Who consumes ultra-processed food? A systematic review of sociodemographic determinants of ultra-processed food consumption from nationally representative samples · Dicken SJ, Qamar S, Batterham RL

Methods

This systematic review was pre-registered on PROSPERO (CRD42022360199), and conducted according to Preferred Reporting Items for Systematic Reviews Meta-Analysis (PRISMA) guidelines (18). The checklist is provided in the supplementary materials. The protocol was unchanged from pre-registration.

Eligibility criteria

Eligibility criteria were considered based on the population, exposure, comparator, outcome (PECO) approach: population: nationally representative sample of adults (with or without children) from any country; exposure: any sociodemographic measure; comparator: other levels or strata of a sociodemographic measure; outcome: absolute or relative individual-level consumption of UPF, defined by the NOVA classification.

Studies were included if they were: written in English; an original article published in a peer-reviewed journal; from any date; a cross-sectional or longitudinal observational study; a nationally representative sample of adults (with or without children) from any country; statistically assessing the association between sociodemographic characteristics and UPF intake; with individual-level, relative or absolute UPF intake analysis (e.g. grams/day, %kcal/day). defined by the NOVA classification (4), using detailed food consumption assessment tools.

Papers were excluded if they: were not in English; were not published in a peer-reviewed journal; were a review, ecological study, interventional trial, laboratory study or animal study; were not a nationally representative sample of adults or a subgroup of a nationally representative sample (e.g. elderly adults only or females only); did not provide statistical measures of the association between sociodemographic variables and UPF intake; did not define UPF intake using the NOVA classification (or used NOVA, but the dietary outcome was determined using principal component or latent class analysis); only measured a subgroup of UPFs (e.g. only sugar-sweetened beverages, only sweet UPFs); did not directly assess individual-level dietary intake (e.g. UPF intake based on sales, household purchases).

Search strategy and study selection

Identification of papers was achieved by searching Pubmed/MEDLINE until 07/09/2022. No filters or limits were placed on searches. Pubmed/MEDLINE was searched with the terms: “ultra-processed” OR “ultraprocessed” OR “ultra-processing” OR “ultraprocessing”, producing 1,131 results. Results were obtained and imported into Excel by SJD.

The selection process was independently conducted by two authors (SJD and SQ). To select papers, the authors developed a flowchart for full-text selection from title and abstract review, and a flowchart for full-text review to determine inclusion into the systematic review (provided in the supplementary materials). Flowcharts were based on the inclusion/exclusion criteria. Titles and abstracts were independently screened by both authors for relevance using the flowcharts. Full papers were then retrieved for eligibility analysis. All papers were manually screened, with no automation tools. After full-text screening, both authors met to discuss any disagreements. SJD then screened the references and citations of included papers, which were then agreed upon by two authors (SJD and SQ). As the selection process was predetermined prior to data extraction, all papers from the same nationally representative survey were considered for inclusion, as different papers from the same country and survey may report different sociodemographic predictors and/or cover different time periods. If papers were deemed eligible for the systematic review based on inclusion/exclusion criteria but contained no statistical assessment of UPF intake across sociodemographic variables (e.g. confidence intervals of mean intake across strata, p-values of proportions across quantiles of UPF intake), the authors were contacted to provide statistical detail.

Outcomes included absolute or relative (to total energy or food intake by weight) individual-level UPF intake (such as servings/day, kcal/day, percent (%) kcal/day or grams/day). Studies needed to report at least one unit of measurement for inclusion. The effect measures were the statistical assessments of an association between sociodemographic variables and UPF intake (e.g., for inferential statistics: beta coefficients or odds ratios and confidence intervals and/or p-values, for descriptive statistics: mean intakes and/or intakes across quantiles and confidence intervals and/or p-values, correlations and p-values).

Data extraction

Two authors (SJD and SQ) independently and manually extracted data using a pre-specified template. The data extracted by the authors included: title; authors; date of publication; country; cohort; sampling method; analytical sample size; dietary assessment method; sociodemographic assessment method; primary measure of UPF intake; average UPF intake for the sample; sociodemographic variables; unadjusted or adjusted measures of association; UPF intake across sociodemographic strata and/or statistical measures of association.

Assessment of methodological quality

Risk of bias was assessed using the Newcastle-Ottawa Scale (NOS) method adapted for cross-sectional observational studies (supplementary materials). Risk of bias was independently conducted by two authors (SJD and SQ), with disagreements resolved through discussion. No formal assessments were conducted for risk of reporting bias due to missing results.

A certainty assessment was conducted by considering the risk of bias scores, limitations in included studies, and gaps in the narrative analysis across each grouping (number of countries, range of sociodemographic predictors, range of multivariate analyses).

Data Synthesis

Papers were presented in tabular format, reporting the key characteristics of each study, including risk of bias. No data conversion or handling of missing data was conducted prior to data presentation. Results are presented as a narrative synthesis due to the varied reporting methods and statistics across studies. Results are reported in terms of the number of distinct countries, number of distinct surveys and number of unique analyses. Papers were grouped for narrative analysis according to study characteristics from the variables extracted during data collection to assess heterogeneity of associations: by country, across countries, by sociodemographic variable, by country-level income, and by multivariate analyses. Simple descriptive fractions are provided of which predictors were significantly associated with UPF intake, and which were not. No sensitivity analyses were planned. For papers from the same survey, during the same years and with the same predictors, the multivariate analyses were prioritised in the narrative review.