SEARCH

SEARCH BY CITATION

Keywords:

  • height;
  • cancer incidence;
  • cohort studies

Abstract

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Although the influence of body mass index on cancer risk has been intensively investigated, few epidemiologic studies have examined the association of adult height with risk of cancer. We assessed the association of height with risk of all cancer and of 19 site-specific cancers in the Canadian National Breast Screening Study, a prospective cohort of nearly 90,000 women. Weight and height were measured at enrollment, and information on reproductive and medical history as well as lifestyle exposures was obtained by means of questionnaire. After exclusions, 5,679 incident invasive cancers were identified among 88,256 women. We used Cox proportional hazards model to estimate hazard ratios (HRs) and 95% confidence intervals (95% CI) per 10 cm increase in height. All tests of statistical significance were two sided. All cancers combined and ten specific sites (colorectum, colon, premenopausal breast, postmenopausal breast, endometrium, ovary, kidney, thyroid, melanoma and leukemia) showed statistically significant positive associations with height. The HR for all cancers combined was 1.13 (95% CI: 1.08–1.18), and the magnitude of the associations for specific sites ranged from HR 1.11 (95% CI: 1.03–1.20) for postmenopausal breast cancer to HR 1.51 (95% CI: 1.27–1.80) for melanoma. Our study provides strong support for a positive association of adult height with risk of certain cancers. The underlying biological mechanisms are not clear but may differ by anatomic site.

Over the past three decades, there has been intensive investigation of the associations of cancer with body mass index (BMI), a proxy for the degree of adiposity. Weight gain, most likely due to gain in fat, is a common feature of aging, and obesity has been demonstrated to be associated with increased risk of several types of cancers.1, 2 Cancer risk in association with adult height, which is influenced both by environmental exposures early in life as well as by genetics, has received much less attention. Several early studies reported that attained stature was associated with increased risk of all cancers and of specific cancers, including those of the breast, prostate and colorectum3–5; however, the numbers of cases in these studies were limited and the results were inconsistent.

More recently, the association of height with risk of cancer at various anatomic sites has been examined in several large cohort studies.6–9 Results of these studies suggest that (i) height is positively associated with risk of cancer at a wide range of anatomic sites and (ii) that the magnitude of the association varies considerably by site. However, there is a degree of inconsistency in the results by site across studies, and also by sex, in those studies that enrolled both males and females.7, 8 Furthermore, because both weight and BMI are associated with a number of cancers, and because these factors are also correlated with height, there is a need to tease out the association of height with cancer at specific sites independent of weight/BMI.10, 11 This issue has received scant attention in analyses examining the association of height with cancer risk.4–9

We therefore used data on measured height and weight on 89,835 Canadian women enrolled in the Canadian National Breast Screening Study (CNBSS) prospective cohort to examine the association of height with a wide range of cancer sites, controlling for potential confounding variables and assessing possible effect modification by other variables. We devote special attention to controlling for body weight by using a range of weight-for-height indices.

Material and Methods

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Study population and questionnaire

The CNBSS is a randomized controlled trial of screening for breast cancer, which has been described in detail elsewhere.12, 13 In brief, 89,835 women aged 40–59 were recruited from the general Canadian population between 1980 and 1985. On enrollment into the study, information was obtained from participants on demographic, hormonal, reproductive and lifestyle characteristics using a self-administered questionnaire. Participating women had their weight and height measured at the initial examination. Women were asked whether they were still having menstrual periods. If not, they were asked when their last menstrual period was (day, month and year). They were also asked whether they had had a hysterectomy and whether both ovaries had been removed. In addition, a subcohort of 49,654 women completed a food-frequency questionnaire that assessed intake of 86 food items, including alcoholic beverages and physical activity.14

Ascertainment of index cancers and deaths

Incident cases of cancer and deaths from all causes were ascertained by means of computerized record linkages to the Canadian Cancer Database and to the National Mortality Database, respectively. The linkages to the databases yielded data on cancer incidence and mortality to December 31, 2000 for women in Ontario, December 31, 1998 for women in Quebec, and December 31, 1999 for women in other provinces. For our analyses, study participants were considered at risk from their date of enrollment until the date of diagnosis of cancer at one of 19 different sites, termination of follow-up (the date to which cancer incidence data were available for women in the corresponding province) or death, whichever occurred first. We excluded women with missing information on height or weight (N = 868) and women with a cancer diagnosis preceding study entry (N = 711). After exclusions, during an average of 16.2 years (1,429,734 person-years) of follow-up, there were 5,679 incident invasive cancers in the cohort of 88,256 women. The following numbers of incident cases at specific sites were available for analysis: 1,908 premenopausal breast, 2,316 postmenopausal breast, 780 endometrium, 471 ovary, 1,096 colorectum (758 colon, 327 rectum and 11 colon + rectum), 757 lung (657 ever smokers and 100 never smokers), 182 pancreas, 196 kidney, 167 thyroid, 158 bladder, 91 cervix, 327 melanoma, 138 brain, 112 non-Hodgkin's lymphoma (NHL), 113 myeloma and 155 leukemia. Analyses of endometrial cancer were restricted to women with an intact uterus, and those of ovarian cancer were restricted to women with at least one intact ovary.

Statistical analysis

Cox proportional hazards models, with time-to-event (in days) as the time scale, were use to estimate hazard ratios (HRs) and 95% confidence intervals (95% CI) for the association of height with cancer risk. We examined the association of height categorized into quintiles with risk of all cancers and computed the HR for all sites combined and for each individual site per 10 cm increase in height. For each cancer site, we present the results of age-adjusted and multivariable-adjusted (MV) models. The following covariates were included in the main multivariable model: age at enrollment (continuous), menopausal status (premenopausal, perimenopausal and postmenopausal), years of education (<12, 12 and >12) and pack-years of smoking (continuous). An additional multivariable model included these covariates plus BMI. For breast cancer, analyses were stratified by menopausal status (premenopausal vs. postmenopausal) and included the additional variables: age at menarche (<12, 12, 13 and ≥14 years), parity (0, 1, 2 and ≥3), oral contraceptive use (ever and never), family history of breast cancer in a first-degree relative (yes or no), history of benign breast disease (yes or no) and hormone therapy (ever or never—for postmenopausal women). For cancers of the endometrium and ovary, reproductive factors and ever use of exogenous hormones were also included in the model. For lung cancer, analyses were performed after stratification by smoking status (ever vs. never). Because height is correlated with weight and BMI (r = 0.30 and −0.14, respectively), we compared the results of the age-adjusted model with those of the multivariable model adjusting for covariates other than weight-for-height and models including covariates plus weight-for-height metrics (W/H1.4 and W/H2). (W/H1.4 was used because this scaling of weight-for-height had a correlation of 0 with height in this population.11) In addition, we carried out stratified analyses to determine whether the association of height with cancer risk varied by level of potential effect modifiers, including education level, BMI, smoking status, age at menarche, parity, menopausal status, oral contraceptive use and hormone therapy, focusing on all cancers, cancers of the colorectum, breast and endometrium (sites with the largest number of cases). We formally tested for interactions between height (<162 cm/≥162 cm) and potential effect modifiers on the risk of all cancers by comparing the fit of models with and without the product terms representing the variables of interest using a likelihood ratio test. All statistical significance tests were two sided. All analyses were performed using SAS version 9.1 (SAS Institute, Cary, NC).

Results

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Baseline weight, years of education and % ever used oral contraceptives showed statistically significant positive associations with height, whereas BMI, % postmenopausal, % ever used hormone therapy and % age at menarche <12 years showed statistically significant inverse associations with height (Table 1).

Table 1. Baseline characteristics by quintiles of height in the Canadian National Breast Screening Study
inline image

Risk of all cancers combined increased with increasing height. Compared to women in the lowest quintile (<157 cm), those in the highest quintile (≥167 cm) had a 24% increased risk (HR 1.24, 95% CI: 1.14–1.34, p for trend 0.006) (Table 2). The age-adjusted risk of all cancers per 10 cm increase in height was 1.13 (95% CI: 1.09–1.18). Adjustment for individual covariates and for all covariates combined had little effect on this risk estimate (Table 3).

Table 2. Hazard ratios and 95% confidence intervals for quintiles of height with risk of all cancers in the CNBSS
inline image
Table 3. Effect of adjustment for different potential confounding variables on the association of height with all cancer (per 10 cm increment) in the CNBSS
inline image

Height was significantly positively associated with risk of cancer at ten anatomic sites in both age-adjusted and multivariable models that included BMI (Table 4). HRs and 95% CIs from the multivariable models including BMI were as follows: colorectum (1.13, 95% CI: 1.02–1.24), colon (1.12, 95% CI: 1.00–1.26), premenopausal breast (1.11, 95% CI: 1.03–1.20), postmenopausal breast (1.12, 95% CI: 1.05–1.20), endometrium (1.36, 95% CI: 1.22–1.52), ovary (1.21, 95% CI: 1.04–1.40), kidney (1.28, 95% CI: 1.02–1.60), thyroid (1.39, 95% CI: 1.09–1.78), melanoma (1.51, 95% CI: 1.27–1.80) and leukemia (1.30, 95% CI: 1.01–1.68) (Table 4). In addition, the positive association of height with rectal cancer bordered on statistical significance: 1.15 (95% CI: 0.97–1.37). For two other sites, brain and NHL, risk was elevated (above the HR for all cancers) but nonsignificantly. Figure 1 presents the ranking of sites by the magnitude of the HR.

thumbnail image

Figure 1. Hazard ratios and 95% confidence intervals per 10 cm increase in height for incident cancer at 19 different sites and for all cancers. The dotted line represents the HR per 10 cm increase. HRs are adjusted for BMI and for other covariates shown in Table 4.

Download figure to PowerPoint

Table 4. HRs for the association of height (per 10 cm increase) with specific cancer sites, with adjustment for different covariates in the CNBSS
inline image

In most cases, the results of age-adjusted models and those of MV models were similar, indicating little overall confounding (Table 4). However, for two sites (endometrium and kidney), there were large differences between the results of different models. For endometrial cancer, the MV model containing BMI yielded an HR of 1.36 (95% CI: 1.22–1.52), whereas the HRs from the age-adjusted model and the MV model without BMI were 1.12 and 1.14, respectively. Kidney cancer showed a similar, albeit smaller, difference. For all sites, adjustment for W/H1.4 yielded similar results to those for BMI (data not shown). For the remaining sites that showed associations with height, the association was consistent across all three models.

The association of height with risk of all cancers was generally consistent across strata of other variables (Fig. 2). However, the association appeared weaker among older women, among those with fewer years of education, current smokers, ever users of hormone therapy and among premenopausal women. Only the interactions of height with age, smoking status and menopausal status were statistically significant (p < 0.0001). When a product term for age × height was included as a continuous variable in the model for all cancers combined, the interaction was no longer significant. Some variation in the HR by strata of potential effect modifiers was observed for colorectal, breast and endometrial cancers; however, none of the interactions was statistically significant (data not shown).

thumbnail image

Figure 2. Hazard ratios and 95% confidence intervals per 10 cm increase in height for all incident cancers, stratified by characteristics at enrollment. The dotted line represents the HR per 10 cm increase. HRs are adjusted for age at entry, menopausal status, level of education and pack-years of smoking, where appropriate.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

In the CNBSS cohort of 88,256 women, in whom 5,679 incident invasive cancers were identified over 16.4 years of follow-up, height was positively associated with risk of all cancers and with cancers of the colorectum, colon, breast, endometrium, ovary, kidney, thyroid, melanoma and leukemia. The HR for all cancers per 10 cm increment in height was 1.13 (95% CI: 1.08–1.18), and the magnitude of the associations for specific sites ranged from HR 1.11 (95% CI: 1.03–1.18) for postmenopausal breast cancer to HR 1.51 (95% CI: 1.27–1.80) for melanoma. In general, adjustment for a range of confounding variables had minimal effect on the associations; however, the HRs for endometrial cancer and kidney cancer were sensitive to whether BMI was included in the model. The association of height with risk of all cancers combined was observed within all strata of potential effect modifiers.

The association of height with a range of cancer sites has been examined in only a few cohort studies.6–9 These studies, and several earlier ones,4, 5 have demonstrated a positive association between adult height and risk of all cancers combined and of cancers at specific anatomic sites,7, 9 as well as with all cancer mortality and site-specific mortality.6, 8 By far the largest study, the Million Women Study,9 observed significant positive associations of height with cancers of the colon, rectum, breast, endometrium, ovary, kidney, central nervous system, melanoma, leukemia and NHL. In our study, we observed significant associations for all of these sites, with the exception of brain and NHL, for which there were suggestive positive associations. In addition, we observed a significant positive association with thyroid cancer, on which the Million Women Study did not report. There is also a similarity between our results and those of Sung et al.7 and Batty et al.6, 8

Concerning the magnitude of the associations reported in different studies, there is some degree of consistency. Melanoma, leukemia, kidney cancer and colorectal cancer were among the highest ranking sites across the four studies that included women (Refs.7–9 and our study). Among men in the Whitehall cohort,6 melanoma and kidney cancer were the two top-ranking sites. Among men in the Sung et al.'s study,7 thyroid cancer, lymphoma and melanoma were the highest-ranking sites. Only our study and Sung et al.'s study presented data on thyroid cancer.

Pooled analyses and reports from large cohort studies focusing on specific cancer sites provide additional support for the existence of associations of height with melanoma,15 with cancers of the endometrium,16, 17 ovary,18 breast,19 colorectum,20–22 brain23 and thyroid24, 25 and with lymphatic malignancies.26 However, regarding other sites, including lung, bladder, kidney and prostate, the evidence is less consistent.

Some studies have reported minimal evidence of confounding.7, 9 Sung et al.7 presented the results of models for specific sites with different levels of adjustment, whereas Green et al.9 presented the results obtained by including different covariates for the association of height with all cancers but not with individual cancers. In our study, although a number of baseline characteristics were associated with height, confounding by these covariates appeared to be minimal. However, because weight is correlated with height and because caution has been recommended regarding possible confounding of the association of height with cancer risk by body weight,10, 11 we examined the effect of adjusting for covariates other than BMI vs. including BMI in the multivariable model. For most cancers that showed a positive association with height, the HR was consistent across the three models presented in Table 4. However, for endometrial cancer and kidney cancer (both strongly associated with weight and BMI in our data, p < 0.0001), the risk estimates were markedly affected by the presence/absence of BMI in the MV model. Our results suggest that caution is in order when assessing the association of height with cancers that are also associated with BMI.

The association of height with all cancers was statistically significant within strata of potential effect modifiers, with the exception of current smokers and ever users of hormone therapy, for whom the association was of borderline significance. Only the interactions of height with age, smoking status and menopausal status reached statistical significance. Our results provide confirmation of the finding of Green et al.9 that smoking history is an important effect modifier. In both studies, height was more strongly related to cancer risk among nonsmokers than among smokers. This is important because the prevalence of smoking is declining in Western countries, and therefore the attributable risk for height will likely increase over time.

There has been a secular trend toward increased height in many countries starting in the late 19th century.27 Attained stature is known to be affected both by genetic factors and environmental exposures in early life. These two different components appear to interact, so that environmental exposures may determine whether the genetic potential to reach maximum height is realized.27, 28 Relevant early-life environmental exposures include energy intake and circulating levels of growth hormones.27, 28 It has long been known that experimental animals on a calorie-restricted diet (but balanced in terms of micronutrients) have a reduced cancer incidence and greater longevity compared to controls fed ad libitum.29, 30 A controlled trial of calorie restriction in humans31 and an observational study in which children's energy consumption was assessed through parental interviews and the children were followed into adulthood32 appear to support the animal findings regarding cancer incidence. A high-energy intake in childhood and adolescence is associated both with increased adult height33 and with higher levels of insulin-like growth factor-1 (IGF-1) in early adulthood.34 IGF-1 levels and the IGF-1-to-IGFBP3 ratio have been linked to increased risk of certain cancers, particularly those of the colorectum, premenopausal breast and prostate.35

Other explanations that have been proposed for the positive association between height and cancer risk include the possibility that genes that influence height also influence cancer risk3 and that greater stature is associated with greater organ size and, thus, a greater number of cells per organ that are susceptible to malignant transformation.36 Batty et al.27 have pointed to the latter of these two explanations as a possible explanation for why skin cancer (presumably, predominantly melanoma, as their study focused on mortality) showed the greatest height-malignancy gradient in the original Whitehall study.

The striking differences in the magnitude of the association of height with risk of specific cancers and the heterogeneity of types of cancers associated with height suggest that either there are different mechanisms by which height influences risk or that height may enhance the effect of other risk factors. A possible example of the latter explanation is that in a pooled analysis of eight case–control studies of height and melanoma risk, height was positively associated with melanoma overall; however, in stratified analyses, an association was present only in women aged <50 years.15 Identifying the etiologic factor(s) for which height may be a proxy will be difficult given the challenge of directly assessing early diet and other environmental factors, and growth factors, such as IGF-1, in a sufficiently large cohort, which is then followed for many years for the development of a range of cancers.

Strengths of our study include its prospective nature, substantial numbers of incident cancer cases identified through national registries, information on height and weight measured at baseline and information on a range of covariates. Among study limitations is the fact that information on dietary and alcohol intake and physical activity was only available for roughly half of the cohort members. However, when we carried an analysis including alcohol intake and physical activity, the results were essentially unchanged. Additionally, we were not able to adjust for skin color, eye color or sunlight exposure in the analysis of melanoma risk; however, in a pooled analysis of eight case–control studies, a significant association of height with risk was not affected by adjustment for known risk factors for melanoma.15 We did not have information on parents' height and thus were unable to adjust for this hereditary component of height. Finally, as our cohort was limited to women, we were unable to compare associations of height for those sites common to both men and women.

In conclusion, height measured at baseline was positively associated with increased risk of all cancer as well as with cancer at ten of the 19 specific sites examined. Our study provides important corroboration for the height–cancer association and contributes information regarding sites for which existing data were either equivocal (e.g., ovary) or limited (e.g., thyroid). In addition, it confirms the finding that smoking history modifies the association of height with cancer risk. Although identifying the underlying causes responsible for these associations is challenging, studies examining exposures and circulating levels of growth factors in childhood and adolescence could contribute to an understanding of why height is associated with risk of certain cancers but not others and why there are substantial differences in the magnitude of the association.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Dr. Rohan is supported by a grant from the Breast Cancer Research Foundation.

References

  1. Top of page
  2. Abstract
  3. Material and Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References