published edition

8/14/2019 Published Edition

1/10

Articles

www.thelancet.com Vol 371 February 16, 2008 569

Body-mass index and incidence of cancer: a systematic

review and meta-analysis of prospective observationalstudies

Andrew G Renehan, Margaret Tyson, Matthias Egger, Richard F Heller, Marcel Zwahlen

SummaryBackground Excess bodyweight, expressed as increased body-mass index (BMI), is associated with the risk of somecommon adult cancers. We did a systematic review and meta-analysis to assess the strength of associations betweenBMI and different sites of cancer and to investigate differences in these associations between sex and ethnic groups.

Methods We did electronic searches on Medline and Embase (1966 to November 2007), and searched reports toidentify prospective studies of incident cases of 20 cancer types. We did random-effects meta-analyses and

meta-regressions of study-specific incremental estimates to determine the risk of cancer associated with a 5 kg/mincrease in BMI.

Findings We analysed 221 datasets (141 articles), including 282 137 incident cases. In men, a 5 kg/m increase inBMI was strongly associated with oesophageal adenocarcinoma (RR 152, p


2/10

Articles

570 www.thelancet.com Vol 371 February 16, 2008

for 20 cancer types in 15 sites in the body: colorectal(colon and rectum); gastro-oesopheageal (gastric,oesophageal adenocarcinoma, and squamous cellcarcinoma); hepatobiliary (gallbladder and liver);

leukaemia; lung; malignant melanoma; multiple

myeloma; non-Hodgkin lymphoma; pancreas; renal andthyroid for both genders; and prostate, breast (premeno-pausal and postmenopausal), endometrium, and ovaryfor single genders. Our core search consisted of termsrelated to bodyweight (obesity, adiposity, body massindex, and body size), combined with specific termsfor each cancer site (webappendix 1). If a site-specificdataset had been published more than once, we used themost recent publication. We scrutinised the referencelists of the identified reports,1,2 reviews,4,2426meta-analyses,3,519 and other relevant publications to findadditional pertinent studies. A research librarian whodid an independent search for one site (endometrium)did not find any additional studies that met the inclusion

criteria.We included cohort studies if they determined BMI at

baseline and then recorded incident cancer cases duringfollowup. We specified that every cohort study musteither report risk estimates (relative risks, odds ratios,or hazard ratios) with 95% CIs separately for men,women, or both across at least three categories of BMIor must report sufficient data to estimate these.[A:okay?] We also included casecontrol studies nested insuch cohort studies and control arms from clinicaltrials. We included studies in which height and weight(to calculate BMI) had been self-reported, and those inwhich they had been directly measured.

The eligibility of each study was assessedindependently by two investigators (AGR and MT). Weexcluded studies that were not published as full reports,such as conference abstracts and letters to editors,studies of cancer mortality (rather than incidence),studies of cancer precursors (for example colorectaladenoma), and studies that reported results only for allbreast or all colorectal cancers.

Data extraction and quality assessmentOne investigator (AGR) extracted data, which waschecked by two others (MT and MZ). We used informationabout: study design and patient characteristics, and riskestimates and their 95% CIs (either with one BMI

category as a referent group or expressed as a slope perincremental BMI increase [standardised to 5 kg/mincrements]). We collected data for both minimallyadjusted and maximally adjusted risk estimates, ifavailable. Populations were categorised into five groups:North American (greater than 80% White), Europeanand Australian, AsiaPacific (including Japanesepopulations based in Hawaii), Multi-ethnic, and BlackAmerican.We extracted the mean BMI (and its standarddeviation) by sex for each study, or, if these data weremissing, used sex-specific and population-specific values(webappendix 2).

Methodological quality was assessed according tothree study components which might affect the strengthof the association between BMI and cancer risk: length

_

7

7

4825 citations in Medline and Embase found by electronic search

270 found by othersearch methods

897 given more detailed assessment

438 comprehensively assessedagainst inclusion criteria

221 datasets for final meta-analysis29 colorectal

10 gastro-oesophageal5 hepatobiliary7 leukaemia

13 lung7 melanoma

10 multiple myeloma9 non-Hodgkin lymphoma

16 pancreas17 renal

5 thyroid27 prostate

34 breast19 endometrial13 ovarian

217 did not meet criteria47 included fatal cases only40 were duplicates26 did not distinguish gender, site,

or menopausal status42 did not adequately report

association with BMI

6 used less than three BMI categories40 used a non-scalar definition of

bodyweight16 for other reasons

3658 excluded on first pass

459 excluded on second pass41 reviews or meta-analyses

320 casecontrol studies98 no reported association with BMI

Figure 1: Flow diagram of search strategy and study selectionNumbers refer to datasets, rather than studies. NHL: non-Hodgkin lymphoma.

35000

N

umberofcases(logscale)

Year of publication

1988 1992 1996 2000 2004 2007

10000

1000

100

North American

European and Australian

AsiaPacific

Multi-ethnic

Black American

Figure 2: Datasets by year and population group

Size of circle is proportional to sample size.

See Online for webappendices 1

and 2


3/10

Articles


of follow-up; whether BMI was self-reported ormeasured; and the extent of adjustments for potentialconfounding factors (webappendix 2).

Statistical analysisWe transformed category-specific risk estimates intoestimates of the risk ratio (RR) associated with every5 kg/m increase in BMI by use of the method ofgeneralised least-squares for trend estimation.27 Theseestimates were calculated from the assumption of alinear relation between the natural logarithm of risk ratioand increasing BMI. The value assigned to each BMIcategory was the mid-point for closed categories, and themedian for open categories (assuming a normaldistribution for BMI).28 We combined the risk ratios foreach 5 kg/m increase in BMI by use of random-effectmeta-analysis.29 Unless otherwise stated, we used themost adjusted risk estimate from each study. We assessed

heterogeneity between studies with the I statistic30 as ameasure of the proportion of total variation in estimatesthat is due to heterogeneity, where I values of 25%, 50%,

and 75% correspond to cut-off points for low, moderate,and high degrees of heterogeneity.

We did meta-regression analyses for each site toidentify study-level factors that modify the associationbetween increased BMI and cancer risk, and contributeto heterogeneity between studies.31 For sensitivityanalyses, we repeated our analyses with a fixed-effectsmodel, used minimally adjusted risk ratios, andestimated the median for open BMI categories withdifferent methods.32 We also did influence-analyses toassess the effect of each study on the summary riskestimates.32 Furthermore, we explored threshold effectsacross BMI ranges using splines within the generalisedleast-squares for trend estimation models.27 Publicationbias was examined in funnel plots and with a regression

Number of

datasets*

Population group Number of

cases in men

Number of

cases inwomen

Total

samplesize

Number that

measured BMIdirectly

Median number of

potential cancer-specificconfounders in analysis

Geometric mean

duration offollow-up (years)

North

America

Europe and

Australia

AsiaPacific

Colorectal cancer 29 11 12 6 4 833 139 16 2 (0 to 6) 110 (91133)

Colon 22 440 20 975

Rectum 14 894 9052

Gastro-oesophagealcancers

10 0 8 2 4 673 213 8 2 (1 to 3) 108 (74160)

Gastric 817 325

Oesophageal

adenocarcinoma

1315 735

Oesophageal squamous

cell carcinoma

6201 1114

Hepatobiliary cancers 5 0 3 2 3 319 024 4 1 (1 to 1) 127 (70231)Gallbladder 928 1111

Liver 2039 31

Leukaemia 7 1 5 1 3371 5317 4 757 649 4 1 (1 to 3) 137 (77245)

Lung cancer 13 1 8 4 7426 4273 2 649 345 10 3 (1 to 4) 119 (85166)

Malignant melanoma 7 1 5 1 3492 4786 3 966 859 5 1 (1 to 1) 106 (71157)

Multiple myeloma 10 4 4 1 4273 3664 5 171 374 3 1 (1 to 2) 146 (107200)

Non-Hodgkin lymphoma 9 2 6 1 7041 6248 5 043 747 3 1 (1 to 2) 124 (86178)

Pancreatic cancer 16 4 8 3 2390 2053 3 338 001 6 3 (2 to 5) 94 (77114)

Renal cancer 17 7 7 2 6073 4614 5 473 638 10 2 (1 to 5) 106 (85133)

Thyroid cancer 5 0 4 1 1212 2375 3 303 073 5 1 (1 to 2) 144 (72288)

Prostate cancer 27 12 10 5 70 421 3 029 338 14 2 (1 to 3) 106 (86131)

Breast cancer 34 12 16 5 2 559 829 15 55 (1 to 11) 84 (71100)

Premenopausal 7930Postmenopausal 23 909

Endometrial cancer 19 5 12 1 17 084 3 044 538 12 2 (1 to 6) 106 (82138)

Ovarian cancer 13 4 7 2 12 208 2 703 734 5 3 (1 to 4) 122 (88169)

Data are number, median (range), or geometric mean (95% CI). BMI=body-mass index. * Dataset refers to a site-specific cohort per paper. Several papers reported multiple sites. If a paper reported two separate

cohorts (e.g. one each for men and women) for the same site, these were counted as two datasets. These sites were grouped together in the literature search, since site-specific estimates were frequently

reported in the same article. Totals do not equal sum of population groups since they include multiethnic populations: one each for pancreatic, renal, and endometrial cancers. Totals do not equal sum ofpopulations groups since they include Black American population: one each for multiple myeloma and breast cancer.

Table 1: Baseline characteristics for studies included in meta-analysis


4/10

Articles


asymmetry test.33 We used STATA version 9.0 (College

Station, TX, USA) to analyse data.

Role of the funding sourceThe sponsor of the study had no role in study design,data collection, analysis, interpretation, or writing of the

report. The corresponding author had full access to all

data in the study and had final responsibility for thedecision to submit for publication.

ResultsOf 4285 citations, we identified 221 datasets from141 articles which met the inclusion criteria (seewebappendix 3). Figure 1 shows our search and selectionprocess, and webappendix 4 lists excluded articles withreasons for their exclusion. Agreement betweenobservers on which studies to include was good(Cohens unweighted =086). All papers used in ouranalysis were published in English, except for one thatwas written in Chinese.34Figure 2 shows that more thanhalf the papers (73/141) were published since 2004, that

many of the larger datasets were from recentpublications, and that only a few studies fromAsiaPacific regions, and most with small numbers,were before 2004. The 141 papers reported on 76 studies(67 cohort studies, six nested casecontrol studies, andthree randomised trials). 28 studies were from NorthAmerica, 35 from Europe and Australia, and 11 fromAsiaPacific. One cohort was multi-ethnic (threepapers),3537 and two cohorts (one as a subcohort withinone article) analysed black American populations.38,39

Table 1 summarises the characteristics of includedstudies; the webtable has more details. The analysisincluded 282 137 incident cases (154 333 men and127 804 women), over more than 133 000 000 person-years of follow-up. The geometric mean follow-up percancer site varied from 84 years (breast cancer) to144 years (multiple myeloma). Notably, no NorthAmerican population data contributed to the summariesfor gallbladder, gastric, liver, oesophageal, or thyroidcancers. The proportion of studies in which BMI wasmeasured directly varied by cancer site, as did themedian number of potential confounders that wereincluded in adjusted analyses.

Figures 3 and 4 show the results of meta-analyses ofrisk ratios (per 5 kg/m increase in BMI) in men and inwomen. Separate meta-analyses for each site are inwebappendix 5. In men, increased BMI was strongly

associated with oesophageal adenocarcinoma (RR 152,p


5/10

Articles


increased BMI and leukaemia (117, p=001), and cancersof the thyroid (114, p=00001), postmenopausal breastcancer (112, p


6/10

Articles


effect size (across all sites) between studies in which

women self-reported their measurements and those inwhich weight and height were directly measuredwas 106 (95% CI 102109). In men, the additionaleffect size was 102 (098105). Other study-levelvariables, such as the year of publication, mean age atbaseline, mean BMI at baseline, and the duration offollow-up, had little effect on the association betweenincreased BMI and cancer.

Results were generally consistent when we repeatedanalyses with a fixed-effects model rather than arandom-effects model (webappendix 8). Risk ratioswere somewhat attenuated for liver (112, 105119)and thyroid (119, 108130) cancers in men, and forgallbladder (137, 126148) cancer in women.

Repetition of analyses with minimally adjusted riskratios rather than maximally adjusted estimates did notproduce different results. We examined to what extentresults were affected by the way we assigned midpointsto open upper BMI categories (webappendix 8). Resultsbased on our assigned midpoints did not differ fromthose based on midpoints reported in individual studies(p=098, based on 12 datasets). We repeated analyseswith a gamma distribution to estimate the median foropen upper BMI categories: results were similar tothose from analyses with a normal distribution. Weexplored the possibility of threshold effects across BMIranges by use of splines in generalised-least-squares fortrend estimation models. When we adjusted models forpopulation group, method of BMI determination, andextent of cancer-specific risk-factor adjustment, theassociation between increased BMI and the risk ofendometrial cancer was non-linear: the increase in riskassociated with each 5 kg/m increase in BMI wasgreater above 28 kg/m (RR 304, 231401).

Influence analyses showed that, for most of sites, nosingle studies affected the sex-specific summaryestimates (webappendix 9). Finally, we noted funnelplot asymmetry for colon cancer in men and women:increased BMI had large effects on cancer risk in smallstudies (p=0002 and 0006, respectively) but littleevidence for other sites. To test whether this could have

affected results, we repeated analyses excluding studieswith less than 150 cases (seven in men and eight inwomen). These analyses produced similar summaryestimates and did not affect the strength of evidence fordifferences between men and women in the BMIcancerrisk association for colon cancer (webappendix 10).

DiscussionOur large standardised meta-analysis shows thatincreased BMI is associated with an increased risk ofseveral cancers in adults. Our findings extend theresults of previous reports,1,2 to show associationsbetween increased BMI and cancer risk for less commonmalignancies, and evidence that associations mightdiffer between men and women for some sites, in

Studies Risk ratio* p value

(univariablemodel)

p value

(multivariablemodels)

Men

Colon 22 020 035

North American 6 135 (121150)

European and Australian 10 121 (118124)

AsiaPacific 6 132 (120146)

Rectum 18 054 018




Pancreas 11 004 014




Renal 11 049 050




Prostate 27 0003 020




Women

Colon 19 004 009




Rectum 14 082 082




Pancreas 10 062 076




Premenopausal breast 19 001 0009




Postmenopausal breast 30 004 006




Ovarian 13 040 029


European and Australia 7 103 (098107)


Data are number or RR (95% CI), unless otherwise specified. *Risk ratios per 5 kg/m increase in BMI. Meta-regression

regression analyses with region but no other variables. Meta-regression regression analyses with the method of BMI

determination (measured or self-reported), and the extent of adjustment for risk factors specific to cancer sites. Data

for renal cancer in women not analysed since only one AsiaPacific dataset. Denotes North American White

populations since too few studies of Black Americans for subanalysis.

Table 3: Comparison of risk ratios for different cancer sites between main population groups


7/10

Articles


particular for colon cancer. The magnitudes of

associations between increased BMI and cancer weresimilar across populations for most cancer sites.However, the association was particularly strong forbreast cancer in AsiaPacific populations, which needsconfirmation from further studies.

Meta-analyses of observational studies are prone tobiases and confounding factors that are inherent inthese studies.47 We restricted our analyses to prospectivestudies and the casecontrol studies nested withinthem, and excluded traditional casecontrol studies,which are prone to recall and interviewer bias.48Furthermore, we assessed the methodological quality ofcomponent studies and explored sources of heterogeneitywith meta-regression models. We showed that the

method by which BMI is determined was a source ofheterogeneity in results for cancer risk at several sites,which is consistent with the finding that self-reportedweight is lower than true bodyweight. In turn,underestimation of weight varies with age (increases inolder individuals), and with relative baseline BMI(increases with higher BMI).49Our results indicate thatthis affected associations in female populations.Extensive sensitivity analyses indicated that results wererobust to changes in model assumptions.

Increased BMI was associated with some cancers, butnot others: the specificity of these associations arguesagainst confounding and bias, and for a possible causallink between increased BMI and the risk of developingsome cancers. Summary estimates from maximally andminimally adjusted analyses were similar, indicatingthat for the cancers we studied, the effect of BMI oncancer risk is not confounded by other included factorsAlternatively, important confounding factors might nothave been measured with sufficient precision, or nothave been measured at all in these studies. For somecancer types, smoking seemed to be a major confounderin the association of increased BMI with risk. This wasexemplified in the case of lung cancer, for whichincreased BMI was not associated with cancer risk inthose who have never smoked, but was inverselyassociated in smokers. Similarly, increased BMI had a

strong inverse association with squamous cellcarcinoma of the oesophagus, which is more stronglyassociated with smoking than is oesophagealadenocarcinoma.50 However, we could not determinethe effect of smoking on risk of squamous cellcarcinoma of the oesophagus, since too few studieswere stratified by smoking status. For postmenopausalrisk of breast cancer, studies of postmenopausal womenonly had similar results to those from cohorts with bothpremenopausal and postmenopausal women,suggesting that excess bodyweight is relevant to riskboth before and after menopause. However, the use ofhormone replacement therapy51,52 and mammographicdensity53 could be additional confounders. This needsfurther study.

Anthropometric measures other than increased BMI,

for example waist-to-hip ratio or waist circumference,might be better measures of adiposity in terms ofcancer risk, as is the case for cardiovascular risk.54 Twoprevious meta-analyses (mixed casecontrol and cohortstudies)55,56 reported positive associations betweenwaist-to-hip ratio and premenopausal breast cancer. Inour review, too few studies determined waist-to-hipratio or waist circumference to permit comprehensiveanalyses of such associations across several sites.

Several meta-analyses have quantified associationsbetween BMI and cancer risk at specific sites. Somehave quantified associations separately for men andwomen;3,5,812,1519 others for men and women com-bined.4,14,57 Since our analysis raised the probability of

differences between sexes at several sites, future studiesshould report BMIcancer risk associations separatelyby sex. Inclusion criteria for studies differed and severalreviews included both conventional casecontrol studiesand cohort studies.35,9,10,12,1518 Furthermore, severalmeta-analyses3,5,8,9,11,15,17,18 combined cohort studies ofcancer deaths with studies of cancer incidence.

The WRCF review2 also examined associations forseveral cancer types. However, by contrast with thatreview, we reported our results as sex-specific estimates;addressed differences across populations; and calculatedrisk estimates for several additional cancer types:leukaemia, malignant melanoma, multiple myeloma,non-Hodgkin lymphoma, and thyroid cancer. Moreover,despite attempts to standardise methodologicalprocesses across different centres, selection of studiesfor the WCRF review was inconsistent. For example,studies of cancer mortality were included in the analysesof cancers of the pancreas, endometrium, andgallbladder, but not for oesophageal adenocarcinoma,kidney, colon, or breast cancers. Because obesity hasbeen associated with poor prognosis in, for examplebreast,58 colon,59,60 endometrium,61 ovary,62 and prostate63cancers (and with a favourable prognosis in renal callcarcinoma64), we restricted our analyses to studies ofincident cancers. Our combined risk estimates weregenerally more conservative than estimates from

previous reviews, and provided only weak evidence forthe association of increased BMI with the risk ofgallbladder, 9 pancreatic,5,8 and prostate15 cancers in men,and for ovarian cancer16,17 in women.

Mechanisms that link excess weight and cancer riskare not fully understood, though three hormonalsystemsthe insulin and insulin-like growth factor(IGF) axis, sex steroids, and adipokinesare the moststudied candidates. All three systems are interlinkedthrough insulin; however, their roles might varybetween cancer sites. The insulincancer hypothesispostulates that chronic hyperinsulinaemia decreasesconcentrations of IGF binding protein-1 and IGFbinding protein-2, which increases bioavailable or freeIGF-I with concomitant changes in the cellular

See Online for webappendix 8




8/10

Articles


environment (mitogenesis and anti-apoptosis) that

favour tumour formation.

65

Circulating total IGF-I,which is a major determinant of free IGF-Iconcentrations, is also associated with an increased riskof colorectal and prostate cancers, and withpremenopausal rather than postmenopausal breastcancer.66 Mean circulating concentrations of total IGF-Iare higher in men than in women,67 which could partlyexplain the differences between sexeseg, in colorectalcancer risk.

For postmenopausal breast cancer, the increase inrisk might be explained by the higher rates of conversionof androgenic precursors to oestradiol throughincreased aromatase enzyme activity in adipose tissue.More than one system might affect the risk of

endometrial cancer: increased oestradiol not onlyincreases endometrial cell proliferation and inhibitsapoptosis, but might also stimulate the local synthesisof IGF-I in endometrial tissue.25 Furthermore, chronichyperinsulinaemia might promote tumorigenesis inoestrogen-sensitive tissues, since it reduces bloodconcentrations of sex-hormone-binding globulin, andin turn, increases bioavailable oestrogen.25 Adiposity isinversely related to testosterone concentrations inmen,68 but positively related in women;69 which couldbe relevant to sex differences in the association betweenBMI and cancer risk.

Adiponectin is the most abundant adipokine. It issecreted mainly from visceral fat adipocytes, and isinversely correlated with BMI. Mean circulatingconcentrations are higher in women than men. Interms of tumour development, this insulin-sensitisingagent is antiangiogenic and anti-inflammatory, andinhibits tumour growth in animals.70 Inverseassociations between adiponectin concentrations andcancer risk have been reported in some studies inpeople.70 Beyond these mechanisms, other candidatesystems include obesity-related inflammatory cytokines,altered immune response, oxidative stresses, thenuclear factor B system,65 hypertension and lipidperoxidation for renal cancer,71 and acid-reflux foroesophageal adenocarcinoma. We do not yet know what

mechanisms might link the less common malignancieswith obesity.

In US adults, 71% of men and 62% of women areoverweight or obese (with a BMI of more than25 kg/m);72 in the UK, 65% of men and 56% of womenare overweight or obese.73 Moreover, these prevalencesare expected to increase, in the UK, for example, to75% of men and 58% of women by 2010.73 Excessbodyweight could therefore contribute to a substantiallylarger burden of cancer in such populations. We havemodelled a 5 kg/m increase in BMI, which correspondsto weight gains of about 15 kg in men and 13 kg inwomen who have an average BMI of 23 kg/m. Many ofthe observed associations between increased BMI andcancer risk are for cancers that are not related to

smoking. Conceivably, as cigarette smoking (which is

the largest cause of cancers in developed countries)decreases, excess bodyweight could become thedominant lifestyle factor that contributes to canceroccurrence in such countries.

Important questions remain about the cumulativeeffects of excess bodyweight over several decades, theeffect of key weight-change periods in the l ife-course ofindividuals, and interactions with other risk factors.74Other unresolved questions relate to the mostappropriate measure of adiposity in terms of cancerrisk, the mechanisms that underpin sex differences,and differences across ethnicities. Finally, we need toknow whether effective interventions to reduce BMI inadult populations will reduce cancer risks. This

knowledge will allow the formulation of public-health strategies to prevent obesity-related cancersworldwide.

Contributors

AGR contributed to protocol design, data extraction, quality assessment,statistical analysis, and writing the report. MT contributed to protocoldesign, data extraction, and writing the report. ME contributed toprotocol design, quality assessment, statistical analysis, and revision ofthe report. MZ contributed to data extraction, statistical analysis, andwriting the report. RFH contributed to interpretation of data andrevision of the report. All authors have seen and approved the finalversion.

Conflict of interest statement

AGR has received hospitality from Diagnostic Systems Laboratoriesand from several pharmaceutical companies that make hormone

replacements, and a lecture honorarium from Eli-Lilly.Acknowledgments

This study was partly funded by an award to AGR from the BritishMedical Association. We thank Dr Jian Mei Hou, from the PatersonInstitute for Cancer Research, Manchester, for translation of the paperin Chinese; Anne Webb, from the Kostoris Library, at the ChristieHospital, Manchester, for her independent literature search; andDr Chien-An Sun for sharing unpublished data of risk estimatesseparately for premenopausal and postmenopausal breast cancer fromthe published Taiwan Breast Cancer Study.

References1 IARC. Weight control and physical activity. eds. Vainio H,

Bianchini F. Lyon, France: International Agency for Research inCancer Press, 2002.

2 WCRF. World Cancer Research Fund. Food, Nutrition, PhysicalActivity, and the Prevention of Cancer: a Global Perspective.2nd edn. Washington, USA: American Institute for Cancer

Research, 2007.3 Bergstrm A, Hsieh CC, Lindblad P, Lu CM, Cook NR, Wolk A.

Obesity and renal cell cancer a quantitative review. Br J Cancer2001; 85: 98490.

4 Bergstrom A, Pisani P, Tenet V, Wolk A, Adami HO. Overweightas an avoidable cause of cancer in Europe. Int J Cancer2001;91: 42130.

5 Berrington de Gonzalez A, Sweetland S, Spencer E. Ameta-analysis of obesity and the risk of pancreatic cancer.Br J Cancer2003; 89: 51923.

6 Dal Maso L, La Vecchia C, Franceschi S, et al. A pooled analysis ofthyroid cancer studies. V. Anthropometric factors. Cancer Causes Control2000; 11: 13744.

7 Kubo A, Corley DA. Body mass index and adenocarcinomas of theesophagus or gastric cardia: a systematic review andmeta-analysis. Cancer Epidemiol Biomarkers Prev2006; 15: 87278.

8 Larsson SC, Orsini N, Wolk A. Body mass index and pancreatic

cancer risk: a meta-analysis of prospective studies. Int J Cancer2007; 120: 199398.


9/10

Articles


9 Larsson SC, Wolk A. Obesity and the risk of gallbladder cancer: ameta-analysis. Br J Cancer2007; 96: 145761.

10 Larsson SC, Wolk A. Obesity and risk of non-Hodgkinslymphoma: A meta-analysis. Int J Cancer2007; 122: 156470.

11 Larsson SC, Wolk A. Obesity and colon and rectal cancer risk: ameta-analysis of prospective studies. Am J Clin Nutr2007; 86: 55665.

12 Larsson SC, Wolk A. Overweight, obesity and risk of liver cancer:a meta-analysis of cohort studies. Br J Cancer2007; 97: 100508.

13 Larsson SC, Wolk A. Body mass index and risk of multiplemyeloma: A meta-analysis. Int J Cancer2007; 121: 251216.

14 Larsson SC, Wolk A. Overweight and obesity and incidence ofleukemia: a meta-analysis of cohort studies. Int J Cancer2007;published online Nov 2007. DOI: 10.1002/ijc.23176.

15 MacInnis RJ, English DR. Body size and composition andprostate cancer risk: systematic review and meta-regressionanalysis. Cancer Causes Control2006; 17: 9891003.

16 Olsen CM, Green AC, Whiteman DC, Sadeghi S, Kolahdooz F,Webb PM. Obesity and the risk of epithelial ovarian cancer: asystematic review and meta-analysis. Eur J Cancer2007;

43: 690709.17 Purdie DM, Bain CJ, Webb PM, Whiteman DC, Pirozzo S,

Green AC. Body size and ovarian cancer: case-control study andsystematic review (Australia). Cancer Causes Control2001;12: 85563.

18 Ursin G, Longnecker MP, Haile RW, Greenland S. Ameta-analysis of body mass index and risk of premenopausalbreast cancer. Epidemiology1995; 6: 13741.

19 van den Brandt PA, Spiegelman D, Yaun SS, et al. Pooled analysisof prospective cohort studies on height, weight, and breast cancerrisk. Am J Epidemiol2000; 152: 51427.

20 Reeves GK, Pirie K, Beral V, Green J, Spencer E, Bull D. Cancerincidence and mortality in relation to body mass index in theMillion Women Study: cohort study. BMJ2007; published onlineNov 2007. DOI: 10.1136/bmj.39367.495995.AE.

21 Iwasaki M, Otani T, Inoue M, Sasazuki S, Tsugane S. Body sizeand risk for breast cancer in relation to estrogen andprogesterone receptor status in Japan. Ann Epidemiol2007;

17: 30412.22 Kuriyama S, Tsubono Y, Hozawa A, et al. Obesity and risk of

cancer in Japan. Int J Cancer2005; 113: 14857.

23 Oh SW, Yoon YS, Shin SA. Effects of excess weight on cancerincidences depending on cancer sites and histologic findingsamong men: Korea National Health Insurance Corporation Study.J Clin Oncol2005; 23: 474254.

24 Bianchini F, Kaaks R, Vainio H. Overweight, obesity, and cancerrisk. Lancet Oncol2002; 3: 56574.

25 Calle EE, Kaaks R. Overweight, obesity and cancer:epidemiological evidence and proposed mechanisms.Nat Rev Cancer2004; 4: 57991.

26 Kaaks R, Lukanova A, Kurzer MS. Obesity, endogenoushormones, and endometrial cancer risk: A synthetic review.Cancer Epidemiol Biomarkers Prev2002; 11: 153143.

27 Orsini N, Bellocco R, Greenland S. Generalized least squares fortrend estimation of summarized dose-response data. Stata J2006;

6: 4057.28 Haftenberger M, Lahmann PH, Panico S, et al. Overweight,obesity and fat distribution in 50 to 64-year-old participants in theEuropean Prospective Investigation into Cancer and Nutrition(EPIC). Public Health Nutr2002; 5: 114762.

29 DerSimonian R, Laird N. Meta-analysis in clinical trials.Control Clin Trials 1986; 7: 17788.

30 Higgins JP, Thompson SG. Quantifying heterogeneity in ameta-analysis. Stat Med2002; 21: 153958.

31 Thompson SG, Higgins JP. How should meta-regression analysesbe undertaken and interpreted? Stat Med2002; 21: 155973.

32 Tobias A. Assessing the influence of a single study inmeta-analysis. Stata Tech Bull1999; 47: 1517.

33 Egger M, Davey Smith G, Schneider M, Minder C. Bias inmeta-analysis detected by a simple, graphical test. BMJ1997;315: 62934.

34 Li HL, Gao YT, Li Q, Liu DK. [Anthropometry and female breastcancer: a prospective cohort study in urban Shanghai].

Zhonghua Liu Xing Bing Xue Za Zhi 2006; 27: 48893.

35 Nothlings U, Wilkens LR, Murphy SP, Hankin JH,Henderson BE, Kolonel LN. Body mass index and physical activity

as risk factors for pancreatic cancer: the Multiethnic Cohort Study. Cancer Causes Control2007; 18: 16575.

36 Setiawan VW, Pike MC, Kolonel LN, Nomura AM, Goodman MT,Henderson BE. Racial/ethnic differences in endometrial cancerrisk: the multiethnic cohort study. Am J Epidemiol2007; 165: 26270.

37 Setiawan VW, Stram DO, Nomura AM, Kolonel LN,Henderson BE. Risk factors for renal cell cancer: the multiethniccohort. Am J Epidemiol2007; 166: 93240.

38 Friedman GD, Herrinton LJ. Obesity and multiple myeloma.Cancer Causes Control1994; 5: 47983.

39 Palmer JR, Adams-Campbell LL, Boggs DA, Wise LA,Rosenberg L. A prospective study of body size and breast cancerin black women. Cancer Epidemiol Biomarkers Prev2007;16: 1795802.

40 Canoy D, Wareham N, Luben R, et al. Cigarette smoking and fatdistribution in 21,828 British men and women: apopulation-based study. Obes Res 2005; 13: 146675.

41 Kanashiki M, Sairenchi T, Saito Y, Ishikawa H, Satoh H,Sekizawa K. Body mass index and lung cancer: a case-controlstudy of subjects participating in a mass-screening program.Chest2005; 128: 14906.

42 Kark JD, Yaari S, Rasooly I, Goldbourt U. Are lean smokers atincreased risk of lung cancer? The Israel Civil Servant CancerStudy. Arch Intern Med1995; 155: 240916.

43 Olson JE, Yang P, Schmitz K, Vierkant RA, Cerhan JR, Sellers TA.Differential association of body mass index and fat distributionwith three major histologic types of lung cancer: evidence from acohort of older women. Am J Epidemiol2002; 156: 60615.

44 Samanic C, Chow WH, Gridley G, Jarvholm B, Fraumeni JF, Jr.Relation of body mass index to cancer risk in 362,552 Swedishmen. Cancer Causes Control2006; 17: 9019.

45 Bjorge T, Tretli S, Engeland A. Relation of height and body massindex to renal cell carcinoma in two million Norwegian men andwomen. Am J Epidemiol2004; 160: 116876.

46 Engeland A, Tretli S, Austad G, Bjorge T. Height and body mass

index in relation to colorectal and gallbladder cancer in twomillion Norwegian men and women. Cancer Causes Control2005;16: 98796.

47 Egger M, Schneider M, Davey Smith G. Spurious precision?Meta-analysis of observational studies. BMJ1998; 316: 1404.

48 Schlesselman JJ. Case-control studies. Design, conduct, analysis.New York, USA: Oxford University Press, 1982.

49 Kuczmarski MF, Kuczmarski RJ, Najjar M. Effects of age onvalidity of self-reported height, weight, and body mass index:findings from the Third National Health and NutritionExamination Survey, 19881994.J Am Diet Assoc2001; 101: 2836.

50 IARC. Tobacco Smoke and Involuntary Smoking. IARCMonographs on the Evaluation of Carcinogenic Risks to Humans.Lyon, France: International Agency for Research on Cancer, 2004.

51 Lahmann PH, Hoffmann K, Allen N, et al. Body size and breastcancer risk: Findings from the European ProspectiveInvestigation into Cancer and Nutrition (EPIC). Int J Cancer2004;111: 76271.

52 Mellemkjaer L, Bigaard J, Tjonneland A, et al. Body compositionand breast cancer in postmenopausal women: a Danishprospective cohort study. Obesity (Silver Spring) 2006; 14: 185462.

53 Boyd NF, Martin LJ, Sun L, et al. Body size, mammographicdensity, and breast cancer risk. Cancer Epidemiol Biomarkers Prev2006; 15: 208692.

54 Yusuf S, Hawken S, Ounpuu S, et al. Obesity and the risk ofmyocardial infarction in 27,000 participants from 52 countries: acase-control study. Lancet2005; 366: 164049.

55 Connolly BS, Barnett C, Vogt KN, Li T, Stone J, Boyd NF. Ameta-analysis of published literature on waist-to-hip ratio and riskof breast cancer. Nutr Cancer2003; 44: 12738.

56 Harvie M, Hooper L, Howell AH. Central obesity and breastcancer risk: a systematic review. Obes Rev2003; 4: 15773.

57 Hampel H, Abraham NS, El-Serag HB. Meta-analysis: obesity andthe risk for gastroesophageal reflux disease and its complications.Ann Intern Med2005; 143: 199211.

58 Carmichael AR. Obesity and prognosis of breast cancer. Obes Rev2006; 7: 33340.


10/10

Articles

578 www thelancet com Vol 371 February 16 2008

59 Dignam JJ, Polite BN, Yothers G, et al. Body mass index andoutcomes in patients who receive adjuvant chemotherapy for

colon cancer.J Natl Cancer Inst2006; 98: 164754.60 Haydon AM, Macinnis RJ, English DR, Morris H, Giles GG.

Physical activity, insulin-like growth factor 1, insulin-like growthfactor binding protein 3, and survival from colorectal cancer. Gut2006; 55: 68994.

61 von Gruenigen VE, Tian C, Frasure H, Waggoner S, Keys H,Barakat RR. Treatment effects, disease recurrence, and survival inobese women with early endometrial carcinoma: a GynecologicOncology Group study. Cancer2006; 107: 278691.

62 KjaerbyeThygesen A, Frederiksen K, Hogdall EV, et al. Smokingand overweight: negative prognostic factors in stage III epithelialovarian cancer. Cancer Epidemiol Biomarkers Prev2006;15: 798803.

63 Gong Z, Agalliu I, Lin DW, Stanford JL, Kristal AR. Obesity isassociated with increased risks of prostate cancer metastasis anddeath after initial cancer diagnosis in middle-aged men. Cancer2007; 109: 1192202.

64 Awakura Y, Nakamura E, Ito N, et al. Influence of body massindex on prognosis of Japanese patients with renal cell carcinoma.Urology2007; 70: 5054.

65 Renehan AG, Frystyk J, Flyvbjerg A. Obesity and cancer risk: therole of the insulinIGF axis. Trends Endocrinol Metab 2006;17: 32836.

66 Renehan AG, Zwahlen M, Minder C, ODwyer ST, Shalet SM,Egger M. Insulin-like growth factor (IGF)-I, IGF bindingprotein-3, and cancer risk: systematic review and meta-regressionanalysis. Lancet2004; 363: 134653.

67 Juul A, Bang P, Hertel NT, et al. Serum insulin-l ike growthfactor-I in 1030 healthy children, adolescents, and adults: relation

to age, sex, stage of puberty, testicular size, and body mass index.J Clin Endocrinol Metab 1994; 78: 74452.

68 Derby CA, Zilber S, Brambilla D, Morales KH, McKinlay JB. Bodymass index, waist circumference and waist to hip ratio andchange in sex steroid hormones: the Massachusetts Male AgeingStudy. Clin Endocrinol (Oxf) 2006; 65: 12531.

69 Key TJ, Appleby PN, Reeves GK, et al. Body mass index, serumsex hormones, and breast cancer risk in postmenopausal women.J Natl Cancer Inst2003; 95: 121826.

70 Rose DP, Komninou D, Stephenson GD. Obesity, adipocytokines,and insulin resistance in breast cancer. Obes Rev2004; 5: 15365.

71 GagoDominguez M, Castelao JE, Yuan JM, Ross RK, Yu MC.Lipid peroxidation: a novel and unifying concept of the etiology ofrenal cell carcinoma (United States). Cancer Causes Control2002;13: 28793.

72 Ogden CL, Carroll MD, Curtin LR, McDowell MA, Tabak CJ,Flegal KM. Prevalence of overweight and obesity in the UnitedStates, 19992004.JAMA 2006; 295: 154955.

73 Zaninotto P, Wardle H, Stamatakis E, Mindell J, Head J.Forecasting Obesity to 2010. London, UK: National Centre forSocial Research, Department of Health, 2006. http://www.dh.gov.uk/en/Publicationsandstatistics/Publications/PublicationsStatistics/DH_4138630 (accessed Oct 4, 2007).

74 Ben-Shlomo Y, Kuh D. A life course approach to chronic diseaseepidemiology: conceptual models, empirical challenges andinterdisciplinary perspectives. Int J Epidemiol2002; 31: 28593.

published edition

Documents