Novel approaches for identifying unhealthy food and beverages for policy: a cross-sectional study of purchases of packaged products in Latin America

Summary

Background

This research aims to facilitate the identification of unhealthy food and beverages by examining approaches that progressively integrate nutrient criteria and specific markers of ultra-processing to strengthen and support the development and implementation of regulations targeting unhealthy foods within Latin America.

Methods

This cross-sectional study examined urban representative household purchase data for packaged foods and beverages from Kantar Chile, Mexico, and Peru in 2018–2019. Five approaches identifying products as unhealthy were compared: the Nova classification system to identify ultra-processed foods (UPFs); products high in saturated fat, sodium, and sugar (HFSS) based on the Pan American Health Organization (PAHO) criteria; and three approaches combining HFSS criteria with the presence of cosmetic additives (e.g., non-nutritive sweeteners, colors, flavors, other additives). Proportions of products targeted for regulation under each approach were compared within and between countries.

Findings

The proportion of unhealthy products identified across the different approaches ranged from 57 to 81% for beverages and 50–77% for foods. The most comprehensive identification of unhealthy products for policy targeting was found when products met either HFSS criteria or contained any cosmetic additive.

Interpretation

A combined approach that identifies products meeting either PAHO’s nutrient thresholds or containing UPF additive markers can substantially increase the proportion of foods and beverages identified as potentially unhealthy for policy targeting, compared with using either approach alone. Policymakers should select an identification approach that is practical to implement within their country’s existing technical capacities and policy frameworks, while prioritizing broad coverage of all potentially unhealthy products.

Funding

Bloomberg Philanthropies.

Research in context.

Evidence before this study

There is strong evidence from systematic and umbrella reviews linking the high consumption of ultra-processed foods (UPFs) with adverse health outcomes, including overweight and obesity, type 2 diabetes, cardiometabolic diseases, cancer, and all-cause mortality, as well as poorer diet quality. However, how UPFs should be identified for regulatory purposes remains unresolved.

We searched Google Scholar, PubMed/MEDLINE, and Web of Science without publication date restrictions. The initial search was conducted in September 2024 and updated in January 2026, using combinations of “ultra-processed food”, “UPF”, “Nova classification, “food additive”, “non-nutritive sweetener”, “NNS”, “HFSS”, “nutrient profile model”, “front-of pack labeling”, “food policy”, and “Latin America”. Searches included publications in English and Spanish. We also reviewed relevant regulatory and grey literature, including PAHO reports and its Nutrient Profile Model, Codex Alimentarius food additive standards, the US FDA food additive inventory, and national labeling regulations.

We found that while the processing-based Nova classification system is widely used in research, most regulatory policies rely on nutrient-based definitions (e.g., PAHO NPM). Although emerging approaches incorporate additives and other processing markers, evidence assessing how combinations of nutrient and processing criteria perform across countries, or how alternative definitions affect the proportion of household purchases that may be targeted by regulation, remains limited and heterogeneous.

Added value of this study

This study provides new evidence on how different approaches to identifying unhealthy and ultra-processed foods translate into the classification of packaged food and beverage purchases in Latin American countries. We show that nutrient-based and processing based criteria identify substantially different shares of household purchases, and that incorporating identifiable markers of processing can capture products that would otherwise be missed by nutrient-based criteria alone. By quantifying these differences across Chile, Mexico, and Peru, our findings address an important evidence gap between conceptual definitions of UPFs and their operationalization for regulatory policies. The results provide evidence to inform more consistent and context-adaptable criteria for identifying products targeted by food policies in Latin America.

Implications of all the available evidence

Implementing regulatory policies to decrease UPF consumption and prevent associated health consequences is a priority in global public health. To do so effectively, a clear, objective, and replicable definition is needed. Using data from three Latin American countries with similar food policies but different cultural and socioeconomic contexts, this paper demonstrates the usefulness of approaches that combine nutrient and processing elements for policymakers. It also offers a replicable framework to enhance regulatory consistency across Latin America.

Introduction

High consumption of ultra-processed foods and beverages (UPFs) has been associated with an increased risk of multiple adverse health outcomes, including obesity, type 2 diabetes, cardiovascular diseases, metabolic syndrome, mental health disorders, multiple cancers, and all-cause mortality.1, 2, 3, 4 The global increase in UPFs consumption is particularly concerning for public health, given their unfavorable nutrient profile, which tends to be low in micronutrients and fiber but energy-dense and high in saturated fat, sodium, and sugar (HFSS).^4,5 In Latin America, where UPFs account for up to 44% of the total energy intake among children and adolescents^6,7 and represent 16.8% of the global UPFs market share,⁸ multiple countries have adopted policies to curb their consumption, such as taxes on sugar-sweetened beverages and HFSS foods, marketing bans, and front-of-package (FOP) warning label laws. Countries in Latin America implementing FOP warning labels, such as Argentina, Brazil, Chile, Colombia, Ecuador, Mexico, Peru, Uruguay, and Venezuela, have policies utilizing pre-established thresholds for HFSS foods and beverages, following a country-specific developed nutrient criteria (i.e., Chilean or Brazilian), or the Pan American Health Organization’s Nutrient Profile Model (PAHO NPM).^9,10

While the adverse health outcomes associated with UPFs are often explained in part by their energy density and poor nutrient composition, evidence suggests that additional factors may contribute to their negative health associations. The changes in chemical and physical properties resulting from industrial ultra-processing, along with specific ingredients and by-products, may interact in complex ways, potentially resulting in a “cocktail effect” of combined or synergistic exposures that could be associated with greater risks for human health than exposure to individual harmful food components alone.^3,11,12 Moreover, the composition of UPFs, which combine high levels of sugar, saturated fat, sodium, and additives serving cosmetic purposes, is formulated to be hyperpalatable, making these products highly appealing and prone to compulsive consumption, with some evidence suggesting they may be associated with addictive-like eating behaviours.13, 14, 15, 16

One challenge in studying UPFs and creating policies to address their consumption is defining them accurately. The most common approach, the Nova classification system, categorizes foods into four mutually exclusive groups based on their level of processing and is widely used in research to identify UPFs. Nova defines UPFs as industrial formulations primarily made from chemically altered ingredients and additives that enhance taste, texture, appearance, and shelf life while containing little to no whole foods.¹⁷ While the Nova classification is commonly used in research examining the relationship between diet quality and health outcomes, it has been argued that its descriptive and qualitative nature lacks precision and may be too subjective, difficult to monitor, and ambiguous for direct use in policymaking.^18,19 The definition also does not encompass the levels of health risk-associated nutrients such as saturated fat, sugars, and sodium, which are prevalent in most ultra-processed foods, and may serve as practical markers of both UPFs and the potential mechanisms through which they possibly operate (i.e., hyperpalatability and increased energy density). These critiques raise concerns about the operationalization of UPF definitions in regulatory contexts. To address this, novel approaches are needed that combine both nutritional and processing components.

Recent research from the US has suggested that a combined approach considering HFSS nutrient criteria as well as elements of the Nova classification system for UPFs may offer a practical approach for policymakers.²⁰ Studies using nutrition and ingredient information from Chilean and Brazilian foods have also suggested a combined approach, where the presence of additive markers of ultra-processing and the excess of at least one critical nutrient could prevent UPFs from being “missed” or misclassified using traditional HFSS-based nutrient profiling approaches.21, 22, 23

The objective of this research was to examine alternative approaches to identifying unhealthy foods and beverages that integrate HFSS nutrient criteria with ingredient and additive markers of UPFs using nationally representative samples of food and beverage products purchased in three Latin American countries. The aim was to facilitate the identification of products for regulatory purposes to strengthen and support the development and implementation of effective food policy interventions and frameworks within the region.

Methods

Study design and population

This retrospective cross-sectional study used household purchase data from Kantar World Panel Chile, Mexico, and Peru. Kantar datasets contain detailed information on packaged food and beverage purchases at the barcode or Universal Product Code (UPC) level, and demographic and socio-economic information from longitudinal household panels followed in each country. To compare similar years across each country, the dataset for Chile included purchase data from July 2018 to June 2019²⁴ from 2030 households representative of urban areas with >20,000 inhabitants across six regions. In Peru, 2019 purchase data were included from 4408 households, representing 67% of the urban population in Lima and 14 other cities/regions.²⁵ For Mexico, the 2019 panel included 9885 households, representative of urban areas with >20,000 inhabitants across five regions. The selected years were before COVID-19 affected food purchases.

Detailed demographic and sample characteristics for each country panel are presented in Supplementary Table S1. Country-specific household-level annual weights, provided by Kantar, were used to ensure national representativeness. Data collection methods were similar in Chile, Mexico, and Peru, and all datasets contain item-level information, including UPC, product name, brand, description, volume, price per unit, and date of purchase. Detailed Kantar data collection methodology and descriptions of all collected variables are available elsewhere.^26,27 From the total number of food and beverage purchases recorded in Chile, Peru, and Mexico, 34%, 24%, and 30% were excluded, respectively, due to missing data on price, volume, or nutrition label, or would not be subject to regulations (i.e., baby food, culinary ingredients, alcohol, government-sponsored products, supplements, or plain water; details in Table 1).

Table 1.

Household panel sample size, number of food and beverage purchases, and inclusion/exclusion criteria in Chile, Peru and Mexico.

Number of households	Chile	Peru	Mexico
	2030	4408	9885
Purchases
Total number of purchases–food and beverages	1,241,535	3,261,206	7,553,490
Total number of purchases–food	926,397	2,172,302	4,622,072
Total number of purchases–beverages	315,138	1,088,904	2,931,418
Purchases excluded total	422,322	773,739	2,239,580
Price = 0	10,873
Bulk (no volume/reported in counts)	289,195	600,886	1,717,601
Baby food/formula	10,140	4597	42,121
Culinary ingredients	72,818
Alcohol		16,487	77,825
Government sponsored products		2339
Supplements			15
Plain water	12,492	69,256	401,979
No NFLa	26,804	80,174	39
Number of purchases included in analyses -food and beverages	819,213	2,487,467	5,313,910
Number of purchases included in analyses -food	568,237	1,509,401	2,945,291
Number of purchases included in analyses -beverages	250,976	978,066	2,368,619
Unique products
Total number of unique products–food and beverages	8686	11,878	25,953
Total number of unique products–food	6362	8637	16,430
Total number of unique products–beverages	2324	3241	9523
Unique products excluded total	1968	2483	2636
Price = 0	23
Bulk (no volume/reported in counts)	1038	1410	577
Baby food/formula	166	207	696
Culinary ingredients	447
Alcohol		66	770
Government sponsored products		13
Supplements			1
Plain water	90	335	590
No NFLa	204	452	2
Number of unique products included in analyses -food and beverages	6718	9395	23,317
Number of unique products included in analyses -food	4894	6834	15,864
Number of unique products included in analyses -beverages	1824	2561	7453

^aNo Nutrition facts label (NFL) for following reasons: no NFL listed on label at least one key nutrient missing for HFSS assignment not enough information included to match to correct NFL.

Nutrient and ingredient information

Each uniquely barcoded product reported as purchased was linked to time-specific nutrition label data and ingredient information using country-specific nutrition databases. Data handling was conducted by a team of trained nutritionists using highly standardized methods. All three countries collected photographs of large numbers of packaged food and beverages and entered the nutrition label data into a database, as described in prior publications.24, 25, 26 In all countries, nutrition label information was linked to purchase data using the product’s barcode. When a barcode match was unavailable, product characteristics (e.g., brand, product description) were used to complete the match. When in-country nutrition label data were unavailable, Kantar purchases were matched to nutrition data from Mintel’s Global New Products Database for Latin America for each country. Ingredient search terms were conducted in both English and Spanish, and trained nutritionists systematically reviewed the nutrition label assignments as described in previous studies.^27,28 Table 1 shows the final sample sizes for the total number of purchases and the number of unique products included in this analysis.

UPFs and HFSS identification approaches

We tested five approaches to identify products for regulatory purposes.

Approach 1: Nova

We used the Nova classification system¹³ to classify foods by their level of processing. Nova Group 1 includes unprocessed foods and minimally processed foods that usually contain no added substances to the original food. Nova Group 2 includes processed culinary ingredients obtained directly from Group 1 foods or nature and are used for seasoning or cooking. Nova Group 3 includes processed products that contain added substances (e.g., salt, sugar, or any product from Group 2), primarily intending to increase their durability and enhance their sensory qualities. Nova Group 4 items are UPFs that have undergone multiple industrial processes, including fractioning of whole foods, chemical modifications, and assembling using industrial techniques with the addition of flavors, colors, and other cosmetic additives to create a final hyperpalatable product.^17,20 Nova dictates that 12 Codex additive classes and the full category of flavors are considered cosmetic additives and markers of UPFs.¹³ In this study, products were assigned as Nova Group 4 (UPFs) based on assessment of ingredient lists that contained cosmetic additives (listed in Supplementary Table S2) or industrialized ingredients (i.e., ingredients of no culinary use and not used in home kitchens²⁹; listed in Supplementary Table S3). Cosmetic additives were classified according to all functions for which they are approved under the Codex Alimentarius,³⁰ and the FDA “Substances Added to Food” inventory, which includes additives and flavoring substances evaluated under FDA regulations as well as FEMA (Flavor and Extract Manufacturers Association) and JECFA (Joint Expert Committee on Food Additives)³¹ assessments, unless a specific function was explicitly indicated on the product label, in which case the additive was classified only under the stated function.

Approach 2: PAHO-based HFSS

Products were considered “HFSS” if they contained added sugar, saturated fat, or sodium ingredients and exceeded PAHO limits for saturated fats (≥10% of energy from saturated fat) and sugar (≥10% total energy from total sugar). Although the PAHO NPM is based on free sugars, we used total sugar as a proxy in combination with the presence of an added sugar ingredient, due to the absence of free sugar data. For sodium, the Mexican FOP regulation³² thresholds (≥350 mg of sodium per 100 g/ml for foods and caloric beverages, and ≥45 mg of sodium per 100 ml for non-caloric beverages) were selected, as they are broadly aligned with PAHO guidance on sodium thresholds for warning labels while reflecting country-specific adaptations relevant to the study context. These thresholds express sodium content relative to product volume (mg/100 g or mL) rather than per calorie (≥1 mg/kcal), which may more accurately identify products with high absolute sodium density. For example, when using a per-calorie metric, beverages that are low in calories and contain small amounts of sodium (e.g., a zero-calorie diet soda with 35 mg of sodium in 100 mL) would still be classified as high in sodium even if their absolute sodium content is low. On the other hand, high-calorie products (e.g., margarine) that contain high amounts of sodium may not be classified as high in sodium when using a per-calorie metric.

Approach 3: PAHO-based HFSS or contains non-nutritive sweeteners (NNS)

Approach 3 identified products that were HFSS, as defined by PAHO’s NPM, or contained NNS. In this study, NNS were comprehensively identified based on three sources: 1) additives classified under the functional class “Sweetener” in the Codex Alimentarius,³³ 2) substances listed under the Substances Added to Foods inventory,³⁴ and 3) the US Food and Drug Administration’s (FDA) list of approved high-intensity sweeteners.³⁵ Sugar alcohols, including sorbitol, xylitol, lactitol, mannitol, erythritol, and maltitol, as well as monk fruit extract, were also included in the NNS search terms, as recommended in the current literature.³⁶

Approach 4: PAHO-based HFSS or contains added colors or added flavors

Approach 4 identified products that were considered HFSS under PAHO’s NPM or contained added colors or added flavors according to Codex, FEMA (Flavor and Extract Manufacturers Association), and JECFA (Joint Expert Committee on Food Additives).³⁷ The search term list encompassed both specific chemical names (as listed in the Substances Added to Food inventory) and commonly used generic terms such as “artificial flavor,” “natural flavor,” “flavor,” and “flavoring,” including all alternate spellings (e.g., “flavors,” “flavorings”), reflecting how these typically appear on ingredient lists.

Approach 5: PAHO-based HFSS or any cosmetic additive

In this last approach, products were identified if they met the criteria for the PAHO-based HFSS or contained any cosmetic additive according to Nova (Supplementary Table S2).

Statistical analysis

For each country, analyses were conducted at the household level. The unit of analysis was the household, and the outcome was the percentage of total purchased volume classified as unhealthy foods and beverages under each of the five described approaches. For each household, this percentage was calculated by dividing the volume, in grams or milliliters, of products classified as unhealthy under a given approach by the total volume of products purchased by that household. Weighted means of the household proportions were then calculated using country-specific household weights.

Because each household contributes observations under multiple classification approaches, resulting in correlated measures, differences in the mean proportions across approaches were evaluated using linear mixed-effects models with model specification as a fixed effect and household as a random effect. Least squares means were estimated for each approach and compared using Bonferroni-adjusted pairwise comparisons.

To compare the proportion of purchased products identified as unhealthy using the Nova 4 classification and the PAHO-based HFSS criteria, we classified all foods and beverages into four mutually exclusive categories: (1) classified as both Nova 4 and HFSS; (2) classified as Nova 4 only; (3) classified as HFSS only; and (4) classified as neither. The proportions classified as Nova 4, as HFSS, and their overlap (i.e., products classified by both approaches) are presented in Venn diagrams for ease of interpretation. To determine concordance between Nova and HFSS classifications, Kappa coefficients were calculated using dichotomous NOVA 4 (yes/no) and HFSS (yes/no) classifications. All analyses were conducted using SAS v9.4. This study followed STROBE guidelines.

Ethics approval

This study used secondary analysis of de-identified data and was formally determined to not constitute human subjects research (NHSR) by the University of North Carolina at Chapel Hill Institutional Review Board. Therefore, IRB approval and informed consent for the secondary analysis were not required.

Role of the funding source

The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the study.

Results

Does PAHO-based HFSS encompass UPFs, and vice versa?

Fig. 1 presents the proportion of food and beverage total volume purchases identified for policy regulation comparing Nova 4 and PAHO-based HFSS for each country. Among beverages, the proportion of purchases identified as both Nova 4 and HFSS (i.e., overlap = met criteria for both approaches) ranged from 48% in Chile to 59% in Peru. However, although almost all beverages classified as HFSS were also classified as Nova 4, one-fifth to one-third of Nova 4 beverages did not meet HFSS criteria (36% in Chile, 36% in Peru, and 19% in Mexico Nova 4 only). For foods, the overlap between the Nova 4 and PAHO-based HFSS approaches ranged from 19% in Peru to 57% in Chile, while one-fifth to over half of Nova 4 foods did not meet HFSS criteria (20% in Chile, 54% in Peru, and 24% in Mexico Nova 4 only). These findings indicate that HFSS PAHO-based criteria does not capture a large portion of Nova 4 products.

Fig. 1.

Proportionoffood and beverage purchases identified for policy regulation comparing Nova 4 and PAHO-based HFSS in Chile, Peru, and Mexico. Each Venn diagram presents the proportion of purchases classified as: (1) Both Nova 4 & HFSS (overlap or shaded area); (2) Nova 4 only (area with diagonal hatch marks); and (3) HFSS only (not Nova 4; dotted area) by country for beverages and foods. Purchases neither Nova 4 nor HFSS are not presented in this figure. These estimates represent survey-weighted mean percentages of household purchase volume. All proportions are presented in Supplementary Tables S4 and S5.

Across all countries, expanding the HFSS definition to include NNS, added flavors or added colors, or additional cosmetic additives consistently increased the proportion of household purchases classified as UPFs (Table 2, top panel). Considering the combined results across countries, using only the Nova approach, 76% of purchased products (79% of beverages (Supplementary Table S4) and 71% of foods (Supplementary Table S5)) would be targeted for regulation; however, 4% of purchases identified as HFSS were not captured using Nova alone (Table 2, HFSS only). In comparison, using only the HFSS approach would target 54% of purchased products (57% of beverages (Supplementary Table S4) and 50% of foods (Supplementary Table S5)) for regulation, while 25% of products identified by the Nova Group 4 criteria would be “missed” (Table 2, Nova 4 only). The proportion of products targeted for policy intervention under each alternative approach (Table 2, top panel) for all three countries combined ranged from 59% for Approach 3 (HFSS or NNS) to 79% for Approach 5 (HFSS or any cosmetic additive).

Table 2.

Proportion of foods and beverages identified for policy regulation using Nova, PAHO-based HFSS, or presence of additives.

Classification	N. products purchased	Approach 1: Nova Group 4	Approach 2: HFSS	Nova and HFSS Kappa Coefficientb	Approach 3: HFSS or NNS	Approach 4: HFSS or Colors or Flavors	Approach 5: HFSS or any Additivea
All countries combined
% of products identified	8,620,590	76%	54%c	0.25	59%c,d	72%c,d	79%c,d
Both Nova 4 & HFSS approach (overlap)			51%		55%	68%	75%
Nova 4 only			25%		21%	8%	1%
HFSS approach only (not Nova 4)			4%		4%	4%	4%
Not UPF (not Nova 4 or HFSS approach)			20%		20%	20%	20%
Chile
% of products identified	819,213	81%	54%c	0.29	70%c,d	77%c,d	82%c,d
Both Nova 4 & HFSS approach (overlap)			53%		69%	76%	81%
Nova 4 only			28%		12%	5%	0%
HFSS approach only (not Nova 4)			1%		1%	1%	1%
Not UPF (not Nova 4 or HFSS approach)			18%		18%	18%	18%
Peru
% of products identified	2,487,467	84%	42%c	0.06	45%c,d	70%c,d	86%c,d
Both Nova 4 & HFSS approach (overlap)			41%		44%	69%	84%
Nova 4 only			43%		41%	15%	0%
HFSS approach only (not Nova 4)			1%		1%	1%	1%
Not UPF (not Nova 4 or HFSS approach)			14%		14%	14%	14%
Mexico
% of products identified	5,313,910	72%	58%c	0.32	60c,d%	71%c,d	77%c,d
Both Nova 4 & HFSS approach (overlap)			53%		55%	65%	71%
Nova 4 only			20%		18%	7%	1%
HFSS approach only (not Nova 4)			5%		5%	5%	5%
Not UPF (not Nova 4 or HFSS approach)			22%		22%	22%	22%

These estimates represent survey-weighted mean percentages of household purchase volume. Nova definition refers to products categorized into group 4 of the Nova classification. HFSS (High in Saturated Fat, Sodium and Sugar).

NNS = Non-nutritive sweeteners.

^aAny additive included in the 12 Codex categories (anti-foaming agent, foaming agent, bulking agent, gelling agent, thickener, carbonating agent, color, emulsifier, emulsifying salt, flavor or flavor enhancer, glazing agent, or sweetener [NNS or sugar alcohol]).

^bKappa coefficients were calculated at the product level using dichotomous NOVA 4 (yes/no) and HFSS (yes/no) classifications. Values range from −1 to 1, with values near 0 indicating agreement by chance and higher values indicating stronger agreement. Thus, low values reflect limited concordance between approaches.

^cP < 0.0001 comparing Nova % UPF to Approaches 2–5.

^dP < 0.0001 comparing HFSS to Approaches 3–5.

Approach 3 (PAHO-based HFSS or contains NNS)

Under this approach, 59% of food and beverage products purchased by households in Chile, Peru, and Mexico would be classified as unhealthy (55% overlap with the Nova approach); however, 21% of products classified as UPFs under Nova were not captured using this approach. Approach 3 resulted in an additional 4% of products being identified for policy intervention compared to the classic HFSS approach, that is, 4% of products contained NNS but were not considered HFSS (Table 2). Interestingly, this effect was more pronounced in beverages, resulting in an 8% increase in the proportion of unhealthy beverages identified compared to the classic HFSS criteria, while no increases were observed for foods (Supplementary Tables S4 and S5). A notable difference among the three countries under Approach 3 was that in Peru, a much higher proportion of UPF products were “missed” (41%) compared to Mexico (18%) and Chile (12%), consistently observed in food and beverage products (Table 2, Supplementary Tables S4 and S5).

Approach 4 (PAHO-based HFSS or contains added colors or added flavors)

Under Approach 4, 72% of food and beverage products purchased by households across Chile, Peru, and Mexico would be targeted for intervention (68% overlap with Nova). The addition of flavors and colors to the HFSS criteria resulted in only 8% of overall food and beverage product purchases considered UPFs that would be “missed” for policy intervention, less so among foods (6%) compared to beverages (11%) (Table 2, Supplementary Tables S4 and S5). Like Approach 3, Peru had different results for beverages compared to Chile and Mexico, with 27% of UPFs “missed” compared to 8% and 7%, respectively (Supplementary Table S4), indicating that many beverage purchases included products containing additives other than colors or flavors. In contrast, Mexico had the highest proportion of foods for regulation “missed” under Approach 3 (8%), compared to Chile and Peru (1% each) (Supplementary Table S5).

Approach 5 (PAHO-based HFSS or any cosmetic additive)

Under this approach, 79% of food and beverage products purchased overall by Chilean, Peruvian, and Mexican households would be targeted for policy intervention (75% overlap with Nova and 4% unique to this approach). Only 1% of UPFs would be “missed” using this approach (Table 2). While the proportion of beverage products classified as both UPF and HFSS varied by about 7 percentage points between approaches 4 and 5 in Chile (76% vs. 84%) and Mexico (68% vs. 74%), a much larger difference was observed in Peru (68% vs. 95%). This can be explained, in part, with further examination of the proportion of household beverage purchases that contained specific additives (Supplementary Table S6). Peru showed a disproportionate use of additive types, such as bulking agents, emulsifiers, gelling agents, glazers, and thickeners, in beverage products compared with Chile and Mexico. Another notable difference was that Chile had a higher proportion of food purchases classified as UPF and HFSS under Approach 5 than Peru and Mexico (77% compared to 73% and 67%, respectively) (Supplementary Table S5). These results are primarily explained by the additive content shown in Supplementary Table S6, which indicates that Chile had up to 44 percentage points more food purchases containing additives such as flavors, flavor enhancers, emulsifiers, colors, anti-foaming agents, bulking agents, gelling agents, glazing agents, and thickeners compared to Peru and Mexico. Overall, additive use was common across countries (Supplementary Table S6).

Discussion

This study shows that it is possible to identify unhealthy products within national food supplies using publicly available information, and outlines alternative approaches that could support Latin American policymakers more accurately identify unhealthy food and beverages for regulation. Analysis of a year’s nationally representative household purchase data from Chile, Peru, and Mexico shows that relying solely on either the Nova or PAHO-based HFSS classification systems leads to substantial gaps in identifying unhealthy products.

Using the Nova system alone classified 76% of the purchased products as UPFs, while the classic HFSS model identified 54% as unhealthy. However, each system missed products that the other captured. Nova failed to identify 4% of purchases that would be considered HFSS, and the HFSS approach missed 25% of purchases that Nova considered to be UPFs. In fact, across the three countries examined, the current HFSS definition missed between 20 and 43% of UPFs as defined by Nova. Beverages were particularly affected; many were classified as UPFs, but not as HFSS, meaning they would fall outside regulatory scope if only nutrient thresholds were applied. Conversely, although fewer in number, some unhealthy products were identified using the HFSS nutrient criteria but not by Nova. The low agreement observed between Nova and HFSS classifications (e.g., Kappa = 0.06 in Peru) suggests that these approaches alone capture different dimensions of unhealthy products classification, reflecting their distinct underlying criteria. This limited overlap underscores the need for Latin American countries to consider regulatory frameworks that incorporate UPF markers alongside nutrient-based criteria. Our analysis also found that adding NNS to the classic HFSS definition identified only a small proportion of additional unhealthy products, suggesting that other additive markers may be more effective in identifying unhealthy foods and beverages. Overall, the findings highlight that a combined regulatory approach integrating both nutrient thresholds and UPF-specific additive markers increases the proportion of products identified as potentially unhealthy compared with using either system alone.

To address the misalignments, we examined three alternative approaches that progressively integrated markers of ultra-processing, such as NNS, added colors and flavors, or any cosmetic additives, into classic HFSS criteria to expand policy options beyond relying solely on either Nova or HFSS. While Nova offers a comprehensive conceptual framework, applying it requires detailed ingredient information that is not always available. In contrast, HFSS thresholds are more straightforward for policymakers to operationalize. By gradually incorporating additive-based UPF markers into the HFSS criteria, we examined how much additional unhealthy product identification could be achieved using data that countries could realistically access.

One concern with ingredient-based extensions to HFSS is that focusing only on a narrow set of ingredients, such as colors, flavors, or emulsifiers (mainly for cosmetic purposes), may oversimplify the concept of “ultra-processing.” Such an approach risks ignoring broader issues like corporate power, marketing, and food system dynamics central to Nova but difficult to operationalize. Our purpose was not to redefine what unhealthy products are, but to combine conceptual and operational definitions to develop pragmatic regulatory tools that reflect the data availability in each country and prioritize feasibility of implementation.

Our findings show that combining UPF markers with HFSS nutrient thresholds substantially narrows the gap between and across HFSS and Nova classifications in Chile, Peru, and Mexico. Under the PAHO-based HFSS definition, 54% of food and beverage products would be targeted for policy intervention, increasing to 79% under Approach 5 (the most comprehensive approach). While Approach 5 provides the broadest coverage in some countries and may be best suited for beverages, Approach 4, combining HFSS and the Codex classes of colors and flavors, may be sufficient in foods, as it substantially reduces the gap between and across HFSS and Nova while relying on fewer indicators, which would be easier for implementation. Policymakers should select the method most suitable for their context. For example, in Peru, more unhealthy beverages were missed by Approach 4, which likely reflects the beverage market dynamics and reformulation efforts following policy implementation, as previous evidence suggests that front-of-package regulations in the region have prompted changes in product composition, including reductions in sugars and potential shifts and increases in additive use. In this context, beverages may contain additives beyond color or flavor, highlighting the need for a broader additive-based criterion, such as Approach 5, to adequately capture the diversity of ultra-processed drinks.

Our findings align with previous efforts to refine unhealthy foods and beverages identification using additive-based criteria. Fagundes Grilo et al.³⁸ tested five scenarios that combined food byproducts and additives to estimate the prevalence of unhealthy products in the Brazilian food supply, identifying approximately 65% of UPFs as defined by Nova. Though useful when nutrient data are unavailable, their work underscores the limitations of relying solely on ingredients or additives. In contrast, our study highlights the added value of combining nutrient thresholds with markers of ultra-processing, which ensures a more complete and accurate identification of unhealthy products.

Similarly, Canella et al. combined the PAHO NPM with the presence of cosmetic additives to analyze packaged foods available in Brazil. Their analyses showed that identifying products with at least one critical nutrient in excess and at least one cosmetic additive aligns with the Nova classification and could effectively capture most unhealthy foods and beverages.²³ In line with these results, Approach 4 in our analysis (HFSS or colors or flavors) missed only 7% of unhealthy beverages in Mexico; 8% in Chile, and 27% in Peru. The larger gap observed in Peru likely reflects beverages containing cosmetic additives beyond colors and flavors. For foods, Approach 4 was similarly effective, missing only 1% of unhealthy foods in Chile and Peru, and 8% in Mexico. These findings reinforce the potential of using colors and/or flavors, combined with HFSS criteria, to identify the majority of UPFs in Latin America. It is important to note that the overlap observed in our study is lower than that reported in previous studies (e.g., Canella et al.), which may reflect differences in data sources (purchases vs. products available at supermarkets), nutrient profiles and thresholds (adapted PAHO vs. original PAHO NPM for sodium and sugars), additive markers, as well as country contexts and time periods.

In our study, Approach 5 provided the most comprehensive strategy for the identification of unhealthy foods and beverages by combining HFSS and the presence of any cosmetic additive. Nonetheless, the intermediate approaches (3 & 4) may be more affordable, faster, and easier to implement. This makes them especially practical for countries with limited technical capacity, data infrastructure, or resources for ongoing product monitoring. While these approaches may support product identification, further validation would be needed to assess their effectiveness in improving dietary intake or health outcomes.

Adopting any of these approaches requires ongoing access to label data, standardized procedures to identify cosmetic additives, and maintaining up-to-date product and ingredient registries and nomenclature. Potential barriers include insufficient data management systems, limited technical expertise, and industry resistance. Even once a suitable approach is selected, an additional challenge is how to communicate the risks associated with additives and ultra-processing. While Chile, Peru, and Mexico currently use front-of-pack warning labels for nutrient excesses, no comparable tools indicate the presence of ultra-processing or cosmetic additives. Addressing this gap may require testing communication strategies, such as expanded FOP labels or public education campaigns, to help consumers understand why additives and processing levels matter for their health.

This study has several limitations. Although the PAHO NPM is widely used in Latin America, other nutrient profiling models (such as Chile’s national NPM) may yield different results. A recent US study, however, found comparable results using the Chilean NPM combined with similar alternative approaches examined in this study.²⁰ Furthermore, more lenient nutrient profile models would likely decrease the overlap between HFSS and UPFs and could lead to reductions in the proportion of potentially unhealthy products.39, 40, 41, 42 A second limitation is the use of total sugar rather than free sugar, as recommended by PAHO. Because this method does not distinguish between intrinsic and free sugars, our use of total sugars as a proxy for free sugars may have introduced misclassification bias and could have led to overestimation of HFSS products. This may have modestly over- or underestimated the proportions of HFSS but offers a practical, conservative proxy in the absence of free sugar data. Furthermore, for purposes of approaches comparison, this misclassification would have been consistent across the data.

Additionally, additives were classified based on all their approved functions; however, some serve multiple roles (e.g., as both preservatives and flavor enhancers). When no specific function was indicated on the product label, these additives were classified under all applicable functions, which may have introduced some degree of overestimation or misclassification and influenced our estimates. This misclassification could vary depending on each country’s regulations and whether it requires declaring the functional class or not. Furthermore, not all additives in use are included in existing regulatory lists (e.g., FDA additive lists), which may have led to an undercounting of certain unhealthy products, although the extent of this potential bias is unknown. Our datasets have some limitations, including that Kantar data only contains information on packaged products with barcodes, excluding all items sold in bulk, at street stands, or restaurants, thus data may not capture unpackaged foods or foods consumed outside the home. However, as unprocessed and minimally processed foods are generally not the focus of current food regulations in the countries studied, their omission is unlikely to substantially affect our results on the identification or comparison of unhealthy products. Additionally, we used volume-based measures to provide a proxy for consumption and the relative contribution of products to purchases. However, we acknowledge that volume may not capture the full range of products available or potential future shifts in consumption. Thus, unit-based approaches could complement these analyses in the future. We also do not attempt to explain the large cross-country differences in UPF availability, as we acknowledge that these are likely influenced by food culture and dietary traditions, such as Mexico’s substantial reliance on minimally processed staples, beyond differences in income alone. Furthermore, the data are representative of urban populations in areas with more than 20,000 inhabitants, which may limit the generalizability of the findings to rural settings. Also, because this analysis is based on data from three countries, the findings may not be directly generalizable to other contexts with different food environments, regulatory frameworks, or consumption patterns. Finally, our analysis is limited to product classification and does not assess dietary intake or health outcomes.

Despite these limitations, our research also has several strengths. To our knowledge, this is the first study to compare classic and novel UPFs and HFSS classification approaches using nationally representative food purchase nutrient and ingredient data from multiple Latin American countries. Moreover, our stepwise framework combining PAHO-based HFSS thresholds with markers of ultra-processing offers a replicable methodology that researchers and policymakers can use to strengthen product identification for regulation or reformulation.

Conclusions

Combining PAHO’s nutrient thresholds with markers of ultra-processing substantially increases the proportion of foods and beverages identified as unhealthy compared with using either approach alone. Using HFSS or Nova alone leaves major gaps; hybrid approaches provide more complete and consistent identification of unhealthy foods and beverages. Policymakers should adopt approaches tailored to their food supply while ensuring that different policies (e.g., taxes, warning labels, marketing restrictions) target the same unhealthy products. A unified regional approach could improve regulatory coherence, close loopholes, and increase public health impact across Latin America. Further research is needed to examine the implications of applying these approaches at the food group level, to validate them using dietary intake and health outcomes, and to assess their applicability beyond the countries studied.

Contributors

BMP and LSP conceptualized the study; DRM led all statistical analyses; DRM, LSP, and EKD directly accessed and verified the underlying data; LSP and EKD drafted the paper with BMP’s input at all stages; LSP and EKD reviewed and edited the draft; and BMP, LST, SWN, DRM, CZ, and FD-C all reviewed and edited the final draft and approved the final paper. All authors had full access to the data, approved the final manuscript, and had final responsibility for the decision to submit for publication.

Data sharing statement

We are unable to share data from this study as the data were used under license from Europanel (formerly Kantar World Panel). Interested parties can contact understand@europanel.com (nuria.ros@europanel.com) to inquire about accessing this proprietary data.

Declaration of interests

We declare no competing interests.