Asia Pacific J Clin Nutr (1994) 3, 19-31

Development of the Melbourne FFQ: a food frequency questionnaire for use in an Australian prospective study involving an ethnically diverse cohort

Paul Ireland¹, Damien Jolley¹, Graham Giles¹, Kerin O'Dea², John Powles³, Ingrid Rutishauser², Mark L. Wahlqvist⁴ and Joanne Williams¹

1. Cancer Epidemiology Centre, Anti-Cancer Council of Victoria, Carlton, Australia;
2. School of Nutrition and Public Health, Deakin University, Geelong, Australia;
3. Department of Community Medicine, Institute of Public Health, University of Cambridge, Cambridge, UK;
4. Department of Medicine, Monash Medical Centre, Monash University, Clayton, Victoria, Australia.

Objective. To develop an optically scannable food frequency questionnaire (FFQ), 'The Melbourne FFQ', suitable for classifying Australian-, Greek- and Italian-born individuals into quantiles of intake for a range of foods and nutrients. The FFQ would provide the primary measure of dietary exposure in a prospective cohort study.

Design. The FFQ was modelled on that used for the (US) Nurses' Health Study. Food items were chosen on the basis of their relative contribution to the intake of a range of nutrients computed from weighed food records.

Setting. Metropolitan Melbourne, Australia, a city of 3 million people, of whom 75.5% were born in Australia, 2.7% were born in Italy and 1.7% were born in Greece.

Participants. Weighed Food Survey (1987-1989): A volunteer sample of 810 healthy middle-aged (40-69 years) men and women of whom 35% were born in Greece, 33% were born in Italy, and 32% were born in Australia. Melbourne Collaborative Cohort Study (1990-1993): A volunteer sample of 17 949 healthy men and women aged between 40 and 69 years of whom 61% were born in Australia, 21% were born in Italy and 17% were born in Greece.

Results. A 121 item FFQ was developed, together with a customized nutrient database. The optical scanning format was generally well received with the majority of subjects requiring no assistance. The FFQ appeared to overestimate the consumption of fruit and vegetables.

Conclusions. The Melbourne FFQ provides a convenient method of measuring habitual dietary intake in a large population setting. A separate study is required to assess how well the instrument characterizes diet at the level of the individual.

---

Introduction

Large, longitudinal epidemiological studies of diet and health require accurate ordinal classification of individuals with respect to selected characteristics of their habitual diet. This can only be achieved economically by the use of a food frequency questionnaire (FFQ)¹ but at the time the Melbourn 1000 e Collaborative Cohort Study (MCCS) was conceived², no Australian FFQ had been validated. Although one had been shown to have an acceptable degree of repeatability³, it would have been improvident or venturesome to use that instrument in the MCCS which was designed to take advantage of the breadth of dietary exposures likely to be accomplished by the inclusion in the cohort of a large proportion of migrants from Greece and Italy⁴.

The Weighed Food Survey (WFS) was thus undertaken with one aim being to develop an FFQ capable of correctly classifying men and women from Greek, Italian and Anglo-Celtic Australian backgrounds into quantiles of intake for a range of foods and nutrients suspected of having a role in the pathogenesis of cancer, heart disease, stroke, diabetes and premature death. Additional considerations were that the FFQ needed sufficient detail to allow quantitative assessment of dietary intake, yet be simple enough to enable self-administration in any of three languages in a format suitable for optical scanning⁵. Four other issues needed to be resolved in the development of the FFQ: the choice of frequency response options; whether questions were to be asked about portion sizes; which food items were to be included; and which nutrient database was to be used in the analysis. The first two points had straightforward solutions. The nine frequency response options used in the Nurses Cohort Study⁵ were adopted. For simplicity, and because frequency was more discriminatory than portion size in the Nurses' Health Study⁶, no information about usual portion size was sought.

Selection of the food list and the nutrient database were more complex matters. The inclusion of subjects from different ethnic backgrounds presented problems when formulating the list of food items required to classify subjects according to their usual diet. A fundamental tenet of the MCCS was that the cohort had to be considered as a single entity, rather than a collection of different ethnic sub-cohorts. Following this reasoning, an instrument was required that enabled the most accurate possible ranking of individuals by exposure irrespective of their ethnic background. Nonetheless, the food list could have become forbidding in length if it included every item that a proportion of subjects from each ethnic group might have been expected to eat. Additionally, the questionnaire could have lost face validity if subjects were asked how often they ate numerous items with which they were unfamiliar. Some information was available regarding the eating patterns of Italian- and Greek-born Australians^7,8. As with the 1983 National Dietary Survey of Adults (NDSA)^9,l0, these studies were useful for the purpose of identifying relevant foods but were unable to provide measures of individual variability (or dispersion) as they all relied on 24-hour recall data. In addition, the published analyses of the NDSA did not distinguish the specific country of birth of southern European-born migrants, presumably because the numbers were too small. It was therefore considered necessary to obtain records of weighed food intake from a sample of Italian-, Greek- and Australian-born men and women to use in the formulation of the Melbourne FFQ food list.

Methods

Study population and recruitment

As there is only limited geographic clustering of Melbourne residents by birthplace, and as there is no population register of residents by place of birth, it was considered impossible to recruit random samples. It was also deemed desirable to obtain samples of persons likely to want to participate in a long-term study of health. The WFS population, therefore, consisted of a 1000 volunteer sample of 810 healthy men and women aged between 40 and 69 years who were living within the Melbourne Statistical Division and were born in Australia or had entered Australia on an Italian or Greek passport referred to throughout as Australian-, Italian-, and Greek-born. (The latter included some ethnic Greeks born in Egypt and Cyprus.) The same eligibility criteria applied to subjects enrolled in the MCCS.

Assistance with recruitment was provided by established network within the local Italian and Greek communities. Talks were given to church congregations, regional clubs and people attending centres providing assistance to migrants. Articles were written in the ethnic and commercial radio programmes and awareness was spread further by word of mouth. Most of the Australian-born subjects in the WFS responded to an advertisement in a major metropolitan daily newspaper whereas most of the participants in the MCCS responded to personally addressed invitation letters produced from the rolls of the Australian Electoral Commission.

Weighed food records

Upon enrolment into the WFS, subjects were visited at home by a bilingual research assistant who demonstrated the food weighing technique and explained how the diet record booklets were to be complete. Subjects were asked to weigh individual food items separately and record the weight of foods in the form that they were eaten. Serve size was recorded each time a particular food or drink was consumed. Subjects were also asked to provide recipes for cooked dishes.

Weighed food records (WFRs) were kept on two occasions, each of four days duration, at least six weeks apart. To ensure that each day of the week was covered, subjects were randomized to begin their first 4-day WFR on either a Sunday or a Wednesday. Subjects who completed the first WFR from Sunday to Wednesday, completed the second from Wednesday to Saturday and vice versa. The completed WFRs were returned by mail in pre-paid envelopes. The WFRs of the Greek-born subjects were conducted between November 1987 and July 1988, those of the Italian-born subjects were conducted between May and December 1988 and the Australian-born subjects completed their WFRs between October 1988 and March 1989.

Most of the WFRs of the Italian- and Greek-born subjects required translation prior to coding, as did the recipes which were used to estimate the nutrient content of items not already available on the nutrient database. The WFRs were coded as precisely as the information provided would allow, but a standard item was coded in cases where the description was not specific. For example, if the description was simply 'roast chicken', the standard item 'roast chicken, meat and skin' was assumed.

At the time the WFRs were coded (December 1989 to February 1990) the Australian Food Composition Tables (NUTTAB) had only just been released¹¹. The list of foods for which complete data was available was inadequate so we used the British nutrient database McCance and Widdowson's The Composition of Foods¹², which we supplemented with certain local foods¹¹. The nutrient content of some further items was estimated from recipes provided by the Italian- and Greek-born subjects. These values were added to the database together with the nutrient content of some Greek composite dishes supplied by the author of the Greek Food Composition Table¹³.

Selecting items for inclusion on the FFQ

When compiling the list of food items for inclusion on the FFQ, it was necessary to combine similar foods and drinks. Decisions regarding food combinations were largely based on those use 1000 d in previous US^14,15 and UK^l6 studies. Additionally, a cluster analysis was performed on the nutrient database to provide a further objective means of collapsing nutritionally comparable items into a smaller number of common groupings. We refer to the abridged food classifications as categories, rather than groups, because some comprise a single food. Altogether 911 unique food items were coded in the WFRs. Each item was assigned to one of 168 discrete WFR categories. These WFR categories were then ranked, separately for each nutrient, according to their contribution to the overall intake within each of the six sex-ethnicity strata (Tables 1 and 2). In principle, a WFR category was included on the FFQ if it contributed to the first 80% of the cumulative intake for at least one nutrient for at least one sex-ethnicity stratum^17,18. The nutrients of interest were: energy; fat; saturated fat; monounsaturated fat; polyunsaturated fat; carbohydrate; sugars; starch; dietary fibre; protein; cholesterol; sodium; potassium; calcium; iron; zinc; retinol; carotene; vitamin C; vitamin E; and folic acid. The process of selecting the 121 FFQ food categories is outlined in Figure 1.

Table 1. Percentage of energy intake from major' food sources: Weighed Food Survey, 1987-1989.

< 1000 tr> 1000 1000 1000 1000 < 1000 td align="center" valign="top" width="11%">100

WFR category	Australian-born		Greek-born		Italian-born
	females	males	females	males	females	males
(Number of subjects)	163	99	151	130	147	120
white bread	4.7	4.9	8.9	11.0	10.5	13.5
pasta or noodles	1.4	1.3	3.1	3.5	7.4	8.1
Whole wheat or rye bread	5.6	6.3	3.2	4.1	2.6	2.5
cheeses (excluding fetta)	2.5	2.6	1.9	2.3	4.1	3.7
cakes	4.0	3.9	2.3	2.0	2.5	1.8
beef or veal, grilled or fried	1.6	2.2	2.0	3.4	3.5	3.3
milk, full cream	3.3	2.9	3.0	2.9	1.7	1.6
chicken, roast or fried	2.3	2.0	3.1	3.0	2.6	1.9
lentil or bean soup			2.6	2.9	3.9	4.0
olive oil (as a seasoning)			2.8	2.5	4.5	3.5
lamb, chops or roast	2.4	2.1	2.7	4.0	0.7	0.9
biscuits, plain	1.3	1.5	2.0	1.2	3.0	2.3
sugar	1.7	2.0	1.3	1.4	2.1	2.1
margarine, polyunsaturated	3.7	3.8	0.7		1.3	1.0
potatoes, fried	1.5	1.8	1.9	2.3	0.8	0.9
potatoes, not fried	1.9	2.0	1.1	1.1	1.3	1.1
apples (fresh, stewed or juice)	1.1	1.0	1.6	1.2	2.2	1.4
savoury pastries	1.3	1.8	2.6	2.7
wine, red				1.6	1.2	5.4
salad vegetables with dressing		0.6	2.5	2.7	0.9	0.8
breakfast cereals (sweetened)	1.2	1.0	0.7	0.9	1.6	1.9
beer		3.7		2.1		0.7
fish, steamed, grilled or baked	0.8		1.9	1.5	1.1	1.2
coffee (including espresso and Greek style)			2.3	1.9	1.3	0.9
spinach or other leafy greens			1.9	1.8	1.6	1.1
rissoles or meatloaf	1.0	1.2	1.2	1.3	0.9	0.9
butter	2.1	1.6	0.7		0.8	0.9
biscuits, sweet	1.2	1.1	1.4	1.2		0.8
muesli	2.4	2.5		0.7
crackers or crispbreads	1.7	1.1	1.2		1.4
fish, fried	1.1	0.7	1.8	1.5
milk, reduced fat (1.5%)	2.1	1.3			1.0	0.7
desserts or puddings	1.9	1.8	0.7		0.7
wine, white	1.4	1.1			0.9	1.1
vegetable oils (as a seasoning)					2.6	2.0
pizza		0.6			2.0	1.9
bananas	1.5	1.4		0.6		0.7
cola or other soft drink	0.7	1.1	0.9	0.9	0.7
milk, skimmed	2.3	1.8
mixed dishes with rice			1.7	1.1	0.7	0.7
soups or broths (without beans or lentils)			1.2	1.0	1.0	0.8
fetta cheese			2.1	1.9
mixed vegetable dishes			2.2	1.5
oranges or mandarins			0.9	0.7	1.2	1.0
mixed dishes with beef	0.9	1.3	0.6			0.7
salami				0.7	0.9	1.5
frankfurters or sausages	1.1	1.9
bran-based breakfast cereal	1.3	0.9			0.7
orange juice	1.0	1.0	0.8
ice cream	1.3	1.2
smoked or canned fish	0.8	0.6			0.7
chocolate	1.2	1.0
breakfast cereals (unsweetened)	0.8	1.3
grapes			1.1	1.0
buns or doughnuts	0.9	1.2
pears			0.6		0.7	0.7
peanuts or peanut butter	0.9	1.0
cabbage rolls or stuffed vine leaves			1.0	0.7
cocoa or coffee substitutes	0.9	0.7
oatmeal porridge	0.8	0.7
liqueurs or fortified wines	0.6	0.8
marmalade or other jams	0.6	0.8
rice, boiled	0.7				0.7
honey or syrups			0.6	0.7
nuts, other than peanuts	0.6		0.7
peaches or nectarines	0.6			0.6
sweet pastries			0.8
mixed dishes with lamb			0.7
capsicum (including stuffed peppers)			0.7
tomato			0.7
mixed dishes with egg	0.7
eggs, boiled or poached	0.7
yoghurt	0.7
mayonnaise	0.7
pork, chops or roast				0.7
Other	13.4	12.2	10.8	7.5	9.5	8.8
Total	100	100	100	100	100

*The food categories for which values are indicated provide 80% of the cumulative intake within each sex-ethnicity stratum.

 

Table 2. Percentage of beta-carotene intake from major* food sources: Weighed Food Survey, 1987-1989.

WFR category	Australian-born		Greek-born		Italian-born
	females	males	females	males	females	males
Number of subjects	163	99	151	130	147	120
carrots	43.6	46.1	28.5	18.4	24.6	20.7
spinach or other leafy greens	4.8	4.9	28.6	37.0	31.1	24.0
broccoli or cauliflower	4.4	3.7	4.1	3.8	17.5	19.9
lettuce or other salad greens	6.8	6.1	4.8	5.4	7.8	9.7
tomato	5.1	5.4	5.9	6.4		2.1
cantaloupe or honeydew	5.0	3.7	6.6	5.1
mixed dishes with beef	4.3	4.4				5.2
pumpkin	5.2	4.9
figs			3.0	2.9
lentil or bean soup (including minestrone)		2.9
apricots	2.8
peaches or nectarines
Other	16.1	10.3	11.9	12.9	10.1	14.7
Total	100	100	100	100	100	100

*The food categories for which values are indicated provide 80% of the cumulative intake within each sex-ethnicity stratum.

Figure 1. Flow chart outlining the process of selecting categories for inclusion in the Melbourne FFQ and producing the sex-ethnic-specific nutrient databases used in its analysis.

a WFR categones that did not satisly the criteria for inclusion in the FFO (ie not among the lirst 80% cumulative intake lor any nutnent for any sex-ethnicity stratum) but were able to be reassigned appropriately to another category b WFR ca 1000 tegones that made a negligible contribution to nutrient intake and which compnsed items that could not readily be reassigned to other categories c In the MCCS quesbons regarding the use ol these important sources ol nutrients were asked separately Irom the FFQ This includes alcoholic drinks sugar oils nutnent supplements and milk added to hot drinks and cereal d Post-analysis collapsing and splining ol selected WFR categories to achieve greater specdicity and clarity on the FFO

Creation of sex-ethnic specific FFQ nutrient databases

Once the 121 FFQ food categories were determined, a method was required for assigning nutrient values to a serve (or portion) of each category. We chose to customize the nutrient database to take account of sex and ethnic variations in food intake. We were interested in two sources of inter-stratum variation, which were the differences in portion size and the differences in the relative contribution made by various items within a category. The imputed nutrient values per portion were obtained from the WFR data as outlined below.

Of the 911 food items coded in the WFR analysis, 715 were assigned to the 121 FFQ food categories. Ninety of the remaining items related to alcoholic drinks, sugar, oils and condiments. These items were excluded because questions were to be asked about their use separately from the FFQ. The remaining food items did not correspond to any of the FFQ food categories. Collectively they made a negligible contribution to nutrient intake with most having had fewer than 10 serves in total over the 6480 person days of diet recording.

For each FFQ food category, the nutrient content per portion was averaged for each ethnic group (by sex). Within food categories, a proportion of food items had zero content for one or more nutrients. Geometric means across non-zero entries were used due to the naturally skewed distributions, but numbers of zero nutrient values were also recorded. Ethnic group and sex strata were combined in cases which showed little heterogeneity or for which there were fewer than 50 serves in total. Arithmetic mean nutrient content was computed after weighting for the number of portions with zero nutrient content.

FFQ administration and analysis

The FFQ was completed by 17 949 subjects attending the MCCS between November 1990 and April 1993. The English version is illustrated in the appendix All language versions align identically so that only one optical scanning program was required for data entry. Although the FFQ was designed to be self-administered, approximately 25% of subjects required at least some assistance. The FFQs were scanned while the subjects were in attendance so that gross errors, such as the omission or duplication of frequency responses, were rectified immediately.

Average daily nutrient intake for each FFQ response was calculated by matrix multiplication with the nutrient table appropriate for the respondent's ethnicity and sex. The nine frequency response options were converted to daily equivalents as follows:

Frequency response label	Daily equivalent
never or less than once per month	0
1-3 per month	0.067
1 per week	0.15
2-4 per week	0.43
5-6 per week	0.80
1 per day	1
2-3 per day	2.5
4-5 per day	4.5
6+ per day	8

The cut-off points for improbable energy intake were those used in the Health Professionals Follow-up Study^l9 and the Nurses' Health Study²⁰. This involved excluding values for women with intake below 2100kJ/day, men below 3360kJ/day, and men and women with intake above 16 800kJ/day.

Energy adjustment

Energy-adjusted nutrient intakes for both the WFRs and the FFQs were computed as the residuals from the regression model on a log scale plus the expected nutrient intake for the mean energy intake of the study population²¹. The regression analyses were specific for each of the six sex-ethnicity sub-populations. This method of energy adjustment alters the distribution of nutrient intake values within the population but does not substantially change median values. The other purpose for adjusting energy intake is to facilitate comparison of nutrient intakes in situations where there are differences in total consumption between groups, so median nutrient intakes were expressed per MJ of energy derived from protein, carbohydrate and fat. This enabled a standardized comparison between the FFQs and the WFRs, which also included alcohol.

Results

The criteria for selecting categories for listing on the F FQ were chosen to capture the breadth of dietary exposure likely to be experienced in a cohort of men and women from diverse culinary backgrounds. Cross-cultural dietary heterogeneity was evident when ranking food sources to the intake of particular nutrients calculated from the WFRs. The proportion of energy intake contributed by major food sources for each of the sex-ethnicity strata is presented in Table 1. A blank value indicates that the category was not among those that contributed to the first 80% of the cumulative energy intake for that particular stratum; it does not necessarily indicate that the food source made no contribution to energy intake, indeed in each stratum the sum of the 'other' categories was less than 20%. On the basis of energy alone, 75 categories satisfied the criterion for inclusion on the FFQ. As with the other ubiquitous nutrients, eg protein, fat and carbohydrate, many categories each con 1000 tributed a relatively small proportion (0.5-1.0%) to total energy intake. In contrast, more than 80% of the beta-carotene intake was derived from only 12 categories (Table 2).

Twenty-two WFR categories did not satisfy the criteria for inclusion on the Melbourne FFQ (Figure 1). Of these, 14 were eliminated because they could not be reassigned sensibly to another category. These rejected categories were: poultry other than chicken; mixed dishes with pork; mixed dishes with fish or seafood; turnips or swedes; globe artichoke; asparagus; okra; radishes; sweet potato; cherries; yeast; seeds; bean sprouts; and polenta. The Melbourne FFQ does not include 19 WFR categories relating to alcoholic drinks, dietary supplements, oils, sugar, milk added to breakfast cereal, tea and coffee because questions concerning their use were asked separately. For the purposes of clarity and greater specificity, some modifications were made to the list of chosen categories. A net reduction of 10 was achieved by the post-analysis collapsing of 29 WFR categories (including eight that did not satisfy the model) into nine FFQ categories and the splitting of 10 WFR categories into 20 FFQ categories. For example, the category 'broccoli and cauliflower' was split into separate categories on the FFQ because these two vegetables differ substantially in their beta-carotene content. The combined category was quantitatively the third most important source of beta-carotene in the analysis of the WFRs (Table 2).

Median daily intake data for a range of nutrients in the WFS and the MCCS are presented in Table 3. The energy values calculated from the FFQs in the MCCS do not include energy derived from alcoholic drinks. Energyadjusted nutrient intake values were therefore calculated to facilitate a standardized comparison between the two dietary intake methods (Table 4). The FFQ energy values were not identical in Table 3 and 4. Table 3 presents median values for energy as a nutrient in its own right, whereas in Table 4 energy values were computed as the sum of energy derived from the population median intake of carbohydrate (16 kJ/g), fat (37 kJ/g) and protein (17 kJ/g). The energy-adjusted intakes of dietary fibre, beta-carotene and vitamin C in the MCCS were consistently higher than those in the WFS. On the other hand, energy-adjusted calcium intakes were consistently lower in the MCCS.

Discussion

Not surprisingly, the WFRs indicated differences between the ethnic groups in the proportion of nutrients derived from different food sources (Tables 1 and 2). In particular, men and women born in Australia reported eating more carrots, wholegrain bread, breakfast cereals, butter and margarine and less legumes and leafy green vegetables than did their Italian- and Greek-born counterparts. Those born in Greece ate more savoury pastries, salads, fish and fetta cheese, whereas the Italian-born ate more pasta and pizza and less lamb. Within each of the ethnic groups, the most notable difference between the sexes involved alcoholic drinks, particularly beer and red wine.

Table 3. Median daily nutrient intake in the Melbourne Collaborative Cohort Study (FFQ, 1990-1993) and the Weighed Food Survey (WFR, 1987-1989).

	Greek-born					Italian-born					Australian-born
	females			males		females			males		females			males
	WFR	FFQ		WFR	FFQ	WFR	FFQ		WFR	FFQ	WFR	FFQ		WFR	FFQ
Number of subjects	151	1620		130	1273	147	2057		120	1613	163	6522		99	4202
Subjects excludeda (n)		95			141		55			92		92			187
Total energy (kJ)b	6680	8300		8790	9370	6910	7160		7470	7800	7400	7100		10500	8320
Protein (g)	71	105		98	121	74	79		100	91	76	71		97	84
Carbohydrate (g)	177	226		228	263	174	222		236	241	207	204		291	238
Fibre (g)	17	30		21	33	18	26		24	26	20	23		26	23
Fat (g)	70	78		91	85	65	54		84	58	72	66		91	80
Retinol (m g)	135	176		158	174	184	165		215	157	302	313		346	322
Beta-carotene (m g)	2470	6360		2590	5930	2670	5000		2840	4180	2940	5190		3510	4280
Vitamin C (mg)	68	139		64	129	53	98		65	93	81	103		102	92
Calcium (mg)	578	673		726	708	636	604		820	609	771	583		951	616
Iron (mg)	9	15		12	17	11	13		16	14	12	13		17	15

^a. subjects were excluded from the analysis if their estimated energy intake computed from the FFQ was above 16 800 kJ/day or below 2100 kJ/day (women) or 3360 kJ/day (men). ^b FFQ does not include sugar and alcoholic drinks.

Table 4. Median daily energy-adjusted^a nutrient intake in the Melbourne Collaborative Cohort Study (FFQ, 1990-1993) and the Weighed Food Survey (WFR, 1987-1989).

	Greek-born				Italian-born				Australian-born
	females		males		females		males		females		males
	WFR	FFQ	WFR	FFQ	WFR	FFQ	WFR	FFQ