The global, regional, and national burden of colorectal cancer and its attributable risk factors in 195 countries and territories, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017

Summary Background Data about the global, regional, and country-specific variations in the levels and trends of colorectal cancer are required to understand the impact of this disease and the trends in its burden to help policy makers allocate resources. Here we provide a status report on the incidence, mortality, and disability caused by colorectal cancer in 195 countries and territories between 1990 and 2017. Methods Vital registration, sample vital registration, verbal autopsy, and cancer registry data were used to generate incidence, death, and disability-adjusted life-year (DALY) estimates of colorectal cancer at the global, regional, and national levels. We also determined the association between development levels and colorectal cancer age-standardised DALY rates, and calculated DALYs attributable to risk factors that had evidence of causation with colorectal cancer. All of the estimates are reported as counts and age-standardised rates per 100 000 person-years, with some estimates also presented by sex and 5-year age groups. Findings In 2017, there were 1·8 million (95% UI 1·8–1·9) incident cases of colorectal cancer globally, with an age-standardised incidence rate of 23·2 (22·7–23·7) per 100 000 person-years that increased by 9·5% (4·5–13·5) between 1990 and 2017. Globally, colorectal cancer accounted for 896 000 (876 300–915 700) deaths in 2017, with an age-standardised death rate of 11·5 (11·3–11·8) per 100 000 person-years, which decreased between 1990 and 2017 (−13·5% [–18·4 to −10·0]). Colorectal cancer was also responsible for 19·0 million (18·5–19·5) DALYs globally in 2017, with an age-standardised rate of 235·7 (229·7–242·0) DALYs per 100 000 person-years, which decreased between 1990 and 2017 (−14·5% [–20·4 to −10·3]). Slovakia, the Netherlands, and New Zealand had the highest age-standardised incidence rates in 2017. Greenland, Hungary, and Slovakia had the highest age-standardised death rates in 2017. Numbers of incident cases and deaths were higher among males than females up to the ages of 80–84 years, with the highest rates observed in the oldest age group (≥95 years) for both sexes in 2017. There was a non-linear association between the Socio-demographic Index and the Healthcare Access and Quality Index and age-standardised DALY rates. In 2017, the three largest contributors to DALYs at the global level, for both sexes, were diet low in calcium (20·5% [12·9–28·9]), alcohol use (15·2% [12·1–18·3]), and diet low in milk (14·3% [5·1–24·8]). Interpretation There is substantial global variation in the burden of colorectal cancer. Although the overall colorectal cancer age-standardised death rate has been decreasing at the global level, the increasing age-standardised incidence rate in most countries poses a major public health challenge across the world. The results of this study could be useful for policy makers to carry out cost-effective interventions and to reduce exposure to modifiable risk factors, particularly in countries with high incidence or increasing burden. Funding Bill & Melinda Gates Foundation.


Introduction
In 2016, cancer accounted for more than 213 million disability-adjusted life-years (DALYs) and 8·9 million deaths globally. 1,2 The burden of cancer is usually reported in aggregated form, 1,3 but cancer-specific reports allow a more detailed exploration of the problem by providing information that is useful for the development and evaluation of cancer-specific prevention programmes, screening strategies, treatment, and resource allocation. An understanding of the geographical and temporal trends in colorectal cancer is important because it was the second leading cause of death (age-standardised and all ages) among cancers globally in 2017 and the 16th leading cause of death among all diseases and injuries. 4 Trends in the burden of colorectal cancer have been subject to substantial changes across the world because of the expansion of screening programmes, with wide recommendation of colonoscopy in the late 1990s, as well as changes in risk factors associated with colorectal cancer. 5,6 Whereas colorectal cancer age-standardised death rates have stabilised or declined in many high-income countries, which historically had the highest burden of colorectal cancer in the world, 7 the burden is increasing in most low-income and middle-income countries, 8 possibly as a result of ageing populations, urbanisation, and increased prevalence of westernised lifestyle risk factors, such as alcohol consumption, obesity, smoking, and suboptimal diet. 9,10 The global burden of colorectal cancer attributable to various modifiable risk factors has not been described elsewhere and is an important estimate to report because it has implications for policy making and prevention efforts.
Studies reporting the global burden of colorectal cancer have been published previously but have several limitations. Specifically, previous estimates reported the global burden of colorectal cancer in terms of incidence and mortality but did not aim to calculate important measures such as years of life lost (YLLs), years lived with disability (YLDs), and DALYs. 3,7,[11][12][13][14] Moreover, although the burden of colorectal cancer and trends associated with this disease have been reported up to 2018, the temporal trends occur at 4-year or 6-year intervals for most countries and 95% uncertainty intervals (UIs) have been provided only for the most recent global estimates in 2018. 7,[11][12][13]15 Finally, the association between countries' development status and colorectal cancer burden has previously been described using Global Cancer Incidence, Mortality and Prevalence (GLOBOCAN) data from only a subset of countries. 16 We aimed to report the incidence, mortality, and disability due to colorectal cancer and its attributable risk factors from 1990 to 2017 in 195 countries and territories, by age, sex, Socio-demographic Index (SDI; a composite of sociodemographic factors), and Healthcare Access and Quality (HAQ) Index, an indicator of health system performance.

Overview
This study is part of the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD), which covers seven super-regions, consisting of 21 regions containing 195 countries and territories. The most up-to-date iteration, GBD 2017, reported estimates for 359 diseases and injuries; 282 causes of death; and 84 behavioural, environmental and occupational, and metabolic risk factors. The general methodology used and updates to the methodology have been previously presented in GBD 2017 papers. 4,[17][18][19][20][21] Briefly, the mortality-to-incidence ratio (MIR) estimation was updated from GBD 2016, with use of the HAQ Index rather than the SDI in the data cleaning and modelling process, and the spatiotemporal Gaussian process regression approach was also updated. Covariate inputs for the Cause of Death Ensemble model (CODEm) were updated and changed on the basis of recommendations from GBD collaborators. The rates were standardised according to the GBD world population and reported per 100 000 person-years. 17 The method for propagating uncertainty in this paper is similar to that used in previous GBD 2017 papers. 4,19 The distribution of every step in the computation process is stored in 1000 draws that are used for every other step in the process. The distributions are characterised from the sampling error of data inputs, the uncertainty of the model coefficients, MIRs, and age-specific death rates. GBD assumes that uncertainty in the MIR is independent of uncertainty in the estimated age-specific death rates. Final estimates were computed using the mean estimate across 1000 draws, and the 95% UIs were specified on the basis of the 25th and 975th ranked values across all 1000 draws. The GBD study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER).

Data sources
All cancers coded as C18-C21, D01.0-D01.2, and D12-D12.8 in the 10th revision of the International Classification of Diseases were considered to be colorectal cancer. 19 Vital registration (18 857 site-years of data), sample vital registration (761 site-years), verbal autopsy (660 site-years), and cancer registry (4474 site-years) data from GBD 2017 were used in this study. 4 Vital registration is the system by which governments record the vital events of their residents, including causes of death. In sample vital registration, vital events are recorded in nationally representative cluster samples to estimate birth rates, deaths rates, and causes of death for the total population in countries where high coverage of vital registration is not available. Verbal autopsy is a method by which trained interviewers collect information about the signs, symptoms, and demo graphic characteristics of a recently deceased person from an individual familiar with the deceased to determine individuals' causes of death and cause-specific mortality fractions in populations without a complete vital registration system. Finally, a cancer registry gathers data on every person with cancer in a defined population, usually comprising residents in a well defined geographical region. The details on data quality rating for 195 countries and territories are provided in the appendix (pp [11][12][13][14][15][16][17]. More detailed information about the data sources used for each country can be found on the GBD 2017 Data Input Sources Tool website.

Mortality estimates
Mortality data from vital registration, sample vital registration, and verbal autopsy were sparse. Therefore, incidence data from cancer registries were converted into mortality data by modelling the MIRs independently. We modelled MIRs using the locations that had both incidence and mortality data available for the same year. The initial MIR model used a linear-step mixed-effects model with logit link functions, as well as the HAQ Index, age, and sex as covariates. The resulting estimates were then smoothed over space and time, and adjusted with spatiotemporal Gaussian process regression. 18 We used the observed mortality (from vital registration and verbal autopsy) and mortality estimates (computed from the MIRs and incidence data) as inputs for a CODEm. 4 Country-level covariates used for the CODEm and the assumed directions are described in the appendix (p 18). We used CODEm to select which predictors produce the best fit to the data. We used the CoDCorrect algorithm to adjust the sum of predicted single-cause mortalities in an age-sex-location-year group to be consistent with the results from all-cause mortality estimation. 4

Non-fatal estimates
The final mortality estimates were divided by the MIR to compute colorectal cancer incidence. 19 Colorectal cancer prevalence was calculated by estimating 10-year survival based on MIRs and adjusting for expected background mortality. The cohort members who had survived more than 10 years were assumed to be cured, and one of the two sequelae were assigned to them: the diagnosis and primary therapy phase or the controlled phase. The controlled phase included all patients who survived more than 10 years and who had finished primary therapy. The prevalence for the cohort in which people died during the 10-year period was categorised into four sequelae (appendix p 20). The diagnosis and primary therapy phase was defined as 4·0 months, the metastatic phase as 9·7 months, and terminal phase as 1 month. 22,23 The remaining time was assigned to the controlled phase. The duration of sequela one (diagnosis and primary therapy) described by Allgar and colleagues 22 was used and 2 months were added to account for the average treatment duration. Duration of sequela two (controlled phase) was 10 years for the survivors minus the duration of the other sequelae. Duration of sequela three (metastatic phase) was based on Surveillance, Epidemiology, and End Results (SEER) data for median survival of patients with stage IV disease. A duration of 1 month for sequela four (terminal phase) was used for all cancers. 22 To estimate procedure-related disability for all locations by age, sex, and year, we used hospital data on the proportion of patients that undergo ostomies (ie, the procedure proportion) as our input for a DisMod-MR 2.1 proportion model. 19 We determined through a literature review that an average of 58% of all ostomies are for colorectal cancer, so we multiplied the all-cause ostomies by 0·58. [24][25][26] We applied these procedure proportions to the number of incident cases of colorectal cancer and multiplied that by the proportion of the incident population that had survived for 10 years. This process gave us the number of incident cases of colorectal cancer that involved an ostomy procedure and survived beyond 10 years. We then input these cases into DisMod-MR 2.1. This model produced estimates of incidence and lifetime prevalent cases of people with colorectal cancer-related stomas who have survived beyond 10 years. 19 See Following this process, to estimate the sequela-specific YLDs, procedure sequelae prevalence and general sequela prevalence rates were multiplied by the sequelaspecific disability weight. The disability weights for four sequelae and one procedure can be found in the appendix (p 19). 19 The disability weights ranged from 0 (perfect health) to 1 (equivalent to death). GBD uses different disability weights for the four phases of

SDI and HAQ Index
We used the GBD 2017 SDI and GBD 2016 HAQ Index to determine the association a country's development level had with colorectal cancer age-standardised DALY rates. Examining the association of development level (SDI) and health system performance (HAQ Index) with colorectal cancer burden is important because these factors affect the prevalence of cancer risk factors. In GBD 2017, the SDI was revised to better reflect the development status of each country. 4,[18][19][20][21] The SDI ranges from 0 (worst) to 1 (best) and incorporates the total fertility rate in women under the age of 25 years, mean education for individuals aged 15 years and older, and lag-distributed income per person. The HAQ Index reflects the personal health-care access and quality for 195 countries and territories and was calculated on the basis of amenable mortality (ie, deaths from causes that should not occur in the presence of effective medical care). The HAQ Index ranges from 0 (worst) to 100 (best). Further details on the HAQ Index are presented elsewhere. 27

Risk factors
We selected risk factors that had evidence of causation with colorectal cancer. We extracted the relative risks and exposure estimates from all available data sources. We calculated a population attributable fraction as the proportional reduction in a health outcome that would occur if exposure to a risk factor was reduced to the theoretical minimum exposure level. We reported the proportion of DALYs due to colorectal cancer that were attributable to smoking, high body-mass index, high fasting plasma glucose, low physical activity, and five dietary risks (diets low in calcium, milk, and fibre, and diets high in red meat and processed meat). Details on definitions of these risk factors and their relative risk for colorectal cancer, prevalence of risk factors, and methods for quantifying the proportion of the burden of colorectal cancer attributable to these risk factors are described elsewhere. 18 The DALYs due to colorectal cancer that were attributable to each risk factor were estimated by multiplying the total DALYs for colorectal cancer by the population attributable fraction for the risk-outcome pair for each age group, sex, location, and year.

Role of the funding source
The funder of the study had no role in study design; the collection, analysis, or interpretation of the data; or the writing of the report. The corresponding authors had full access to the data and had responsibility for final submission of the manuscript.

Figure 1: The age-standardised incidence (A) and death (B) rates of colorectal cancer for 21 GBD regions by sex, 2017
Error bars indicate 95% uncertainty intervals.  [21][22][23][24][25][26][27][28][29]. Percentage change increments in age-standardised incidence rates of colorectal cancer from 1990 to 2017 were higher among males in most regions except Andean Latin America and south Asia ( figure 2A). Similarly, percentage change increments for colorectal cancer age-standardised death rates in this period were highest in males in most regions, except for south Asia (figure 2B). In 2017, the highest number of incident cases were found in east Asia, western Europe, and high-income North America (table; appendix p 1). The highest numbers of deaths were in east Asia, western Europe, and high-income North America in 2017 (appendix pp 2, 21-28).
In Percentage change in age-standardised incidence rate (per 100 000 person-years)

Figure 2: The percentage change in age-standardised incidence (A) and death (B) rates of colorectal cancer for 21 GBD regions by sex, 1990-2017
Error bars indicate 95% uncertainty intervals. Our study found that, in 2017, the incidence rate increased in a non-linear manner with increasing age and was higher in males than in females across all age groups (figure 4). The difference in incidence rates between males and females increased with each increasing age group up to the ages of 85-89 years, after which the gap started to decrease again. The number of incident cases was also higher in males than in females up to the ages of 80-84 years and peaked at ages 65-69 years ( figure 4). A relatively similar pattern was also observed for death rates and death counts (appendix p 3). The highest rates of incidence and death observed were in the oldest age group (≥95 years) for both sexes in 2017. The pattern for DALY rates was slightly different, such that the age-standardised DALY rate started decreasing after the ages of 80-84 years for males and after the ages of 85-89 years for females (appendix p 4). The number of DALYs was also higher in males than in females up to the ages of 80-84 years, and then females had slightly higher numbers of DALYs for the older age groups. The number of DALYs followed a normal distribution and peaked at ages 65-69 years (appendix p 4). Decomposition of the DALY rate into YLLs and YLDs showed that YLLs were the primary contributor to DALYs, with the 2017 YLL rate peaking at the ages of 80-84 years (appendix p 5). Figure 5 presents the global and regional-level observed age-standardised DALY rates from 1990 to 2017 versus the expected level based only on the SDI values of the global regions. The expected pattern was non-linear in nature, peaking at an SDI value of approximately 0·75, before decreasing with increasing SDI values. However, there were large regional differences. Australasia, central Europe, western Europe, and high-income North America showed the largest decreases in observed agestandardised DALY rates with increases in SDI value, whereas the Caribbean and central Latin American regions showed increases in observed age-standardised DALY rates with increasing SDI value. The observed agestandardised DALY rate for some regions, such as southern sub-Saharan Africa, initially increased and then decreased with an improvement in SDI value over time. At the global level, the age-standardised DALY rate dropped below the expected level for 2015-17. Figure 6 shows the national-level observed agestandardised DALY rates and their association with the SDI and HAQ Index. The expected patterns were nonlinear in nature, peaking at an SDI value of approximately 0·81 and HAQ Index value of approximately 84, before decreasing with increasing SDI and HAQ Index values. However, there were large national differences. Several countries, including Hungary, Greenland, Slovakia, Serbia, and Brunei, had a higher than expected agestandardised DALY rate, whereas others, such as Iraq, Maldives, Sri Lanka, Kuwait, and Oman, had much lower than expected age-standardised DALY rates based only on the SDI. This pattern was also observed based on the HAQ Index.

Figure 4: Global number of incident cases and incidence rate of colorectal cancer per 100 000 person-years by age and sex, 2017
Error bars indicate the 95% uncertainty interval for incident cases. Shading indicates the 95% uncertainty interval for the incidence rate.

Discussion
From 1990 to 2017, the age-standardised incidence rates of colorectal cancer increased globally, with substantial regional and national heterogeneity. By contrast, the agestandardised death and DALY rates decreased across the study period. On the basis of our DALY estimates, colorectal cancer is the 36th leading cause of disease burden globally for 2017, and is the fourth leading cause of cancer burden, behind only lung cancer, liver cancer, and stomach cancer. The most recent GLOBOCAN report 3 3 we found that the highest age-standardised incidence rates in 2017 were in Australasia, high-income Asia Pacific, and high-income North America, and the highest age-standardised death rates were found in central Europe, eastern Europe, and southern Latin America.
We also investigated heterogeneous trends in agestandardised incidence, death, and DALY rates from 1990 to 2017 at the national level. Most countries showed an increase in the age-standardised incidence rate of colorectal cancer during 1990-2017, such that only Australasia and high-income North America experienced a decrease in age-standardised incidence rate at the regional level. One potential explanation for this global increase in age-standardised incidence is that the introduction of screening tests might have led to increased detection and thus increased incidence, but this increase might be short-lived because of the removal of precancerous polyps during colonoscopies. 5 Similarly, in countries where screening programmes were established two or three decades ago, reductions in death rates were observed that support the benefits attributable to screening interventions. 28 Improving survival by adopting the best practices in cancer treatment and management can also lead to reduced death rates. On the basis of the data from high-income countries, several factors might have contributed to the decrease in the number of deaths due to colorectal cancer, such as enhanced access to screening colonoscopy and early stage detection, as well as improved surgical techniques, radiotherapy, chemotherapy, targeted therapy, and palliative care. [29][30][31][32] Key interventions to decrease deaths from colorectal cancer include the removal of polyps and early detection interventions, such as colonoscopy, flexible sigmoidoscopy, faecal occult blood testing, and faecal immunochemical testing. Previous research has investigated the association between a country's development level and the incidence and mortality rates of colorectal cancer, using the Human Development Index. 7 Because one of the components of the Human Development Index is health related, it is not optimal to use this index when comparing the health outcomes of countries. To avoid this problem, we used the SDI, which does not contain any health-related measures. Our analysis of the association between the SDI and age-standardised DALY rate of colorectal cancer produced results that have not been previously reported. In some regions, such as Australasia and central Europe, the age-standardised DALY rates were higher than expected from 1990 to 2017, whereas in other regions they were lower than expected, such as in central Latin America and south Asia. Several regions also fluctuated  between higher and lower than expected age-standardised DALY rates during the study period. Therefore, regional trends in age-standardised rates of deaths, DALYs, and incidence should not just be considered in isolation.
Instead, their observed rates should be compared with their expected rates to determine whether regions have managed colorectal cancer better or worse than expected. Additionally, dividing regions or countries into developed and undeveloped regions to evaluate the association between development and colorectal cancer burden might be an oversimplification, since the findings from this study suggest that the association is complex and non-linear in nature. Furthermore, at the national level, age-standardised DALY rates of countries on both ends of the SDI range were at higher than expected levels based on the SDI, so countries at all development levels need to enhance their prevention programmes.
Our report indicates that the colorectal cancer burden attributable to risk factors is different in males and females, and this difference should be considered in national policy makers' prevention programmes. Alcohol use, smoking, and diets low in calcium, milk, and fibre had considerable attributable colorectal cancer burden in males. By contrast, dietary risks, but not alcohol use or smoking, were found to have considerable attributable burden in females. The results of this study highlight the role of certain dietary risk factors, which are responsible for a greater burden than smoking or alcohol use globally. Specifically, diet low in calcium has been previously described as a risk factor for colorectal cancer. 33 However, the large burden attributable to this dietary risk factor has not been described before at the global level. This large burden attributable to a diet low in calcium is likely due in part to the high prevalence of this risk factor.   The results of our study underscore the importance of improving diet through public health interventions.
A previous study showed that alcohol use is responsible for nearly 10% of global deaths in the population aged 15-49 years and will lead to remarkable health loss in the absence of appropriate policy action. 34 The same study also reported that the safest level of alcohol consumption is zero, which is in contrast to current health guidelines. 34 Decreasing population-level alcohol consumption should be considered in prevention strategies to effectively minimise the corresponding health loss. 34 Smoking is another important risk factor. The global prevalence of smoking has decreased by 28% in males and 34% in females since 1990. 18 Taxation, advertising bans, and educational programmes about smoking and its toll on health are suggested as strategies to decrease smoking prevalence more substantially. 18 Diets low in calcium, milk, and fibre should also be addressed in colorectal cancer prevention strategies along with addressing low physical activity. To improve diet, increase physical activity, and reduce smoking, the American Heart Association suggests following population-based approaches, such as media and educational campaigns; labelling and consumer infor mation; taxation, subsidies, and other economic incentives; school and workplace approaches; local environmental changes; and direct restrictions and mandates. 35 Our findings indicate that although high body-mass index was not among the top three risk factors for attributable DALYs, it is an important risk factor that has a higher attributable percentage of DALYs due to colorectal cancer in males than in females. One study showed that the prevalence of obesity has doubled in more than 70 countries and has continuously increased in most other countries since 1980. 36 Effective prevention programmes are needed to decrease exposure to this important risk factor through appropriate strategies, such as restricting the advertisement of unhealthy foods, using taxation to reduce consumption of unhealthy foods, providing subsidies to increase intake of healthy foods, and using supply-chain incentives to increase the production of healthy foods. 37 Fasting plasma glucose can also be controlled mainly through physical activity and healthy diets. 38 This study had several limitations, including the fact that some of the variations in age-standardised incidence and mortality rates might be due to detection biases as well as changes in screening protocols. For example, the low age-standardised incidence and death rates in Iraq might be due to low detection rates. In addition, a major limitation of cancer burden research is the scarcity of data for many countries. Although different data sources, such as cancer registries, vital registration systems, and verbal autopsies, are used to produce cancer estimates, some countries do not have any of these sources available so their estimates are based on predictive covariates or trends from neighbouring countries. Moreover, estimates for the most recent years are usually based on past trends and covariates because there is a lag in data availability. As GBD is an iterative study, additional data sources for different locations will be added in future rounds and make the estimates more data driven, particularly in data-sparse locations.
This study found large country and regional variations in the burden of colorectal cancer in 2017. Whereas agestandardised incidence rates increased in most countries and territories over the measurement period, agestandardised death rates decreased at the global level, and in particular in high SDI countries, possibly due to fast improvement in diagnostics and interventions in these countries. Further research is required to expand our knowledge of additional factors associated with colorectal cancer incidence and to improve early detection and treatment of this disease, especially in developing countries. Clearly, colorectal cancer remains a substantial public health challenge across the globe. The results of GBD 2017 can be valuable for policy makers to implement cost-effective interventions and address modifiable risk factors and for researchers to design and carry out further research on proper modalities for prevention, early detection, and treatment of colorectal cancer.