Discussion
This systematic review included 55 nationally representative studies, spanning 32 countries and three decades of dietary intake. Average UPF intake varied greatly across countries, from 14-16% of total energy intake in Italy, Romania and Colombia, to 61% in the Netherlands. Intakes also varied greatly within countries, with several sociodemographic factors being independently associated with UPF intake.
Age (being highest in either younger adults, or adolescents and children) demonstrated a consistent inverse association with UPF intake, with a large magnitude difference in UPF intake with age. Other sociodemographic characteristics associated with large magnitude differences in UPF intake across strata included race/ethnicity, income, country of birth, region of the country, rural/urban living and food insecurity. Despite only a few studies reporting on race/ethnicity, ethnic differences in UPF intake were large, with 10-20% absolute differences as a proportion of total energy. Similar magnitude differences were seen in the small number of countries reporting UPF intake based on country of birth or food insecurity, except no difference in UPF intake seen in Portugal across levels of food insecurity.
Living in an urban residence and being unmarried, single, separated or divorced were typically associated with a higher UPF intake, whereas education level, income and socioeconomic status showed varying directions of association with UPF intake, depending on country. Gender was generally not significantly associated with UPF intake in most countries, and neither was the number of individuals in the household. At least one multivariate adjusted association was reported in 13 countries, showing largely unchanged estimates from the crude associations. These findings indicate that the significant and large differences in UPF intake across levels of sociodemographic variables are independent of other sociodemographic variables.
The results from this systematic review confirm and contrast with the findings from previous reviews. Similar to the results in this systematic review, a systematic review that included non-nationally representative samples suggested minimal differences in UPF intake with gender, but with higher intakes in younger ages (15). One review suggested links between a younger age, urban residence, male gender, lower education level, lower household income and food insecurity with higher UPF intake (13). Another review suggested links between age, gender, education and income with UPF intake, and also a varying association of socioeconomic status depending on country-level income (14). This review also identified urban living as an important predictor of greater UPF intake, which is also in line with the findings from global UPF sales and household purchases (3,5). But, male gender was not consistently associated with higher UPF intakes, and the association between education level and UPF intake varied across countries. Indeed, in UMICs, higher education levels, incomes and socioeconomic status tended to have higher UPF intakes, but the association between education level and UPF intake across HICs varied. In the US, differences in UPF intake across levels of education and income also showed contrasting associations across ethnic groups, and in Portugal, the association between education level and UPF intake differed based on age. These findings indicate the need for more detailed assessments in other countries to tease apart the sociodemographic inter-relationships with UPF intake. Race/ethnicity, region of the country and country of birth were all significant predictors of UPF intake in this systematic review, but had largely been unconsidered in previous reviews. Given the lack of studies assessing the association between food insecurity and UPF intake, and the large differences in food insecurity across ethnic groups (83), these highlight important and understudied sociodemographic associations for the consumption of UPF intake.
Meta-analyses demonstrate significantly increased risks of poor health with each 10% increment in UPF intake as a proportion of total energy in the diet, including a 15% increased risk of all-cause mortality (84), 7% higher risk of overweight, 6% higher risk of obesity, 5% higher risk of abdominal obesity (85) and 15% higher risk of type 2 diabetes (86). Meta analyses comparing the highest vs. lowest quantiles of UPF intake from prospective cohort studies also show a 25% higher risk of all-cause mortality, 29% higher risk of cardiovascular disease incidence and mortality, 11% higher risk of overweight or obesity, 20% higher risk of depression (9) and 23% higher risk of hypertension (87). The magnitude of the absolute differences in UPF within several sociodemographic variables reported in this systematic review were typically in the range of 5-20% of total energy intake (e.g. across age, race/ethnicity, income, country of birth and region of the country) and independent of other sociodemographic variables. These independent differences in UPF intake correspond with the differences in UPF exposure that are associated with significantly increased risks of non-communicable disease reported in meta-analyses. The findings from this systematic review highlight existing health inequalities across sociodemographic subgroups associated with increased UPF intake, and future sociodemographic differences in non-communicable disease incidence. Furthermore, by considering each of the independent sociodemographic associations produces striking differences in mean UPF intake across specific sociodemographic populations. For example in Mexico, the average male adult, living in a rural residence in the south of the country, with a low socioeconomic status and from a household with no head of household education consumes 39.3% less UPF as a proportion of total energy than the average female pre-school child, living in an urban residence in North Mexico with a medium-high socioeconomic status and from a household with a head of the household with a college graduate education (25.2% vs. 64.5%) (54). Given the growing epidemics of obesity and cardiometabolic disease, these results present an alarming picture regarding the incidence of future adiposity-related cardiometabolic disease.
Notably, no studies included in this review were from Africa or Asia, accounting for over half of the global population. Non-nationally representative samples from Asian countries also demonstrate sociodemographic associations with UPF intake, supplementing the global picture of national UPF intake patterns. With almost a fifth of the global population, results from the China Health and Nutrition Survey across 1997-2009 (CHNS; 12,451 adults sampled using a random, complex multistage methodology across nine of 31 Chinese provinces, but not considered to be nationally representative (88)) showed that the highest UPF consumers (≥50g UPF/day) vs. non-consumers vs. (0g/day) were more likely to be male, have a higher income, higher education level and live in an urban residence (22), similar to the UMICs in this review. Age was not significantly associated with UPF intake. However, another CNHS study reported similar findings, except age was positively associated with UPF intake, in contrast to the findings here, (89). Other Asian samples report differences in UPF intake with a younger age, higher education level and race/ethnicity, with variable gender associations. In Jakarta, Indonesia, a younger age was associated with increased UPF intake (90), and in a multicentric study from Iran, being under 40 (vs. 40 or older) and having a higher level of education were associated with higher intakes of UPF, but there was no difference in UPF intake with gender (91). In middle-aged Japanese adults, people who were never married, living alone, in regular full-time work and with a lower income had higher UPF intakes, but no difference was seen with age, gender or number of children (92). In the Singaporean multi-ethnic cohort study, the highest vs. lowest quartiles of UPF intake (85.9% vs 51.8% of total energy) were more likely to be younger, male, be of Malay or other ethnicity, and less likely to be of Indian or Chinese ethnicity (23). In the Taiwan, a higher UPF intake was seen with a younger age and a higher education level, but gender was not significant (24). Improving the understanding the sociodemographic patterns of individual-level UPF intake in African and Asian nations represents a key research focus for understanding the global implications of UPF consumption.
Strengths and limitations
This review builds upon previous reviews by systematically reporting sociodemographic predictors of individual-level UPF intake from nationally representative samples, reducing biases resulting from convenience sampling, increasing the generalisability of results to each nation. The use of food consumption surveys strengthens the confidence in assessing individual-level UPF intake, compared with household purchase data. This review reported all sociodemographic predictors from included studies, including those which were not associated with UPF intake, and provided absolute and relative quantitative estimates of differences in UPF intake across sociodemographic strata. Multivariate analyses were conducted in several countries in this review, allowing for the identification of independent sociodemographic associations, separate to other sociodemographic characteristics and health behaviours.
However, some limitations must be considered. First, the lack of complete reporting of all sociodemographic predictors precludes determining the most important sociodemographic determinants of UPF intake. Within each country, most studies considered age and gender, but other sociodemographic predictors were not consistently reported across all studies, and multivariate adjustments in assessing associations were not performed in all countries. In addition, different metrics and stratifications were used across papers to report the same sociodemographic factor (e.g. different methods of assessing socioeconomic status), limiting comparability. Other sociodemographic influences were also not considered, such as work shift pattern. Second, the use of food consumption surveys was a criterion for inclusion, but are time intensive and costly. Combined with the requirement for nationally representative sampling methodology, the inclusion criteria biased towards including higher income countries with the resources to conduct such studies. Meaning, many countries were not included in this review, particularly those from middle- and low-income countries. Third, UPF intake was self-reported, as was the assessment of sociodemographic variables, which may have introduced recall bias or bias from misreporting. The dietary assessment method used to assess UPF intake is an important factor. Accurate classification requires extensive details on a food item, which is typically more than what is provided in an FFQ. FFQs also do not allow for the ability to categorise similar foods into different NOVA groups (for example, a single item in the FFQ for all breads or lasagne). Multiple-day, 24-hour recalls or food diaries have greater potential to identify UPFs and estimate habitual UPF intake from detailed, open-ended, questions, connected to a database of several thousand food items. All but one study in this analysis used 24-hour recalls or food diaries with an extensive food database coded into NOVA to estimate UPF intake, and most assessed intakes over multiple days. Only one cohort assessed UPF intake with a standalone FFQ. This reduced the risk of misclassification that may occur with less detailed FFQs. However, misclassification is still possible, and the classification of specific foods into NOVA processing groups may vary across studies from coding error. Some studies used a single 24-hour recall, which, whilst providing more granular detail than an FFQ, may not accurately reflect habitual intake. Fourth, despite the NOVA classification being used extensively in research and implemented in national dietary guidelines or health organisation recommendations, there is still disagreement regarding the utility and validity of measuring UPF intake using the NOVA classification. Fifth, bias may have been introduced through excluding participants with missing data for complete case analyses in the adjusted models, or in reports of unadjusted cohort characteristics, depending on the specific outcome of interest in the study. However, agreement amongst samples was high, and the consistency within each country indicates that the impact of exclusion bias across samples was minimal. Most studies did not compare the characteristics of excluded and included samples, resulting in lower risk of bias scores, making it unclear as to the extent of this bias. Adjustment in multivariate models for confounding factors can allow for the determination of independent associations, but not all studies considered all potential sociodemographic factors in multivariate models. Furthermore, some studies also adjusted for non-sociodemographic factors, such as health behaviours, diet quality or BMI, which may bias the estimates for sociodemographic variables. Fifth, the methodological rigour required to meet the inclusion criteria of this systematic review increased the certainty of the evidence examining the relationship between sociodemographic factors and individual-level UPF intake during specific times or across time in nationally representative samples. However, it is not possible to ascertain the extent to which differences within surveys that span different time periods are due to bias, changes in measurement methodology, different analytical approaches, or from genuine changes in UPF intake. Finally, the narrative synthesis limits the ability to determine overall associations within countries and across sociodemographic variables. It was not possible to synthesise papers quantitatively, due to differences in the reporting of sociodemographic associations (e.g. reporting mean UPF intakes or associations across quantiles of UPF intake), and whether papers analysed a single cross-sectional sample, repeated cross-sectional samples, or trends over time.
Implications and further research
The systematic assessment of sociodemographic determinants of UPF intake in this review indicates that UPF intake is unevenly distributed within nations, varying based on previously identified factors such as age and income, but also with previously unconsidered factors such as race/ethnicity or by region of each country. Furthermore, sociodemographic predictors such as gender appear to be less important than previously suggested, and others, such as education or income, are more nuanced than previously suggested. The results here have implications for health policy and research, indicating that certain groups may obtain greater benefit from policy action and targeted/directed interventions. Importantly, the associations between UPF intake and sociodemographic factors are likely to be a reflection of social injustice, and the adverse associations linked with UPF a result of such inequalities. These findings indicate a need to consider social, cultural and geographical influences on UPF intake, and how barriers to reducing UPF intake and accessing MPF may vary across sociodemographic populations within each country. Whether a given food may be considered as UPF can vary across countries, depending on the typical processing method. For example, breads are typically considered as PF in Australia, but as UPF in the UK. Such cultural differences need to be taken into account when considering public health interventions regarding UPF intake. Little is still known about the actual consumption of UPF within many populations (93). There is a need for more nationally representative samples assessing individual-level dietary intake that perform multivariate adjustments of a wider range of sociodemographic predictors of UPF intake, particularly in middle- and low-income countries, and in Africa and Asia.
CC BY 4.0. "Who consumes ultra-processed food? A systematic review of sociodemographic determinants of ultra-processed food consumption from nationally representative samples" by Dicken SJ, Qamar S, Batterham RL, Nutrition research reviews (2024). https://doi.org/10.1017/S0954422423000240. Licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Peer-reviewed and open access; full text from PubMed Central (PMC7617046). Headings and layout adapted for reading; figures and tables are in the original.