Development of Substituted Fatty Acid Food Composition Table for the Use in Nutritional Epidemiologic Studies for Japanese Populations : Its Methodological Backgrounds and the Evaluation

Results of dietary assessment are influenced by quality of food composition tables used for nutrient calculation. The Japanese food composition table has considerable missing values for fatty acid compositions. Substitution is often used for filling missing values. We examined reliability of the following 4 major substitution methods using available values of arbitrarily selected 83 sets of foods from the published fatty acid composition table of Japanese foods: by a different part of the same specie, by a similar specie, by a same specie in the United States' Department of Agriculture food composition table, and by recipe. The mean correlation coefficients of food pairs were 0.97, 0.96, 0.84, and 0.80 respectively. Next, we substituted fatty acid compositions for the 794 missing foods using the 4 substitution methods, and developed the table with 1245 foods including those listed in the original (non-substituted) fatty acid composition table. Lastly, we calculated fatty acid intake levels with the original (non-substituted) and the developed (substituted) tables using 28or 14-day dietary records of 211 men and women as a sample data, and compared the results. The intakes of all five fatty acid groups increased. The increase was most marked in saturated fatty acids (26% in men and 31 % in women in crude values). As a consequence, polyunsaturated to saturated fatty acid ratio decreased from 1.15 to 1.01 in men and from 1.13 to 0.96 in women. The use of the developed fatty acid food composition table may increase the reliability on nutrition-disease association in future nutritional epidemiologic studies for Japanese populations. J Epidemiol, 1999 ; 9 : 190-207

Substitution is often used for filling missing values.We examined reliability of the following 4 major substitution methods using available values of arbitrarily selected 83 sets of foods from the published fatty acid composition table of Japanese foods: by a different part of the same specie, by a similar specie, by a same specie in the United States' Department of Agriculture food composition table, and by recipe.The mean correlation coefficients of food pairs were 0.97, 0.96, 0.84, and 0.80 respectively.Next, we substituted fatty acid compositions for the 794 missing foods using the 4 substitution methods, and developed the table with 1245 foods including those listed in the original (non-substituted) fatty acid composition table.Lastly, we calculated fatty acid intake levels with the original (non-substituted) and the developed (substituted) tables using 28-or 14-day dietary records of 211 men and women as a sample data, and compared the results.The intakes of all five fatty acid groups increased.The increase was most marked in saturated fatty acids (26% in men and 31 % in women in crude values).As a consequence, polyunsaturated to saturated fatty acid ratio decreased from 1.15 to 1.01 in men and from 1.13 to 0.96 in women.The use of the developed fatty acid food composition table may increase the reliability on nutrition-disease association in future nutritional epidemiologic studies for Japanese populations.J Epidemiol, 1999 ; 9 : 190-207 dietary assessment, epidemiology, nutrition, fat, fatty acids Results of dietary assessments are influenced by quality of food composition tables used for nutrient calculation 1).The role of fatty acids has recently been of great interest in cancer and cardiovascular disease 2-4).The Japanese food composition table has only 471 foods (29%) for fatty acid compositions out of 1621 food items listed in the standard food composition table of Japanese foods 5,6).It makes difficult to estimate fatty acid intake levels and to perform the related studies in nutritional epidemiologic studies for Japanese populations.
Substitution is often used for foods without the data.For example, when one nutrient of food A is missing, the value is substituted by those of food B if the values is available for food B and the two foods are nutritionally comparable.But the theo-retical background is not fully understood for the substitution of foods for missing values on fatty acid food compositions.To our knowledge, no food composition table for fatty acids with substitution is available for Japanese foods.
Firstly, we evaluated 4 major substitution methods using data available in the published fatty acid food composition table for Japanese foods.Secondly, we substituted missing values of the fatty acid food composition table by the substitution methods, and developed a fatty acid food composition table for the use in nutritional epidemiologic studies.Lastly, we applied the developed (substituted) food composition table for the calculation of fatty acid intakes of a sample population, and compared the results with those calculated with the original (non-substituted) composition table 5).

Methods of the evaluation
The following 4 major substitution methods were evaluated by the comparison of two foods as a set whose fatty acid compositions were available in the published fatty acid composition table of Japanese foods 5).Method A: comparison between different food products within a specie.Comparisons between different parts of one specie of animal, between cultivated one and the wild counterpart, between products harvested in different countries, and between a processed food and its raw material were included in this method.Method B: comparison between similar species.Method C: comparison between a food in the Japanese table and the same food in the United States7) Department of Agriculture (USDA) food composition tables ~.Method D: comparison between values of a food and of the estimated recipe for highly-processed or cooked foods.Method A was examined mainly by meats, fishes shellfishes, cereals, pulses, and milks.When processed foods were compared with their raw materials, foods with relatively simple food-processing such as drying or salting were chosen.Especially foods with food-processing without adding fat were evaluated in this method.Method B was examined mainly by leafy vegetables, fruits, cereals, meats, and fishes.Method C was examined mainly by cereals, vegetables, fruits, and cereal products, confectioneries, and seasonings.Method D was examined by confectioneries, cereal products, and seasonings.The ingredients listed in the footnotes of the standard food composition table were considered for preparing the recipes 6).Cooking books commonly available in Japanese markets were also referred.Three dietitians prepared all the recipes used in this study under the supervision of one dietitian (MK).In order to ensure the quality of food composition of the recipes, composition of five nutrients such as water, fat, protein, carbohydrate, and sodium, were adjusted within 5% of listed values in the standard food composition table 6).

Foods used in the evaluation
We arbitrarily selected 83 sets of foods consisted of 102 foods of which fatty acid compositions were available in the food composition table of Japanese foods for fatty acids 5).The selection was based on the foods appeared in the data of 28-or 14-day dietary records surveyed in 1994 or 1995 for 221 persons aged 44-63 years, who were a sub-sample of Japan Public Health Center-based prospective Study (JPHC Study) living in Iwate, Akita, Nagano, and Okinawa prefectures 8).At least one food of a pair was selected from those foods as much as possible.In order to cover wide variety of foods and cooking/processing methods within each food group, some foods were included even if the contribution was relatively small in the sample data.Sugar and sweeteners, mushrooms (fungi), and seaweeds (algae) were not included because fat contents are generally quite low, i.e., less than 1.0% in weight, and their contribution to the fat intake was almost negligible in the above data.

Statistical analysis
Fatty acid contents (g/1000g total fat) was compared between a set of two foods.A fatty acid included only in one food was excluded from the analysis.The Pearson product moment correlations (r) were calculated.Because the contents of some fatty acids were extremely high compared to other fatty acids in each food, the distribution of the points for fatty acids in a scattergram were positively skewed in most of foods examined.They were log (natural) transformed before analysis to achieve normal distribution.Because some values were less than 1.0, we added 1.0 to all values before log-transformation.Our purpose was to quantify the correlations between two foods rather than test hypothesis, therefore p-values were not presented.

Development (substitution) of a fatty acid food composition table
According to the results for the evaluation for substitution methods, we substituted fatty acid compositions for foods without the data.For a food without fatty acid composition, one reference food with the fatty acid composition or recipe was selected for substitution.Because the mean correlation coefficient by sample food pairs was the highest in Method A, and Methods B, C, and D followed in this order (see Result section more in detail), we gave a priority of substitution in this order, i.e., first we sought a substitution food by Method A, if a suitable substitution food was not found, we next sought a substitution food by Method B, and so on.For foods, mostly confectioneries, with no alternative method other than recipe, we collected as much information as possible before the development of the table was started.After several recipes were collected, we excluded them for home-cooking and for special delicacies, and selected them for ordinary market-use.
When one reference food was selected, the fatty acid composition was considered as those of a target food after adjustment for total fat content.Thirty-six specific types of fatty acids, besides groups of fatty acids such as saturated fatty acids (SFA), monounsaturated fatty acids (MUFA), and polyunsaturated fatty acids (PUFA), were included in the table .
Considering the contribution to fat intakes, we selected foods for substitution in the following two ways.First , we selected all foods with total fat >_ 1.0% in weight.Second, we selected food with total fat > * (trace) and which appeared at least once in the dietary record of the sample population .Among them, no reliable substitution method was found for 25 foods , and the values remained missing (see Result section more in detail).Prepared foods categorized to, so called, "food group 18" in the standard food composition table were not included in the substitution in this study 6).

Application of the developed composition table
We calculated fatty acid intake levels using the data of the sample population used in the evaluation of substitution methods and in the development of the fatty acid composition table.The fatty acid intake levels calculated using the original (nonsubstituted) and the developed (substituted) food composition tables were compared at a group level.Because the developed fatty acid food composition table did not have values for prepared foods (food group 18), these foods were not used for the data preparation from the recorded data.In the both analyses, the 5 foods were replaced as follows for the calculations because more than one foods were matched in the fatty acid food composition table: vegetable oil, mixed [5-lg] for vegetable oil , margarine, soft type [5-7a] for margarine [5][6][7], peanuts, Virginia type, dried [6-25a1] for peanuts, dried [6-25a], red sea bream, wild, raw [8-110al] for red sea bream, raw [8-110a], and tiger prawn, wild, raw [8-219a1] for tiger prawn, raw [8-219a].The SFA, MUFA, PUFA, n-3 PUFA, and n-6 PUFA, were included in this analysis.Their ratios such as PUFA to SFA (P/S ratio) and n-6 PUFA to n-3 PUFA (n-6/n-3 ratio) were also included.They were also examined by food group.

Evaluation of substitution methods
Table 1 shows the Pearson correlations between 39 food sets paired by Method A. In this analysis, the reliability was evaluated for food sets with different products from one specie for cereals, pulses, meats, fishes, and milks, and those with different parts from one specie for meats.The level of correlation was generally high (mean=0.97)except for two sets (r=0.72 and 0.86 in fishes).Table 2 shows the Pearson correlations between 11 food sets paired by Method B (similar species).The level of correlation was also generally high (mean=0.96).
Although a wide variety of food sets were tried to select in both analyses, it was not necessarily accomplished because of a limited number of foods covered in the fatty acid food composition table.Table 3 shows the correlations between food sets paired by Methods C and/or D. The number of fatty acids included in the analysis was in general fewer than those in Table 1.The correlation coefficients for foods between two, the Japanese and the USDA, food composition tables was 0.80 except 3 foods in confectioneries, 2 in vegetables, 1 in fruits, and 1 in seasonings.The correlation coefficients between foods and their estimated recipes ranged between 0.50 and 1.00.The correlation coefficients was in general 0.90 for con-fectioneries in both methods of comparison.The mean correlation coefficients of Methods C and D were 0.84 (n=32) and 0.80 (n=14) respectively.Two examples, one for a good correlation and the other for a poor correlation, are shown in Figure 1.

Development of a fatty acid composition table
Table 4 shows the number of foods in the standard food composition table of Japanese foods by food group, listing in the fatty acid food composition table, appearance in the dietary record in a sample population, and total fat contents of foods.Among 1605 foods listed in the standard food composition table, 1103 foods appeared in the dietary record.Among them, the values of fatty acid were missing in 652 foods (59% of the appeared foods).The 521 foods without values of fatty acids had 1.0% fat in weight.Among 510 foods with fat more than trace (0) and less than >_ 1.0% in weight, 298 foods appeared in the dietary record.As a result, 819 foods were selected for substitution of fatty acid compositions.
The food compositions of 5 fatty acid groups, i.e., SFA, MUFA, PUFA, n-3 PUFA, and n-6 PUFA, for the 794 substi-Table 1. Correlations for fatty acid compositions (g/1000 g total fat) between selected two foods by method A (between different food p rodncts within a specie): All values were loge-transformed before analysis.Table 2. Correlations for fatty acid compositions (g/1000 g total fat) between selected two foods by method B (between similar species): All values were loge-transformed before analysis.
1 Food code used in the food composition tables of Japanese foods, 4th edition (ref.6).
tuted foods are listed in Appendix.The data are available, but for non-profit use only, in the R-TMRNET homepage of the authors' affiliation (http://www.east.ncc.gojp/epi) without permission.The authors however request that this article is cited when a study in which the data, even the part, was used is published or open to the public.

Application of the developed composition table
Table 6 shows the means (standard deviations) of fatty acid intakes of the sample populations.Mean total fat intake was 59.1 and 52.9g/day and 23.1 and 26.4%E (percentage of total energy) in men and women respectively.The coverage rate of total fatty acids to total fat increased from 73 to 88% and from 72 to 88% in men and women respectively.The increase was most marked in SFA (26% in men and 31% in women in crude values) compared to other fatty acids.As a consequence, P/S ratio decreased from 1.15 to 1.01 in men and from 1.13 to 0.96 in women.The n-6/n-3 ratio was stable.Table 7 shows mean fatty acid intakes by food group in men and women combined.Confectioneries and meats were two major contributors to the increase in all fatty acids examined except for n-3 PUFA.In lesser extent, milks were a major contributor to the increase in SFA, and fishes for PUFA, mostly for n-3 PUFA.Cereals con-tributed, although not so marked, to the increase in all fatty acids.

DISCUSSION
Substitution is often used for completion of food composition tables for missing values 9).Several study groups have been developing food composition tables with this method for their own use 10).But, to our knowledge, neither scientific nor systematic evaluation for substitution methods has been reported so far for Japanese food composition tables.It seems to be a problem because results of dietary surveys are influenced by quality of food composition tables used for nutrient calculations 1).
The contribution of specific fatty acids to the etiology of cancer and cardiovascular disease is recently of great interest 2-4) Food composition tables are however far from optimal for fatty acid compositions of Japanese foods at the present time because of the considerable missing values.Some other nutrients have same problems.Some study groups have developed food composition tables for the use in their own epidemiologic studies 10-12) Some studies measured compositions of nutrients of interest 13).Some others tried to compile indepen-Table 3. Correlations for fatty acid compositions (g/1000 g total fat) between selected foods and the same foods selected from the USDA food composition tables and/or from recipes All values were loge-transformed before analysis.
Figure 1.Correlations for fatty acid compositions (g/1000g total fat) between two similar foods.All values were log, transformed before analysis.Each point represents each fatty acid.Some of the names were specified.
(B) Example of poor correlation.
Between tomato in the original table  and in the USDA table.
dently measured food compositions 14).Recipes have also been used for the substitution 9).The correlations examined in Tables 1 and 2 were quite reliable for meats, fishes, and other raw foods with some exceptions.Even though for cooked and processed foods, it seemed to be acceptable if fat or fat containing food was not added during the food processing or cooking.On the other hand, the substitution seemed less reliable for the processed foods with more than one material were used (Table 3).The example was white bread.Although inconclusive, values obtained from the USDA food composition tables may be acceptable when no alternative method was available within Japanese food composition tables.Some recipes showed poor correlations with their original foods such as white bread and bread type rolls (soft rolls).In these foods, a type of fat used in the original food was unclear.Therefore we estimated the type of fat from the footnote in the standard food composition table 6).However, because the quantity was not available in it, we estimated the quantity from the contents of other nutrients such as energy, water, protein, carbohydrate, and sodium.It apparently has limitations for the estimation on what kind of materials and how much of them are used.The results suggested that three substitution methods, i.e., Methods A, B, and C, examined in this study could be used for the substitution for most of foods.Note: The 819 foods in the shadowed area were selected for substitution of fatty acid composition in this study.1 The 5 foods were replaced as follows because more than one foods were matched in the fatty acid food composition table: vegetable oil, mixed [5-1g] for vegetable oil , margarine, soft type [5-7a] for margarine [5][6][7], peanuts, Virginia type , dried [6-25a11 for peanuts, dried [6-25a], red sea bream, wild, raw [8-110al] for red sea bream, raw [8-110a], tiger prawn, wild, raw [8-219a1] for tiger prawn, raw [8-219a].
2 The following 4 foods with data in the fatty acid food composition table only were not considered in this analysis: beef separable fat , kidney leef [9-23i], daikon, Japanese radish, sprouts [12-56-2], Chinese chive, blanched [12-93-2], and garlic , flower stalk [12-95-2].3 Data were based on the 28or 14-day dietary records in 4 areas of Japan (221 middle-aged men and women) who were a subsample of the JPHC study.See text in detail.
However, the reliability of Method D, i.e., substitution by recipe, was not satisfactory enough.
At the second part of this study, we developed a substituted fatty acid food composition table according to the results for the evaluation of the substitution methods.However, we did not use fully systematic and validated method for the selection of the substitutes.It might bias the results.Further studies are necessary for the development of more systematic method of food groupings for a selection of substituted foods for this purpose.
At the last part of this study, we applied the developed table using sample dietary data obtained from 4 areas in Japan.The coverage rate of total fatty acids to total fat increased from 73 to 88% in men and 72 to 88% in women.Among recent dietary surveys conducted in Japan, the coverage rate varied from 73% in men and 77% in women in Takayama 15) and 82% in women in four areas 16) to 89-92% in women in Okayama 17).In this study the P/S ratio also changed considerably from 1 .15 to 1.01 in men and from 1.13 to 0.96 in women by the substitution.Although the comparison is difficult because of the different population characteristics and methodologies between the previous and the present studies, the difference may partly be attributable to a different food composition tables used in each study.Sasaki et al. reported that a difference between P/S ratio aggregated from several dietary surveys and the estimated one from a food disappearance data was more marked in Japan than in most of Western countries examined 18) , and proposed that it would be attributable to a possible underestimation of SFA intake in dietary surveys 19) .Our results showed that underestimation occurred especially in SFA compared to the other fatty acids.Table 5.The number (percentage in parenthesis) of foods by substitution method of fatty acid composition and food group.
In conclusion, we developed a fatty acid food composition table by 4 major substitution methods for the use in nutritional epidemiologic studies for Japanese populations.The use of the developed fatty acid food composition table may be useful to estimate fatty acid intake levels for Japanese populations and to examine nutrition-disease association in future nutritional epidemiologic studies.But the validity is inconclusive.Further validation, for example using biomarkers, is necessary for the table developed in this study 22,29).Abbreviations: SD=standard deviation, %E=percentage of total energy, SFA=saturated fatty acids, MUFA=monounsaturated fatty acids, PUFA= polyunsaturated fatty acids. 1 The 5 foods were replaced as follows because more than one foods were matched in the fatty acid food composition table: vegetable oil, mixed [5-1g] for vegetable oil , margarine, soft type [5-7a] for margarine [5][6][7], peanuts, Virginia type, dried [6-25a1] for peanuts, dried [6-25a], red sea bream, wild, raw [8-110al] for red sea bream, raw [8-1l0a], and tiger prawn, wild, raw [8-219a1] for tiger prawn, raw [8-219a].
Results of dietary assessmentare influenced by quality of food composition tables used for nutrient calculation.The Japanese food composition table has considerable missing values for fatty acid compositions.

Table 4 .
The number of foods in the standard food composition tables of Japanese foods by food group, listing in the fatty acid food composition table.annearance in the dietary record in a samnle nonulation, and total fat contents of foods.

Table 6 .
Means (standard deviations) of fatty acid intakes calculated with the original and the developed fatty acid food composition tables: Data were 28-or 14-day dietary records of 211 residents aged 44-63 years in Iwate, Akita, Nagano, and Okinawa prefectures surveyed in 1994 or 1995.