Individuals may report the consumption of single foods, beverages, and recipes during dietary assessment. Recipes, also known as composite dishes or mixed dishes, represent the combination of ingredients from different food sources (1). Depending on the study’s objectives, detailed information from recipes should be gathered during the data collection phase of dietary assessment, which would include collecting details about ingredients and their quantities, the cooking method, the total amount prepared, the number of people served, and the amount of the portion consumed by the participant (or served and leftover). However, in practice, obtaining such detailed information is not always straightforward. Individuals may not recall the details because they were not involved in the recipe preparation or simply do not know or remember. In addition, data collection methods may not have been specifically designed to capture such information (2, 3).

Consequently, during the analysis of dietary survey data, in particular for surveys that did not collect detailed recipe information, it may be necessary to post-disaggregate recipes into ingredients. This step can be crucial for accurately estimating the composition of recipes and their precise contribution to overall food consumption and nutrient intake (4–7). Non-disaggregated recipes can lead to inaccuracies in dietary intake estimates by failing to account for the individual contributions of each ingredient, resulting in either overestimation or underestimation. For example, Sui and colleagues (8) found that disaggregating recipes in the Australian national dietary survey significantly influenced nutrient and food group estimates, resulting in lower intakes of red meat (9%), poultry (25%), and fish (18%) but higher estimates of processed meat (17%). After disaggregation, contributions of foods to total energy (25%), protein (49%), saturated fat (29%), iron (26%), and zinc (38%) were notably higher.

Recipe disaggregation involves identifying all the ingredients and their quantities (and, therefore, proportions) in a recipe. In dietary surveys, this means preparing a food consumption dataset at the ingredient level. Recipe disaggregation also facilitates data harmonization and sharing (8), especially at the international level (9). These are some of the reasons why the European Food Safety Authority (EFSA) has recommended recipe disaggregation in the EU-Menu methodology for national individual dietary studies (10).

Although seen as necessary, recipe disaggregation is not always performed in dietary surveys. There are different reasons for this, including the lack of standardized guidance, recipe source information and the cost-efficiency of applying the procedure during data analysis.

Other non-nutrient dietary indicators may be of interest and will also require recipe disaggregation at the ingredient level to allow proper use and comparability of the data, including assessments focused at the food group level, as well as food safety and environmental evaluations. For instance, the consumption of vegetables may be significantly underestimated, or the classification of foods according to food processing categories may be imprecise if recipes are not broken down into individual ingredients (7, 11, 12). Different studies using the technique of classifying foods into processing categories have highlighted the challenges and limitations imposed by not disaggregating recipes (11, 12). These often arise from the lack of detailed information collected and analyzed during dietary assessments, such as brand names, place of production, and other consumption details. Moreover, more refined analysis can be performed if data are also available at the food item/ingredient level.

International recommendations exist on how to calculate the nutrient values of recipes (13–15). Thus, focusing on the disaggregation of recipes to estimate nutrient intakes, INFOODS (the International Network of Food Data Systems) recommends using the mixed recipe method (15). In this recipe calculation method, the yield factor is applied at the recipe level and nutrient retention factors at the ingredient level, i.e., the retention factor-adjusted nutrient values of the different ingredients are summed and divided by the yield factor for considering the weight change of the recipe during cooking. This method, however, does not express the results for each ingredient, i.e., grams of consumed amounts and nutrient values for each ingredient, which may be required in some dietary analyses. In this sense, the so-called ingredient method may be a possible alternative approach (14).

Additionally, the risk of biased results during recipe disaggregation is non-negligible. To mitigate this risk, using standardized techniques and methodologies for recipe disaggregation can be applied. This includes implementing systematic procedures for ingredient identification and quantification. Given the above, this investigation aimed to assess the influence of recipe disaggregation of a recent national dietary survey in Saint Kitts and Nevis. Furthermore, it seeks to provide recommendations for future assessments and possible guidelines related to best practices in recipe disaggregation.

This comparative analysis aimed to evaluate the influence of recipe disaggregation on estimates of food group intakes. A simple and pragmatic recipe disaggregation methodology was specifically developed for this purpose, tailored for the National Individual Food Consumption Survey (NIFCS) conducted in Saint Kitts and Nevis. The analysis focused on differences in frequencies of food group consumption before and after recipe disaggregation.

A total of 440 recipes were reported in the 2020–2021 Saint Kitts and Nevis NIFCS. Of these, 65 recipes were reported with ingredient details during the interview, while 375 were reported with no indication of ingredients used. Seventy-eight non-disaggregated recipes were further identified as recipes to be further disaggregated because they were reported frequently or consumed in larger amounts (See Supplementary material for the list of recipes).

From the 78 recipes, three recipes had the ingredients identified from previously existing information, five were adapted from existing information, and 70 new recipes were identified from relevant recipe books (23, 24) and the internet (25–29). The decisions taken were confirmed by nutritionists knowledgeable about the local eating habits in the Caribbean. While there is always a level of uncertainty about such decisions and in the absence of recognized standard recipes for the country, the best possible recipe was chosen. Concerning the determination of fluid content, four recipes contained visible liquid when ready for consumption at the end of preparation. Consequently, their recipe calculation accounted for the quantification of fluids.

With the standard recipes identified, the ingredients of each recipe were itemized, and amounts were described using household measurements that were converted into grams using the food measurement information used in the survey (e.g., 1 tablespoon of milk is equivalent to 15.6 grams). In addition, the physical state of each item was identified before and after preparation (e.g., raw, cooked, canned). Each recipe ingredient was assigned a food code, which was subsequently used to merge survey files, including composition data. Furthermore, edible coefficients and yield factors were identified, and calculations were performed to determine the proportions of each ingredient.

The 78 disaggregated standard recipes generated information for 326 unique ingredients, which were repeated in some recipes, resulting in 899 rows of ingredients in the recipe database. These ingredients were from the following food groups: vegetables and their products (29.1%), spices and condiments (17.4%), cereals and their products (13.4%), milk and milk products (6.8%), meat and meat products (6.5%), fats and oils (6.3%), pulses, seeds, nuts and their products (5.6%), fruits and their products (3.8%), eggs and their products (3.3%), beverages (2.3%), fish, shellfish and their products (2.3%), sweets and sugars (1.6%), roots, tubers, plantains and their products (0.7%), and a composite dish, which referred to chicken broth (0.7%). After incorporating these ingredients into the main survey dataset, 6284 new food (ingredient) rows were generated. These disaggregated recipes contributed to 15.3 percent of the total energy intake of the population, while the remaining non-disaggregated standard recipes (n = 297) contributed to 12.6 percent.

Overall, there was a significant difference in the estimated diet when comparing disaggregated and non-disaggregated recipes. The analysis revealed that the proportions of consumers across the different food groups increased dramatically for some food groups after recipe disaggregation (Table 2), with the most remarkable difference for fats and oils (6.9% before and 44.5% after), pulses, seeds, nuts and their products (18.6% before and 49.2% after), vegetables and their products (49.9% before and 76.6% after), cereals and their products (81.3% before and 94.7% after), and meat and meat products (59.7% before and 71.4% after). Consequently, the reported consumptions were significantly different across all food groups, except for beverages, foods for particular nutritional uses, fruits and their products, savory snacks, and sweets & sugars. As expected, composite dishes were a group that decreased its proportion of consumers and amounts consumed since part of these dishes/recipes were post-disaggregated (83.8% before and 60.1% after).

Furthermore, this analysis highlighted three crucial aspects: firstly, the uncertainty surrounding the selection of standard recipes and sources in step 2; secondly, the uncertainty involved in selecting edible and yield factors in steps 7 and 8; and finally, the potential conflicts arising from the use of different approaches to disaggregate dietary information within the survey (data not shown).

This study aimed to assess the influence of recipe disaggregation in the 2020–2021 Saint Kitts and Nevis NIFCS, as well as provide recommendations for future dietary assessments. The assessment comprised nine steps, leading to the disaggregation of approximately 21% of the total non-disaggregated standard recipes that lacked ingredient details in the survey. The proportion of consumers and amounts of consumption of the following food groups were influenced by the recipe disaggregation: cereals and their products; composite dishes; eggs and their products; fats and oils; fish, shellfish and their products; meat and meat products; milk and milk products; pulses, seeds, nuts and their products; roots, tubers, plantains and their products; spice and condiments; and vegetables and their products.

The influence of recipe disaggregation on the estimates of food groups has been highlighted by previous research (6, 7). However, the steps and decisions taken for breaking down recipes in nutritional surveys are not always clearly explained. Our work contributes to addressing this gap and highlights three key aspects for further consideration. Firstly, we acknowledge the level of uncertainty associated with the selection of standard recipes and sources utilized. Secondly, we recognize the uncertainty surrounding the selection of edible and yield factors. Lastly, we explore the potential conflicts in using different approaches to disaggregate dietary information within the survey.

Selecting which standard recipes to consider in a post-disaggregation method involves an important degree of uncertainty related to the information sources used and the fact that only one version of each recipe is used for the disaggregation work. In the absence of established, nationally recognized standard recipes, the best possible recipe was chosen based on the information available. Different culinary practices are expected across households, and this may be even more complex if we consider recipes eaten away from home (30, 31). Moreover, depending on the size of the region/country being assessed and the level of variation in recipe preparation in different contexts, the variety of recipe possibilities may increase. The degree to which these factors influence the recipe disaggregation procedure and its conclusions is not well understood. Previous work from Zhang et al. (3) assessing the extent of standard recipe modifications in the Dutch National Food Consumption Survey indicated that there seems to be a minor loss in validity for food group and nutrient intake if no recipe function is available in the data collection software, and recipes are always treated as standard. We believe that further work on this topic is warranted to confirm the results with different populations, especially considering different population/country sizes, cultural habits and frequency of recipe preparation. The influence of using standard recipes in Saint Kitts and Nevis instead of recipe variations may not be large, considering the relatively small population size (less than 50,000 habitants) (16).

The selection of appropriate edible and yield factors also adds a substantial degree of uncertainty in the assessment of recipe disaggregation. Ideally, we would have preferred to use local factors, but these were not available and, therefore, pragmatic choices were made. The lack of factors is even more problematic to determine the recipe total yield factors used in the procedure for recipes with visible fluid. One could argue that the available and often used references for these factors (14, 32–34) are sufficient and enable the application of the recipe disaggregation procedure. Nonetheless, we reinforce the need for improved and culturally-specific yield and edible factors and the need to assess the influence of “universal” factors in assessments within and across countries.

The procedure for recipe disaggregation described in this manuscript focused on enabling a better assessment of food consumption reported in the 2020–2021 Saint Kitts and Nevis NIFCS, using an adapted ingredient method. Nevertheless, this is not the only procedure to disaggregate and calculate recipes and to consider ingredient information in dietary assessments, as at least three procedure types exist (i.e., recipe method, ingredient method, and mixed method) (15). A recipe calculation method had already been considered for assessing the nutrient composition during the planning phase of the NIFCS. Using the myfood24^® software, all composite dishes identified in the system were linked to food composition information, and the composition of new recipes was calculated using the mixed recipe method. While the ingredient method applies the yield factor at the ingredient level, the mixed method applies it at the recipe level. The mean of the yield factors of the ingredients may not be the same as the true yield factor of the recipe. Possible inconsistencies in using the two procedures within the same survey (mixed method in the planning phase for some recipes, and ingredient method in the data analysis phase for other recipes) should be further explored. Furthermore, the situation observed in the NIFCS survey may not be the same as for other surveys, where food composition data may be linked only after the dietary data is collected, incorporating nutrient quantification within the recipe disaggregation procedure or in parallel. We anticipate that data analysis capable of integrating both procedures and accommodating various user needs in dietary assessment would be preferable.

Some limitations of this assessment are to be acknowledged. First, not all non-disaggregated standard recipes reported in the survey were disaggregated due to resource constraints. Nonetheless, the disaggregated recipes (n = 78) represented 15.3% of the total energy intake of the recalled data, a proportion similar to the remaining non-disaggregated ones (n = 297; 12.6% of the total energy intake). Therefore, it is expected that the remaining recipes that have not been disaggregated in the survey would have a similar influence on the estimates of food consumption. The huge amount of additional work might not justify their non-disaggregation. Further evaluations to assess the effects of these decisions or other proposals are, however, needed. Second, the distribution of the food group consumption have not been adjusted for within-person variability. Because different consumption frequencies were observed for the other food groups under assessment, modeling of the variability would require different model types and this imposed technical challenges. Given that, we opted not to adjust the data for within-person variability. Lastly, the food coding used in this assessment was specific to the country. However, to facilitate data harmonization and dissemination, the FoodEx2 system—a standardized tool for classifying and describing food (35)—was applied to the St. Kitts and Nevis national dietary survey data. This data was subsequently shared on the FAO/WHO GIFT platform, where all datasets are coded using FoodEx2. This system automatically assigns individual food items into food groups. Furthermore, FAO has been supporting EFSA in expanding the FoodEx2 catalogue to make it suitable for global use, beyond its initial focus on European countries. As part of this effort, several food items not commonly consumed in Europe were added to the catalogue.

In conclusion, the study conducted as part of the 2020–2021 Saint Kitts and Nevis NIFCS demonstrates the significance of recipe disaggregation in enhancing our understanding of dietary intake. The findings revealed substantial changes in the proportion of consumers across different food groups after recipe disaggregation, notably for fats and oils, pulses, vegetables, cereals, and meat. These adjustments underscore the importance of considering disaggregated recipes in dietary assessments to obtain a more accurate representation of food consumption. In addition, we believe that the method used was a feasible and optimal recipe disaggregation method, which fits the purpose of our assessment. There are, however, some limitations and challenges to be acknowledged, which may be further considered in the future: (1) the definition of standard recipes is not straightforward. There may be many variations of the same recipe within one region and/or country, and the lack of precision in the dietary collection of recipe information imposes uncertainty about the chosen version of the recipes and their disaggregated amounts; (2) there is a lack of country-specific edible and yield factors. Specifically, there is an important gap in existing recipe yield factors, which hampers the quantification of ingredients from recipes with visible fluids; and (3) the use of different approaches to disaggregate and calculate recipes within the same survey may lead to possible inconsistencies, and this should be further explored.