The economic burden of cancers attributable to tobacco smoking, excess weight, alcohol use, and physical inactivity in Canada

Objectives The purpose of the present study was to calculate the proportion of cancers in Canada attributable to tobacco smoking ( ts ), alcohol use ( au ), excess weight ( ew ), and physical inactivity ( pia ); to explore variation in the proportions of those risk factors ( rf s) over time by sex and province; to estimate the economic burden of cancer attributable to the 4 rf s; and to calculate the potential reduction in cancers and economic burden if all provinces achieved rf prevalence rates equivalent to the best in Canada. We used a previously developed approach based on population-attributable fractions ( paf s) to estimate the cancer-related economic burden associated with the four rf s. Sex-specific relative risk and age- and sex-specific prevalence data were used in the modelling. The economic burden was adjusted for potential double counting of cases and costs. billion in direct and $8.0 billion in indirect costs. Applying the lowest rf rates to each province would result in an annual reduction of 6204 cancers (13.2% of the potentially avoidable cancers) and a reduction in economic burden of $1.2 billion. Conclusions Despite substantial reductions in the prevalence and intensity of ts , ts remains the dominant risk factor from the perspective of cancer prevention in Canada, although ew and au are becoming increasingly important rf s.


INTRODUCTION
Assessing the economics of cancers and their management has a long history in Canada. In the early 1990s, Statistics Canada developed the Population Health Model to "assist in the evaluation of cancer control interventions and policy decision-making" 1 , with a focus on lung 2,3 , breast [4][5][6] , and colorectal cancers 7 . More recently, the Canadian Partnership Against Cancer developed the Cancer Risk Management Model to "gain insight into the cost/benefit of cancer control strategies to help guide and strengthen decision-making" 8 , with a focus on lung 9 and colorectal cancers 10 . Other Canadian researchers have focused on prostate cancer 11,12 or the 21 most common cancers (brain, female breast, cervix, colorectal, corpus uteri, esophagus, gastric, head-and-neck, leukemia, liver, lung, lymphoma, melanoma, multiple myeloma, ovary, pancreas, prostate, renal, testis, thyroid, and urinary bladder) 13,14 . To our knowledge, however, no Canadian research on the economic relationship between cancers and modifiable risk factors (rfs) exists.
A considerable proportion of cancers are attributable to modifiable rfs and are therefore potentially preventable. Early work by Doll and Peto suggested that most cancers in the United States might be attributable to modifiable rfs such as tobacco smoking and diet 15 . Subsequent analyses have fine-tuned that estimate. Danaei et al. estimated that 37% of cancers in high-income countries are attributable to smoking, alcohol use, overweight and obesity, physical inactivity, low fruit and vegetable intake, urban air pollution, unsafe sex, contaminated injections in health-care settings, and indoor smoke from household use of solid fuels 16 . Parkin and colleagues found that 14 rfs are responsible for 42.7% of cancers in the United Kingdom 17 . Of those 14 rfs, 4 (smoking, excess weight, alcohol use, and physical inactivity) cause 70% of preventable cancers.
The purpose of the present study was fourfold: n to identify the proportion of cancers in Canada that are attributable to the rfs of tobacco smoking, excess weight, alcohol use, and physical inactivity; n to determine whether the proportion varies by sex or province, or over time; n to estimate the cancer-related economic burden attributable to the 4 rfs; and n to determine the potential reduction in cancer cases and economic burden if all provinces achieved prevalence rates equivalent to the best in Canada for the 4 rfs.

METHODS
Our approach was based on population-attributable fractions (pafs) and utilized our previously published model [18][19][20] to estimate the cancer-related economic burden associated with the 4 rfs.

Relative Risk
Sources and values for the relative risks (rrs) associated with tobacco smoking 21 , excess weight 22 , and physical inactivity 23

RF Exposure
The analysis of Canada's population exposure to tobacco smoking, alcohol use, overweight or obesity, and physical inactivity used data from the Canadian Community Health Survey (cchs) in 2000-2001 and 2011-2012 a,b . The territories were not included in the provincial-level analysis, but were included in the analysis of Canada as a whole. Individuals were considered overweight if their body mass index was between 25 kg/m 2 and 29.99 kg/m 2 and obese if their body mass index was 30 kg/m 2 or greater, calculated based on self-reported height and weight. For youth 12-17 years of age, the Cole system of body mass index was used to determine rates of overweight and obesity 29 . Tobacco smokers were grouped into light (<10 cigarettes daily or occasional, non-daily smoking), moderate (10-19 cigarettes daily), or heavy (≥20 cigarettes daily) categories. Physical inactivity rates were based on categorization of individuals in the cchs as "inactive" based on average daily leisure energy expenditure over the preceding 3 months. Respondents were classified as physically inactive if their daily leisure energy expenditure was less than 1.5 kcal/kg (1.5 metabolic equivalents). We made one adjustment to the base cchs data: specifically, we estimated the rates of overweight, obesity, and physical inactivity for children less than 12 years of age based on the sex-specific rates for 12-to 14-year-olds in the cchs. We assumed that children under the age of 12 did not smoke.
In The prevalence of alcohol use was calculated for all individuals 15 years of age and older, and we assumed that no individuals younger than 15 consumed alcohol. For consistency with age groups used in the literature, age categories for alcohol use were also adjusted from those used with the other rfs. The resulting age groups were 15-29, 30-44, 45-59, 60-69, 70-79, and 80 years and up.
The 2005 iteration of the cchs asked respondents to state the number of drinks (defined as 1 bottle or can of beer, 1 glass of draft, 1 glass of wine or a wine cooler, or 1 cocktail with 45 mL liquor) that they had consumed on each of the past 7 days. Using those responses, each individual's average daily consumption was calculated, based on the assumption that a standard drink contains 13.6 g ethanol 31 .
The [2000][2001]2005, and 2011-2012 iterations of the cchs all collected data on the frequency of drinking occasions in the preceding 12 months. Those data were used to group respondents into categories of drinking frequency: less than once monthly, once monthly, 2-3 times monthly, once weekly, 2-3 times weekly, 4-6 times weekly, or every day. Using only the 2005 cchs, we determined the weighted proportion of individuals reporting a particular average daily consumption, given their drinking frequency in the preceding 12 months. By applying those proportions to the weighted number of individuals reporting a given drinking frequency in 2000-2001 or 2011-2012, we were able to estimate the distribution of average daily consumption by individuals in those years. Average daily consumption was then used to classify all drinkers into the average daily consumption categories (that is, abstainer or very light, or category i, ii, or iii).
However, self-reported alcohol use tends to be underestimated. Individuals either report fewer drinks than were actually consumed or are unaware of the amount of alcohol present in their drinks [31][32][33] . Over-pouring is also a common occurrence, particularly among college-aged adults 34 . As a result, respondents tend either to underreport the number of "standard drinks" or to report their "standard drinks" using a much higher estimate of grams of ethanol than researchers assume.
To account for underestimation, it was necessary to adjust the usage values based on cchs data to more accurately reflect the number of drinks that individuals were consuming. To estimate the degree to which results were underreported, we compared the cchs results in 2005 to those of Taylor et al. 30 , which were taken from the 2003-2004 Canadian Addiction Survey and adjusted for underreporting. We assumed that the number of individuals in the abstainer or very light category would not be susceptible to underreporting (that is, underreporting because of inaccurate estimation of drink size would be negligible if only 0 or 1 drinks were consumed monthly). Those values were therefore kept the same. However, category ii and category iii largely underrepresent the true proportions; we therefore scaled up the proportion of individuals in categories ii and iii to match those of Taylor

Multiple Exposure Levels
The most basic paf calculation, derived from a single rf prevalence and disease-related rr, uses the formula where E is the proportion of the population exposed to the rf of interest (the prevalence), and rr is the relative risk of disease developing in the exposed group. Equation 1 was then used to calculate the paf of physical inactivity. However, more sophisticated approaches are required to calculate the paf when a polytomous rf is involved, as is the case for excess weight, tobacco smoking, and alcohol use. Overweight and obesity should be regarded as a trichotomous exposure to excess body weight because 3

Calculating and Adjusting Costs
We used a prevalence-based cost-of-illness approach to estimate the economic burden (direct and indirect costs) associated with the rfs. The cost estimates are expressed in 2013 Canadian dollars. Direct costs, including hospital care, physician services, other health care professionals (excluding dental services), drugs, health research, and "other" health care expenditures, were extracted from the National Health Expenditure Database 36 . Hospital care, physician care, and drug costs were allocated to each comorbidity, stratified by sex, based on 2008 data from the Economic Burden of Illness in Canada online tool 37 .
The 2008 Economic Burden of Illness in Canada tool does not allocate costs for other health care professionals (excluding dental services), health research, or "other" health care expenditures. Those expenditures were therefore estimated by allocating costs using a proportional distribution the same as that for hospital, physician, and drug costs.
All direct care costs were multiplied by the calculated rf-, sex-, and comorbidity-specific pafs to calculate the direct care costs attributable to a given rf. By completing the analysis at that level of detail, results were able to be segmented from a number of perspectives, including an assessment of direct care costs by cost category, sex, level of rf exposure, and specific diseases.

Adjusting Direct Costs in a Multifactorial System
To adjust for double counting, we used the following formula to calculate the combined paf in a multifactorial system 18 : where paf TS is the crude paf for cost of tobacco smoking, paf EW is the crude paf for cost of excess weight, paf PIA is the crude paf for cost of physical inactivity, and paf AU is the crude paf for cost of alcohol use. A disaggregation step was applied at the end of the direct costing process to assign an economic burden to each rf. In that step, the crude cost for each rf was divided by the sum of the costs for all rfs (that is, the crude total cost for the combined system), thereby generating a ratio that was then applied to the adjusted total cost.

Indirect Costs
We applied the method used in the 1998 Economic Burden of Illness in Canada (a modified human capital approach) to calculate indirect costs (premature mortality, shortand long-term disability) 38 . To make that calculation, we determined the ratio of direct to indirect costs for each diagnostic category within the 1998 Economic Burden of Illness in Canada, stratified by the specific category of indirect cost (that is, short-term disability, long-term disability, and premature mortality) 38 . To generate the equivalent indirect cost data, the pertinent ratios (by diagnostic category and specific indirect cost category) were applied to the previously identified direct costs within each diagnostic category attributable to individual rfs.

Provincial-Level Analysis
After calculating the adjusted economic burden attributable to the 4 rfs in each province, we took the sex-and age-specific prevalence rates for each rf from the province with the lowest overall prevalence rate per rf and applied those to the populations of each remaining province. Thus, the differences in annual incident cancers and in the related economic burden were calculated for each province based on actual prevalence rates and the rates from the comparator province.

Sensitivity Analysis
The point estimates for rr were used in the base model. As reflected by the 95% confidence intervals (cis), some degree of uncertainty is attached to the point estimates. To assess the effect of that uncertainty on the results, we used the lower and upper bounds of the 95% ci for the rr associated with each rf and disease in a sensitivity analysis.

RESULTS
In Canada in 2013, 27.7% of incident cancer cases (95% ci: 22.6% to 32.9%) were attributable to the rfs of tobacco smoking (15.2%; 95% ci: 13.7% to 16.9%), excess weight (5.1%; 95% ci: 3.8% to 6.4%), alcohol use (3.9%; 95% ci: 2.4% to 5.3%), and physical inactivity (3.5%; 95% ci: 2.7% to 4.3%; Table i). The proportion and the effect of each rf varied by sex, with 25.6% of cancers in women (95% ci: 21.1% to 30.2%) and 29.8% of cancers in men (95% ci: 24.0% to 35.6%) being attributable to the 4 rfs. The effects of smoking and alcohol use are higher in men than in women, and the effects of excess weight and physical inactivity are higher in women ( . Despite that decline, the proportion of cancers attributable to tobacco smoking continues to be higher than those for the other 3 rfs combined. The overall decline was not observed for all rfs. The proportion of cancers attributable to excess weight and alcohol use increased from 2000 to 2013. The changes in proportions over time are mirrored in the prevalence of the rfs. The prevalence of tobacco smoking in Canada has declined to 17.5% in 2013 from 21.6% in 2000; at the same time, the prevalence of physical inactivity declined to 43.6% from 49.0%. On the other hand, the prevalence of obesity increased to 15.4% from 12.6%, and the prevalence of hazardous or harmful alcohol use increased to 9.9% from 7.5%. The proportion of incident cancer cases attributable to the 4 rfs also varies substantially by province, from a low of 23.7% in British Columbia (95% ci: 19.0% to 28.5%) to a high of 32.3% in Quebec (95% ci: 26.7% to 37.9%; Table ii).
British Columbia had the lowest prevalence of tobacco smoking, excess weight, and physical inactivity in Canada 20 ; Prince Edward Island had the lowest prevalence of hazardous and harmful alcohol use. Because of the relatively small population sample from Prince Edward Island, we combined its age-and sex-specific prevalence rates with those for New Brunswick, the province with the second-lowest proportion of cancers attributable to alcohol (Table ii).
Applying the sex-and age-specific prevalence rates for tobacco smoking, excess weight, and physical inactivity from British Columbia and for alcohol use from Prince Edward Island and New Brunswick to the populations of all other provinces would result in a reduction of 6204 (13.2%) potentially avoidable cancers and a reduction of $1.2 billion in economic burden annually ( Table v). The proportion of cancers attributable to the 4 rfs that could potentially be

DISCUSSION
Approximately 47,000 of 170,000 new cancers diagnosed in Canada each year (27.7%) are caused by tobacco smoking, excess weight, alcohol use, and physical inactivity-a proportion that has declined from 30.1% in 2000. The greatest proportion of the decline is attributable to a reduction in the prevalence and intensity of tobacco smoking and, to a lesser degree, to improvements in physical activity. However, those improvements are offset to some degree by increases in the prevalence of excess weight and hazardous and harmful alcohol use. The estimated annual economic burden attributable to those 47,000 cancers is $9.6 billion. The proportion of cancers attributable to the 4 rfs varies by sex (25.6% for women vs. 29.8% for men). The effects of smoking and alcohol use are higher in men than in women, and the effects of excess weight and physical inactivity are higher in women. A higher proportion of Canadian men tend to be heavy smokers (6.4% vs. 2.8% of women) and heavy or harmful users of alcohol (6.0% vs. 2.3% of women). In women, the risk of breast, uterine, and ovarian cancers attributable to excess weight accounts for the difference between women and men. The risk of breast cancer also accounts for the difference between women and men with respect to physical inactivity.
The estimated proportion of cancers attributable to the 4 rfs also varies by province: from 23.7% in British Columbia to 32.2% in Newfoundland and Labrador. Variation between the provinces tends to reflect differences in the prevalence of the rfs between provinces 20 . Among the provinces, Quebec has the highest prevalence of smoking, contributing substantially to a higher observed proportion of cancers. British Columbia has the lowest rates of smoking, excess weight, and physical inactivity, leading to its lower observed proportion of cancers. However, the prevalence of hazardous or harmful alcohol use in British Columbia, at 9.9% of the population, is the second-highest in the country (range: 7.3% in Prince Edward Island to 11.9% in Quebec). Only in Quebec is the proportion of cancers attributable to alcohol use higher than it is in British Columbia (4.4% vs. 4.0% respectively; Table ii).
If age-and sex-specific prevalence rates from the provinces with the lowest prevalences of the 4 rfs were to be applied to populations living in the other provinces, the result would be an annual reduction of 6204 (13.2%) potentially avoidable cancers and a reduction of $1.2 billion in economic burden annually.
Despite substantial reductions in the prevalence and intensity of tobacco smoking in Canada, tobacco smoking remains the dominant rf from the perspective of cancer prevention 20 . Between 2000 and 2013, the prevalence of tobacco smoking in Canada declined to 17.5% from 21.6%. Just as importantly, the prevalence of heavy smoking declined to 4.5% from 7.6%. Nevertheless, of the cancers caused by the 4 rfs, almost 55% are still caused by tobacco smoking. Although excess weight and alcohol use have a lesser effect on cancers than does tobacco smoking, the proportion of the cancers caused by the 4 rfs that is attributable to the former 2 rfs is increasing: in the case of excess weight, to 18.4% in 2013 from 15.3% in 2000; and in the case of alcohol use, to 14.1% from 11.6%.
In the United Kingdom in 2010, 19.4% of cancers were attributable to tobacco smoking, 5.5% to excess weight, 4.0% to alcohol use, and 1.0% to physical inactivity 17 . The equivalent proportions in the current study are 15.2%, 5.1%, 3.9%, and 3.5% respectively. The higher proportion of cancers attributable to tobacco smoking in the United Kingdom is likely linked to a higher prevalence of tobacco smoking in that country, where 22% of men and 19% of women 16 years of age and older were current tobacco smokers 39 in 2012 (compared with 18.8% and 13.4% respectively in Canada).
The U.K. study also estimated that just 1.0% of cancers are attributable to physical inactivity, compared with Canada's 3.5%. The U.K. study calculated change in risk per metabolic equivalent below the optimal physical activity Our study has a number of limitations. First, calculating the prevalence of alcohol use by sex, age, and consumption category is particularly challenging given current data availability and issues of underreporting. Despite our best efforts to adjust for underreporting, the actual prevalence of alcohol use could vary from our estimates. Second, the method of scaling up from direct to indirect costs depends on the assumption that the ratios of costs have not changed over time. Third, in generating disease-specific rrs, the sources for the rrs associated with smoking and physical inactivity adjust for known confounding factors. However, the meta-analyses for the rrs associated with overweight and obesity did not include physical inactivity as a potentially confounding rf, which might lead to an overestimate of the economic burden attributable to excess weight. Finally, the inclusion of indirect costs in any economic analysis is controversial, given that the various available approaches generate very different results. We used a modified human capital approach because that approach places an economic value on time lost because of disability and premature mortality. Using the friction cost method, with its narrow focus on production losses, could reduce the indirect costs from $8.0 billion (95% ci: $6.4 billion to $9.4 billion) to just $204 million (95% ci: $166 million to $240 million] 20 .