Article Text

Download PDFPDF

Research report
Risk factors for colonic and rectal cancer mortality: evidence from 40 years’ follow-up in the Whitehall I study
  1. David S Morrison1,
  2. George David Batty2,
  3. Mika Kivimaki3,
  4. George Davey Smith4,
  5. Michael Marmot3,
  6. Martin Shipley3
  1. 1West of Scotland Cancer Surveillance Unit, University of Glasgow, Glasgow, UK
  2. 2MRC Social and Public Health Sciences Unit, Glasgow, UK
  3. 3Department of Epidemiology and Public Health, University College London, London, UK
  4. 4Department of Social Medicine, University of Bristol, UK
  1. Correspondence to Dr David S Morrison, West of Scotland Cancer Surveillance Unit, Division of Community Based Sciences, University of Glasgow, 1 Lilybank Gardens, Glasgow G12 8RZ, UK; david.morrison{at}


Background Modifiable behavioural risk factors—including exercise, obesity and smoking—have been causally associated with colorectal cancer mortality. However, results have been inconsistent and undiagnosed cancers may affect baseline risk factors, distorting the temporal relationship that is observed between them.

Objective To determine whether risk factors for colorectal cancers available in the Whitehall I study were predictive of colonic or rectal cancer mortality.

Methods Prospective cohort study over 40 years on Whitehall I men aged 40–69 on entry between 1967 and 1970. Associations between baseline risk factors and cause-specific mortality were tested with Cox proportional hazards models. Events within the first 10 years of follow-up were excluded to minimise ‘reverse causality.’

Results 329 colon and 121 rectal cancer deaths occurred among 17 949 men followed up for a total of 472 523 person-years. Age and smoking were associated with increased mortality from colorectal cancers. Compared with never-smokers, current smoking was associated with age-adjusted HRs for colon and rectal cancers of 1.45 (95% CI 1.03 to 2.03) and 1.97 (95% CI 1.02 to 3.80), respectively. A significant effect of current smoking on rectal cancer mortality was only apparent after events in the first 10 years of follow-up were excluded. No convincing evidence was found that body mass index, diabetes mellitus, blood pressure or physical activity were associated with colorectal cancer mortality.

Conclusion Smoking significantly increases mortality from colorectal cancer and its decreasing prevalence in the UK may partly explain falling mortality from the disease. Changes in health behaviours in response to early cancer symptoms may result in differential misclassification or ‘reverse causality’ unless early events are excluded. Although many individual cohort studies have not shown significant relationships between behavioural risk factors and colorectal cancer mortality, their contribution to meta-analyses remains important.

  • Cancer
  • smoking
  • health behaviour
  • risk prediction
  • colorectal
  • neoplasms
  • risk reduction behaviour

Statistics from

Colorectal cancers are the fourth most common malignancies in men world wide.1 Incidence is higher in industrialised countries and is relatively uncommon in Africa and much of Asia.1 The incidence of both colon and rectal cancers was similar in the UK in the early 1970s but subsequently colon cancers increased by 30% while the incidence of rectal cancers increased by 6% between 1971 and 1997.2 Over a similar period obesity3 and alcohol consumption4 also increased. These observations led to hypotheses that causal factors for colonic and rectal cancers might be found in ‘Western lifestyle’ factors, in particular, smoking and the metabolic syndrome of dyslipidaemia, hypertension, truncal obesity and type II diabetes mellitus.5 6 There is evidence that obesity7 8 and adult height increase the incidence of colorectal cancers, while physical activity is protective.9 The associations between these risk factors and colorectal cancers vary by site. For example, physical activity has a clearer protective effect for colonic than for rectal cancers.9 However, the results of individual studies are often inconsistent. A recent systematic review found that among 60 cohort studies on body fatness and colorectal cancers, five reported a lower risk with increased body fatness and only about half reported a statistically significant positive association between overweight and cancer risk.9 There is inconsistent evidence about the nature of associations between smoking and colorectal cancer despite a recent meta-analysis.10 Although current smokers have higher colorectal cancer mortality than never smokers, the absence of any significant association between former smokers and colorectal cancer mortality or between smoking and site-specific cancer mortality, suggests that further research is needed. Male sex and increasing age are also associated with greater risks of colorectal cancer incidence.2 Socioeconomic deprivation is associated with higher mortality as a result of both higher incidence and poorer survival11–13 in more deprived populations.

There are important potential methodological artefacts when studying the effects of many of these factors on subsequent cancer risk. Undiagnosed cancers may affect an individual's behaviour or metabolism, such that baseline variables are a consequence of the cancer rather than having an influence on future risk of developing the disease. One method of reducing the effects of ‘reverse causality’ in prospective studies is to exclude events that occur soon after entry, although smaller studies and those with shorter periods of follow-up may not permit this adjustment.

The first Whitehall study was designed to examine risk factors for cardiorespiratory diseases and included baseline anthropometric, behavioural and socioeconomic information that is relevant to understanding risk factors for colorectal cancer mortality. In addition, full death record follow-up data allow deaths from colonic and rectal cancers to be distinguished. Our aim was to determine whether risk factors for colorectal cancers available in the Whitehall I study were predictive of colonic or rectal cancer mortality. We excluded deaths within 10 years to adjust for the potential effects of reverse causality and reported site-specific risks.


The Whitehall study is a prospective cohort study that collected data on 19 019 male employees, aged 40–69 years when examined between 1967 and 1970, from selected civil service departments within two miles of Whitehall, representing a 77% response.14 Entry into the study involved completion of a self-reported questionnaire on health and health-related behaviours and a medical examination, which have been described in detail elsewhere.14 The questionnaire included information on civil service employment grade (an indicator of socioeconomic status),15 smoking habits16 and physical activity.17 18 Blood pressure,19 height,20 weight21 and fasting plasma cholesterol22 (mmol/l) were all determined using standard protocols.14 Diabetes was defined as blood glucose of ≥11.1 mmol/l (≥200 mg/100 ml) and/or a positive response to the questionnaire enquiry “are you, or have you been, diabetic?”; and impaired glucose tolerance as 5.4–11.0 mmol/l (96–199 mg/100 ml). All other men were denoted normoglycaemic. Body mass index (BMI) was calculated by dividing weight (kg) by height (m) squared and categorised using WHO ranges of <18.50 (underweight); 18.50–24.99 (normal); 25.00–29.99 (overweight); and ≥30.00 (obese). Travel activity and leisure activity were collected on different subgroups of the cohort such that approximately two-thirds had travel activity and one-third had leisure activity.17 18 For this study we created an overall physical activity variable that used both of these measures so that information would be available on all the subjects. The three levels of physical activity (sedentary, moderate and high) correspond to 0–9, 10–19 and ≥20 min/day, respectively, for those with travel activity and to inactive, moderately active and active for those with leisure activity.

A total of 18 880 men (99.3% of participants in the baseline survey) were traced using the UK National Health Service Central Registry and followed up for 40 years until 31 October 2008. Colon and rectal cancer deaths were ascertained from death certificates and coded according to the International Classification of Diseases (revisions 8, 9 and 10). Colon cancer was defined by International Classification of Diseases Revisions 8 and 9 (ICD-8/9) code 153 (excluding 153.5, appendiceal cancers) and ICD-10 code C18 (excluding C18.1). Rectal cancer was defined as ICD-8/9 codes 154.0 and 154.1 and ICD-10 codes C19-C20. We excluded 44 men in whom the cause of death was unknown and a further 887 men had some missing data for one or more of the following variables: height (three men), BMI (three), cholesterol (703), diastolic blood pressure (seven), systolic blood pressure (six), diabetes (133), physical activity (41), socioeconomic position (none) and smoking status (eight). The present analyses are based on the 17 949 men (95.1%) who had complete data on all measures. Eight hundred and thirty-nine men employed in the British Council and Diplomatic Service were excluded from the analyses of employment grade as their employment grade classification was not comparable to the rest of the sample. These men have been retained in the analyses of the other risk factors by classifying them as a separate group in the multiply adjusted analyses. The person-years of follow-up for each man was partitioned by age at risk using 5-year age groups. We used Cox proportional hazards models23 to summarise the relationship between each risk factor and mortality from colon and rectal cancer, which produced HRs with accompanying 95% CIs. These analyses were first adjusted for age at risk and then fully adjusted for all potential confounding factors. Further analyses excluded deaths within the first 10 years of follow-up to determine whether morbidity-associated changes in behavioural or anthropometric measures might be responsible for reverse causality. The p values for the overall association between risk factors and each cancer outcome were calculated using likelihood ratio tests comparing models with and without the risk factor of interest.


Three hundred and twenty-nine colon and 121 rectal cancer deaths occurred among 17 949 men followed up for a total of 472 523 person-years. After exclusion of deaths within the first 10 years of follow-up, there were 276 colon and 103 rectal cancer deaths among 15 975 men.

Table 1 shows HRs for both colonic and rectal cancers after adjustment for age only; after adjustment for all risk factors and after adjustment for all risk factors after exclusion of deaths in the first 10 years of follow-up, respectively. Increasing age was associated with greater hazards of death from both colonic and rectal cancers in univariate and adjusted models. Absolute mortality rates for all colorectal cancers increased steeply with age at risk from 26.6 per 100 000 person-years in 50–59 year olds to 242.3 per 100 000 person-years in the 80+ year olds. Current smokers were at increased risk of both colonic and rectal cancer death, although an effect was only clearly seen for rectal cancer after 10-year exclusions. Former smokers were at increased risk of rectal cancer and this estimate was unchanged in the multiply-adjusted model and after early deaths were removed. There was heterogeneity in the age adjusted HRs for rectal cancer across the tertiles of plasma cholesterol (p=0.02) but no overall trend in the hazard across the whole cholesterol distribution. While colon cancer risks differed between physical activity groups in the age-adjusted model, the relationship was not consistent. There was no evidence for an association between diabetes mellitus and colorectal cancer mortality.

Table 1

Hazards ratios (95% CI) for the relation of baseline characteristics with colon and rectal cancer deaths in the original Whitehall Study


We found evidence in the original Whitehall study that being a current smoker increased the hazard of death from colorectal cancers. Current smokers were at 50% greater risk from colon cancer and had twice the risk of death from rectal cancer. We found evidence of a stronger association between current smoking and rectal cancers than between smoking and colon cancers, as previously reported.10 The effect of current smoking on rectal cancer only became clear after events in the first 10 years were excluded but it did not affect the association with colon cancer. This suggests that differential misclassification of smokers may have occurred because individuals with early symptoms of the disease stopped smoking as a result. Such a misclassification might lead to a ‘reverse causal’ relationship biasing the unadjusted observation toward the null. The effect on survival may be more pronounced for rectal cancers because patients are aware of symptoms at an earlier stage than for colonic cancers.24 The induction period of smoking on colorectal cancer is usually at least 30 years and this may explain why former smokers remained at increased risk. It is less clear why the risk was greater than among current smokers, however. We were unable to confirm an association between employment grade, physical activity, diabetes mellitus, systolic or diastolic blood pressures and colorectal cancer mortality.

Our results are confined to observations on mortality because information on cancer incidence was not available in the original Whitehall study. Mortality rates are necessarily products of both incidence and death, and risk factors for each may differ. While there is some evidence for the effects of age and modifiable behavioural risk factors on incidence, there is much less research on the effects of diet, body fatness and physical activity on colorectal cancer death (or its complement, survival) and it remains inconclusive.9 Smoking may be causally related to colorectal cancer incidence through both systemic circulatory exposure to carcinogenic compounds and to direct mucosal exposure from their ingestion.25 A satisfactory explanation for its greater effects on rectal cancer than on colon cancers10 is lacking. The steeper fall in smoking prevalence in the UK since 197026 may have contributed to reductions in mortality from colorectal cancer in the past decade as well as explaining overall higher colorectal cancer mortality rates in Scotland and Northern Ireland, as both countries have high prevalences of smoking.27 Smoking may also worsen survival in patients with colorectal cancer by reducing patients' tolerance of effective treatments—although data remain sparse28—or by causing other, non-cancer related diseases.29

The mechanisms by which metabolic risk factors operate on colorectal cancers are still unclear. Body and abdominal fatness may increase risk through systemic effects in which insulin, insulin-like growth factor-1 and oestrogen levels encourage carcinogenesis and discourage apoptosis.6 25 Physical activity is predominantly assumed to protect against colorectal cancers through its systemic metabolic effects, including reductions in blood pressure and insulin resistance, but, like fibre, it increases gut motility.9 However, we were unable to find a convincing relationship between either BMI or an intermediate pathway such as hypertension, hypercholesterolaemia, or diabetes mellitus and colorectal cancer mortality. The absence of an observed relationship between diabetes and mortality may have been due to the very low prevalence of diabetes in the Whitehall population; the relationship is most clearly seen among populations with higher prevalences of diabetes.30 Intermediately raised plasma cholesterol increased the hazard of death from rectal cancer by 80% but no association was found in the highest cholesterol tertile nor for colon cancer deaths. A convincing relationship between cholesterol and colorectal cancer risk was therefore not seen. The original Whitehall study was among six studies whose meta-analysis showed an unexpected association between low blood cholesterol and increased risk of colon cancer mortality. The conclusions of the original report in 1974, that the association did not appear to be related to the time interval between cholesterol assay and death,31 was subsequently found to be confined to the first 2 years after examination.32 The longer follow-up period and larger number of deaths in our analysis allowed us to extend the exclusion period and demonstrate more conclusively that any short-term association between low baseline cholesterol and colorectal cancer risk was likely to be because of the effects of the cancer on cholesterol and not the other way round.

We found no consistent association between self-reported physical activity and either colon or rectal cancer mortality, although any reductions in hazard were more apparent for rectal rather than colonic cancers, in contrast to previously published findings where the effect was more clearly seen for cancers of the colon.9 We did not find any evidence for differences in colorectal cancer risk between the employment grades even though socioeconomic deprivation may increase risks of colorectal cancers through a variety of behavioural mechanisms.2

Strengths and weaknesses

Our study is based on a large sample size with a high follow-up proportion. However, its weaknesses include unavailability of incidence data (discussed above), information biases from self-reported questionnaire data and limited information on exercise. It is possible that participants over-reported their exercise patterns and under-reported smoking. Both might have biased any observed relationships towards the null because of a loss of ability to discriminate between risk groups. The true effects of smoking might therefore be larger than observed and the absence of a clear relationship between physical activity and cancer risk might also mask true effects. In addition, it is likely that many of the exposures assessed at the start of follow-up will have changed over the course of 40 years, hence reducing the power to detect risk factor effects. The classification of physical activity was based on information from only two questions and misclassification may have biased the observed findings towards the null result we report. Follow-up data on risk factors were collected at a 25-year follow-up survey, but there are insufficient numbers of colorectal events among these men in the past 15 years since this resurvey to produce meaningful analyses.

The relative burden of colorectal cancer is projected to increase.33 Increasing age is the strongest risk factor for colorectal cancer incidence and mortality but it is not modifiable. This makes the role of other, modifiable, behaviours of greater public health importance. It has been estimated that modifications to diet, alcohol consumption, exercise and obesity have the potential to reduce colorectal cancer incidence by one-third in men and by one-fifth in women in the UK.34 35 However, these estimates are uncertain and dependent on the accuracy of the causal assumption underlying the modelling.

Smoking is an important modifiable risk factor for colorectal cancer mortality. Its decreasing prevalence in the UK may partly explain falling mortality from the disease, and its increasing prevalence in developing countries may contribute to future rises in colorectal cancer occurrence. Although we found no convincing evidence for an association between metabolic factors and colorectal cancer, the increasing prevalence of obesity and low levels of physical activity in many Western countries may make them important determinants of future colorectal cancer mortality. Further work is needed to examine these relationships among more recent populations with higher prevalences of metabolic risk factors.

What is already known on this subject

  • Several modifiable risk factors for colorectal cancer have been described, including smoking, physical activity and obesity. Their effects on colon and rectal sites vary.

  • Individuals may reduce their risk factors in response to early symptoms—for example, they may stop smoking or lose weight—so that the relationship between risk factors and cancer is no longer clear.

What this study adds

  • Smoking increases the risk of colon cancer mortality by about 50% and nearly doubles the risk of rectal cancer mortality. The effect of smoking on rectal cancer was only clear after events in the first 10 years of follow-up were removed, suggesting that early symptoms may have encouraged individuals to stop smoking.

  • It is important to adjust for the bias that may occur if early disease affects reported risk factors.

  • Smoking may be a significant determinant of national colorectal cancer mortality rates.



  • The original screening of the Whitehall study was funded by the Department of Health and Social Security and the Tobacco Research Council. DB is a Wellcome Trust fellow; MMis an MRC research professor.

  • Funding MS is supported by the British Heart Foundation and MK by the Academy of Finland, Finland, the National Heart, Lung, and Blood Institute (R01HL036310-20A2) and the National Institute on Ageing (R01AG034454-01), NIH, USA.

  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.