Obesity and overweight in relation to organ-specific cancer mortality in London (UK): findings from the original Whitehall study

Transcription

1 (2005) 29, & 2005 Nature Publishing Group All rights reserved /05 $ PAPER Obesity and overweight in relation to organ-specific cancer mortality in London (UK): findings from the original Whitehall study GD Batty 1,2 *, MJ Shipley 3, RJ Jarrett 4, E Breeze 3, MG Marmot 3 and G Davey Smith 5 1 MRC Social and Public Health Sciences Unit, University of Glasgow, 4 Lilybank Gardens, Glasgow, UK; 2 Department of Psychology, University of Edinburgh, UK; 3 Department of Epidemiology and Public Health, University College London, London, UK; 4 Bishopsthorpe Road, London, UK; and 5 Department of Social Medicine, University of Bristol, Bristol, UK OBJECTIVE: To examine the relation of obesity and overweight with organ-specific cancer mortality. METHODS: In the Whitehall prospective cohort study of London-based government employees, middle-age men participated in a medical examination between 1967 and Subjects were followed up for cause-specific mortality for up to 35 y (median: interquartile range (25th 75th centile); 28.1 y: ). RESULTS: There were over 3000 cancer deaths in this cohort. There was a raised risk of mortality from carcinoma of the rectum, bladder, colon, and liver, and for lymphoma in obese or overweight men following adjustment for range of covariates, which included socioeconomic position and physical activity. These relationships held after exclusion of deaths occurring in the first 20 y of follow-up. CONCLUSION: Avoidance of obesity and overweight in adult life may reduce the risk of developing some cancers. (2005) 29, doi: /sj.ijo ; published online 28 June 2005 Keywords: overweight; cancer; Whitehall; cohort study Introduction Although elevated rates of cancer mortality in individuals with higher body weight were first documented almost a century ago, 1 most attention has focused on the role of obesity in the aetiology of coronary heart disease (CHD). In large-scale prospective studies, findings are generally consistent: a positive association is apparent such that obese and overweight persons, as indexed by their body mass index (BMI), experience a higher risk of CHD than their leaner counterparts. 2 This relation may be largely ascribed to mediation via the established CHD risk indicators of blood pressure, blood lipids and glucose tolerance. 2 In the last two decades, the cohort studies on which these observations are based have matured, so accumulating sufficient events to allow investigators to examine the link between obesity and some organ-specific cancers. There is a consensus that obesity and overweight are associated with an increased risk of cancer of the breast (in postmenopausal *Correspondence: Dr GD Batty, MRC Social and Public Health Sciences Unit, University of Glasgow, 4 Lilybank Gardens, Glasgow G12 8RZE, UK. david-b@msoc.mrc.gla.ac.uk Received 4 August 2004; revised 7 January 2005; accepted 16 March 2005; published online 28 June 2005 women), endometrium, kidney, colon (strongest in men), oesophagus and pancreas. 3,4 However, studies examining the influence of obesity and overweight on other malignanciesfprostate, liver, stomach, bladder, lymphoma and leukaemiafreveal inconclusive findings. 3 5 These discrepant results may be explained by variability in definition of obesity and overweight across reports, so complicating comparison; and a failure to adjust for important covariates, such as socioeconomic position 6 and physical activity. 7 Additionally, in cohort studies, the presence of subclinical malignancy at baseline may lead to low body weight. It is likely, therefore, that the positive obesity/overweight cancer gradient seen for some sites would, in fact, be steeper if deaths occurring in the early years of follow-up were excluded from analyses. However, the few extant prospective cohort studies have a sufficiently high number of cancer cases with which to examine this issue of reverse causality. In the Whitehall study, over middle-aged Londonbased government employees participated in a medical examinationinthelate1960s,whichincludedanassessment of their BMI and a range of covariate data. 8 In an extended (maximum 35 y) mortality surveillance of this cohort, there have been over 3000 cancer deaths, enabling us to address these issues of data scarcity and methodological shortcomings.

2 1268 In earlier (r15 y)follow-ups of this cohort,raised risks of total mortality, 9,10 cardiovascular disease 9,11 and total cancers 9 were reported in overweight and obese groups. In this most recent follow-up we examine, for the first time, the link between obesity and a range of organ-specific malignancies. Materials and methods In the Whitehall study, data were collected on nonindustrial London-based male government employees aged from 40 to 64 y when examined between September 1967 and January 1970, representing a 74% response. This involved the completion of a study questionnaire and participation in a medical examination, both of which have been described in detail elsewhere. 8 In brief, the questionnaire included enquiries regarding civil service employment grade (an indicator of socio-economic position), 12 smoking habits, 13 intermittent claudication, 14 angina, 15,16 chronic bronchitis, 17 marital status, 18 physical activity, 19 unexplained weight loss in the preceding year and the use of drug therapy for heart problem or high blood pressure. 8 Forced expiratory volume in one second (FEV 1 ) adjusted for height, 20 ischaemia, 21 fasting plasma cholesterol, 22 postchallenge 2-h blood glucose, 23 blood pressure, 24 and triceps skinfold thickness 8 were determined using standardised protocols. In addition, in a representative sample of the cohort, 1669 men participated in a dietary survey. This involved the completion of a 3-day semiquantitative record of all food and drink consumed. 25 Assessment of obesity and overweight Height was measured with the subject wearing shoes and standing with his back to a measuring rod; readings were taken to the nearest 1 2 in. (approximately 12.7 mm) below.8 Weight was recorded with the subject wearing shoes but with jacket removed; readings were taken to the nearest 1 2 lb (227 g). 8 Following conversion from imperial to metric units, BMI (weight (kg) divided by height squared (m 2 )) was computed. To facilitate comparability of our findings with those from other studies, we defined underweight (o18.5 to 25.0 kg/m 2 ), normal weight (BMI 18.5 to o25.0 kg/m 2 ), overweight ( kg/m 2 ) and obesity (Z30.0 kg/m 2 ), according to criteria advanced by the World Health Organisation. 29 Ascertainment of cancer mortality The records of men (99.1% of subjects) were traced and flagged using the procedures of the National Health Service Central Registry (NHSCR) until 31 December Among the men who died, 91.6% of death certificates were coded according to the eighth revision of the International Classification of Diseases (ICD), % according to the ninth revision 31 and 1.4% according to the tenth revision. 32 The category of all malignant neoplasms (ICD8: ; ICD9: ; ICD10: C00 C97)Freferred to as all-cancers Fwas divided into individual organs. These were: oesophagus (ICD8/9: 150; ICD10: C15); stomach (ICD8/9: 151; ICD10: C16); colon (ICD8/9: 153; ICD10: C18); rectum (ICD8/9: 154; ICD10: C19); liver (ICD8/9: ; ICD10: C22-C24); pancreas (ICD8/9: 157; ICD10: C25); trachea, bronchus and lung (ICD8/9: 162; ICD10: C33 C34; referred to as lung cancer ); prostate (ICD8/9: 185; ICD10: C61); bladder (ICD8/9: 188; ICD10: C67); kidney (ICD8/9: 189; ICD10: C64 C66, C68); brain (ICD8/9: 191; ICD10: C71); lymphoma (ICD8/9: ; ICD10: C81 C90); and leukaemia (ICD8: ; ICD9: ; ICD 10: C91 C95). Data analyses A total of men identified in the NHSCR had data for BMI and all potential covariates. The cause of death for 41 of these was unknown and they were excluded from all analyses. In addition, we excluded 204 men classified as underweight (see later explanation) leaving an analytical sample of men (92.9% of those recruited). In analyses of baseline characteristics according to the level of obesity and overweight, the prevalence of the former was adjusted for age (5 y age groups) by the direct standardisation method. Trends in proportions were tested for statistical significance using the Mantel Haenszel test; for continuous variables, least-squares means were used to present the age-adjusted means, and tests for trend across obesity, overweight and normal weight groups were computed by fitting a linear trend term. In the examination of the relation of dietary characteristics with weight in a subsample of the present cohort, the distribution of alcohol intake was highly skewed; therefore, analysis was conducted on the logarithmically transformed data after adding 0.5 to all data points to overcome values of zero. Models fitted with a BMI by follow-up time interaction term confirmed that the proportional hazards assumption was not violated. Thus, hazard ratios and accompanying confidence intervals were computed for the relation of obesity/overweight with each mortality outcome using Cox s proportional hazards regression model 33 with follow-up period as the time scale. These models were initially adjusted for age and then for other potential covariates. For the purposes of statistical adjustment, age, triceps skinfold thickness, plasma cholesterol, height-adjusted FEV 1 and systolic blood pressure were fitted as continuous variables; while unexplained weight loss in the last year (two levels), employment grade (5), marital status (4), blood pressurelowering medication (2), blood glucose levels (3), disease at study entry (2) and physical activity (6) were fitted as categorical variables. During the baseline study, the physical activity enquiries on the questionnaire were modified. Levels of this behaviour were therefore determined from either an item about travel activity 19 (administered to approximately two-thirds of men) or from leisure activities 34 (administered

3 to the remainder). Analyses of the obesity/overweight cancer relation indicated that there was no confounding effect due to questionnaire type. Smoking status was grouped into four categories (never, ex-smoker, current pipe or cigar smoker, current cigarette smoker) together with additional adjustment for the number of cigarettes smoked per day in current smokers. Existing disease at study entry was defined as a positive response to enquiries regarding a range of health conditions: myocardial ischaemia, intermittent claudication, physician-diagnosed heart problems or high blood pressure (one question), dyspnoea and bronchitis. The existence of ischaemia was determined from ischaemic signs on an ECG trace, or positive responses to either the Rose angina questionnaire or a report of severe pain across the front of the chest lasting 30 min or more. 15 Men with diabetes comprised those who gave a positive response to the questionnaire enquiry are you, or have you been, diabetic?, or those who had blood glucose level 2 h after the glucose load of Z11.1 mmol/l (Z200 mg/100 ml). A blood glucose of mmol/l ( mg/100 ml) was used to designate participants with glucose intolerance, with all remaining men termed normoglycaemic. 23 To address the problem of reverse causalityffor some cancers, tumour presence may lead to weight loss, so attenuating the obesity cancer relationshipfwe took three approaches. Firstly, we dropped the underweight group from our analyses on the understanding that this group would contain some men with undetected cancer that may have resulted in weight loss. Secondly, we adjusted for unexplained weight loss in the preceding year and existing illness at study entry. Finally, in subsequent analyses, we excluded deaths in the first 10 and 20 y of mortality surveillance. In so doing, we reasoned that a significant proportion of deaths attributable to cancer, if present at study induction, would have occurred within this time frame. 35 All statistical analyses were conducted using SAS computer software. 36 Results In Table 1 the relation of obesity and overweight with baseline characteristics are presented. Men with obesity comprised 4.2% (717) of the analytical sample. The most unfavourable level of each baseline characteristic was apparent in the obese and overweight study participants, the only exception being cigarette smoking where normal weight men had the highest prevalence. In Table 2 the relations of overweight and obesity to a range of dietary characteristics, assessed in a subgroup of the study population, are depicted. Higher quantities of alcohol consumption were apparent in the men classified as obese, while lower levels of fibre and fat (total and saturated) intake were seen. In men there were (63.7%) deaths over a maximum of 35 y follow-up. Of these, 3051 deaths were ascribed to all-cancers, the most common site being that of the lung (26% of all cancer fatalities). Of all cancers, 498 occurred within the first 10 y of follow-up and 1401 within the first 20 y of follow-up. In Tables 3 and 4 the relationships between obesity, overweight and mortality from various cancer sites are presented. The suggestion of a raised rate of total cancers in the obese and overweight groups in the age- and multiply-adjusted analyses was essentially eliminated when deaths occurring in the first 20 y of follow-up were excluded from the analysis. Owing to the low prevalence of obesity in the present study population, there were few cancer deaths in this group for some anatomical sites; some of our findings should therefore 1269 Table 1 Obesity and overweight in relation to baseline characteristics ( ) a Normal weight ( kg/m 2 ) Overweight ( kg/m 2 ) Obese (Z30 kg/m 2 ) Number (%) 9288 (54.3) 7097 (41.5) 717 (4.2) Mean (s.e.) Age (y) 51.1 (0.1) 52.0 (0.1) 52.6 (0.2) Triceps skinfold thickness (mm) 39.4 (0.2) 50.3 (0.2) 64.0 (0.5) Plasma cholesterol (mmol/l) 5.04 (0.01) 5.21 (0.01) 5.15 (0.05) FEV b 1 (l/s) 3.13 (0.01) 3.16 (0.01) 3.09 (0.02) Systolic blood pressure (mmhg) (0.2) (0.2) (0.7) Percent (s.e.) Physically inactive 15.3 (0.4) 16.2 (0.4) 23.8 (1.6) Unintentional weight loss in last year 3.1 (0.2) 0.9 (0.1) 0.3 (0.2) Current cigarette smoker 44.1 (0.5) 37.4 (0.6) 36.9 (1.8) Low work grade 23.3 (0.4) 23.0 (0.5) 30.1 (1.7) No partner 12.4 (0.3) 10.7 (0.4) 14.2 (1.3) Disease at study entry c 19.6 (0.4) 21.5 (0.5) 24.7 (1.6) Blood pressure-lowering medication 1.2 (0.1) 1.9 (0.2) 2.2 (0.5) Glucose intolerance c 4.7 (0.2) 5.8 (0.3) 7.9 (1.0) Diabetic c 1.2 (0.1) 1.2 (0.1) 2.1 (0.5) a Adjusted for age (age is unadjusted). b FEV 1 ¼ forced expiratory volume in one second (adjusted for height). c See Materials and methods section for definitions.

4 1270 Table 2 Overweight and obesity in relation to self-reported dietary intake over a 3-day period (N ¼ 1652) Dietary characteristic (g) Normal ( kg/m 2 ) Overweight ( kg/m 2 ) Obese (Z30 kg/m 2 ) P-value for trend Number of men F Alcohol a 3.70 b (3.21,4.18) 4.07 (3.51,4.70) 4.26 (2.64,6.73) 0.24 Total fibre 16.7 (16.4,17.1) 16.2 (15.8,16.6) 16.2 (15.1,17.4) 0.05 Total fat (110.6,114.5) (107.0,111.3) (103.2,117.0) 0.05 Saturated fat 58.1 (57.1,59.1) 55.9 (54.8,57.0) 56.2 (52.6,59.8) Monounsaturated fat 46.1 (45.3,46.9) 45.0 (44.1,45.9) 45.6 (42.6,48.5) 0.16 Polyunsaturated fat 7.96 (7.77,8.15) 7.97 (7.77,8.18) 8.03 (7.37,8.70) 0.75 a Means for alcohol are geometric means (see Materials and methods). b Results are mean (95% CI). viewed with caution. In an age-adjusted analysis, the lowest risk of lung cancer deaths was apparent in obese men; however, following adjustment for confounding factors which included smokingfthe highest prevalence of this behaviour was evident in the normal weight groupfthis relation was lost. There was a suggestion of elevated rates of carcinoma of colon and lymphoma in the obese and overweight groups following full adjustment and exclusion of deaths in the first 20 y of follow-up, although statistical significance at conventional levels was not always evident. Of the individual cancer sites depicted in Table 4, there were raised rates of carcinoma of liver, rectum and bladder in the obese or overweight groups, although most confidence intervals included unity. In contrast, there was a suggestion of a non-significant inverse weight kidney cancer relation although confidence intervals were wide owing to a low number of cases. There was little evidence of an obesity/ overweight malignancy gradient for any other cancer site featured in Tables 3 and 4. When we included the underweight group in age-adjusted analyses with total cancers as the end point of interest (data not shown), we found an elevated rate in this group (HR underweight vs normal weight (95% CI): 1.61 (1.21, 2.13)) that was heavily attenuated when remaining covariates were added to the multivariable model (1.22 (0.91, 1.62)). There were too few site-specific cancer cases in the underweight group (N ¼ 204 men) to facilitate further analyses. Discussion The main finding of this study was an elevated risk of mortality from carcinoma of the rectum, bladder, colon, and liver, and for lymphoma in men who were obese or overweight in comparison to those in the normal weight group, although statistical significance was not always apparent. There was little evidence of an weight cancer gradient for other malignancies. Comparison with other studies Some investigators combine colon and rectal cancer into a single disease category when examining their aetiology; however, owing to their dissimilar epidemiology, 37 separation is warranted. Previous studies in men suggest a direct obesity/overweight colon cancer association, 3,4 as apparent herein, although this is not a universal result. 38 We also found a positive non-significant gradient for rectal cancer, an association that has previously been suggested to be null. 39,40 Three other studies 26,41,42 have reported an excess occurrence of malignant neoplasm of the liver in overweight persons. While we made the same observation in the present study, these findings should be viewed with caution owing to the very low number of cases; a caveat that should also be applied to some of our other results. In a recent metaanalysis of observational studies, adiposity was associated with increased pancreatic cancer risk. 28 That this effect was modestfhr (95% CI); 1.19 (1.10, 1.29)Fimplicates unmeasured or residual confounding as a likely alternative explanation. 28 Indeed, in the present analyses in which we utilised a wide range of confounding factors, we found no evidence of an association between BMI and this cancer. Plausible mechanisms Several mechanisms have been invoked for the higher occurrence of some cancers in the obese or overweight. Differences in incidence of some cancers across body weight categories may reflect differences in dietary characteristics, such as fat and fibre intake. 43 In the Whitehall study, a small representative subgroup of 1669 participants completed a 3- day dietary record during the period of baseline examination. 25 While there were insufficient site-specific cancer cases to examine the potential confounding effect of dietary characteristics on the obesity/overweight cancer association in this group, when we related intake to overweight and obesity, the differences across these groups were small and not always most favourable in the leanest subjects. It is unlikely, therefore, that these small and inconsistent differences in diet were large enough to explain differences in later cancer risk across the body weight groups. A more specific explanation for the obesity/overweight cancer relation that has received much recent attention concerns the role of insulin and insulin-like growth factors. These hypotheses suggests that obesity precipitates insulin resistance and the resulting prolonged hyperinsulinaemia in

5 Table 3 Mortality rates and hazard ratios for selected site specific cancer deaths in relation to obesity and overweight 1271 Type of cancer (total no. of deaths) Normal ( kg/m 2 ) Overweight ( kg/m 2 ) Obese (Z30 kg/m 2 ) P for trend Number of subjects F All cancers (3051) Mortality rate a (no. of deaths) 7.66 (1648) 7.53 (1279) 7.88 (124) F Age adjusted b 1.0 (ref) 1.00 (0.93,1.07) 1.14 (0.95,1.37) 0.69 Multiply adjusted c (0.97,1.13) 1.21 (1.00,1.48) 0.08 Multiply adjustedfexcluding first 10 y d (0.95,1.13) 1.17 (0.94,1.45) 0.20 Multiply adjustedfexcluding first 20 y e (0.98,1.22) 1.09 (0.81,1.47) 0.13 Lung (783) Mortality rate (no. of deaths) 2.20 (450) 1.85 (308) 1.55 (25) Age adjusted (0.73, 0.98) 0.77 (0.51, 1.15) Multiply adjusted (0.85, 1.16) 1.00 (0.65, 1.53) 0.98 Multiply adjustedfexcluding first 10 y (0.80, 1.15) 1.15 (0.72, 1.83) 0.98 Multiply adjustedfexcluding first 20 y (0.79, 1.30) 1.07 (0.53, 2.16) 0.85 Prostate (431) Mortality rate (no. of deaths) 1.14 (243) 1.02 (175) 0.81 (13) Age adjusted (0.76, 1.12) 0.88 (0.50, 1.54) 0.42 Multiply adjusted (0.75, 1.13) 0.91 (0.51, 1.63) 0.45 Multiply adjustedfexcluding first 10 y (0.74, 1.13) 0.80 (0.43, 1.51) 0.33 Multiply adjustedfexcluding first 20 y (0.74, 1.22) 0.76 (0.35, 1.67) 0.54 Colon (279) Mortality rate (no. of deaths) 0.64 (139) 0.72 (122) 1.18 (18) Age adjusted (0.89, 1.46) 2.00 (1.22, 3.26) 0.03 Multiply adjusted (0.93, 1.56) 2.21 (1.29, 3.79) 0.02 Multiply adjustedfexcluding first 10 y (0.90, 1.60) 2.14 (1.15, 3.99) 0.05 Multiply adjustedfexcluding first 20 y (1.01, 2.01) 1.37 (0.53, 3.56) 0.07 Stomach (190) Mortality rate (no. of deaths) 0.48 (100) 0.49 (81) 0.56 (9) Age adjusted (0.76, 1.37) 1.27 (0.64, 2.52) 0.70 Multiply adjusted (0.76, 1.44) 1.23 (0.59, 2.58) 0.60 Multiply adjustedfexcluding first 10 y (0.78, 1.62) 1.80 (0.81, 3.99) 0.22 Multiply adjustedfexcluding first 20 y (0.55, 1.51) 1.30 (0.38, 4.51) 0.95 Lymphoma (158) Mortality rate (no. of deaths) 0.33 (77) 0.42 (74) 0.44 (7) Age adjusted (0.94, 1.79) 1.46 (0.67, 3.16) 0.08 Multiply adjusted (0.95, 1.89) 1.49 (0.65, 3.42) 0.10 Multiply adjustedfexcluding first 10 y (0.86, 1.81) 1.61 (0.69, 3.74) 0.18 Multiply adjustedfexcluding first 20 y (0.84, 2.09) 1.92 (0.70, 5.25) 0.15 Pancreas (147) Mortality rate (no. of deaths) 0.36 (75) 0.41 (69) 0.20 (3) Age adjusted (0.86, 1.66) 0.61 (0.19, 1.93) 0.77 Multiply adjusted (0.83, 1.68) 0.58 (0.18, 1.91) 0.80 Multiply adjustedfexcluding first 10 y (0.77, 1.66) 0.43 (0.10, 1.84) 0.90 Multiply adjustedfexcluding first 20 y (0.76, 2.03) 0.86 (0.19, 3.76) 0.57 a Age-adjusted mortality rates per 1000 person-years; b Results are hazards ratio (95% CI). c Multiply adjusted modelfadjusted for age, plus employment grade, physical activity, smoking habit, marital status, disease at entry, weight loss in the last year, blood pressure-lowering medication, height adjusted FEV 1, triceps skinfold thickness, systolic blood pressure, plasma cholesterol, glucose intolerance and diabetes status. d Deaths in the first 10 y of surveillance are excluded. e Deaths in the first 20 y of surveillance are excluded. itself acts as a tumour growth promoter. 44 Additionally, higher levels of insulin-like growth factor (IGF-I)Fa multifunctional, circulating peptide that encourages tumour growth through its mitosis and antiapoptosis properties 45 Fare positively associated with prostate and colorectal cancer risk, 46 both energy-related malignancies. While IGF- I is also positively related to BMI, the gradient is in fact nonlinear (at obese levels it decreases). 47 Although we found an elevated risk of colon and rectal cancer mortality in the obese and overweight groups herein, there was no evidence of a relationship with prostate cancer. Study strengths and limitations The strengths of the present study include its sizefsuperior to most, if not all, 48 reports in the literaturefwhich

6 1272 Table 4 Mortality rates and hazard ratios for selected site-specific cancer deaths in relation to obesity and overweight Type of cancer (total no. of deaths) Normal ( kg/m 2 ) Overweight ( kg/m 2 ) Obese (Z30 kg/m 2 ) P for trend Number of subjects F Bladder (144) Mortality rate a (no. of deaths) 0.36 (75) 0.38 (65) 0.26 (4) Age adjusted b 1.0 (ref) 1.11 (0.80, 1.55) 0.83 (0.30, 2.28) 0.80 Multiply adjusted c (0.84, 1.73) 0.94 (0.33, 2.69) 0.46 Multiply adjustedfexcluding first 10 y d (0.86, 1.85) 0.57 (0.13, 2.41) 0.51 Multiply adjustedfexcluding first 20 y e (1.06, 2.65) 1.19 (0.27, 5.18) 0.05 Oesophagus (121) Mortality rate (no. of deaths) 0.25 (58) 0.34 (58) 0.33 (5) Age adjusted 1.0 b 1.33 (0.92, 1.91) 1.40 (0.56, 3.49) 0.16 Multiply adjusted (0.94, 2.06) 1.48 (0.56, 3.90) 0.13 Multiply adjustedfexcluding first 10 y (0.84, 1.90) 0.84 (0.25, 2.81) 0.51 Multiply adjustedfexcluding first 20 y (0.69, 1.81) 0.00 f 0.75 Leukaemia (108) Mortality rate (no. of deaths) 0.28 (61) 0.27 (45) 0.13 (2) Age adjusted (0.65, 1.40) 0.50 (0.12, 2.04) 0.48 Multiply adjusted (0.61, 1.41) 0.44 (0.10, 1.91) 0.42 Multiply adjustedfexcluding first 10 y (0.59, 1.48) 0.53 (0.12, 2.32) 0.53 Multiply adjustedfexcluding first 20 y (0.68, 2.03) 0.48 (0.06, 3.72) 0.89 Rectum (104) Mortality rate (no. of deaths) 0.24 (52) 0.26 (46) 0.40 (6) Age adjusted (0.79, 1.74) 1.80 (0.77, 4.19) 0.20 Multiply adjusted (0.78, 1.83) 1.97 (0.79, 4.91) 0.21 Multiply adjustedfexcluding first 10 y (0.90, 2.29) 1.81 (0.61, 5.42) 0.09 Multiply adjustedfexcluding first 20 y (0.82, 2.59) 2.92 (0.82, 10.4) 0.08 Kidney (61) Mortality rate (no. of deaths) 0.16 (36) 0.11 (20) 0.34 (5) Age adjusted (0.42, 1.27) 2.12 (0.83, 5.41) 0.99 Multiply adjusted (0.32, 1.04) 1.20 (0.41, 3.52) 0.31 Multiply adjustedfexcluding first 10 y (0.29, 1.07) 1.14 (0.31, 4.23) 0.24 Multiply adjustedfexcluding first 20 y (0.28, 1.35) 0.46 (0.05, 3.88) 0.20 Brain (55) Mortality rate (no. of deaths) 0.12 (30) 0.14 (24) 0.05 (1) F Age adjusted (0.63, 1.85) 0.50 (0.07, 3.66) 0.94 Multiply adjusted (0.53, 1.69) 0.42 (0.05, 3.32) 0.59 Multiply adjustedfexcluding first 10 y (0.32, 1.66) 0.00 f 0.24 Multiply adjustedfexcluding first 20 y (0.20, 1.85) 0.00 f 0.24 Liver (51) Mortality rate (no. of deaths) 0.11 (27) 0.10 (18) 0.37 (6) Age adjusted (0.50, 1.65) 3.55 (1.46, 8.63) 0.29 Multiply adjusted (0.53, 1.88) 3.76 (1.36, 10.4) 0.17 Multiply adjustedfexcluding first 10 y (0.48, 2.01) 3.02 (0.80, 11.4) 0.50 Multiply adjustedfexcluding first 20 y (0.39, 2.06) 3.88 (0.96, 15.7) 0.48 a Age-adjusted mortality rates per 1000 person-years. b Results are hazards ratio (95% CI). c Multiply adjusted modelfadjusted for age, plus employment grade, physical activity, smoking habit, marital status, disease at entry, weight loss in the last year, blood pressure-lowering medication, height adjusted FEV 1, triceps skinfold thickness, systolic blood pressure, plasma cholesterol, glucose intolerance and diabetes status. d Deaths in the first 10 y of surveillance are excluded. e Deaths in the first 20 y of surveillance are excluded. f No cases with which to compute a hazards ratio. facilitated an examination of reverse causality; its prospective design; the measurement of a range of covariate data including physical activity and socio-economic position; and the definition of obesity and overweight which matches WHO criteria. These strengths notwithstanding, this study is not without its weaknesses. The assessment of obesity and overweight was based on BMI, an imperfect measure of adiposity. Although skinfold thickness was measured in the Whitehall study participants, readings were only taken at the triceps, rendering the data of little practical use. In contrast to many previous reports that focus on a single cancer site rather than a range (perhaps in the interest of generating a greater number publications), we related obesity and overweight to 16 mortality outcomes. Clearly, not all of the

7 relationships we examined were hypothesis driven. In so doing we wished to gain insight into specificity of association, 49 an important determinant of causality in observational epidemiology. 50 However, one disadvantage of this approach is that some of our findings may have arisen by chance. Additionally, the cancer outcomes reported herein were based on mortality surveillance. Thus, our results reflect the combined effect of weight on survival and incidence. The suggestion has been made that the relation of weight to survival may be higher than that for incidence of some cancers (eg breast). 48 In conclusion, an elevated risk of mortality from carcinoma of the rectum, bladder, colon, and liver, and for lymphoma in men who were obese or overweight in comparison to those in the normal weight group was apparent in this cohort of London-based government employees. Avoidance of obesity and overweight in adult life may therefore reduce the risk of developing these cancers. However, prevention and treatment of obesity is unlikely to be achieved through attempts to alter physical activity and dietary habits at the level of the individual. 51 Rather, fundamental changes in environmental structure that includes ready access to amenities conducive to physical activity (eg parkland) and retail outlets that provide a range of micronutrient-dense food sources is crucial. 52 Contributions David Batty generated the idea for this paper; Martin Shipley conducted all data analyses. David Batty wrote the first draft of the manuscript to which co-authors contributed. Acknowledgements We are grateful to the civil servants who gave of their time to take part in the baseline study. The original screening of participants in the Whitehall study was funded by the Department of Health and Social Security and the Tobacco Research Council. Martin Shipley and Elizabeth Breeze are supported by the British Heart Foundation; Michael Marmot by the UK Medical Research Council (MRC). When work on this report began, David Batty was financed by the UK MRC at the London School of Hygiene and Tropical Medicine and subsequently by a University Senior Research Fellowship at the University of Copenhagen. He is now the recipient of a Wellcome Advanced Training Fellowship. References 1 Actuarial Society of America and Association of Life Insurance Directors. Medico-actuarial mortality investigation. Spector Company: New York; Whitlock G, Lewington S, Ni Mhurchu C. Coronary heart disease and body mass index: a systematic review of the evidence from larger prospective cohort studies. Seminars in vascular medicine (classical and emerging risk factors for cardiovascular disease, Part Vol 4, pp World Cancer Research Fund/American Institute for Cancer Research. Food, nutrition and the prevention of cancer: a global perspective. American Institute for Cancer Research: Washington; IARC handbooks of cancer prevention. Weight control and physical activity Vol 6. International Agency for Research on Cancer: Lyons, France; Bianchini F, Kaaks R, Vainio H. Overweight, obesity, and cancer risk. Lancet Oncol 2002; 3: Batty GD. Confounding effect of socioeconomic position in the study of height in relation to prostate cancer risk (letter). Br J Cancer 2004; 90: Batty GD, Thune I. Does physical activity prevent cancer? Evidence suggests protection against colon cancer and probably breast cancer (editorial). BMJ 2000; 321: Reid DD, Hamilton PJS, McCartney P, Rose G, Jarrett RJ, Keen H, Rose G. Cardiorespiratory disease and diabetes among middleaged male civil servants. Lancet 1974; i: Jarrett RJ, Shipley MJ, Rose G. Weight and mortality in the Whitehall Study. Br Med J (Clin Res Edn) 1982; 285: Jarrett RJ. Is there an ideal body weight? Br Med J (Clin Res Edn) 1986; 293: Fitzgerald AP, Jarrett RJ. Body weight and coronary heart disease mortality: an analysis in relation to age and smoking habit. 15 years follow-up data from the Whitehall Study. Int J Obes Relat Metab Disord 1992; 16: Davey Smith G, Leon D, Shipley MJ, Rose G. Socioeconomic differentials in cancer among men. Int J Epidemiol 1991; 20: Reid DD, Hamilton PJ, McCartney P, Rose G, Jarrett RJ, Keen H. Smoking and other risk factors for coronary heart-disease in British civil servants. Lancet 1976; 2: Davey Smith G, Shipley MJ, Rose G. Intermittent claudication, heart disease risk factors, and mortality. The Whitehall Study. Circulation 1990; 82: Rose G, McCartney P, Reid DD. Self-administration of a questionnaire on chest pain and intermittent claudication. Br J Prev Soc Med 1977; 31: Rose GA. The diagnosis of ischaemic heart pain and intermittent claudication in field studies. Bull World Health Organ 1962; 27: Committee on the aetiology of chronic bronchitis. Definition and classification of chronic bronchitis. Lancet 1965; i: Ben Shlomo Y, Davey Smith G, Shipley M, Marmot MG. Magnitude and causes of mortality differences between married and unmarried men. J Epidemiol Commun Health 1993; 47: Batty GD, Shipley M, Marmot M, Davey Smith G. Physical activity and cause-specific mortality in men: further evidence from the Whitehall study. Eur J Epidemiol 2002; 17: Batty GD, Shipley MJ, Marmot M, Davey Smith G. Physical activity and cause-specific mortality in men with type 2 diabetes/ impaired glucose tolerance: evidence from the Whitehall study. Diabetes Med 2002; 19: Rose G, Hamilton PS, Keen H, Reid DD, McCartney P, Jarrett RJ. Myocardial ischaemia, risk factors and death from coronary heart-disease. Lancet 1977; 1: Davey Smith G, Shipley MJ, Marmot MG, Rose G. Plasma cholesterol concentration and mortality. The Whitehall Study. JAMA 1992; 267: Batty GD, Shipley M, Marmot M, Davey Smith G. Diabetes status and post-load plasma glucose concentration in relation to sitespecific cancer mortality: findings from the original Whitehall study. Cancer Causes Control 2004; 15: Batty GD, Shipley MJ, Marmot MG, Davey Smith G. Blood pressure and site-specific cancer mortality: evidence from the original Whitehall study. Br J Cancer 2003; 89: Keen H, Thomas BJ, Jarrett RJ, Fuller JH. Nutrient intake, adiposity, and diabetes. Br Med J 1979; 1:

8 Calle EE, Rodriguez C, Walker-Thurmond K, Thun MJ. Overweight, obesity, and mortality from cancer in a prospectively studied cohort of US adults. N Engl J Med 2003; 348: Henley SJ, Flanders WD, Manatunga A, Thun MJ. Leanness and lung cancer risk: fact or artifact? Epidemiology 2002; 13: Berrington dg, Sweetland S, Spencer E. A meta-analysis of obesity and the risk of pancreatic cancer. Br J Cancer 2003; 89: World Health Organisation. Physical status: the use and interpretation of anthropometry: report of a WHO expert committee WHO Technical Report Series. WHO: Geneva; Anon. Manual of the international statistical classification of diseases, injuries, and causes of death (8th revision). World Health Organisation: Geneva; Anon. Manual of the international statistical classification of diseases, injuries, and causes of death (9th revision). WHO: Geneva; Anon. International statistical classification of diseases and related health problems (10th revision). WHO: Geneva; Cox DR. Regression models and life-tables. J Roy Stat Soc Ser B 1972; 34: Davey Smith G, Shipley MJ, Batty GD, Morris JN, Marmot M. Physical activity and cause-specific mortality in the Whitehall study. Public Health 2000; 114: Coleman MP, Rachet B, Woods LM, Mitry E, Riga M, Cooper N, Quinn MJ, Brenner H, Esteve J. Trends and socioeconomic inequalities in cancer survival in England and Wales up to Br J Cancer 2004; 90: SAS Institute Inc. SAS/STAT(R) user s guide Version 6, volumes 1 & 2. SAS Institute Inc.: Cary, NC; Haenszel W, Correa P. Cancer of the large intestine: epidemiologic findings. Dis Colon Rectum 1973; 16: Tamakoshi K, Wakai K, Kojima M, Watanabe Y, Hayakawa N, Toyoshima H, Yatsuya H, Kondo T, Tokudome S, Hashimoto S, Suzuki K, Ito Y, Tamakoshi A, JACC Study Group. A prospective study of body size and colon cancer mortality in Japan: The JACC Study. Int J Obes Relat Metab Disord 2004; 28: Nomura A, Heilbrun LK, Stemmermann GN. Body mass index as a predictor of cancer in men. J Natl Cancer Inst 1985; 74: Chyou PH, Nomura AM, Stemmermann GN. A prospective study of weight, body mass index and other anthropometric measurements in relation to site-specific cancers. Int J Cancer 1994; 57: Wolk A, Gridley G, Svensson M, Nyren O, McLaughlin JK, Fraumeni JF, Adam HO. A prospective study of obesity and cancer risk (Sweden). Cancer Causes Control 2001; 12: Moller H, Mellemgaard A, Lindvig K, Olsen JH. Obesity and cancer risk: a Danish record-linkage study. Eur J Cancer 1994; 30A: Hill MJ. Nutrition and human cancer. Ann NY Acad Sci 1997; 833: Giovannucci E. Insulin and colon cancer. Cancer Causes Control 1995; 6: Jones JI, Clemmons DR. Insulin-like growth factors and their binding proteins: biological actions. Endocr Rev 1995; 16: Renehan AG, Zwahlen M, Minder C, O Dwyer ST, Shalet SM, Egger M. Insulin-like growth factor (IGF)-I, IGF binding protein- 3, and cancer risk: systematic review and meta-regression analysis. Lancet 2004; 363: Thissen JP, Ketelslegers JM, Underwood LE. Nutritional regulation of the insulin-like growth factors. Endocr Rev 1994; 15: Calle EE, Rodriguez C, Walker-Thurmond K, Thun MJ. Overweight, obesity, and mortality from cancer in a prospectively studied cohort of US adults. N Engl J Med 2003; 348: Weiss NS. Can the specificity of an association be rehabilitated as a basis for supporting a causal hypothesis? Epidemiology 2002; 13: Bradford Hill A. The environment and disease: association or causation? Proc Roy Soc Med 1965; 58: Batty GD, Lee IM. Physical activity and coronary heart disease. BMJ 2004; 328: Davey Smith G, Gunnell D, Ben-Shlomo Y. Life-course approaches to socio-economic differentials in cause-specific adult mortality. In: Leon D, Walt G (eds) Poverty, inequality and health. An international perspective OUP: Oxford.