Cancer incidence attributable to insufficient fruit and vegetable consumption in Alberta in 2012

Source of relative risk data

Using the method described by Parkin and Boyd,⁴ the excess relative risk (RR) for a decrease in 1 g of vegetable or fruit consumed was estimated by assuming a log-linear relationship in equation 1:

Equation 1: Risk per gram per day = ln(1/RRx)/x

where x is the exposure level (in grams per day) and RRx the RR for x grams per day. We selected RRs from published reports by international collaborative panels where available and from the peer-reviewed literature (Appendix 1, Supplementary Figure 1, available at www.cmajopen.ca/content/4/4/E760/suppl/DC1).

As has been previously described,⁶ the theoretical latency period was conceptualized as the time between the start of exposure and cancer diagnosis, and the measured latency period was conceptualized as the time between exposure measurement and cancer diagnosis. For the analyses related to fruit and vegetable consumption described here, we conceptualize this latency period as the time between a measurement of the prevalence of insufficient fruit and vegetable consumption and cancer diagnoses that could theoretically be attributable to low fruit and vegetable intake. We endeavoured to quantify the measured latency period for each individual cancer site on the basis of follow-up periods from high quality cohort studies^1,7-22 and refer to this simply as the latency period for simplicity. As previously described,⁶ the midpoint of the latency period suggested by cohort studies was chosen.

Assessment of fruit and vegetable intake

Fruit and vegetable consumption in Alberta was estimated from data from the Alberta sampling portion of the Canadian Community Health Survey (CCHS), a population survey conducted by Statistics Canada. In Alberta, the CCHS is considered representative of the general Canadian population excluding people living on reserves or other Aboriginal settlements, full-time members of the Canadian Forces and people in institutions. Total vegetable consumption was estimated by combining the number of times respondents reported consuming green salad, carrots, and other vegetables not including salad, carrots or potatoes. Total fruit consumption was estimated as the number of times respondents reported consuming fruit and fruit juice. For juice specifically, any reported frequency of consumption of fruit juice was equivalent to one serving of fruit, as suggested in previous work.^1,23 Specific questions from the CCHS used to assess fruit and vegetable consumption are shown in Appendix 1, Supplementary Table 1. For both fruit and vegetable consumption, each reported "time" of consumption in the CCHS data was estimated as 1 serving.

The deficit in fruit and vegetable consumption relative to the 5 servings (400 g) per day recommended by the World Cancer Research Fund was calculated using a similar method to that used by Parkin and Boyd.⁴ First, the mean number of servings of fruits and vegetables consumed by men and women in each of 4 age groups (20-34, 35-49, 50-64, 65+) was estimated for each year of survey data used in this analysis. The ratio between fruit and vegetable consumption in these groups was then used to partition the 5 servings suggested as necessary for cancer prevention by the World Cancer Research Fund guideline to preserve the vegetable:fruit ratios seen in the prevalence data. The ratio between the World Cancer Research Fund guideline of 5 servings of fruits and vegetables and the total number of servings of fruits and vegetables consumed within each age group and sex (calculated by summing the mean totals for fruits and vegetables) was calculated. The mean consumption for fruits and vegetables was then multiplied by this ratio to calculate the level of consumption of each that would be consistent with the guideline level. These values were converted into grams by multiplying the number of servings by 80 g per serving, consistent with the guideline, where 5 servings is equivalent to 400 g.

Combined fruit and vegetable consumption was then classified into 7 categories based on the total number of servings reported to be consumed per day (0, < 1, < 2, < 3, < 4, < 5, ≥ 5). The proportion of the population in each category within each age and sex group was estimated, and the deficit in grams from the level of consumption for each of fruits and vegetables was estimated by subtracting the mean level of consumption in grams within each consumption category from the guideline amount.

Statistical analyses

Population attributable risks associated with fruit and vegetable consumption were estimated individually for each of the following cancers using the method described by Parkin and Boyd⁴: oral cavity and pharynx, esophagus, stomach, larynx and lung (fruit only). Relative risks were calculated for fruit and vegetable consumption separately using equation 2:

Equation 2: RR = exp(Rg × Gx)

where Rg represents the RR for a deficit of 1 g per day of fruit or vegetable consumption and Gx represents the deficit in consumption for consumption category x. The benefits of fruit and vegetable consumption were considered multiplicative so that:

Equation 3: RR(f and v) = RR(f) × RR(v)

These combined RR estimates were then used to estimate population attributable risks according to the formula:

Equation 4: PAR = [(p1 × ERR1) + (p2 × ERR2) + (p3 × ERR3) + (p4 × ERR4) + (p5 × ERR5) + (p6 × ERR6) + (p7 × ERR7)] / {1 + [(p1 × ERR1) + (p2 × ERR2) + (p3 × ERR3) + (p4 × ERR4) + (p5 × ERR5) + (p6 × ERR6) + (p7 × ERR7)]}

where px is the proportion of the population within consumption category x and ERRx is the excess RR, calculated as 1 - RR(f and v), within consumption category x. To estimate the total number of cancers at each site overall and by age-group and sex that were attributable to insufficient fruit and vegetable consumption, population attributable risks were applied to cancer incidence data obtained from the Alberta Cancer Registry for 2012. The total proportion of cancer attributable to insufficient fruit and vegetable intake at each individual cancer site was estimated as the total number of excess attributable cases for that cancer across all age groups divided by the total number of observed cancers at that site for both men and women, individually and combined.

Monte Carlo methods were used to estimate 95% confidence intervals (CIs) around point estimates of population attributable risk.²⁴ Prevalence and risk estimates and their associated confidence intervals were used to parameterize a probability distribution from which 10 000 random samples were drawn. A binomial probability distribution was assumed for the prevalence exposure, a lognormal distribution for risk, and a Poisson distribution for cancer incidence. The 95% CIs were then determined by the 2.5th and 97.5th percentiles of the distribution of simulated population attributable risk and estimates of excess attributable cases. Similar techniques were used by 2 previous studies that estimated population attributable risk.^24,25 All analyses were conducted using RStudio (version 0.98.1080, R Studio, Inc.).