Article Text

Download PDFPDF

Adult height is associated with risk of cancer and mortality in 5.5 million Swedish women and men
  1. Emelie Benyi1,
  2. Marie Linder2,
  3. Johanna Adami3,
  4. Helle Kieler2,
  5. Mårten Palme4,
  6. Lars Sävendahl1
  1. 1 Department of Women's and Children's Health, Karolinska Institutet, Stockholm, Sweden
  2. 2 Department of Medicine-CPE (Centre for Pharmacoepidemiology), Karolinska Institutet, Stockholm, Sweden
  3. 3 Sophiahemmet University, Stockholm, Sweden
  4. 4 Department of Economics, Stockholm University, Stockholm, Sweden
  1. Correspondence to Dr Emelie Benyi, Department of Women's and Children's Health, Karolinska Institutet, Stockholm SE-171 64, Sweden; emelie.benyi{at}ki.se

Abstract

Background Previous studies have indicated that taller individuals are at greater risk of developing cancer. Death from cancer and other specific causes have also been linked to height, but the results have been inconclusive. We aimed to shed further light on the associations between height, cancer incidence and mortality.

Methods We conducted a nationwide, population-based prospective cohort study, including 5.5 million Swedish women and men (aged 20–74). They were followed over a period of up to 54 years. Heights were retrieved from national registers (mainly the Passport Register where heights are most often self-reported). The risks of overall and specific cancers, as well as overall and cause-specific mortality, were presented as HR with 95% CIs per 10 cm increase in height.

Results A total of 278 299 cases of cancer and 139 393 cases of death were identified. For overall cancer, HR was 1.19 (1.18–1.20) in women and 1.11 (1.10–1.12) in men for every 10 cm increase in height. All 15 specific cancer types were positively associated with height—most strongly for malignant melanoma in both genders, with HRs of 1.39 (1.35–1.43) in women and 1.34 (1.30–1.38) in men. For overall mortality, HR was 0.98 (0.97–0.99) in women and 0.91 (0.90–0.92) in men for every 10 cm increase in height. Cancer mortality was increased in taller individuals, with HR 1.15 (1.13–1.17) in women and 1.05 (1.03–1.07) in men for every 10 cm increase in height, whereas shorter individuals had increased overall mortality due to a number of other causes, such as cardiovascular disease.

Conclusion Overall and specific cancer risks, particularly malignant melanoma, were positively associated with height. Cancer mortality also increased with height. In contrast, overall mortality was decreased with height, particularly in men due to inverse associations with height for other causes of death.

  • tall stature
  • tumour
  • cause of death

Statistics from Altmetric.com

Introduction

Various anthropometric variables have been associated with cancer. Height has been linked to increased risks of cancers of the colon,1 rectum,1 breast,2 endometrium,3 ovary,4 kidney,5 central nervous system (CNS)6 and prostate,7 and to malignant melanoma,8 non-Hodgkin’s lymphoma9 and leukaemia.10 The World Cancer Research Fund stated in their 2018 report that there is convincing strong evidence that adult height is positively associated with colorectal, ovarian and breast cancer and that there is probable convincing evidence that birth weight is positively associated with breast cancer.11 They think that neither height nor birth weight may directly affect cancer risk, but they could be considered markers for genetic, nutritional and other environmental factors that affect growth in utero and during childhood and adolescence.

Adult height may be related to cancer mortality as well as overall mortality. Previous studies on the association between height and all-cause, cardiovascular and cancer mortality show conflicting results. Many studies show positive associations between height and cancer mortality, whereas cardiovascular mortality is negatively associated with height.12–17 A meta-analysis of 1 million individuals showed an increase in cancer mortality of 4% per 6.5 cm increase in height.14 Another meta-analysis reported a 50% higher relative risk for cardiovascular mortality in the shortest group compared with the tallest.17 However, the literature for overall mortality shows a large variation with regard to the direction of any height association.13–16 In order to shed further light on associations between height, cancer incidence and mortality, we conducted a nationwide, population-based prospective cohort study of more than 5 million men and women in Sweden.

Materials and methods

Study design and cohort

This is a prospective cohort study using the Swedish national registers to assess associations between height, cancer incidence and mortality.

The study population consisted of all individuals in the Swedish population born between 1938 and 1991 for whom adult height data were available in any of the national registers. Inclusion in the national health registers is a legal requirement, and they cover almost 100% of the population. Residents are included in the registers by virtue of informed consent effectively given on their behalf by the Parliament.18 The unique Swedish personal identity number assigned to each Swedish resident was used to link all register data related to each individual, since all registers include these numbers.19

Height variables

We collected information concerning height from the Medical Birth Register, the Conscription Register and the Passport Register. The Medical Birth Register, which started in 1973, includes information from antenatal, obstetric and neonatal records about women in childbirth, as well as information about the child, for example, measured birth length. Height data are most often self-reported in this register. The Conscription Register includes information obtained in connection with military recruitment. Enrolment was mandatory for men and enforced by law in Sweden until July 2010. All men, except those with severe disabilities, were enrolled and subsequently evaluated for military service, generally at the age of 18. Data on height from the Conscription Register were obtained for birth cohorts born between 1951 and 1981. The Swedish National Police Agency provided height data extracted from the Passport Register for all individuals in the Swedish population with a passport issued after 1991. The individual’s height was measured or self-reported at the time of the passport application. Information on birth length is available in the Medical Birth Register. All newborns were measured in the delivery unit and data reported to the register.

Adult height was defined as the individual’s height in centimetres (cm) which was appearing most frequently in the combined registers (the Medical Birth Register, the Conscription Register and the Passport Register). If two or more heights were equally frequent, the higher was chosen unless the difference between them exceeded 5 cm—in which case, the height information was considered unreliable and the individual was excluded. We chose the highest value since adult height is commonly reduced while ageing. Heights below 100 cm in both genders or above 225 and 240 cm for women and men, respectively, were considered unlikely to be correct and were excluded. Only heights from age 20 for men and 18 for women were included, to ensure that the individuals had reached their adult height. As a consequence of this, the Conscription Register will only give a small contribution as a large proportion of the men are measured before 20 years of age.

Other registers

Outcome information was collected from the Cancer Register and the Cause of Death Register. Diseases, complications and causes of death are coded according to the International Classification of Diseases, revisions 7 through 10 (tables 1 and 2).20

Table 1

ICD codes used to identify the cancer outcomes

Table 2

ICD codes used to identify the mortality outcomes

The nationwide Cancer Register started in 1958. It is mandatory to report to this register as well as other national health registers. All new cases of cancer and certain precancerous tumours must be reported by the healthcare givers. The register covers clinical and morphological diagnosis, tumour staging and date of diagnosis. Only the primary tumour is registered. If the location of the primary tumour is unknown, information about metastases is available. According to a study from 2009, the coverage of reported tumours is 96.3%.21 Approximately 99% of these tumours are verified morphologically.22

The Cause of Death Register contains dates and causes of death (excluding stillbirths) for all Swedish residents. Deaths occurring outside Sweden are included unless the individual has emigrated.

Statistics Sweden provided socioeconomic data from the Longitudinal Integration Database for Health Insurance and Labour Market Studies. The database, which started in 1990, includes information on employment, income or compensation from unemployment, parental leave, sick leave, studies, pensions, education level and place of residency for individuals from 16 years of age.

Cancer incidence and mortality follow-up

All malignant tumours reported to the Swedish Cancer Register between 1958 and 2011 were included in the assessment of total cancer risk. We also analysed 15 specific cancer sites which have previously been suggested to be associated with height.

All deaths registered from 1958 to 2011 were included in order to determine mortality rates in association with height. Total mortality was analysed, as well as 17 specific causes of death: accidental, circulatory, congenital, digestive, endocrine, genitourinary, haematological, ill defined, infectious, mental, musculoskeletal, neoplastic, nervous, perinatal, pregnancy related, respiratory and skin related.

Statistical analysis

We used Cox regression stratified by birth year (and adjusted for education and income in both cancer incidence and mortality analyses) to estimate HRs with 95% CIs for height. All analyses were carried out separately for men and women. The timescale used was age.

For cancer outcomes, individuals were censored at outcome, death, emigration or end of study, whichever occurred first. For specific causes of death, individuals were censored at death from other than the studied cause, emigration or end of study, whichever occurred first. Survival time was the time from entry into the study to a specific cancer or death from specified cause. As a sensitivity analysis we estimated an additional model using adult height, birth year and the interaction between them as explanatory variables. Since the clear majority of the CIs for additional HRs did not differ we do not present the results from this sensitivity analysis.

We performed separate analyses adjusting for education and income, since it has been suggested that these factors are linked to cancer incidence and mortality. We used the maximum income category (quartiles) over life for each individual in the Cox model. For education we had access to seven country-specific yearly education levels. The maximum education for each individual was retrieved and used as delivered. To handle missing values for maximum income and education we introduced a missing category for each of these variables. Moreover, we conducted a sensitivity analysis on the effect of smoking on the association between height and cancer among those in the female cohort for whom smoking information was available in the Medical Birth Register.

A subanalysis was performed for cancer incidence in association with birth length in individuals born between 1973 and 1991 for whom this information was recorded in the Medical Birth Register.

We also performed a subanalysis on cancer incidence in association with so-called ‘catch-up’ growth, described as an increase in standardised height from birth to adult height. The standardised height was defined within birth cohorts as the individual’s height minus the mean height divided by the SD. In the ‘catch-up’ analysis we added a covariate calculated as the difference between standardised adult height and standardised birth length to the Cox regression models. The catch-up covariate will have a high value for individuals born short but growing tall and a low value for individuals born tall, but becoming short adults. The covariate was categorised into three categories: shrunk 2 or more (smallest to −2), absolute difference within 2 (−1.99 to 1.99) and grown 2 or more (2 to biggest). This covariate was used to model time to cancer as a function of adult height and catch-up growth categories (shrunk, constant, grown). We report the HR with 95% CIs for the catch-up covariate to illustrate its influence on the cancer outcomes. Subanalyses of birth length and catch-up growth were added to distinguish between prenatal and postnatal factors and how they influence cancer risk.

The analysis of birth length and the so-called ‘catch-up’ analysis were restricted to individuals who contributed with both birth length and adult height.

The log-linearity of the HR with respect to the continuous predictor height was assessed by ranking height into 10 categories, constructing nine indicator variables and running the Cox models using these as covariates. The nine parameter estimates were plotted against the means of the height categories where the reference category was added with estimate equal to zero. Outcomes with plot patterns deviating from linearity were deemed to not fulfil the log-linear assumption. We chose 10 categories to enhance the visual interpretation of linearity.

We assessed the proportional hazards assumption by inspection of Kaplan-Meier curves. Finally, we also checked the adequacy of the Cox regression model by performing the graphical and numerical methods of Lin et al.23 In some of the analyses the number of cases is very small. However, we decided to present results for all outcomes since it might be of interest for the reader. All analyses were carried out using SAS V.9.4 (SAS Institute). Graphs were created using SAS and R.

Results

Characteristics of the cohort

The original cohort consisted of 2 795 119 women and 3 036 548 men born between 1938 and 1991 (figure 1). For 9930 women, information on adult height or death was missing or considered unreliable, and 47 528 women had emigrated. For men, 216 636 had missing or unreliable information on adult height or death, and 42 084 had emigrated.

Figure 1

Flow chart of the study population. F, females; M, males.

A total of 2 737 661 women and 2 777 828 men were included in the analysis and followed for a total of over 70 million person-years. During the follow-up, 152 481 women (5.6%) and 125 818 (4.5%) men developed cancer, and 54 668 (2.0%) women and 84 725 (3.1%) men died (table 3).

Table 3

Number of cancer cases, overall deaths and cancer deaths per birth cohort

Birth length was available for 659 280 women and 683 716 men. Information on smoking was available in 1 708 069 women.

Height and cancer incidence

The risks of developing cancer in relation to each 10 cm increase in height are presented as crude HRs and 95% CIs (figure 2). We found that the risks for developing malignant melanoma and Hodgkin’s lymphoma increased by 30%–40% among men and women. For women, we found 20%–30% risk increases for breast and ovarian cancer, Hodgkin’s lymphoma and leukaemia, and 10%–20% risk increases for cervix, kidney and colon cancer, and for myeloma. The risks for uterus, rectum and CNS cancer were increased by under 10%. For men, the risks of developing CNS, kidney and prostate cancer, non-Hodgkin’s lymphoma and myeloma increased by 10%–20% for every 10 cm increase in height. Colorectal cancer risk was increased by under 10%. A comparison of overall cancer risks between different percentiles of height is also illustrated in a Kaplan-Meyer curve (figure 3). In multivariable analyses, adjustments for education and income had no significant impact on the results, neither had smoking in the sensitivity analysis among women recorded in the Medical Birth Register. We therefore present crude estimates. When checking the adequacy of the proportional hazard model, the log-linear assumption may have been violated for melanoma (both sexes), myeloma (females), non-Hodgkin's lymphoma (males) and prostate cancer.23 By inspection of plots of estimates versus means some look slightly curved. The proportional hazards assumption seems adequate through the ocular inspection of the Kaplan-Meier curves.

Figure 2

HRs for overall and specific cancer risk. The associations between height and cancer risk are presented as crude HRs with 95% CIs per 10 cm increase in height for overall cancer as well as for 14 specific cancer types in females and 11 specific cancer types in males. HRs are stratified by age. CNS refers to all tumours in the central nervous system. Skin refers to non-melanoma skin cancer. Uterus refers to both endometrial cancer and sarcoma of the uterus. **P<0.01.

Figure 3

Kaplan-Meyer survival curves for total cancer risk. Women and men are presented separately. The curves represent the 1st, 25th, 50th, 75th and the 99th centiles in height.

Height and mortality

The risk of dying from cancer increased with height in both genders, whereas almost all other causes—of which cardiovascular disease was the most prevalent—instead decreased with height. The overall risk of dying during the follow-up period was lower in taller individuals (figure 4).

Figure 4

HRs for overall and cause-specific mortality. The associations between height and mortality are presented as crude HRs with 95% CIs per 10 cm increase in height for all-cause mortality as well as 17 specific causes of death in females and 16 specific causes of death in males . All risk estimates are stratified by age. Perinatal refers to deaths from long-term complications of events occurring in the perinatal period. Values below 0.1 and above 1.6 are truncated. *p<0.05; **p<0.01.

Birth length and ‘catch-up’ growth

No clear differences were found when comparing the height and cancer associations with regard to birth length and adult height, respectively (online supplementary figure 1). In this analysis there were 6830 cancer cases. Overall mortality appeared to be more strongly negatively associated with adult height than birth length, but differences within each cause of death could not be confirmed (online supplementary figure 2).

Supplemental material

Supplemental material

We also specifically analysed if the magnitude of the increase in height from birth to adulthood (so-called ‘catch-up’ growth) had an effect on cancer risk. In total, 7416 cancer cases were included in this analysis. No such additional effect was found (data not shown).

Discussion

This population-based cohort study with virtually complete long-term follow-up, the largest of its kind including both women and men, shows an association between height and cancer risk in the Swedish population. We report total cancer risk to be increased by 19% in women and 11% in men for every 10 cm increase in height. Cancer mortality was also positively associated with increasing height. In contrast, a number of other causes of deaths were negatively associated with increasing height. Overall mortality was lower in taller individuals, particularly in men, which may be explained by socioeconomic factors not controlled for.

Our results are in line with previous studies of cancer in association with height.2–11 24–26 A link between height and cancer risk was reported in a study of approximately 1.3 million women in the UK. They found a 16% increase in the risk of developing cancer for every 10 cm increase in height, which is close to our own finding of 19%.10 Two previous cohort studies reported increases of overall cancer per 10 cm in height of 13% and 14%, respectively, in women.25 26 One of these studies also included men, for whom the overall risk increased by 8% per 10 cm increase in height.26

With regard to specific cancer sites, we found that height was positively associated with CNS, colon, kidney, rectal and skin cancer as well as Hodgkin’s lymphoma, leukaemia and malignant melanoma in both women and men. It was also associated with breast, cervix and ovarian cancer in women and prostate cancer and myeloma in men. The strongest association with height in our study was seen for malignant melanoma, where the risk increased by 39% in women and 34% in men per 10 cm in height. Malignant melanoma also appears to have a strong positive association with height in many other studies.8 27 We also report that breast cancer, the most common type of cancer among Swedish women, increased by 23% and cancer in the uterus by 8% per 10 cm increase in height. There are a number of other smaller cohort studies on height and breast cancer with results in line with our own.2 24 28

We found that the log-linear assumption might be violated for melanoma (both sexes), myeloma (females), non-Hodgkin's lymphoma (males) and prostate cancer which could affect the results for these cancer sites. However, we chose to use the same model for all cancer forms, to enhance interpretation within study and comparability to other studies using Cox regression. The test of log-linearity by categorising the data is sensitive to choice of number of groups and cut-points.

Overall mortality was negatively associated with height, an association that was stronger in men. Previous studies of height in association with mortality have been inconclusive. In a meta-analysis of approximately 1 million individuals and 170 000 deaths, for every 6.5 cm increase in height, HR for all-cause mortality was 0.97 (95% CI 0.96 to 0.99).14 HR for death from cancer was 1.04 (95% CI 1.03 to 1.06) per 6.5 cm.14 In our study, the risk of death from cancer during follow-up increased by 15% per 10 cm increase in height in women and 5% in men. Some other studies have found a modest or no height-associated increase in cancer and overall mortality.12 16

In our study, the risk of death from cardiovascular disease showed a stronger relationship with height in women than men (approximately 20% vs 10%), but the association was negative, that is, shorter individuals were more likely to die of cardiovascular disease. A meta-analysis of more than 3 million individuals reported relative risks in shorter compared with taller individuals of 1.35 (95% CI 1.25 to 1.44) for all-cause mortality and 1.55 (95% CI 1.37 to 1.74) for cardiovascular mortality.17 For perinatal mortality, there were too few cases (five females and three males) to draw conclusions regarding any association with height.

The combined evidence for an association between height and cancer is now very strong. The next step is to establish the mechanisms behind this association. Every step in improving the understanding of the complex aetiology of cancer is a potential step closer to finding effective cancer therapies. One explanation behind a height-associated increase in cancer risk may be that a larger body has a higher number of cells, which increases the chance of malignant transformation. This theory fits with melanoma being most affected by height since the skin is the largest organ of the body. It is also possible that height affects cancer sites differently through separate mechanisms. However, it may not be the height itself that increases cancer risk but instead that taller people are more exposed to growth factors during childhood and adolescence as earlier hypothesised.29 One such growth factor is insulin-like growth factor-1 (IGF-1) which has been implicated in breast cancer biology.30 In contrast, circulating concentrations of IGF-1 measured in adulthood were recently found to not be associated with the risk of melanoma in a large international nested control study.31 Another possible explanation may be that taller individuals have a higher caloric intake, which has previously been linked with cancer.32

Some studies suggest that leg length plays a more significant role than overall height with regard to cancer risk, with certain points during childhood and adolescence being particularly critical.33 34 Whitley et al reported a slight increase in cancer risk with increasing childhood stature but no single anthropometric measure was of particular importance at any specific time point.35

While looking for factors related to childhood growth, we also studied whether catch-up growth was associated with cancer. Our results do not suggest that birth length or catch-up growth has a separate effect on cancer risk. However, the risk estimates were imprecise due to the fact that we only had access to data on birth length from the Medical Birth Register for 1.3 million individuals born after 1973 and to the low cancer rates in young individuals. With regard to the analysis of birth length and mortality, the risk estimates pointed in the same direction as for adult height but the association was weaker for birth length which could indicate that environmental factors or other factors related to postnatal growth may play an important role in the association between height and mortality. As for the cancer analysis, poor precision due to limited follow-up is a problem. The number of cases within some of the specific cancer sites or causes of death is very small.

A major strength of our study is the large size of the cohort with virtually complete long-term follow-up in extensive national registers which makes it unique compared with earlier studies. This reduces the risk of selection bias. Another advantage of our cohort is that it included both females and males. In addition, we performed a mortality analysis of the same cohort. A limitation of the study is that in some of the analyses there were very few cases which make these results very uncertain due to low statistical power. Moreover, there was a lack of complete data on certain lifestyle factors. Even though we had information on age, socioeconomic factors and in a subgroup also smoking, additional information on possible confounders would have added value. These include information on smoking for the whole cohort including men as well as other lifestyle factors such as sun exposure as it is an important risk factor for malignant melanoma and could possibly also be linked to height. A larger body surface may increase the risk of skin tumours, and socioeconomic status, which is known to be associated with height, may also be associated with levels of sun exposure through, for example, vacation habits. It would also have been desirable to include socioeconomic data before 1990 as well as information on premenopausal and postmenopausal status and cancer screening, which was not possible. However, we think that such influences would probably more likely be intermediate factors rather than confounders and hence not affect the association between height per se and cancer incidence or mortality.36 Also, in epidemiological studies, there is often a need to choose between high statistical power and detailed information. The national registers provide very good coverage, although there is a slight risk of unreported or misclassified cases. Furthermore, even though the study included the vast majority of the Swedish population, the fact that it did not include the relatively small number of individuals who had not given birth, not participated in military conscription, nor been issued a passport after 1991 could potentially give rise to selection bias. The height data were mostly self-reported rather than measured by a professional, which could potentially introduce a bias. However, it has been shown that self-reported heights correlate well with measured.37 Though self-reported heights might be incorrect, we presume that they would not be differential, for example, not differ between taller and shorter individuals, and considering the large cohort size they would unlikely alter the results significantly.

Conclusion

We found an association between adult height and total cancer incidence and cancer-specific mortality in the Swedish population. Malignant melanoma was the specific type of cancer most strongly associated with increasing height. The overall risk of death was somewhat lower in tall individuals, especially men. Several of the investigated causes of death, including cardiovascular disease, were lower with increasing height.

Our results suggest a link between height and cancer and between height and mortality for specific causes. Further research of the different mechanisms behind height-associated morbidity and mortality outcomes would be desirable.

What is already known on this subject

  • The majority of previous reports of height in association with cancer incidence and mortality indicate that height is positively associated with the risk of developing as well as dying from different types of cancer.

  • Cardiovascular mortality was negatively associated with height in most studies whereas the association with overall mortality was less clear.

  • Some studies found it to be negatively associated with height whereas a number of studies reported a positive association. However, the larger of these studies are meta-analyses which have their limitations, publication biases and heterogeneity of results. Furthermore, few published studies include both women and men.

What this study adds

  • Our study, one of few including both women and men, provides evidence of an association between height and cancer in a larger cohort than ever studied before and including far more cases of cancer than any previous cohort study of this subject. The size of the cohort gives a high precision in our estimates and an opportunity to perform subanalyses without losing statistical power. These analyses include any effect of smoking on the association between height and cancer and how birth length affects cancer risk.

  • Our results are in line with some of the largest previously published studies with regard to the risks of overall as well as site-specific cancers such as malignant melanoma and breast cancer.

  • We shed more light on the association between height and overall as well as cause-specific mortality in women and men separately.

Acknowledgments

The authors thank Dr David Hägg, Dr Laura Pazzagli and Dr Ondrej Soucek for graphic assistance.

References

Footnotes

  • Contributors EB, ML, JA, HK, MP and LS planned the study. ML performed the main statistical analyses. EB contributed in the statistical analyses. EB retrieved register data and wrote the first draft of the manuscript. ML, JA, HK, MP and LS contributed in writing the manuscript. Dr David Hägg, Dr Laura Pazzagli and Dr Ondrej Soucek contributed in creating figures. EB and ML had full access to all data in the study and all authors had final responsibility for the decision to submit for publication.

  • Funding The study was funded by the Swedish Research Council (grant number 2015-02406), Stockholm City Council, Stiftelsen Samariten, HKH Kronprinsessan Lovisas Förening för Barnasjukvård and Stiftelsen Frimurare Barnhuset i Stockholm.

  • Disclaimer The study sponsors had no involvement in the study design, writing of the report nor the decision to submit the paper for publication.

  • Competing interests None declared.

  • Patient consent for publication Not required.

  • Ethics approval The Regional Ethics Board at Karolinska Institutet in Stockholm approved all study protocols (approval number 2011/1267-31).

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

  • Data availability statement All data relevant to the study are included in the article or uploaded as supplementary information.

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.