The burden of diet-related diseases remains a substantial population health problem worldwide(Reference Afshin, Sur and Fay1). Within Canada, where we conducted the current study, poor diet is the third largest contributor to death and disability, accounting for an estimated 48 000 deaths and over 860 000 years of healthy life lost in 2016(2).
The cost of food is a key influence on diet(Reference Lee, Mhurchu and Sacks3–Reference Gittelsohn, Trude and Kim7). A growing literature has examined how diet cost predicts dietary intakes and diet quality(Reference Lee, Mhurchu and Sacks3,Reference Rao, Afshin and Singh5) , including population-representative studies in the USA, Sweden and Spain, and cohort studies in the USA, UK and France(Reference Rehm, Monsivais and Drewnowski8–Reference Bernstein, Bloom and Rosner12). To date, this literature suggests that higher diet costs tend to be associated with better dietary quality, as indicated by fruit and vegetable intake, energy density and holistic healthy eating measures(Reference Rao, Afshin and Singh5,Reference Rehm, Monsivais and Drewnowski8–Reference Lee, Ralston and Truby13) . Diet cost literature within Canada is highly limited, with no population-representative studies to date. Two Canadian studies have reported the prices of margarine and oils in relation to their nutritional value, and another examined the association between diet cost and quality in grade 5 children living in Alberta(Reference Dachner, Ricciuto and Kirkpatrick14–Reference Bukambu, Lieffers and Ekwaru16). Some related research has examined expenditure patterns in a nationally representative sample of households, which can help to inform inferences about diet costs(Reference Kirkpatrick and Tarasuk17). Together, these studies set a foundation for the modern advancement of diet costing and food pricing methodology in Canada, with a focus on price collection(Reference Dachner, Ricciuto and Kirkpatrick14–Reference Kirkpatrick and Tarasuk17).
The majority of diet cost research involves methodology matching intake data from population-based dietary surveillance surveys to food prices from pre-collected food supply data, such as national food price monitoring databases, rather than a proximal measure of individual/household spending collected contemporaneously or linked directly to respondent intake data records(Reference Rao, Afshin and Singh5,Reference Rehm, Monsivais and Drewnowski8–Reference Lee, Ralston and Truby13,Reference Bukambu, Lieffers and Ekwaru16) . No studies have examined the measurement error introduced from using food supply prices in this way, and its potential influence on estimating diet costs. Therefore, the aim of the current study was to examine the coverage of dietary intakes reported in the 2015 Canadian Community Health Survey-Nutrition (CCHS-N) by the Canadian Consumer Price Index (CPI) to price dietary intakes and relatedly the nutritional significance of measurement error. In particular, we were interested in understanding the extent and quality of coverage of CPI food price data to provide prices for main food groups and food components that contribute to dietary intake. Our objectives were to (i) describe the coverage of the CPI food price data on the overall dietary intakes of Canadians, and by food components and food groups and (ii) describe mean nutrient intakes based on the degree of CPI coverage within individuals’ total diets.
Methods
Data sources
We matched 24-h dietary recall data from the 2015 CCHS-N to the 2015 Canadian CPI(18,19) . The 2015 CCHS-N is the most recent nationally representative dietary surveillance survey in Canada, administered by Statistics Canada(18). In 2015, 20 487 respondents aged one and older, residing in private dwellings in one of the ten Canadian provinces completed the CCHS-N(18). For respondents under 6 years old, a proxy (parent/guardian) completed the CCHS-N, for those between 6 and 11 years old, the survey was proxy-assisted, and those 12 years old and older completed the survey themselves(18). All respondents completed a first 24 h-dietary recall along with a socio-demographic and health questionnaire(18). A subset of respondents completed a second 24-h recall 3 to 10 d later(18). Diet recalls were completed using the Automated Multiple-Pass Method(18). In the 2015 CCHS-N, respondents reported intakes of over 2500 unique food items in the first 24-h dietary recall which was then combined with energy and nutrient information using the Canadian Nutrient File and collapsed into 156 Bureau of Nutritional Sciences food product categories(18). The CCHS-N excludes individuals living in the territories, on reserves and other Aboriginal settlements, full-time members of the Canadian Forces and institutionalised populations(18). The current analysis uses the reported intakes from the first recall only(18).
The Canadian CPI, administered and maintained by Statistics Canada, collects and monitors monthly prices of consumer goods in eight aggregate categories, of which one is food(19). The CPI publishes the percent change in monthly prices at various Census geographies and the average monthly retail price for select items provincially and/or nationally(19). In 2015, national prices were collected for forty-nine food items through field collection or scanner data, and the average price of each product was weighted by the population area to derive an average Canadian retail price for that product(19).
Matching the Canadian Community Health Survey-Nutrition and Canadian Consumer Price Index
We matched each of the 156 Bureau of Nutritional Sciences food product categories, excluding the water category, to the 2015 CPI average retail items ($/g) as either a match or non-match. Where further product information was needed from the CCHS-N (e.g. ‘other fruit’ category) to match to the CPI, the most consumed (g) item within that food category reported from the CCHS-N was used. This process is consistent with CPI methodology where the representative product is selected based on consumer popularity(20). We used mean gram weight for the food items consumed in the 2015 CCHS-N as a proxy for consumer popularity. Water was excluded from the current analysis as it could not be reliably estimated.
The process by which food items were matched was informed by the diet cost literature wherein foods are sometimes matched based on food group or a product type that might be reasonably substituted at retail purchase, as an exact match is not always possible(Reference Rao, Afshin and Singh5,Reference Rehm, Monsivais and Drewnowski8–Reference Lee, Ralston and Truby13) . One member of the team first attempted to find an exact match for items in the CCHS-N using the CPI. For example, ‘Beef: Ground’ (BNSD22C) in the CCHS-N was an exact match to ‘Ground Beef’ in the CPI. If an exact item match did not exist, then two members of the team worked together to find a near-exact match by selecting a comparable product considering the food group, price, and then finally the nutrient composition of the proposed match item, in that order. Some near-matches were straightforward, for example, ‘Sausage’ (BNSD30A) in the CCHS-N was matched to ‘Wieners’ in the CPI. Other matches involved more discussion by the team and reference to the literature, for example, for the three cheese products listed (BNSD 14B, 14C, 14D), we identified that the CPI item of processed cheese provided a comparable match despite the three different % BF ranges for each of the three products. A judgement was also made about the risk of bias introduced by having no match at all (e.g. using this criterion, margarine in the CCHS-N was matched to butter in the CPI). Exact and near-exact matches were coded as a match. If no comparable product was identified in that process, then it was coded as a non-match. For instance, all baked products were non-matches, as the closet match was bread in the CPI and that was not deemed a comparable product on the basis of food group, price or nutrient composition. Finally, all non-match items were re-evaluated once more by three members of the team, comparing to other diet cost literature. The final matches are described in Table 1. Altogether, despite the potential for overmatching (see limitations), this matching process is consistent with previous methods used in the diet cost literature(Reference Bernstein, Bloom and Rosner12,Reference Rehm, Monsivais and Drewnowski21) .
Table 1 Description of food groups and food items in the Canadian Community Health Survey-Nutrition (CCHS-N) 2015 by Bureau of Nutritional Sciences categories, matched to the 2015 Canadian Consumer Price Index (CPI) based on food group, price and nutrient composition
*No match available in the CPI.
The proportion of individuals’ total intakes covered by the CPI prices (matched items) was described as a percent of the gram weight of all foods and beverages consumed (excluding water), energy, macronutrients, total sugar, sodium and total dietary fibre, recognising the impact of these food components on non-communicable disease risk(Reference Afshin, Sur and Fay1). Additionally, to gain an understanding of CPI coverage in relation to food items, the proportion of foods consumed (g) within eighteen food groups with matches to CPI was calculated for each respondent(Reference Afshin, Sur and Fay1). Food groups were created based on Kirkpatrick and colleagues’ food groupings that were used to assess the top food sources of energy, sodium, sugars and saturated fats in Canadians’ diets(Reference Kirkpatrick, Raffoul and Lee22).
Analyses
Proportions and means were generated to describe CPI coverage in relation to total intake (g), energy (kcal), macronutrients (g), total sugar (g), sodium (mg), total dietary fibre (g) and food groups (g). Quartiles of CPI coverage were then assigned based on the proportion of total individual intake (g) captured by the CPI. Group mean intakes of energy and food components were compared across quartiles, applying a Bonferroni correction to obtain a Bonferroni corrected P-value (α = 0·05).
The analysis was conducted on STATA/MP 16.1 statistical software. The CCHS-N Public Use Microdata File was accessed through the Data Liberation Initiative at Dalhousie University. Accompanying Statistics Canada sampling weights were used to create weighted point estimates which represented the 34·5 million Canadians based on the Canadian Dietary Reference Intake age-sex groups and the 2011 Canadian Census population size(18). The Canadian Dietary Reference Intake age-sex groups refer to Health Canada groupings used to determine the recommended intakes of nutrients and energy for a healthy individual based on their age and sex(23). Bootstrap replication weights (B 500) provided by Statistics Canada were used to generate 95 % CI that took into account the multi-stage survey design used by the CCHS-N(18).
Results
Canadian Consumer Price Index coverage overall and by food groups and food components
For overall CPI coverage of dietary intakes, we found that sixty-seven of the 155 items (43·2 %) in the 2015 CCHS-N matched to food prices in the 2015 CPI, accounting for an average of 76·3 % (95 % CI: 75·9, 76·7) of total gram weight consumed by Canadians. A detailed description of the food groups, food items and matches to the CPI is presented in Table 1.
Table 2 shows the extent of CPI coverage by food groups and food components. We found that fats and oils, processed meats, breakfast cereals and baby food groups had the best CPI coverage among the eighteen food groups. Fats and oils were consumed by 92·9 % (95 % CI 92·2, 93·6) of Canadians on their recall day, processed meats were consumed by 37·3 % (95 % CI 35·9, 38·7), breakfast cereals were consumed by 32·2 % (95 % CI 31·0, 33·3) and baby foods were consumed by 0·6 % (95 % CI 0·4, 0·8). In contrast, baked products, alcoholic beverages and snack food groups had the poorest CPI coverage (Table 2). Baked products, alcoholic beverages and snack foods were consumed by 27·3 % (95 % CI: 26·1, 28·5), 21·9 % (95 % CI: 20·8, 23·0) and 26·4 % (95 % CI 25·3, 27·7) of the population on their recall day.
Table 2 Mean and median proportion of Canadians’ dietary intakes in Canadian Community Health Survey-Nutrition (CCHS-N) 2015 covered by the Canadian Consumer Price Index (CPI), by total grams of each food group and overall, adjusted for Canadian Dietary Reference Intake (DRI) age-sex groups(23)
Examining the variation in coverage further, Table 2 shows the computed median proportion of CPI coverage among food groups, which indicates that the majority of the data is left skewed, suggesting that there are outliers with poor CPI match that are influencing the mean estimate.
In Canadian Dietary Reference Intake age-sex group adjusted analyses, the CPI coverage for gram weight, energy and food components ranged from an average of 65·8 % (95 % CI 65·2, 66·4) for sodium intake to 82·3 % (95 % CI 81·9, 82·7) for protein intake (Table 3). In Table 3, the estimates of gram weight, energy and nutrients of study are consistently left skewed, with median intakes ranging from 68·3 % for sodium intake to 87·4 % for protein intake.
Table 3 Mean and median proportion of Canadians’ dietary intakes covered by the Canadian Consumer Price Index (CPI), by gram weight, energy and food components adjusted for DRI age-sex groups, 2015
Respondent energy and nutrient intakes by Canadian Consumer Price Index coverage
The mean intake of energy and food components for each quartile of the population, when stratified by total CPI coverage (gram weight captured by the CPI), is shown in Table 4. When comparing the proportion of each age group captured within quartile 1 and quartile 4, there were significant differences only for specific age groupings of children and adults, wherein there was a greater proportion of children, ages 6 to 12, and adults, ages 18 to 54, in quartile 1 as compared with quartile 4.
Table 4 Canadians’ dietary intakes of energy and food components by quartile of Canadian Consumer Price Index match, 2015
Comparison to quartile 4: *P < 0·05; **P < 0·01; ***P < 0·001.
† Age groups: Young children (1–5 years old), children (6–12 years old), adolescents (13–17 years old), adults (18–54 years old), older adults (55+ years old)(23).
Respondents with the poorest quartile of CPI match (quartile 1) had a significantly greater mean intake of gram weight (g), energy (kcal) and dietary fibre (g) as compared with those with the best CPI match (quartile four). Conversely, those with the poorest CPI match had significantly lower mean sodium (mg) as compared with those with the best CPI match. Those with the poorest CPI match also had a significantly lower percentage CPI capture of energy, protein, fat, carbohydrates, total sugar, sodium and total dietary fibre as compared with those with the best CPI match (Table 4).
Discussion
Summary of results
We found that the 2015 Canadian CPI covered 76·3 % of Canadians’ total dietary intake by weight (g) without water as assessed in the 2015 CCHS-N. This overall extent of CPI coverage is comparable to that of a diet cost study using Statistics Sweden data which identified that food price data covered 71 % of food intakes(Reference Ryden and Hagfors9). CPI coverage among food groups was heavily left skewed with the mean and median CPI match ranging from 0·0 % to 100·0 %. Indeed, the majority of food groups in our analysis had a median of 100 % coverage, namely, many staple foods and food groups that are a priority to economic monitoring about the food supply. By gram weight, energy and food components, however, there was differential coverage, wherein protein intake was best covered by the CPI and sodium was the poorest covered.
Finally, in our analysis, we found that the poorest quartile of CPI capture occurred among individuals with high intakes of food by gram weight, energy and fibre. Those in the lowest quartile of CPI match also have a significantly lower percentage of their total diet priced for energy, protein, fat carbohydrates, total sugar, sodium and dietary fibre as compared with those with the best CPI match.
Together, these findings indicate that a majority of Canadians’ dietary intake, including many staple food groups and commonly consumed foods, is well covered by using CPI food supply data to price dietary intakes. Yet using CPI prices to estimate broader pricing patterns across diets and populations may result in under coverage bias and measurement error. These types of error have not yet been well quantified, nor their potential presence consistently described in the literature on diet cost methodology. Thus, our analysis suggests that the CPI is a fair starting point for pricing dietary intakes in Canada, particularly given the lack of more proximal measures and population-based data on what Canadians pay for what they consume. Yet, our coverage analysis also demonstrates that further research and refinement of price data sources for population-representative analyses of the effects of food price on diet is needed. The development of robust price data sources is especially important for specific food groups where CPI does not provide prices for all items within the 2015 CCHS-N. As we will discuss further below, a number of these food groups are of key concern to the burden of diet-related noncommunicable diseases risk, and differences in CPI coverage by food groups could result in differential misclassification bias when analysing the influence of diet costs on dietary intake. Moreover, our findings are illustrative that similar problems relating to differential coverage of food supply databases may affect the results and interpretation of existing diet cost analyses in the literature. These sources of error have neither been comprehensively described nor consistently evaluated in diet cost research to date.
Canadian Consumer Price Index coverage for staple foods, v. undercoverage for sweets and snacks
On average, staple food groups that were more commonly consumed within the population had better CPI price coverage than less commonly consumed food groups. Products including grains (96·0 %), fats and oils (92·9 %), milk and dairy products (91·6 %) and meat products (72·5 %) were food groups consumed by the population, and the mean proportion of CPI coverage in these food groups ranged from 83·6 % for grains to 100·0 % for fats and oils. The median proportion of CPI coverage for all of these items was 100·0 %. In these cases, the CPI proves to be a useful measure to cost commonly consumed food group items.
Our analysis showed that several other commonly consumed food groups, however, were not at all well covered by CPI prices. We found that vegetables (92·9 %), additions (93·1 %) and sweet (80·5 %) food groups that were similarly commonly consumed had CPI coverage by food group of just 60·1 %, 25·7 % and 44·8 %, respectively. Furthermore, among all food groups, those with < 50 % CPI coverage were snack foods, baked products, sweets, additions and alcoholic beverages.
A recent analysis of 2015 CCHS-N by Kirkpatrick and colleagues identified that foods such as muffins, quick breads, biscuits, dairy dessert, chocolate and cookies were among the top twenty contributors to energy, sodium, saturated fat and sugar intake(Reference Kirkpatrick, Raffoul and Lee22). These food items fall precisely within the food groups poorly covered in price data by the CPI(Reference Tugault-Lafleur and Black24–Reference Wang, Chiavaroli and Roke27). Hence, the poor coverage of food groups of nutritional concern could result in error when pricing diets and studying the association between diet costs and dietary intake, with implications for our understanding of how food price affects diet quality and consequently diet-related diseases risk.
Could we simply use other available price data to fill the gaps?
Our analysis highlights the need to fill the gaps in dietary coverage by CPI. In the absence of purpose-built data sets to price dietary intakes in our context in Canada, a potential solution for addressing these gaps in price coverage is to use other supermarket audit or scanner data for food groups not well covered. However, these approaches may also introduce error in analysis and interpretation resulting from heterogeneity in price data collection(Reference Bernstein, Bloom and Rosner12,Reference Timmins, Hulme and Cade28,Reference Andrieu, Darmon and Drewnowski29) . Another potential source of regional food price data in Canada is the National Nutritious Food Basket, collected annually by provinces and territories(30). The National Nutritious Food Basket is a surveillance tool used to assess the cost and affordability of healthy eating in a given year, based on a market basket composed of 60–70 indicator food items selected to be consistent with Canada’s Food Guide and commonly consumed foods as measured in CCHS-N intake data(30–33). Briefly, the regional National Nutritious Food Basket presents substantial limitations for population-based diet costing work(31–33). Sub-national jurisdictions have discretion to modify the design and implementation of the food costing method including store selection, community inclusion, food price calculation and food item inclusion for regional purposes. Furthermore, neither comprehensive regional food price lists nor item lists are typically publicly shared, although many governments do make them available upon request. Like the CPI, the National Nutritious Food Basket does not collect food prices for snacks and sweet foods and thus could not be used to fill in price gaps for missing CPI products.
Additionally, in the USA in 2008, the United States Department of Agriculture (USDA) introduced a cross-sectional food price list intended for use with analyses of dietary data in 2008(Reference Carlson, Lino and Juan34). The national food price database was designed to correspond to the dietary intakes of the 2001–2002 National Health and Nutrition Examination Survey and was updated once for the 2003–2004 National Health and Nutrition Examination Survey(Reference Darmon and Drewnowski4,Reference Carlson, Lino and Juan34) . One well-designed diet cost study among USA women identified that of the 467 food products in reported intakes, just twenty-seven items were not easily matched to the USDA food price database(Reference Bernstein, Bloom and Rosner12). In this example, Bernstein and colleagues identified the process of obtaining the prices of their remaining twenty-seven unmatched items which included pricing these items from retail food outlets(Reference Bernstein, Bloom and Rosner12).
In our analysis, eighty-eight of the 155 items were not initially easily matched, although they accounted for a minority of dietary intake by gram weight. To estimate diet costs, the prices of the remaining eighty-eight items could be derived using Bernstein’s method for missing items. However, the analysis in the present paper highlights that CPI coverage varies by food group, with processed meats and fats and oil having better CPI coverage than snacks and baked products for example. Therefore, it is clear that attention should be given to identifying the quantity, type and nutritional quality of unmatched food items to assess potential biases that may be introduced when estimating diet costs and the association to dietary intake.
Another issue in using heterogeneous sources for price data in diet cost estimation is that different methodological standards have been developed for measurement of prices in different nutrition sub-fields. Collecting price information from multiple sources can increase the specificity of price data for diet cost analyses. For instance, the missing food supply data in diet cost studies have often been addressed using supermarket audit data(Reference Bernstein, Bloom and Rosner12,Reference Timmins, Hulme and Cade28) . Other studies purposefully combine price data from multiple sources including through market research and CPI measures(Reference Bernstein, Bloom and Rosner12,Reference Andrieu, Darmon and Drewnowski29) .
For example, Rehm and colleagues (2011) assessed diet costs among American adults from 2001 to 2002 and used the aforementioned USDA food price database as the main source(Reference Rehm, Monsivais and Drewnowski21). Where prices for alcoholic beverages, namely beer, wine, whisky and vodka were not available through the USDA database, they were available through the American CPI(Reference Rehm, Monsivais and Drewnowski21). In this case, the researchers took into account the systematic error due to the difference purposes of each database when supplementing the USDA database with alcohol prices from the CPI(Reference Rehm, Monsivais and Drewnowski21). In order to correct for this potential error, a subset of foods – fresh celery, ground coffee and white bread prices – were compared between the USDA database and the 2001–2002 CPI(Reference Rehm, Monsivais and Drewnowski21). The researchers found that the prices were 10·4 % lower per edible portion in the database as compared with the CPI data(Reference Rehm, Monsivais and Drewnowski21). Hence, the prices of alcoholic beverages in the CPI were deflated accordingly(Reference Rehm, Monsivais and Drewnowski21).
As a final example, in retail food audits, the least expensive, smallest package size of the audited food item is often selected(Reference McKinnon, Reedy and Morrissette35,Reference Lytle and Sokol36) . In contrast, food supply databases such as the CPI typically select an average price, average package size(19). As such, retail audit prices at the population level used to price intakes may skew high in comparison with food supply prices, since many consumers will have actually paid a great deal less for the same volume/weight of food than the measured retail smallest package price. Certain subpopulations of consumers may be more sensitive to volume discounts, as well as volume discounts for certain food groups(Reference Gittelsohn, Trude and Kim37, Reference Bennett, Zorbas and Huse38). Volume discounts are also more prevalent in certain types of food environments such as hypermarket stores(Reference Bennett, Zorbas and Huse38–Reference Riesenberg, Backholer and Zorbas41).
Recommendations for food price data in diet cost analyses moving forward
The current study thus highlights four considerations to aid in future diet cost methodology and its interpretation. First, diet cost research should consistently quantify and report the extent and quality of match between food supply prices and intake data used. In the literature to date, the range of numbers of prices typically used for diet cost analysis are wide and the proportion of match is unclear. Identifying and reporting the details about how dietary intakes are priced allows for the assessment of potential biases that may be introduced.
Second, researchers should examine alternatives to using a single source of food supply data, beyond addressing missing values. Comparing different methods to derive diet costs can improve the estimation of diet costs and the examination of the association with dietary intakes.
Third, food supply data sets can present a risk of misclassification bias in matching intake data to estimate diet costs. For example, in the CPI, we found greater differential product coverage in milk and egg and beef products, in comparison with non-alcoholic beverages, fruits and vegetable products (Table 1). This differential coverage could influence the costing of specific food items/food groups and contribute to differential misclassification bias in attributing dietary costs to subpopulations. Publishing consistent methods to assign product matches from food price databases to dietary intakes could help to identify such potential biases in diet costing.
Fourth, researchers can consider the potential for systematic error, where prices for certain food categories are omitted altogether. For instance, the CPI did not provide any price information for alcoholic beverages despite alcoholic beverages being among the top contributors to energy intake within beverages in the 2015 CCHS-N(Reference Jones, Kirkpatrick and Hammond26). In the 2015 CCHS-N, alcoholic beverages were consumed in the past 24 h by 21·8 % of respondents which accounted for an average of 3·2 % of total 24-h energy intake and 4·5 % of total intake based on total grams consumed. Diet costs analyses in the UK, USA and Japan identified that diet costs were positively associated with alcohol consumption among nationally representative samples of adults(Reference Rehm, Monsivais and Drewnowski8,Reference Timmins, Hulme and Cade28,Reference Okubo, Murakami and Sasaki42) . Bernstein and colleagues (2010) in the USA have been one of the few diet cost studies to describe the inclusion of alcoholic beverages(Reference Bernstein, Bloom and Rosner12); in their case, they supplemented their main food supply price data set with retail audit data for remaining food items, including alcoholic beverages(Reference Bernstein, Bloom and Rosner12). Hence, some consideration should be given to the inclusion/exclusion of certain categories, such as alcoholic beverages, given the potential contribution to dietary intake and diet costs.
Strengths and limitations
This paper presents an assessment of measurement error to inform diet cost analysis based on Canada’s only nationally representative dietary intake surveillance data source. There are three main limitations. First, food items in the CPI are most consistent with the level of detail presented in the CCHS-N BNS groups which are already collapsed into 156 food categories, from over 2500 unique food items reported as consumed. The use of the CPI limited our ability to price all 2500 food items. Additionally, some items in the CCHS-N that were matched with CPI items were not an exact match, but instead a comparable match despite differences in food group, price or nutrient profiles (e.g. matching margarine from the CCHS-N to butter in the CPI). Hence, the CPI coverage estimates may be an overestimation of the food price information that is currently available in Canada to match to the CCHS-N(Reference Nardocci, Leclerc and Louzada43). This may also influence the association between diet costs and dietary intake or inferences about diet costs and health outcomes because some food items that were nutritionally different in the CCHS-N were matched to the same CPI item. Similarly, we further aggregated the 156 food categories into eighteen food groups which may hinder our understanding of CPI coverage of specific food items within the CCHS-N. Second, most diet cost methodology usually makes the assumption that intakes reported in dietary surveillance are purchased through retail food outlets and that sources of food supply price data are representative of what is available and accessible to consumers at retail. For example, within Canada, approximately 70 cents of every household food dollar is spent in a retail food store(44). Food prices within the CPI are collected from retail food outlets, while dietary intake data in the CCHS-N recorded food items that were both purchased from retail food outlets and restaurant/quick service settings (a variable for location of purchase does not exist in this data set). We may have artificially overestimated the CPI coverage and underestimated food prices in instances where a substantial proportion of individuals’ intake comes from non-retailed foods. Third, the dietary intake data that were matched to the CPI were taken from a single 24-h dietary recall per respondent. A single 24-h recall provides rich detail about dietary intake on the reported day and is sufficient to estimate group means(Reference Baranowski and Willett45). However, it should be noted that the current analysis does not account for episodically consumed foods. Addressing this was not necessary in the current analysis, but statistical methods have been developed to estimate usual intakes including episodically consumed items where it is needed to assess compliance with guidelines or correlate intakes with health outcomes(Reference Tooze, Midthune and Dodd46–Reference Dodd, Guenther and Freedman48).
Conclusion
Given the proliferation of diet costs studies using food supply prices over the last 20 years, the current study highlights potential sources of error and bias in using food supply data to price dietary intakes(Reference Rao, Afshin and Singh5). Researchers and decision makers should pursue opportunities for tailored data sets to price dietary intakes when using food supply data prices alone or in combination with other sources (e.g. retail audit data).
Acknowledgements
Acknowledgements: Our thanks to Sai Choi Chua for providing statistical computing and data support throughout this project. Financial support: C.M. received funding for this research from Canadian Institutes of Health Research (CIHR) (PJT #162373) and is supported by the Canada Research Chairs program; GL is supported by a CIHR Canada Graduate Scholarships Doctoral Award (FRN #170778). Conflict of interest: There are no conflicts of interest. Authorship: G.L. led the data analysis, interpretation and writing of the manuscript and co-led the formulation of the research question and study design; C.M. co-led the formulation of the research question and study design and contributed to the data analysis, interpretation and editing of the manuscript, and holds oversight for funding of this study. V.T. and Y.Y. contributed to the data interpretation and editing of the manuscript. Ethics of human subject participation: This research did not require institutional ethics review board approval.
Open access
