Time Trends in Income-related Differences in Food Group Intakes: The National Health and Nutrition Survey, Japan in 2010, 2014, and 2018

Background We aimed to clarify whether differences in food group intake according to household income have changed over the last decade in Japanese people aged 20 years or older. Methods This cross-sectional study was based on the 2010, 2014, and 2018 National Health and Nutrition Surveys in Japan. Food intake was assessed using a 1-day semi-weighed household dietary record. The participants were categorized into three groups based on their income. The mean of each food intake according to the income group was estimated by adjusting for age, occupation, and number of participants from the same household. The significance of the interaction terms between income and survey year was evaluated to assess the change in income-related differences in food intake over time. Results Cereal intake was lower in the middle- and the highest-income groups than in the lowest-income group, regardless of sex, and the interaction between income and year was nonsignificant for cereal intake. In the former two surveys, vegetable intake was higher among the highest-income women, while in the 2018 survey, the vegetable intake decreased in the women in the middle- and the highest-income groups. The interaction between income and year was significant for vegetable intake among the women. For other foods, the differences in intake among the income groups did not significantly change over time. Conclusion The tendency for lower cereal intake in the higher-income groups was consistent over time in both the sexes, and the tendency for higher vegetable intake in the highest income women disappeared over time.


INTRODUCTION
Several studies have reported inequalities in the risk of noncommunicable diseases and mortality according to income level. 1,2iet is considered an important contributor to the income-related health inequality. 3,4Studies conducted in Australia 5 and China 6 in the 1990s, the United States 7 and the United Kingdom 8 in the 2000s, and Japan 9,10 and Brazil 11 in the 2010s, suggested that individuals with a low income consumed a low-quality diet, 7,10 especially less consumption of fruits and vegetables. 5,6,8,9,11owever, other studies suggested inverse association between income and diet quality.For example, during 1990s, individuals with a high income in Portugal had a lower adherence to the healthy diet based on the dietary guidelines for the prevention of chronic disease. 12In addition, in 2005, individuals with a high income in Mozambique had lower frequencies of vegetable consumption. 13hese studies suggest that the association between diet quality and income might depend on the characteristics of the countries where the studies were conducted (eg, stage of economic development and degree of urbanization 14 ).
However, study results may be inconsistent within the same country, since the inequalities in the dietary intakes according to income levels can be changed overtime due to several factors.For example, economic crisis as well as consequent unemployment could negatively impact dietary intake, especially in individuals with low socioeconomic status. 15An increase in food prices has a stronger negative impact on the demand for food in low-income households. 16Moreover, previous studies have suggested that public health promotion campaigns targeting the whole population might further widen socioeconomic inequality in dietary intake [17][18][19] ; this "inequality paradox" is thought to be caused by the difference in acceptance to health promotion campaign among socioeconomic status. 20All of these factors can widen the income-related dietary inequalities.Considering these, understanding time trends in dietary intakes according to income levels is important, in order to update public health promotion strategies and narrow income-related dietary inequalities.For example, interventions for low-income households, such as food subsidy programs for fruits and vegetables, may be effective in some situations, but dietary problems in low-income households can change over time.Therefore, the time trends in income-related inequalities in the prevalence of unhealthy dietary habits should be monitored in each country.
Several countries have conducted repeated cross-sectional surveys to evaluate the temporal trends of income-related dietary inequalities [21][22][23][24][25] and showed different findings among countries.For example, an Australian study showed that, although no dietary change was observed in vegetables across income levels and survey years, people with fruit intake ≥2 servings/day significantly decreased in the highest income level but unchanged in the lowest. 24On the other hand, an American study revealed that, although there was no change in inequalities in income levels over the survey years for most food groups, including vegetables, inequalities widened for the intake of whole fruit and nuts and seeds, with significantly increased intake in the highest income level and unchanged intake in the lowest. 21However, these time trends have not been clarified in the Japanese population, despite findings from cross-sectional studies suggesting the existence of income-related differences in food intake. 9,10,26In the last decade, the National Health Promotion Movement, Health Japan 21 (second term), has been conducted aiming to reduce health disparities; however, the targets of healthy lifestyles including dietary habits (eg, reducing insufficient fruit and vegetable intake) were not set according to household income level.It is necessary to evaluate whether Health Japan 21 (second term) has been effective across all income levels.During the same period, the economic condition changed overtime in Japan.The 2008 Great Recession raised the unemployment rate (from 4% to >5%), and the Great East Japan earthquake 2011 made it stay high. 27oreover, in Japan, the consumption tax increased from 5% to 8% in 2014.Therefore, these environmental changes may worsen the income-related differences in dietary habits in Japan.
The purpose of this study was to determine whether differences in food intake according to household income level increased or decreased over the last decade.This study was based on data from Japanese adults aged 20 years or older who participated in the 2010, 2014, and 2018 National Health and Nutrition Survey Japan (NHNS).

Study participants
This cross-sectional study was based on the NHNS, which is conducted annually in November in the Japanese population aged ≥1 year and consisted of three parts: a dietary survey, a physical examination, and a lifestyle habits questionnaire.Detailed descriptions of survey procedures have been published previously. 28,29In brief, participants were recruited using a two-stage cluster sampling scheme.Households in the 300-unit blocks were randomly selected from the unit blocks of the Comprehensive Survey of Living Conditions (CSLC) for each survey year.Households with heads who were not Japanese, households that were provided with delivered/prepared meals three times a day, and one-person households residing in dormitories provided with meals were excluded from the NHNS.All participants gave verbal informed consent to the local government based on the Health Promotion Act.According to the Ethical Guidelines of Epidemiological Research, approval from the Institutional Review Board was not required.
Information on household income was collected in the 2010 NHNS for the first time, followed by in the 2011, 2014, and 2018 surveys, to obtain the baseline data of income-related differences in lifestyle factors before the Health Japan 21 (second term) started.Of them, this study utilized the household income data for 2010, 2014, and 2018 to equalize the time intervals.Although the individual-level participation rate was not available, the number of households participating in the dietary survey (participation rate) was 3,684 (68.8%) in 2010, 30 3,648 (67.2%) in 2014, 31,32 and 3,268 (64.9%) in 2018. 33,34Men and women aged 20 years or older were included in this study.Among them, pregnant and lactating women, those without information regarding their household income or occupation, or those who did not participate in the dietary survey were excluded from the analyses.

Food group intakes
A 1-day semi-weighed household dietary record was collected on a usual day, excluding Sundays and public holidays. 29Trained fieldworkers visited each household, provided a recording form, and instructed the main record keepers in the household (those who mainly prepared meals) about the survey purpose and how to complete the dietary record.The main record keepers were asked to weigh the foods and beverages and record their names and amounts, including the amount of food wasted and leftovers.When household members shared foods from the same dish, the approximate proportions of foods consumed by each member were also recorded.If weighing was difficult (eg, restaurant meals or school lunch), the main record-keepers were asked to record as much information as possible, including meal ingredients and estimated portion sizes.Information on sex, date of birth, pregnancy or lactation status, and occupation were also collected using the same recording form.Fieldworkers visited each household again, checked the completeness of the recording form, and confirmed missing information, if necessary.Detailed information of the records and data collection have been described in a previous study. 29he following food groups were used in this study: cereals, vegetables, fruits, fish and shellfish, meat, eggs, soy and other beans, milk and dairy products, snacks and confectionaries, alcoholic beverages, and tea, coffee, and other soft drinks.Foods were classified according to the food grouping methods used in the Standard Tables of Food Composition (STFC) in Japan (5 th revised and enlarged edition for 2010 survey, 35 2010 edition for 2014 survey, 36 and 2015 edition including supplementary editions 2016 and 2017 for 2018 survey 37 ).Consumption of vegetables was calculated by summing the intake of green and yellow vegetables, other vegetables, pickled vegetables, mushrooms, and seaweeds.Juice and jam were excluded from the calculation of fruit and vegetable intake, although the inclusion of these items did not change the study results (data not shown).Energy intake was calculated using the STFC in Japan.Energy-adjusted food intake (g/1,000 kcal) was calculated using the density method. 38he validity of dietary intake from this household-based dietary record was investigated in 64 Japanese volunteers (female students taking a dietetics course and their mothers). 39Dietary intakes among students estimated from the 1-day household-based dietary records kept by their mothers were compared to those estimated from the 1-day individually-based dietary records, which were independently conducted by the students.The study suggested that the household-based dietary record underestimated energy intake by 6.2% compared to individually-based dietary records, and the Pearson's correlation coefficient between the two methods was 0.90 for energy.However, the validity of the estimated food intake was not reported in this study.
Occupation was categorized into four categories considering the average income level 40 and the types of work (office work or hard labor); "professional/manager", "sales/service/clerical", "security/transportation/labor", "and non-worker", as well as previous studies. 10,41come and number of household members In the 2010 and 2014 surveys, the four options provided with regards to annual household income were: "<2 million yen," "2 to <6 million yen," "≥6 million yen," and "I don't know". 30,31In the 2018 survey, the five options provided were: "<2 million yen," "2 to <4 million yen," "4 to <6 million yen," "≥6 million yen" and "I don't know". 33Household members who answered "I don't know" were excluded from the analyses.
Information on the number of household members was not collected in the 2010 and 2014 surveys.In the 2018 survey, the person representing the household indicated the number of household members. 33Thus, the number of survey participants from the same household was used as an alternative indicator for the number of household members.In the 2018 survey, the number of household members reported by household representatives was same as survey participants from the same household in 88.5% of the participating households, and one more household member was reported in 7.3% of the participating household.
Only the household representatives were asked to provide information on household income, and the response was allocated to the other members.When several members from the same household provided information on household income, all their responses were considered invalid (ie, missing). 10This is because it was not possible to determine which household members had correctly answered the question.

Statistical analyses
The analyses were performed separately for men and women.Participants were categorized into three groups according to their income: lowest (<2 million yen), middle (2 to <6 million yen), and highest (≥6 million yen).All analyses were conducted using the SAS statistical software (version 9.4; SAS Institute Inc. Cary, NC, USA).P values <0.05 based on the two-sided test statistics were considered statistically significant.

Change in income-related differences in food group intakes over time
Considering that age group, occupation, and cohabitation with other family members might be associated with both dietary habits 10,42,43 and dietary cost, 41 all analyses were adjusted for age (ie, 20-29, 30-39, 40-49, 50-59, 60-69, and ≥70 years), occupation (ie, professional/manager, sales/service/clerical, security/ transportation/labor, and non-worker), and the number of participants from the same household (ie, 1, 2, ≥3).For each survey year, the adjusted mean of each food group intake according to income was estimated using the PROC SURVEYREG analysis with STRATA (for prefecture) and CLUSTER (for unit blocks) statements to properly account for the two-stage cluster sampling scheme used in the NHNS.Tests for differences in food groups between income groups were performed by including the categories of income as dummy variables, using the lowest income category as a reference in the regression models.Differences in food group intake among the survey years were evaluated by adding survey year as dummy variables using 2010 as a reference in the model.The significance of the interaction terms between income and survey year was evaluated to assess the change in income-related differences in food intake over time. 21,24,25Interaction terms between income (dummy variables) and survey year (dummy variables) were added into the models, including income, survey year, age, occupation, and the number of participants from the same household.The overall significance (P < 0.05) of the interaction terms was evaluated using the Wald test.
Considering previous studies suggesting that household members of the same gender share dietary habits, 44 multiple participation from the same household could result in underestimation of the standard error of food intakes according to household income levels.Therefore, we also conducted the same analyses limiting participants to a single man and single woman from the same household.

Change in income-related inequalities in food intakes over time
The adjusted mean intakes of food groups according to income and survey year are presented in Table 3 and Table 4. Regardless of sex, the cereal intake was lower in the middle-and the highestincome groups than in the lowest-income group.Compared with 2010, regardless of sex and income, a lower cereal intake was observed in 2018.For both sexes, the interaction between income and survey year was not significant for cereal intake.In the 2010 and 2014 surveys, vegetable intake was higher in women with the highest income than in women with the lowest income.In women with middle-and the highest-income, vegetable intake was lower in 2018 than in 2010.The interaction between income and survey year was significant for vegetable intake among women (P = 0.04).In the 2010 survey, egg intake was lower in the middle-and the highest-income men than in the lowestincome men.In the 2018 survey, egg intake was higher in the middle-income men than that in the lowest-income men.In men with middle and high incomes, egg intake was higher in 2018 than in 2010.The interaction between income and survey year was significant for egg intake in men (P = 0.002).Except for women in the 2010 survey, the consumption of milk and dairy products was higher in the highest-income group than in the lowest-income group.In men in the lowest-and middle-income groups, milk and dairy product intake was higher in 2018 than in 2010.The interaction between income and survey year was not significant for milk and dairy products.
For other food groups, the differences in intake among income groups did not change significantly over time.Meat intake was higher in the 2014 and 2018 surveys regardless of sex and income.In the middle-and the highest-income groups, the intakes of fruits as well as fish and shellfish were lower in the 2018 survey.These results were not changed after liming participants to one man and one woman from the same household (eTable 2 and eTable 3).

DISCUSSION
To the best of our knowledge, this is the first study to evaluate the time trends of income-related differences in food intake in Japanese adults.Despite several economic changes affected by the 2008 Great Recession, the Great East Japan earthquake 2011, and consumption tax increase in 2014, widening of income-related differences in food intakes was not observed during survey period.Lower cereal intake in the middle and highestincome groups was stable over time.Unexpectedly, in 2018, the vegetable intake decreased among women in the middle-and highest-income groups, and the tendency of a higher vegetable intake in women with the highest income disappeared over time.Egg intake was lower in men with the middle and the highest income as per the 2010 survey, but a higher egg intake was observed in men with the middle income than in those with the lowest income in 2018.The tendency for higher intake of milk and dairy products in men with the highest income was stable over time.
Our study findings that indicate a higher cereal consumption in the low-income group were supported by studies conducted in Japan, 9,45 China, 6 and Korea, 46 although non-significant or opposite relationships have been reported in studies conducted in other countries (eg, the United States, Australia, and Canada). 7,47,48Cereals, particularly refined cereals, are known for their low cost and low nutrient density at the same energy. 49apanese people with a low income have limited food budgets and mainly consume cereal and a low variety of foods. 41Therefore, increasing the cereal intake is an easy way to increase satiety without increasing dietary costs in Japan, and it might result in a low-nutrient-density diet in low-income individuals.Public support is needed to help low-income individuals consume cereals with nutrient-rich foods.
In contrast, the tendency for higher vegetable consumption in women with the highest income disappeared over time.Among women with the middle and the highest income, vegetable intake decreased in 2018.Therefore, income-related differences in vegetable intake might tend to be narrowed, not due to the increased intake among women with a low income, but rather due to the decreased intake among women with a high income.In addition, in 2018, the fruit intake decreased in the higher-income groups.Considering the inverse relationship between fruit and vegetable intake and the risk of cardiovascular disease, cancer, and mortality, 50 these dietary changes may have an unfavorable impact on the health of higher-income groups.
Previous studies revealed inconsistent results for the time trends of income-related differences in the consumption of fruits 21,24 and vegetables 51 ; this inconsistency might partly be explained by the stage of economic development and degree of urbanization.In Japan, working time is slightly increasing in the women with higher household income. 52For example, mean working minutes per day on weekdays was 219 in 2011, 234 in 2016, and 259 in 2021, among married women with household income of 7-9.99 million yen. 52Considering the possible association between longer working hours and lower vegetable intake in women, 53 changes in working hours during the study period may partly explain the decline in vegetable intake in women with a higher income.These results suggest that social support is needed to remove the barriers concerning the consumption of fruits and vegetables regardless of income level.Moreover, changes in dietary habits should be monitored even in the high-income women whose diet quality is suggested to be the highest in the Japanese population. 10lthough, men in the middle-and highest-income groups had low egg intake compared to those with the lowest income in 2010, the opposite tendency of egg intake according to income levels was observed in 2018.Egg intake increased in men in the middle-and high-income groups during the survey period.b Participants were categorized into three groups according to income: lowest (<2 million yen), middle (2 to <6 million yen), and highest (≥6 million yen).c Survey year as dummy variables using 2010 as a reference were included in the model with adjustment for above mentioned variables.
d P-values for the income (dummy variables) × survey year (dummy variables) interaction terms were calculated with adjustment for the above-mentioned variables; a significant interaction meant that the income-related differences in food intake changed over time.e Income as dummy variables using the lowest group as a reference were included in the model with adjustment for above mentioned variables.
Trends of Food Intake in Income Groups c Survey year as dummy variables using 2010 survey as reference were included in the model with adjustment for above mentioned variables.
d P-values for the income (dummy variables) × survey year (dummy variables) interaction terms were calculated with adjustment for the above-mentioned variables; a significant interaction meant that the income-related differences in food intake changed over time.e Income as dummy variables using the lowest group as a reference were included in the model with adjustment for above mentioned variables.Although we do not know the reason of such findings, one assumption would be the sales price changes over these years.
The average price of one pack of eggs (usually ten eggs per pack) in major cities in Japan increased slightly during the study period, comprising 202 yen in 2010, 227 yen in 2014, and 221 yen in 2018. 54This may explain the reason that low-income group did not increase egg consumption given they are more price-sensitive.Since the income-related difference in egg intake was not supported by previous studies, 6,46,55 future studies are needed to confirm our study findings.Moreover, considering that the difference in egg intake among the income categories was around 3 g/1,000 kcal in this study, the probable impact of incomerelated differences in egg intake might have a small impact on health outcomes in the Japanese population.The positive relationship between income and milk and dairy products is supported by previous study findings. 6,11,46,48,56Japanese dairy consumers are more likely to have adequate calcium intake and inadequate saturated fats consumption. 57Further studies using detailed food grouping methods are needed as the relationship between income and low-fat milk intake differs from that of fullfat milk intake. 47his study had several limitations.First, only 65-70% of the sampled households took part in the survey.During the observational period, the household-level participation rate and proportion of low-income individuals slightly decreased.If lowincome individuals with higher or lower dietary intake were less likely to participate in the survey or provide income information, the relationship between the income and food intake might be biased.Second, in the NHNS, information on household income is obtained using only three or four choices.Therefore, equivalized household income cannot be calculated to reflect the difference in household size.Moreover, the scientific basis of the cutoff-values of the income (<2 million yen, 2 to <6 million yen, ≥6 million yen) used in the question is not explained in the NHNS.Additionally, a precise definition of "household income," such as whether to include income sources other than salary income, was not provided in the questionnaire.Collecting information on household income with more detailed choices and clear definitions in the NHNS is considered useful for planning population and high-risk approach interventions in Japan.Third, self-reported dietary assessments are subject to systematic and random errors.Food intake estimated from a 1-day dietary record does not likely represent the habitual food intake of the study participants.More importantly, the ability of household-based dietary records to estimate the energy and nutrient intake has been validated among female university students alone 39 and not in women of other age groups or men.In addition, this study failed to assess the validity of the estimated food intake. 39Furthermore, previous studies have reported that under-reporting of energy intake was more prevalent in individuals with lower income than in those with higher income. 58,59Although food intake was adjusted for energy intake, the difference in the reporting accuracy of dietary intake between income levels may have affected our findings.Fourth, the STFC used for food coding was different among the three surveys, which might have affected the estimation of food groups and energy intake.However, the food grouping methods and cooking status of the calculated food intake did not dramatically change after the 2001 survey.Although no study has compared energy intakes estimated by the 5th and 2010 editions, it was suggested that the percentage difference in energy intake estimated by the 2010 and 2015 editions was small (1%). 60Fifth, although education is strong determinant of future employment and income, 61 the models were not adjusted for education since the information has not been collected in the NHNS.Considering that education might have stronger associations with food intakes than income 62 further studies are needed to determine how to monitor socioeconomic differences in dietary habits in Japanese.Sixth, given economic crisis, as well as related unemployment, could negatively impact dietary intake especially in individuals with low-socioeconomic status, 15 a longer study period before the 2008 Great Recession might clarify the income-related differences in dietary habits of the Japanese population.Future studies using longer period data are needed.
In conclusion, despite economic challenges that could affect the most vulnerable people, widening of income-related differences in food intakes was not observed during the survey period.The tendency for lower cereal intake in the higher-income groups was stable over time in both sexes.Unexpectedly, vegetable intake decreased in women in the middle-and high-income groups in 2018, and the tendency of higher vegetable intake in women with the highest income disappeared over time.Further studies using data over a longer period with a detailed questionnaire on income are needed to determine change of the relationship between income and food intake over time.

•••Figure 1 .
Figure 1.Number of participants included in the present analysis.Part of the excluded participants had missing values for ;2 variables.
Trends of Food Intake in Income Groups who did not meet the inclusion criteria, 6,222 adults in 2010, 5,967 adults in 2014, and 5,037 adults in 2018 were included in the analyses.Regardless of sex and survey year, the prevalence of participants in each age category, occupation, and the number of participants from the same household were significantly different among the income groups (Table1 and Table 2).

Table 2 .
Characteristics of Japanese women according to income a : the 2010, 2014, and 2018 National Health and Nutrition Survey (NHNS), Japan a Participants were categorized into three groups according to income: lowest (<2 million yen), middle (2 to <6 million yen), and highest (≥6 million yen).b Pearson's chi-squared test.c Participants who responded as "smoking every day" or "occasionally smoking".d Defined according to the definition used in the NHNS report (33): participants who responded to the question on drinking frequencies and quantities as a Participants were categorized into three groups according to income: lowest (<2 million yen), middle (2 to <6 million yen), and highest (≥6 million yen).b Pearson's chi-squared test.c Participants who responded as "smoking every day" or "occasionally smoking".d Defined according to the definition used in the NHNS report (33): participants who responded to questions on drinking frequencies and quantities as J Epidemiol 2024;34(2):76-86 j 81