Nutrient clustering, NOVA classification, and nutrient profiling: How do they overlap, and what do they predict about food palatability?

We compared the performance of three food categorisation metrics in predicting palatability (taste pleasantness) using a dataset of 52 foods, each rated virtually (online) by 72 – 224 participants familiar with the foods in question, as described in Appetite 193 (2024) 107124. The metrics were nutrient clustering, NOVA, and nutrient profiling. The first two of these metrics were developed to identify, respectively: ‘hyper-palatable ’ foods (HPFs); and ultra-processed foods (UPFs), which are claimed to be ‘made to be hyper-palatable ’ . The third metric cat-egorises foods as high fat, sugar, salt (HFSS) foods versus non-HFSS foods. There were overlaps, but also significant differences, in categorisation of the foods by the three metrics: of the 52 foods, 35 (67%) were categorised as HPF, and/or UPF, and/or HFSS, and 17 (33%) were categorised as none of these. There was no significant difference in measured palatability between HPFs and non-HPFs, nor between UPFs and non-UPFs ( p ≥ 0.412). HFSS foods were significantly more palatable than non-HFSS foods ( p = 0.049). None of the metrics significantly predicted food reward (desire to eat). These results do not support the use of hypothetical combinations of food ingredients as proxies for palatability, as done explicitly by the nutrient clustering and NOVA metrics. To discover what aspects of food composition predict palatability requires measuring the palatability of a wide range of foods that differ in composition, as we do here.


Introduction
Previously, we reported the lack of systematic differences in food palatability as a function of level of processing as defined by the NOVA system (NOVA is not a synonym) (Rogers et al., 2024).This is surprising, given that it is claimed that ultra-processed foods (UPFs) are 'made to be hyper-palatable' (Monteiro et al., 2017).Relatedly, a nutrient clustering metric, based on the fat, carbohydrate, sugars, and salt (sodium) content of foods, has been developed to define hyper-palatable foods (HPFs) (Fazzino et al., 2019).However, palatability, which as we describe below, refers to taste pleasantness or liking, was not measured in this study, so the metric lacks direct validation.Accordingly, in the present research we investigated whether the HPFs metric does indeed predict palatability.This involved carrying out further analyses of data from our recently reported study (Rogers et al., 2024).The data comprise palatability ratings and nutrient composition data for 52 foods, ranging from cucumber to milk chocolate.When we designed that study in 2020, we were not aware of the research reported by Fazzino et al. (2019).
A further aim of the present research was to investigate similarities and differences (i.e., the overlap) in the categorisation of foods by the nutrient clustering (HPFs) and NOVA (UPFs) metrics.We also compared their categorisation of foods and prediction of palatability with a third metric, namely the nutrient profiling metric, which identifies high fat, sugar, salt (HFSS) foods (Department of Health, 2011).
The HPFs metric defines a food as hyper-palatable if it falls within one or more of three nutrient clusters, as follows: (1) > 25% energy from fat and ≥0.30% sodium by weight, (2) > 20% energy from fat and >20% energy from sugars, and (3) > 40% energy from carbohydrates (after subtraction of fibre and sugars) and ≥0.20% sodium by weight.As explained by Fazzino et al. (2019), these criteria were derived from information contained in 14 peer-reviewed articles which described or listed foods that the authors categorised as, for example, 'tasty', 'palatable', 'hyper-palatable', or 'appealing', and then quantifying the fat, sugars, carbohydrate, and sodium content of those and similar foods. 1  The UPFs metric, is well known.It categorises foods (and drinks) as ultra-processed 'according to the extent and purpose of industrial processing', which includes making the 'final product palatable or often hyper-palatable' (Monteiro et al., 2019, p. 937).UPFs include, for example, 'carbonated drinks; sweet or savoury packaged snacks; ice cream, chocolate, candies (confectionery); mass-produced packaged breads, buns, cookies (biscuits), pastries, cakes and cake mixes; breakfast 'cereals', 'cereal' and 'energy' bars; margarines and spreads; processed cheese; 'energy' drinks; sugared milk drinks, sugared 'fruit' yoghurts and 'fruit' drinks' (Monteiro et al., 2017, p. 17).
The UK Department of Health (2011) nutrient profiling metric sets out a formula for identifying foods high in energy, fat, sugar, and/or salt (HFSS foods).This metric was developed to provide guidance for regulations in England covering television advertising of foods to children, and guidance for regulations on promotional offers (e.g., 'buy one get one free') and the placement of food products in larger food stores.The HFSS metric was not specifically intended to identify foods based on palatability, rather it seeks to identify which foods are 'less healthy' (i.e., foods that fall within the HFSS category).This category comprises foods relatively high in combinations of the following features: energy density, saturated fat, sugars, and/or sodium.These features, however, are mitigated by the protein and fibre content of the food, and/or if it contains fruit, vegetables and/or nuts.So, for example, cashews (unsalted), and dried pitted dates, which have respectively relatively high saturated fat and high sugar content (g/100 g), are not categorised as HFSS, because they are 'nuts' or fruit, and they contain substantial amounts of fibre.(Full details of the HFSS profiling metric are available at https://assets.publishing.service.gov.uk/media/5a7cdac7e5274a2c9a484867/dh_123492.pdf) Because of its nutrient-based criteria, the HFSS foods metric would appear to be more like the HPFs metric than the UPFs metric.
Before, summarising the aims of the present research and specifying our hypotheses, it is important to be clear about the meaning of the term (psychological construct) palatability, and how palatability can be measured.The dictionary definition of palatable, in relation to food, is 'pleasant to the mouth'.This being so, palatability is a private experience of the eater; nonetheless, one which they can communicate in facial expression, verbally, and on a rating scale.In our studies, we ask participants to rate palatability using the following instructions: 'How PLEASANT does this food taste in your mouth?When making this judgement, ignore how much or little of the food you want to eat, and what it would be like to swallow itjust focus on how pleasant it tastes in your mouth right now.' (e.g., Gumussoy & Rogers, 2023;Rogers et al., 2021;Rogers & Hardman, 2015).The purpose of these precise instructions is to separate taste pleasantness (palatability) from the complete experience of ingesting a bite of the food, which we show is also affected by, for example, hunger.In this example, taste pleasantness and hunger determine what we call food reward, which we define as the 'in the moment value of the food to the individual' (Rogers & Hardman, 2015).We measure food reward using ratings of desire to eat and/or ratings of eating enjoyment (Rogers et al., 2021;Rogers & Hardman, 2015).Assuming the pleasure experienced from eating is closely associated with food reward, we can then separate pleasantness of the taste (palatability) of a food from the pleasure of eating itby assessing, respectively, (1) taste pleasantness, and (2) desire to eat or food enjoyment.This definition of palatability as taste pleasantness (or liking, to use a shorter word) is consistent, either fully or largely, with that provided by other authors (e.g., Yeomans, 1998;Young, 1959).
The data we used for the present analyses were collected in a virtual (online) study, in which participants were shown a colour image, together with the name, of each of the foods they evaluated (Rogers et al., 2024).They were asked to imagine taking a bite of the food depicted and tasting it, whilst rating its taste pleasantness, sweetness, saltiness, flavour intensity, and lastly, their desire to eat it.The validity of our virtual food rating methods, which in one form or another has been used successfully in many studies (e.g., Brunstrom et al., 2018;Buckland et al., 2015;DiFeliceantonio et al., 2018;Johnson et al., 2014;Monge et al., 2022;Perszyk et al., 2021;Schulte et al., 2017), was confirmed by various results.First, ratings of sweetness and saltiness of the foods were positively correlated with, respectively, food sugars and salt content.These relationships were of the same magnitude as found in studies in which participants, who in some cases where expert sensory-panellists, tasted the foods in question (e.g., van Langveld et al., 2017;Lease et al., 2016;van Dongen et al., 2012).Second, we found that hunger rated by participants at the start of their online session affected their desire to eat, but not taste pleasantness, which replicates findings for foods tasted and eaten in laboratory studies (e.g., Rogers et al., 2021;Rogers & Hardman, 2015).And third, the data replicated and extended previous research demonstrating a relationship between carbohydrate-to-fat ratio and food reward (DiFeliceantonio et al., 2018;Perszyk et al., 2021).Furthermore, it is self-evident that, if we (humans) are familiar with a food, we find it easy to imagine how it will taste and how much we currently desire it.In everyday life, such expectations motivate appetitive (food-seeking) behaviour, the initiation of consummatory behaviour, and ultimately ingestion of the desired foodunless, of course, we choose to resist our desire to eat it.
In summary, the present research investigated the claim that the nutrient-based HPFs metric identifies hyper-palatable foods.We did this by applying this metric to a dataset of 52 foods comprising data on both food composition and food palatability (taste pleasantness).We also analysed data on food reward value (desire to eat), and separately on food sweetness, saltiness, and flavour intensity.Our primary hypothesis was that foods categorised as hyper-palatable by the HPFs metric would be rated significantly and substantially higher on taste pleasantness than foods not categorised as hyper-palatable by the HPFs metric.Furthermore, we compared the HPFs metric with two other food compositionbased metrics, namely the NOVA and the nutrient profiling metrics, both in respect of their success in predicting food palatability, and their overlap (similarities and differences) with the HPFs metric in the categorisation of the foods.For all three metrics we also compared the different categories (i.e., HPFs versus non-HPFs, UPFs versus non-UPFs, and HFSS foods versus non-HFSS foods) in respect of variables that we found in our previous research (Rogers et al., 2024) to predict palatability, namely carbohydrate-to-fat ratio, fibre content, and taste intensity (mean of sweetness, saltiness, and flavour intensity), together with other nutrient variables, including energy density, and fat, carbohydrate, and sugars content.The purpose of these latter analyses was to further characterise overlaps between the metrics in their categorisation of foods according to nutrient composition, and in doing so to provide insight into their relative power to predict palatability.

Methods
As described above, the present paper reports further analyses of data collected in a previously published article (Rogers et al., 2024).Full details of the methods of the study are available in that article.A summary of those methods, together with the details of the present (new) data analyses, are presented below.
1 Fewer than half of the cited articles report palatability ratings (e.g., 'taste pleasantness', 'liking', or 'appeal' ratings) for the foods in question, and these data do not appear to have been used directly by the authors (Fazzino et al., 2019) to categorise the foods as hyper-palatable versus not hyper-palatable.Rather, it is the qualitative descriptions of the foods provided in these articles that guided their categorisation decisions.P.J. Rogers et al.

Pre-registration
The study was pre-registered on Open Science Framework: htt ps://osf.io/ea5by/.The data analyses presented herein were not specified in this pre-registration document, as we were not aware of the research by Fazzino et al. (2019) when we planned our study in 2020.

Participants
A total of 224 adults, 140 women and 84 men, completed the study.These participants were recruited via social media and word-of-mouth.A criterion for participation was that they should be familiar with foods eaten in the UK.Individuals were asked not to participate in the study if they were following a vegan or vegetarian diet, if they were feeling unwell, suffered from food allergies, or were currently suffering from altered taste or smell perception.The study was advertised simply as 'The food rating study', with no incentive being offered for taking part, other than knowing that engagement with the study would be contributing to the scientific understanding of food likes and dislikes.
The study was conducted according to the guidelines laid down in the World Medical Association Declaration of Helsinki, and all procedures involving human participants were approved by the University of Bristol School of Psychological Science Research Ethics Committee (approval code 140121116006).Written informed consent was obtained from participants prior to them beginning the study survey.On completion of the survey, they confirmed consent for the anonymous use of their data.In both instances, this consisted of the participant checking a box confirming that they had read the relevant information.

Study design and randomisation
Participants evaluated three partly overlapping sets of foods (32, 24, or 24 foods, evaluated by respectively 73, 72, and 79 different participants) via an online survey built using the Qualtrics XM platform (Qualtrics, Provo, UT, USA).These three 'study arms' were designed to investigate effects of food energy density (32 foods, ranging from a minimum energy density of 17 kcal/100 g (cucumber) to a maximum of 614 kcal/100 g (roasted salted peanuts)); NOVA level of processing (24 foods, representing unprocessed (e.g.avocado), minimally processed (e. g., sultanas), processed (e.g., olives), and ultra-processed foods (e.g., sugar ring doughnut)); and carbohydrate-to-fat-ratio (24 foods, containing predominantly carbohydrate (e.g., bagel), or similar amounts of carbohydrate and fat (e.g., blueberry and raisin muffin), or predominantly fat (e.g., pepperoni sausage)).Individual participants (total n = 224) were randomised to complete one of these study arms, and each evaluated all the foods within their allocated study arm.Each food was evaluated on five dependent variables, as described below.Participants were able to complete the survey on a laptop or desktop computer, or on a tablet or smart phone.Participant allocation to food set was randomised.In total, 52 different foods were evaluated by a minimum of 72 participants, and a maximum of 224 participants (i.e., when a food occurred in all three study arms).Within each set of foods, the order in which the foods was presented for evaluation was randomised across participants.The results reported herein are based on analyses of 'consensus' (i.e., mean) scores for each food, for each of the relevant dependent variables (e.g., palatability and desire to eat).In our previous article (Rogers et al., 2024), we use the term 'food level analysis' to refer to this method of analysis.

Food stimuli
We selected the foods to test hypotheses concerning the role of energy density, NOVA level of processing, and carbohydrate-to-fat ratio in determining palatability and food reward (Rogers et al., 2024).In doing so, we selected foods that varied substantially in these processing and nutrient characteristics.Further considerations were that the foods be clearly identifiable to our participants, relatively uniform in composition, and acceptable if eaten on their own.We did not present foods comprising multiple components/items, such as pasta in tomato sauce, pizza, or sandwiches, or meals, because of the greater variability of such products, and the complexity of assessing palatability when an imagined bite of the same food may vary in its composition (e.g., bites of pizza differing in proportion of crust versus topping).
Participants evaluated each of their allocated 24 or 32 foods singly.On each food trial, they were shown a colour image of the food in question, together with the written name of the food (e.g., Avocado; Cherry tomatoes; Cornish dairy ice cream; Crisps, Pringles original; Egg, boiled; Frankfurter sausage; Fresh apple slices; King prawns, cooked; Olives; Parma ham; Strawberry yogurt, Ski; Traditional Cheddar cheese; Wine gums).Each food was presented as a 50 g portion on a light-beige coloured plate measuring 20 cm in diameter.The images were made by taking high-resolution, overhead photographs of the foods under artificial lighting.Details of all 52 foods, including nutrient composition data, and example photographs, are shown in Supplemental Methods and Results Table S1 and Fig. S1.

Measures
On each food trial, participants were instructed to imagine taking a bite of the food depicted and tasting it, following which they evaluated the food on five clickable visual-analogue line scales displayed immediately below the food (which remained visible).These line scales were anchored 'not at all' (=0) and 'extremely' (=100), and they were presented in the following order: palatability ('How PLEASANT does this food taste in your mouth?When making this judgement, ignore how much or little of the food you want to eat, and what it would be like to swallow itjust focus on how pleasant it would taste in your mouth right now.'); sweetness ('How SWEET does this food taste?'); saltiness ('How SALTY does this food taste?'); flavour intensity ('Apart from its sweetness and saltiness, how intense is the FLAVOUR of this food?');and food reward ('How strong is your DESIRE TO EAT some of this food right now').Lastly, participants rated their familiarity with eating the food (I have eaten this food, or a very similar food), for which there were three response options ('Often', 'Sometimes', 'Never').
Palatability (i.e., liking), measured by taste pleasantness ratings was the primary outcome measure for the present analyses.Food reward, measured by desire to eat ratings, was a secondary outcome measure.

Procedure
Participants were instructed to have not eaten within 2 h prior to starting the survey, and to complete it on a single occasion between 11 a. m. and 2 p.m. on any day of the week.Whilst we did not collect information on time of last meal, or on time of day the survey was completed, we have no reason to believe that significant numbers of our participants would have ignored these instructions.The survey began with a repetition of the information sheet.This was followed by confirmation of initial consent, hunger ratings, the food ratings, collection of demographic and trait questionnaire information (not reported in detail here), and confirmation of final consent.
Typically, the entire survey took 30-40 min to complete, with the food ratings section completed within, at most, the first 25 min.We did not include attention-check items in the survey; however, as described in sections 1 and 4.4, food sweetness and saltiness ratings made by participants in the study replicated the relationships with, respectively, food sugar and salt content, found in studies in which foods were tasted under supervised laboratory conditions.These results suggest that the quality of the data we collected was not adversely affected by participant fatigue or inattention.
No information was communicated to participants about the study hypotheses until after the end of the 25-day data collection period, when they were all sent a debriefing document.Data collection started on P.J. Rogers et al.March 10, 2021 and ended on April 03, 2021.

Data analysis
All statistical analyses were carried out using IBM SPSS version 28.0.1.0.For the present analyses, we first categorised the 52 foods according to: (1) the HPFs metric described by Fazzino et al. (2109), including following their procedure of subtracting fibre and simple sugars (naturally occurring and added sugars) from total carbohydrate for the calculation of percent energy from carbohydrate; (2) the UPFs metric as described by Monteiro et al. (2017;2019), following discussion and agreement among three researchers (co-authors of Rogers et al., 2024); and (3) the HFSS foods metric described in the UK Department of Health, Nutrient Profiling Technical Guidance document (2011).For each metric we categorised each food as falling within the designated category or outside the designated category (e.g., either as an HPF, or a non-HPF).
Next, we investigated overlaps and differences between the three metrics by cross tabulating the results for the HPFs versus UPFs metrics, the HPFs versus the HFSS foods metrics, and the UPFs versus HFSS foods metrics.We used the Chi-Square statistic to test for differences in the categorisation of foods by the metrics.We also represented the overlaps and differences between the metrics in the form of a Venn diagram.
Finally, for each metric we compared the foods categorised as falling within the designated category versus outside the designated category (e.g., HPFs versus non-HPFs) for three sets of variables using independent samples t-tests.These variables were: (1) palatability, and food reward, which were our primary and secondary outcome measures; (2) carbohydrate-to-fat energy ratio, 2 food fibre content, and taste intensity (mean of sweetness, saltiness, and flavour intensity ratings), which we found in our previous analyses (Rogers et al., 2024) to be significant independent predictors of palatability in this set of 52 foods; and (3) six further food composition variables, namely, energy density (kcal/100 g), and fat, carbohydrate, sugars, protein, and salt content (g/100 g).We also conducted analyses for palatability, taste intensity, and food composition variables separately for the three HPFs-metric clusters, namely the fat-salt, fat-sugar, and carbohydrate-salt clusters.We report effect sizes, together with t and p values for all these analyses in full in Supplemental Methods and Results Table S2.We set alpha <0.05 (two-tailed) as our criterion for statistical significance.In summary, using independent samples t-tests, we compared the palatability, etc. of our set of 52 foods when those foods were divided according to the HPFs metric (HPFs versus non-HPFs), the NOVA metric (UPF versus non-UPF), and the HFSS metric (HFSS versus non-HFSS foods).For palatability, food reward, and taste intensity, the data for each food were based on mean ratings from a minimum of 72 participants.For food composition variables, the data were calculated from information provided by the food manufacturers and/or retailers.
Finally, we conducted sensitivity analyses in which we first excluded the 20% of foods (rounded to the nearest whole number of foods) with the lowest mean food familiarity scores within each category of the three metrics, namely HPF and non-HPF, UPF and non-UPF, and HFSS and non-HPSS.We then repeated the analyses (independent samples t-tests) comparing HPFs versus non-UPFs, etc. for palatability and food reward in these reduced datasets.The purpose of these analyses was to investigate the possibility of bias arising from less familiar foods being rated less reliably for palatability, and/or food reward.

Participants characteristics
Two-hundred and twenty-four participants (140 women and 84 men) completed the study.Their mean (±SD) age and BMI were 34 ± years, and 24.4 ± 4.9 kg/m 2 , respectively.The proportions of participants with underweight, healthy weight, overweight, and obesity/severe obesity were 5%, 53%, 30%, and 13%, respectively.Participants' hunger ratings made at the start of the survey on the 101-point scale (anchored 'not all' to 'extremely'), revealed on average a moderate degree of hunger (mean 49 ± 26).Further details of participant characteristics, including eating behaviour traits and physical activity levels are available in Rogers et al. (2024) (https://www.sciencedirect.com/science/article/pii/S0195666323025862).

Overlaps and differences among the three food-categorisation metrics
Tables 1a, 1b and 1c show that there was substantial overlap in the categorisation of foods by the three metrics.Nonetheless, as revealed by the Chi-square statistic, there were also significant differences between the metrics.Categorisation diverged for 15 of the 52 foods for the HPF versus UPF comparison (10 HPFs were not categorised as UPFs, and UPFs were not categorised as HPFs) (Table 1a).Categorisation diverged for 12 of the 52 foods for the HPF versus HFSS comparison (5 HPFs were not categorised as HFSS foods, and 7 HFSS foods were not categorised as HPFs) (Table 1b).For the UPF versus HFSS comparison (Table 1c), categorisation diverged for 13 of the 52 foods (3 UPFs were not categorised as HFSS foods, and 10 HFSS foods were not categorised as HPFs).
2 Food carbohydrate-to-fat ratio is calculated from nutrient composition data.
It is the amount of carbohydrate (kcal/100 g) divided by the amount of carbohydrate (kcal/100 g) plus the amount of fat (kcal/100 g).That is, carbohydrate energy/(carbohydrate energy + fat energy).Thus, foods containing similar amounts of carbohydrate and fat energy will have a carbohydrate-to-fat ratio of or close to 0.5 (e.g., sugar-ring doughnut = 0.508), whereas foods containing substantially more carbohydrate than fat energy have a ratio closer to 1 (e.g., sultanas = 0.987), and foods having substantially more fat than carbohydrate energy have a ratio closer to zero (e.g., pepperoni sausage = 0.015).However, our hypothesis, supported by previous empirical results (Rogers et al., 2024;DiFeliceantonio et al., 2018), is that foods containing more equal amounts of carbohydrate and fat are more valued than foods containing predominantly carbohydrate or predominantly fat.Therefore, to reflect (i.e., to appropriately quantify) this relationship for our data analyses, we calculated a transformed measure of carbohydrate-to-fat ratio which scores both predominantly carbohydrate-containing and predominantly fat-containing foods as low, and scores foods with more equal amounts of carbohydrate and fat energy as high.We calculated this transformed carbohydrate-to-fat (energy) ratio using the equation: ratio = 1 -Abs(2x -1), where x = carbohydrate energy/(carbohydrate energy + fat energy).Accordingly, possible scores vary continuously from 0 (contains carbohydrate but no fat energy, or contains fat but no carbohydrate energy), to 1 (contains equal amounts of carbohydrate and fat energy).Hereinafter, we refer to this measure as 'transformed carbohydrate-to-fat ratio'.In additional analyses, we adopted a transformed carbohydrate-to-fat ratio of 0.6 as the cut off to categorise foods as either 'combo' foods (transformed ratio >0.6 to 1) or 'non-combo' foods (transformed ratio ≤0.6 to zero).Therefore, for the raw, untransformed carbohydrate-to-fat ratio measure (x, in the equation shown above), 'combo' foods can have a ratio of between 0.3 and 0.7.The relationship between raw and transformed carbohydrate-to-fat ratio measures, and the criteria for determining 'combo' foods, are illustrated graphically in Supplementary Methods and Results, page 10.

P.J. Rogers et al.
Fig. 1 shows that thirty-five of the 52 foods were categorised as at least one of HPF, UPF, or HFSS, and that 15 of the foods were categorised as HPF, UPF, and HFSS.These 15 foods included Babybel cheese, chocolate mousse (Aero), Cornish dairy ice cream, crisps (Pringles original), flapjack bites, frankfurter sausage, oatcakes, pork and pork liver pâté, and pretzels.Eight foods were categorised as both HPFs and HFSS foods, including freshly baked croissant, olives, salted peanuts, and smoked salmon.Three foods were categorised as HPFs and UPFs, including bagel, and strawberry yogurt.Five foods were categorised as both UPF and HFSS, including sugar ring doughnut, and wine gums.Lastly, two foods (natural yogurt, and pitta bread) were categorised only as HPF, and two foods (sultanas, and traditional buffalo mozzarella cheese) were categorised only as HFSS, with no foods categorised only as UPF.

Palatability, taste intensity, and food composition
The first two data columns of Table 2 show the results for foods categorised as HPFs versus non-HPFs, for the primary outcome of palatability, the predictors of palatability identified in our previous analyses (Rogers et al., 2024), and the six additional compositional characteristics of these categories of foods.The third and fourth data columns do the same for the UPFs metric, and the fifth and sixth columns for the HFSS foods metric.Full details of the statistics, including effect sizes, are available in Supplemental Methods and Results Table S2.
Palatability did not differ significantly between HPFs and non-HPFs, or between UPFs and non-UPFs (the effect sizes were small or very small, 0.230 and 0.117, respectively).There was a significant difference in palatability between HFSS and non-HFSS foods (p = 0.049), with a medium effect size (Cohen's d = 0.566).The results for food reward (desire to eat) showed, if anything, smaller differences than for palatability, with a very small negative effect size for UPFs versus non-UPFs.
The extent to which the HPFs metric predicted palatability differed according to HPFs cluster.Foods categorised as HPFs by the fat-salt cluster were equal in palatability to non-HPFs.By contrast, foods categorised as HPFs by the fat-sugars cluster were more palatable than non-HPFs; whereas foods categorised as HPFs by the carbohydrate-sodium cluster were less palatable than non-HPFs.Further details of these results are provided in Supplemental Methods and Results Table S3.
In sensitivity analyses we investigated differences in palatability (and food reward) after excluding the 20% of foods having the lowest Chi-square (df = 1) = 14.86, p = 0.0001.
b The transformed measure of carbohydrate-to-fat ratio was calculated as described in the data analysis section.Possible scores vary continuously from 0 (contains carbohydrate but no fat energy, or contains fat but no carbohydrate energy) to 1 (contains equal amounts of carbohydrate and fat energy).c Mean of sweetness, saltiness, and flavour intensity, each measured on a 101-point scale anchored not at all (=0) and extremely (=100).d ED = energy density.mean food familiarity scores within each food category, separately for the HPFs, UPFs, and HFSS foods metrics.Results showed the same pattern of effects observed for the full datasets, albeit with somewhat larger effect sizes.The only significant difference was for the palatability of HFSS versus non-HFSS foods (p = 0.029), with the next smallest pvalue being 0.209.Full details are shown in Supplementary Methods and Results Table S4.
Transformed carbohydrate-to-fat ratio differed significantly between HPFs and non-HPFs, with HPFs on average containing carbohydrate and fat in more equal amounts (kcal) than non-HPFs.A similar difference in transformed carbohydrate-to-fat ratio was observed for UPFs versus non-UPFs.Transformed carbohydrate-to-fat ratio did not differ significantly between HFSS and non-HFSS foods, although the difference approached a medium effect size (d = 0.456).(Note that a 50:50 mixture of carbohydrate and fat energy on the transformed measure equals a score of 1, whilst a 0:100 and 100:0 ratio equals a score of 0; and, for example a 30:70 ratio equals a score of 0.6, as does a 70:30 ratio.See footnote 2.) The above pattern of results for carbohydrate-to-fat ratio is confirmed when the results are visualised in terms of 'combo' versus 'non-combo' foods (i.e., foods with a transformed carbohydrate-to-fat energy ratio of, respectively, >0.6 versus ≤0.6, as explained in footnote 2).There were 16 combo foods and 36 non-combo foods.HPFs were approximately twice as likely as non-HPFs to be combo foods (13/ 16 (81%) versus 15/36 (42%), p = 0.008).Notably, all eight of the foods categorised as a HPF by the HPFs fat-sugars cluster were combo foods.The overall results were similar for the UPFs metric: UPFs were approximately twice as likely as non-UPFs to be combo foods (11/16 (69%) versus 12/36 (33%), p = 0.018).For the HFSS metric the proportion of combo foods was distributed substantially, though not significantly, differently between HFSS foods and non-HFSS foods: 12/ 16 (75%) versus 18/36 (50%), p = 0.092.Further details of these crosstabulations between food metric category and carbohydrate-to-fat ratio category are shown in Supplemental Methods and Results Tables S5a, S5b, and S5c.
In contrast to carbohydrate-to-fat ratio, fibre content was very similar for HPFs and non-HPFs, and for UPFs and non-UPFs, but somewhat, though not significantly, lower in HFSS foods than in non-HFSS foods (small effect size d = − 0.353).Taste intensity differed between HFSS and non-HFSS foods (very large effect size, d = − 1.457), and to a lesser extent between UPFs and non-UPFs, and not significantly so between HPFs and non-HPFs.

Prediction of palatability and food reward by the three food categorisation metrics
A key finding from this study is that the hyper-palatable foods (HPFs) metric (Fazzino et al., 2019) did not significantly predict palatability, measured by taste pleasantness ratings.The 28 foods categorised as hyper-palatable by this metric did not differ significantly in taste pleasantness from the 24 foods categorised as non-hyper-palatable.That is, a metric that was developed explicitly to identify HPFs failed to do so in our dataset of 52 foodswhich included HPFs such as strawberry yogurt, olives, bagel, Frankfurter sausage, Babybel cheese, pretzels, flapjack, potato crisps, milk chocolate, and salted peanuts; and non-HPFs such as cherry tomatoes, fresh apple, king prawns, avocado, wine gums, and sugar-ring doughnut. 3This result cannot be explained by a ceiling effect for taste pleasantness ratings, as the mean ratings on the 0-100 point scale, anchored not at all to extremely, were 52.6 (HPFs) versus 50.0 (non-HPFs) (small effect size of d = 0.230, p = 0.412).It is clear, therefore that the metric did not successfully identify 'hyper-palatable' foods (i.e., foods with very high, or extremely high, palatability).
Similarly, palatability did not differ between ultra-processed foods (UPFs) and non-UPFs (23 versus 29 foods).This finding stands in stark contrast to the claim that UPFs are 'made to be hyper-palatable' (Monteiro et al., 2017), and the often-repeated assertion that UPFs are indeed highly palatable or hyper-palatable (e.g., Adams et al., 2020;Dicken et al., 2024;Global Research Food Programme, 2023).On the other hand, our finding is consistent with the results of a randomised cross-over trial comparing the effects of 14 days of ad libitum consumption of UPFs versus unprocessed meals (Hall et al., 2019), which found no significant difference in participant-rated palatability (pleasantness) of the meals provided.It is also consistent with reviews which conclude that it is an exaggeration to claim that UPFs are hyper-palatable (Forde, 2023;Gibney et al., 2017;Valicente et al., 2023).
To reiterate, the present results do not support the claims that HPFs and UPFs are 'hyper-palatable'.The English dictionary definition of the prefix 'hyper' is 'excessive' or 'extreme'.As such, the term hyperpalatable strongly implies that foods claimed to be hyper-palatable will be experienced as considerably more palatable than foods claimed not to be hyper-palatable.In other words, the effect size for this comparison can be expected to be, at least, large (i.e., the effect size is denoted in the name of the concept: hyper-palatability).For the present sets of foods, effect sizes for the differences in palatability between HPFs and non-HPFs, and between UPFs and non-UPFs, were small for the full dataset, or small-medium for the sensitivity analyses datasets (i.e., reduced datasets with foods falling within the lowest quintile of food familiarity scores within each food category deleted).None of the differences were statistically significant (smallest p = 0.209).
In contrast to the results for the HPFs and UPFs metrics, there was a statistically significant (p = 0.049) difference in palatability between high fat, sugar, salt foods (HFSS) and non-HFSS foods (30 versus 22 foods).The effect size for this difference is medium (d = 0.566).(In the sensitivity analysis, the effect size for this comparison increased to medium-to-large (d = 0.707), p = 0.029).The HFSS metric aims to identify foods that are 'less healthy' because of their contribution to fat (especially saturated fat), sugar and salt intake in the diet.Paradoxically, palatability is not mentioned in the technical guidance for categorising HFSS foods (Department of Health, 2011).
For the avoidance of doubt, we also analysed the data we collected for desire to eat each of the 52 foods evaluated by our participants.As described in the Introduction section, desire to eat is a measure of food reward, which is determined independently by taste pleasantness (liking/palatability) and wanting (Rogers et al., 2021;Rogers & Hardman, 2015).Our results showed that food reward did not differ significantly between HPFs and non-HPFs, between UPFs and non-UPFs, or between HFSS foods and non-HFSS foods, which confirms that there are minimal differences in the attractiveness of these different categories of foods.

Overlaps between the three food categorisation metrics, and what features of foods predict palatability
Overall, the results for palatability, and for food reward, were similar for the three metrics, HPFs, UPFs and HFSS foods, in that none of them very clearly identified hyper-palatable (i.e., extremely palatable) foods, or highly rewarding, foods.The similarity in results for the three metrics is unsurprising, in that there was substantial overlap between the categorisation of the 52 foods.As shown in Fig. 1, 15 (29%) of the foods were categorised as HPFs, UPFs, and HFSS foods, and 17 (33%) of the foods were categorised as non-HPFs, non-UPFs, and non-HFSS foods.Nevertheless, there were also significant differences in the categorisation of energy amounts of carbohydrate and fat (46.8% and 45.2%, respectively), it does not meet the criteria to be categorised as a HPF because in contains <20% energy from sugars.P.J. Rogers et al. the foods by the different metrics (Tables 1a-1c), which can be understood in terms of differences in the criteria they use to categorise foods. 4 For example, the HPFs and HFSS foods metrics, which categorise foods based primarily on macronutrient and salt content, show differences in both fat and salt content for HPFs versus non-HPFs, and for HFSS foods versus non-HFSS foods.However, perhaps unexpectedly, sugar content is lower for HPFs versus non-HPFs.This comes about mainly because the HPFs metric includes sugars (% of total calories) in only one of the three clusters of nutrients that define HPFs, namely the fat and sugars cluster.In our dataset there were eight HPFs, for example, strawberry yogurt, blueberry and raisin muffin, and milk chocolate, in this cluster.The HPFs metric, however, categorises foods high in sugars but low in fat, such as sultanas, dried apple slices, and jelly babies (each of which contain more than 50 g/100 g of sugars), as non-HPFs.
The significance of food sugars and salt content for palatability lies in our (human) inborn liking for sweet and salty tastes (Liem, 2017;Rogers et al., 2024;Steiner, 1987;Steiner et al., 2001), though as we showed previously, adding more sugar or more salt does not necessarily translate into higher food palatability, in part at least because there is a non-linear relationship between food sugars and salt content and, respectively, sweetness and saltiness (Moskowitz, 1970;Rogers et al., 2024;Stevens, 1960).Nevertheless, the lower sugars content of HPFs versus non-HPFs undoubtedly explains the lower sweetness of the HPFs (Supplemental Methods and Results Table S2), which in turn contributes to the non-significant difference in taste intensity of HPFs versus non-HPFstaste intensity is the mean of sweetness, saltiness, and flavour intensity ratings.In addition, there was no significant difference in the fibre content of HPFs versus non-HPFs.Together, these two results are telling, because our previous analyses of this dataset showed that taste intensity (positively) and fibre content (negatively) independently predict palatability (Rogers et al., 2024).In other words, it appears that the failure of the HPFs metric to predict palatability judgements is due substantially to its failure to discriminate between two key determinants of food palatability, namely taste intensity and fibre content.On the other hand, HPFs did have a higher transformed carbohydrate-to-fat ratio, and correspondingly were more likely to be 'combo' foods, than non-HPFs.Because combo foods are more palatable (Rogers et al., 2024), this contributed positively to the overall palatability of foods within the HPFs category 5 ; however, that was balanced by the relatively low palatability of other HPFs, for example, in the HPFs carbohydrate-salt cluster.
Like HPFs, UPFs had a higher transformed carbohydrate-to-fat ratio and so were more likely to be combo foods.Additionally, UPFs had higher taste intensity than non-UPFs (medium effect size).Nonetheless, and also like HPFs, UPFs did not differ in fibre content from their non-UPFs counterparts, nor did they differ in palatability.In contrast, to the HPFs and UPFs metrics, the HFSS foods metric did, as discussed above, significantly predict palatability, albeit with a medium effect size.HFSS foods were rated as more palatable than non-HFSS foods.This difference in palatability appears to be driven primarily by the much higher taste intensity of HFSS foods compared with non-HFSS foods (very large effect size, d = 1.457).
The present analyses demonstrate that assumptions made about what determines food palatability can prove unreliable.Neither the HPFs metric nor the UPFs metric successfully predicted food palatability.A major problem for these metrics is that they are not based on substantive data on palatability, but rather on assumptions of the authors concerned, or the authors they reference, as to what types of foods are highly palatable (or 'hyper-palatable') versus what types of foods are less palatable.This method is at variance with palatability being a personal hedonic experience of the eater (e.g., Forde et al., 2023;Rogers, 1990), communicated through taste pleasantness ratings, which is subject to large individual differences.In part, those differences relate to prior exposure to the food (or drink) in question, because palatability is modified via taste-to-postingestive-consequences learning (Riordan & Dwyer, 2019;Schier et al., 2019;Myers, 2018;Yeomans, 2012;Dickinson & Balleine, 2009;Brunstrom, 2007;Yeomans et al., 1998;Sclafani & Nissenbaum, 1988).Thus, palatability is not a fixed property of a food, rather it is shaped by outcomes of the individual's past interactions with that food.Therefore, to obtain reliable estimates of palatability it is necessary to collect data on palatability judgments from a significant number of people who familiar with the foods in question.
That is not to say that palatability cannot be predicted with some degree of confidence from knowledge of food composition, as we found for this dataset (Rogers et al., 2024).In terms of purely compositional variables (i.e., variables that can be calculated from food nutritional data), food fibre content and carbohydrate-to-fat ratio account for 39% of the variance in palatability across our set of 52 foods (Rogers et al., 2024).And, as described above, adding taste intensity increases the prediction to 56% of the variance in palatability (Rogers et al., 2024). 6However, we were only able to identify these significant predictors of palatability because we measured the palatability of the foods in question.The next step is for us, and/or others, to validate our composition-based metric by applying it to the prediction of the palatability of different sets of foodswhich can be done in future studies using the information we have presented herein.

Energy density and energy-to-satiety ratio
It is notable that the foods categorised by each of the three metrics differed in energy density (kcal/100 g), that is HPFs, UPFs, and HFSS foods had significantly higher energy density on average than their non-HPFs, non-UPFs, and non-HFSS foods counterparts.However, as demonstrated by our previous analyses (Rogers et al., 2024), energy density does not predict palatability.Rather than energy density, we argue that energy-to-satiety ratio is a key determinant of palatability, because this coincides with the goal of food choice and intake, which is the ingestion of macronutrients (and micronutrients), not to achieve satiety.Indeed, satiety, limits nutrient ingestion.Unfortunately, 4 Our finding of a substantial overlap, but also differences, in categorisation of foods by the three metrics are consistent with results comparing the HPFs versus UPFs metrics (Sutton et al., 2024), and the UPFs versus HFSS foods metrics (Dicken et al., 2024).Sutton et al. report that, depending on the year of the survey, between 66% and 73% HPFs were also categorised as UPFs, whilst between 60% and 78% of UPFs were categorised as UPFs.The corresponding percentage overlaps for our dataset were 64% (HPFs categorised as UPFs) and 78% (UPFs categorised as HPFs) (Fig. 1).Dicken et al. report a 'partial overlap' between front of pack label (number of green and red 'traffic lights') and UPF categorisation.Based on correspondence between front of pack label and food energy, fat, sugar, and salt content, this indicates significant overlaps between HFSS and UPF categorisations, as we observed in the present study (Fig. 1).The Sutton et al. and Dicken et al. studies analysed much larger food datasets than ours, namely, datasets representing the US food system, and the UK National Diet and Nutrition Survey, respectively.None of those datasets, however, include data on food palatability.
5 HPFs were twice as likely to be combo foods (section 3.3) than non-HPFs.
Indeed, all eight foods in the HPF fat-sugar cluster were combo foods, and that cluster had the highest overall palatability rating (Supplemental Methods and Results Table S3).
6 Creating a metric based on a combination of binary cut-offs for food fibre content, carbohydrate-to-fat ratio, and taste intensity, to categorise our foods as 'higher palatable' (n = 26) versus 'lower palatable' foods (n = 26), we find a highly significant difference in palatability of (mean ± SD) 58.although energy density is convenient to calculate from food composition data, it appears to be only a weak proxy for energy-to-satiety ratio.For example, milk chocolate is ten times more energy dense that fresh apple, but there are good reasons to think that milk chocolate is not, calorie for calorie, ten times less satiating than apple (Mantzavinou & Rogers, 2023;Rogers et al., 2024).Nevertheless, it may be possible to reliably estimate food energy-to-satiety ratio from food composition data.For example, we suggest that the contribution of carbohydrate-to-fat ratio and food fibre content to food energy-to-satiety ratio can explain why these variables predict palatability (independently of taste intensity).Specifically, we propose that energy-to-satiety ratio is higher for foods with more equal energy amounts of carbohydrate and fat, and lower for foods having larger amounts of fibre (Rogers et al., 2024).There is good evidence that foods high in fibre are more satiating (e.g., Salleh et al., 2019;Poutanen et al., 2017;Holt et al., 1995), and we have evidence that carbohydrate-to-fat ratio also affects satiety as predicted (Flynn et al., 2023b(Flynn et al., , 2024)).Future studies could focus on expanding on this evidence, especially for carbohydrate-to-fat ratio.Additionally, they should consider the role of food texture, solid versus semi-solid versus liquid, in influencing per calorie satiation and satiety (e.g., Forde & de Graaf, 2022).
The arguments concerning energy-to-satiety ratio as a determinant of palatability do not deny the role of energy density in affecting energy intake.It is well established that consumption of energy dense versus less energy dense foods in a meal or over the longer-term leads to greater ad libitum energy intake (Robinson et al., 2022;Rolls, 2017).This demonstrates that there is not calorie-for-calorie metering and control of food intake during a meal, or at least not for meals with an overall energy density of less than ≈1.75 kcal/g (Flynn et al., 2022(Flynn et al., , 2023a)).Similarly, incomplete compensation for energy density is observed in studies using the preload test-meal design, and in longer-term feeding trials (Flynn et al., 2022).This effect of food energy density on energy intake likely accounts, at least in part, for the higher daily energy intake of participants consuming a UPFs versus an unprocessed foods diet (Hall et al., 2019).Furthermore, the contribution of HPFs consumption to energy intake in this UPFs trial, and in other dietary trials (Fazzino et al., 2023), may be related to both the higher energy density and the higher (transformed) carbohydrate-to-fat ratio of HPFs, and not because HPFs are more palatable than non-HPFswhich we show here (Table 2) not to be the case. 7

Limitations
At this point it is important to consider the generalisability and limitations of the study and its findings as presented and discussed above.The research was conducted in the UK, with a criterion for participation being that participants should be familiar with foods eaten in the UK.Our sample of participants was not fully representative of the UK population, with a younger mean age of 34 years (18-89 years), and 62.5% of the participants being female.The sample comprised participants with BMIs ranging from underweight to obese, with 43% having overweight or obesity.By comparison, the prevalence of overweight and obesity in the UK in 2021 was 64% (Baker, 2023) -we collected our data in March to April 2021.We did not measure the socio-economic status of our participants; however, it is likely that people with higher socio-economic status of were overrepresented in our sample.
We limited the sample of foods we tested to foods that are relatively uniform in composition, acceptable if eaten on their own, and that we anticipated would be readily identifiable by our participants.We did not include foods comprising multiple components, such as pasta in tomato sauce, sandwiches, or meals, because of the greater variability of such products, and the complexity of assessing palatability when an imagined bite of the same food may vary in its composition.Nonetheless, we did include a wide range of foods representing, for example UPFs and non-UPFs, and for each of the three metrics there were a good number of foods represented in each category (HPFs = 28, non-HPFs = 24; UPFs = 23, non-UPFs = 29; HFSS foods = 30, non-HFSS foods = 22).
As described above (sections 1 and 4.1), we predicted a large effect size for palatability differences in respect of the HPFs versus non-HPFs comparison, and the UPFs versus non-UPFs comparison, because these metrics claim to identify hyper-palatable foods (Fazzino et al., 2019), or foods 'made to be hyper-palatable' (Monteiro et al., 2019).The large effect size is strongly implied by the prefix 'hyper'.There is a question, however, as to whether our study was indeed powered sufficiently to detect at least a large effect size.Our reference here to 'large effect' refers to Cohen's d = 0.8.In other words, a difference of 0.8 x SD in mean palatability ratings between two sets of foods, which for our dataset is 0.8 x 11.3 = 9.0 points.(The foods were rated for palatability on a 0-100 point scale, anchored 'not at all' and 'extremely'.)A power calculation shows that the study had 90% power to detect a difference of this magnitude (one-tail) (Howell, 1997).That is, if the null hypothesis (there is not a large difference in palatability in favour of foods categorised as, for example, HPFs versus non-HPFs) is false, we had an 90% chance of correctly rejecting the null hypothesis.To put our predicted effect size further into perspective, for Cohen's d = 0.8 it can be expected that 79% of non-HPFs foods would have a palatability value falling below the mean palatability of HPFs.The observed percentage for non-HPFs versus HPFs was 62.5%.For the non-UPFs versus UPFs comparison this percentage was 58%, and for the non-HFSS versus HFSS foods comparison it was 73%.Despite our study being well-powered to detect at least a large effect size in palatability between 52 foods grouped according to each of the three metrics we tested, it remains for future studies to extend our findings for these metrics and other metrics, in different and larger sets of foods.
Our method for assessing palatability was to conduct an online (virtual) study in which participants were asked, for each of their assigned foods, to imagine taking a bite of the food and then to rate how pleasant in tastes in their mouth, its sweetness, etc.Given the resources we had available, this enabled us to test more participants and more foods than would have been feasible had we conducted a laboratorybased study in which the foods were tasted in the mouth.It is possible that if we replicated our study using the latter procedure, we would find different results for palatability and come to different conclusions.There are several observations, however, which suggest this is not the case.First, the ratings for (imagined) food sweetness and saltiness replicate the relationships with, respectively, food sugar and salt content, found in studies in which foods were tasted (section 1 above, and Rogers et al., 2024).Second, in a separate study (Gumussoy & Rogers, 2023), we found that imagined palatability and palatability after tasting of a familiar food (peanuts) did not differ.Furthermore, before-and after-tasting palatability ratings also did not differ for a food that was unfamiliar, but acceptable, to the study participants (leblebi, which is roasted chickpeas).It appears, therefore, that participants can judge palatability accurately based on the visual appearance and the name of the food ('Ready to eat snack', peanuts; and 'Ready to eat snack, 7 In the Hall et al. (2019) study, the energy density of the meals provided did not differ between the UPFs diet versus unprocessed foods diet.However, the mean energy density of the foods and beverages consumed did differ between the diets (1.36 versus, 1.09 kcal/g, respectively).In the further analyses of the Hall et al. (2019) dataset and other datasets, energy density and %HPFs consumed both contributed significantly to the prediction of energy intake (Fazzino et al., 2023).Our results show that these variables are partly related, in that HPFs are more energy dense than non-HPFs.Separate from energy density, however, we suggest that HPFs will also be less satiating calorie for calorie due to their higher transformed carbohydrate-to-fat ratio.This complexity may help explain why Fazzino et al. (2023) found a significant energy density by %HPFs interaction effect for energy intake, in addition to the independent effects of energy density and %HPFs.Our reasoning concerning the effect of carbohydrate-to-fat ratio on satiation and satiety is set out in Rogers et al. (2024).P.J. Rogers et al. leblebi'), irrespective of familiarity.Third, and consistent with this second observation, our sensitivity analyses, in which we deleted foods falling within the lowest quintile of food familiarity scores within each food category, replicated the results for the full datasets. 8Finally, at the very least, it can be assumed that imagined palatability will typically have a major impact on eating, by motivating appetitive behaviour and by influencing food choice.Taken together, these observations support the validity of our methods, and they testify to the high quality of our dataset.Nonetheless, clearly it would be desirable to replicate and extend the present research in studies using taste-and-spit methods to assess food palatability.Furthermore, this should be done with foods representative of other cuisines and countries; and with the inclusion of more representative and diverse samples of participants, coupled with analyses of individual differences in responses to different categories of foods (e.g., HPFs versus non-HPFs).We are currently planning to carry out at least one such a study.

Conclusions
In conclusion, the present research found that two metrics which claim to identify 'hyper-palatable' foods did not clearly predict palatability across 52 foods differing substantially in palatability and composition (ingredients).That is, there was not a statistically significant difference in palatability between foods categorised as 'hyperpalatable' and foods categorised as non-hyper-palatable by the HPFs metric; nor was there a significant difference in palatability between ultra-processed foods, which are claimed to be 'made to be hyperpalatable', and non-ultra processed foods, as categorised by the NOVA metric.Correspondingly, the effects sizes for these differences were small, or in sensitivity analyses small-to-medium.On the other hand, there was a significant difference (medium effect size) in palatability between HFSS and non-HFSS foods categorised according to the nutrient profiling metric.Given that the prefix 'hyper' means extreme or excessive, it would be expected that the effect sizes for HPFs versus non-HPFs, and between UPFs and non-UPFs, would be large or very large.These results show that when making claims about 'hyper-palatability' it is necessary to verify those claims by systematically measuring palatability, rather than relying on hypothetical combinations of food ingredients to predict palatability.Nonetheless, even small differences in the palatability of foods may affect food choice and/or food intake in a way that is significant for population health.We measured palatability in a virtual study, using methods which are validated by a variety of observations, as outlined in the Introduction and Limitations sections.We anticipate that our results would be replicated in a study using tasteand-spit methods to assess food palatability, though virtual methods have potential for testing more foods in larger samples of participants.Nonetheless, further studies are required to follow up our research, including in studies using taste-and-spit methods, in studies which test more representative and diverse samples of participants, and in studies which test foods representative of other cuisines and countries.

Fig. 1 .
Fig. 1.Venn diagram showing the overlaps and differences in categorisation of foods by the hyper-palatable food (HPF), ultra-processed food (UPF), and high fat, sugar, salt foods (HFSS) metrics.Of the 52 foods, a total of 35 (shown here on the left) were categorised as at least one of, HPF, UPF, or HFSS.The remining 17 foods were categorised as neither HPF, UPF, nor HFSS.
P.J.Rogers et al.