD, et al. A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, : a systematic analysis for the Global Burden of Disease Study Lancet 2012; 380: Parkin E, O Reilly DA, Sherlock DJ, Manoharan P, Renehan AG. Excess adiposity and survival in patients with colorectal cancer: a systematic review. Obes Rev 2014; 15: For the WHO MONICA investigators see thl.fi/publications/monica/ investigators.htm Vol 16 January