Trends of Average Intake of Macronutrients in 47 Prefectures of Japan from 1975 to 1994-Possible Factors that May Bias the Trend Data -

To describe the geographical patterns and trends of macronutrient intake in Japan for the recent 20 years, we analyzed the data sets of the National Nutrition Surveys (J-NNS). First, we calculated regression coefficients for survey year by prefecture for energy, protein, fat, and carbohydrate intake. These results, however, could be affected by the changes in age and sex distributions of the survey samples. Secondly, as the food consumption data were based on household as a unit, we used proportions of the subjects who belong to age-specific groups and female subjects in a sub-sample for each prefecture in a given year for adjustment by general lineal model. As a result, 1 % increase in the subjects aged 1-4, 10-19 and 65years and female subjects was equivalent to the changes in average energy intake by -5.88, +2.27, -1.45 and -1.62 kcal, respectively. After the adjustment for age and sex, number of significant negative coefficients among 47 prefectures decreased for energy and carbohydrate intake, and that of positive coefficients increased for fat intake. This suggested that unadjusted trend data might lead to an overestimation of decreasing trends of energy and carbohydrate intake , and an underestimation of increasing trend of fat intake in the recent 20 years. J Epidemiol, 1998 ; 8 : 160-167.


INTRODUCTION
The Japan National Nutrition Survey (J-NNS) is an annual survey of food intake conducted by the Ministry of Health and Welfare v.The databases in whole Japan and by prefecture are available for the years from 1975 through 1994, although the first survey was carried out in 1946.There have been several reports in which secular trends in the average intakes of energy and nutrients are exhibited graphically and discussed on the basis of visual sensation 1, Z).The main purpose of the present study is to describe the trends of average intake of energy and macromutrients in 47 prefectures by using a general linear model.
In the survey, food intake had been measured by a three-day food record method for household units till 1994.Therefore, not the age-and sex-specific food intake, but only the per-capita-per-day data were available except for the data from one person households 3), which may not represent the all subjects from various types of household.An increasing trend of elderly population and a decreasing trend of children in Japan m are thought to influence the figures for average foods or nutrients intake expressed as unit per capita per day.The second aim of this study is to test a hypothesis that changes in age and sex distributions in the survey samples may bias the diet trend data.

National Nutrition Survey (J-NNS)
Every year 300 census enumeration districts that include around 30 households 5) are randomly selected from a listing derived from a survey named "Comprehensive Survey of Living Condition of the People on Health and Welfare" 10 .The numbers of the households that participated J-NNS were around 5500-6500 in each year 7), therefore, the response rate that is not published is estimated to be 60-70% and they appear representative of the entire Japanese households.
Before the survey, a staff dietitian from the public health center demonstrates the measures to determine food quantities and teaches the survey methods and procedures to a housewife or another person who cooks usually in the family.During the survey period, the dietitian makes a call at the subject's home at least once a day and reviews the dietary record 1).
The at-home foods and beverages consumed by a family are weighed, meal by meal, for three consecutive days in November excluding Saturday, Sunday, national holiday, and ceremonial and ritual days.The types and amount of foods that were taken outside home are also described by individual.The data are recorded in a form that is appropriate to the way in which they are coded.Then, the nutrient composition of the diet is calculated from the Japanese Standard Table of Food Composition (revised in 1982 and 1990 1)) at the Statistics and Information Department, Ministry of Health and Welfare.Finally, dietary intake per capita per day is obtained by dividing the whole amount consumed in a household by the number of family members aged one year and over.

Statistical analyses
We analyzed the secular trends in the intakes of energy, protein, fat and carbohydrate in whole Japan and by prefecture during the period of 20 years from 1975 through 1994 by general linear models with the SAS REG procedure 9).We computed regression coefficients of the intake of energy or a nutrient to survey year (hereinafter referred to as "crude regression coefficient" or "crude trend").The survey year was coded as 0, 1, 2, ... and 19 for 1975, 1976, 1977... and 1994.In the J-NNS, as mentioned above, the total amount of foods and beverages taken by a family is divided by the number of family members aged 1 year and over to get the per-capita-perday intake of energy or a nutrient.However, the intake in females appears to be lower than that in males of similar age.Furthermore, the amount of foods taken varies between age groups a 10) When we try to analyze the secular trends in the intake of energy or a nutrient by using the J-NNS data, we must consider the effect of age as well as sex.That is particularly because the proportion of the population aged 65 years and over to the total population increased markedly during the observation period '.A considerable range of variation of population aging among prefectures is also present).
We tried to examine the effect of sex, age and prefectural difference of population aging on the secular trend in the intake of energy or a nutrient.First, we adopted a cross-sectional approach to observe inter-prefecture variations in terms of percentages of the survey subjects in the samples and average nutrient intake.Pearson correlation coefficients were calculated by use of these data from 47 prefectures for four periods; 1975-1979, 1980-1984, 1985-1989 and 1990-1994, because the annual sample size is relatively small in several prefectures 1u.
Secondly, we used the following statistical model.We did this analysis by use of the SAS GLM procedure with a classification variable of prefecture to have 47 blocks for one-way ANOVA.
Thereafter, we calculated age-and sex-adjusted regression coefficients of the intake of energy or a nutrient for survey year (= * 6) (hereinafter referred to as "adjusted regression coefficient" or "adjusted trend") by prefecture.Yi = *1X1i + * 2X2i + * 3X3i, + * 4X4i, + * 5X5i, + *6X6i + intercept + * i=0 , 1, 2,..., 19 (survey year) Yi : average intake of energy, protein, fat or carbohydrate X1i :percentage of subjects aged 1-4, X2i :percentage of subjects aged 5-9, X3i : percentage of subjects aged 10-19, X4i : percentage of subjects aged 65 and over, X5i :percentage of female subjects X6i : code for survey year (--0, 1, 2,..., 19) (3 1-6 : partial regression coefficients for X,-X6 The simple regression coefficients and these age-and sexadjusted ones in the multiple regression models for survey year were compared to test the effect of the adjustment for age and sex on the trend data of each prefecture. Additionally, regression coefficients from 1975 to 1985 (early 10 years) and from 1984 to 1994 (late 10 years), and before and after adjustment were separately calculated in terms of total fat intake to show the effect of the adjustment for age and sex distributions in the survey samples.For the comparison with the simple average figures for total fat intake, the number of the survey subjects in each prefecture in a given year was used as a variable for weighted procedure in the general linear model.

RESULTS
The crude regression coefficients and standard error estimated in the simple regression models that were adopted to each prefecture, and adjusted R 2 for the models were shown in appendix 1.
To examine the effect of sex and age on the average dietary intake, we performed the simple correlation analysis between the intake of energy or macronutrients and age or sex.The results are presented in Table 1.The proportions of the elderly subjects, i.e. the percentages of the subjects aged 65 years and over showed mostly negative correlations to the intake of fat, and positive to the intake of carbohydrate.The adverse relationships were observed in the proportions of the subjects aged 5-9 years.
The results of the multiple regression models to test the impact of the changes in age and sex distribution in survey samples to the dietary trend data were shown in Table 2. Partial regression coefficients of total energy intake for percentage of subjects aged 1-4, 5-9, 10-19 and 65-years and female subjects were -5.88, -1.72, +2.27, -1.45 and -1.62, respectively.These results mean, for example, that 1% increase in the subjects aged 1-4, 10-19 and 65-years and female subjects could change average total energy intake by -5.88, +2.27, -1.45 and -1.62 kcal, respectively, which could support our initial assumption.The similar results were also observed for protein, and 1 % increase in the elderly subjects would correspond to -0.298 g and +0.481 g changes in average intake of fat and carbohydrate, respectively.
To adjust these suggestive effects of age and sex distributions of the samples on the secular trend data, partial regression coefficients of survey year were calculated by use of percentage of subjects aged 1-4, 5-9, 10-19 and 65-years and female subjects in the survey samples (appendix 2).Adjusted R 2 values in appendix 2 were much increased from the corresponding values in appendix 1, which suggest that adding these variables to the general linear models would improve the fitness of the models.Regression coefficients and standard errors were also slightly changed in many prefectures.
Table 3 shows the number of positive or negative regression coefficients among 47 prefectures before and after adjustment for age and sex.After the adjustment, number of statistically significant negative coefficients decreased for total energy and carbohydrate intake, while number of statistically significant positive coefficients increased for total fat intake.*: adjustement for age and sex distributions in a given sample Fig. 1 An estimated trend of total fat intake in two periods (1975)(1976)(1977)(1978)(1979)(1980)(1981)(1982)(1983)(1984)(1985)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)) by use of the general linear models to adjust for age and sex distributions in survey samples of the National Nutrition Survey, Japan.
Finally, we illustrated the differences in crude and adjusted regression coefficients for total fat intake in early (1975-85) and late (1984-1994) 10 years.The coefficients for survey year were 0.627 (S.E.= 0.044) and 0.611 (0.048) before and after adjustment, respectively, in the early 10 years, and 0.105 (0.036) and 0.209 (0.042) in the late 10 years.By use of these coefficients for one year's change, an estimated trend of the total fat intake was shown in Figure 1.Not so steep but still increasing tendency of the total fat in the Japanese population was observed.

DISCUSSION
After World War II, Japanese lifestyles, particularly the diet, moved towards westernization in concert with the socioeconomic change a.The J-NNS is the only one nation-wide source of dietary intake.The Survey was initiated in 1946 and the household-based method was employed under the direction of the USA that occupied Japan at that time u.To maintain the comparability of dietary intakes between survey years, the same method had been used until it was changed from the household-based to individual-based one in 1995 ').To obtain the intake per person, the total amount of foods consumed in a family are divided by the number of family members without taking into account their age and sex.This calculation would bias the intake per person.In Japan, an increase in the proportion of the elderly 65 years and over has rapidly occurred.This is owing to not only the rapidity of birth rate decline and the continuity of very low fertility level, but also a decline in overall mortality and consequently the prolongation of life expectancy 4).Thus, we assumed that the annual changes in age and sex distribution of the subjects in the J-NNS would affect the simply computed amounts of energy and nutrients taken per capita.
The findings, as shown in tables 1 and 2, would support the assumption.In several reports in which secular trends in the average intakes of energy and nutrients were exhibited graphically 1, 2), they described that the intakes of fat, protein and energy increased and the intake of carbohydrate decreased during the period of high economic growth from 1960 to 1973.The change in fat consumption was the most significant among the major nutrients.Because the change was much more marked in the high economic growth period than in any other period, the intake of fat was reported to stop increasing with the oil crisis in 1973 as a turning point and to remain almost unchanged since then 2).However, our statistical analysis revealed that the intake of fat tended to increase during the period after the oil crisis.Furthermore, comparing the crude with adjusted partial regression coefficients, the crude coefficients tended to overestimate decreasing trends in energy and carbohydrate intake and to underestimate increasing trends in fat intake.
Another important dimension of population aging is disparity among prefectures 4).As expected, the prefectures within metropolitan regions, including Tokyo, Chiba, Kanagawa, etc, showed lower percentages of the aged subjects in the survey samples, whereas those situated remote from the Tokyo Metropolitan Area, such as Aomori, Fukui, Tottori, Tokushima, Nagasaki, etc. indicate higher percentages (data not shown).The adjusted regression coefficients of fat intake tended to be larger than the crude ones in the prefectures with high scores of the percentage of the aged subjects, although not in all the prefectures (Appendices I and 2), also suggesting that the age effect on fat intake was significant.
The annual reports of J-NNS have shown the dietary data for 12 regional blocks 1.4), but not for 47 prefectures.This is mainly because the sampling method is not designed to have a representative sample for each prefecture and partly because the numbers of households surveyed in some prefectures are very small for statistical analyses 11).For example, in the 1994 survey, the smallest survey sample among 47 prefectures was from Tottori Prefecture (n=17), while the largest was from Tokyo Metropolitan (n=433) [unpublished data].In this case, only one census enumeration district in Tottori Prefecture was selected for the survey and this sample is likely to be biased from the all households in the prefecture.However, the sampling of the survey districts is independently done every year, therefore, the characteristics of the samples may be randomly deviated from those of the population in a given prefecture.Under this assumption, we adopted general linear models for the dietary data by prefecture.Although the issues regarding representativeness of the survey samples for each prefecture is a major weakness of this study, we believe that it is valuable to describe the changes in energy and macronutrients intake by prefecture using the data of J-NNS for further epidemilogical analyses, because the food consumption data from prefectural nutrition surveys are limited 11) especially for trend analyses.

Table 1 .
Correlation coefficients between mean energy and nutrients intake and percentage of the subjects in specific age groups in 47 prefectures.

Table 2 .
Number of positive or negative regression coefficients among 47 prefectures and regression coefficients for the whole model.