A Multiethnic Cohort in Hawaii and Los Angeles: Baseline Characteristics

Abstract

The authors describe the design and implementation of a large multiethnic cohort established to study diet and cancer in the United States. They detail the source of the subjects, sample size, questionnaire development, pilot work, and approaches to future analyses. The cohort consists of 215,251 adult men and women (age 45–75 years at baseline) living in Hawaii and in California (primarily Los Angeles County) with the following ethnic distribution: African-American (16.3%), Latino (22.0%), Japanese-American (26.4%), Native Hawaiian (6.5%), White (22.9%), and other ancestry (5.8%). From 1993 to 1996, participants entered the cohort by completing a 26-page, self-administered mail questionnaire that elicited a quantitative food frequency history, along with demographic and other information. Response rates ranged from 20% in Latinos to 49% in Japanese-Americans. As expected, both within and among ethnic groups, the questionnaire data show substantial variations in dietary intakes (nutrients as well as foods) and in the distributions of non-dietary risk factors (including smoking, alcohol consumption, obesity, and physical activity). When compared with corresponding ethnic-specific cancer incidence rates, the findings provide tentative support for several current dietary hypotheses. As sufficient numbers of cancer cases are identified through surveillance of the cohort, dietary and other hypotheses will be tested in prospective analyses.

Ethnic minorities in the United States have not been well represented in epidemiologic research on diet and cancer, especially in prospective cohort studies. Thus, little is known about the relation of dietary factors to cancer risk in these groups. Furthermore, while diet and other external factors are the predominant determinants of cancer risk in all these groups (1), the extent to which environmental exposures explain interethnic differences in incidence is not known. By including a variety of ethnic groups within a single study, and by using a common data collection methodology in all groups, interethnic comparisons of exposure-disease relations can readily be made.

Hawaii has long presented special opportunities for epidemiologic research because of its diverse ethnic/cultural environment. Over the past two decades, the Los Angeles basin has also attracted large ethnic populations, including a variety of groups from Asia, and Latino migrants from Mexico and Central and South America. Together, these two areas now provide an unmatched resource for epidemiologic initiatives.

Recognizing the need for dietary studies of ethnic minorities in the United States, the unique opportunities offered by Hawaii and southern California for conducting cross-ethnic and cross-cultural studies, and our extensive past research in Hawaii in the area of diet and cancer (2–5), we decided to establish a population-based multiethnic cohort in these two areas. The advantages of a cohort design included avoidance of recall bias in dietary intake data, the possibility to assess dietary changes over time, the ability to study many different disease outcomes (both cancer and non-cancer) simultaneously, and the potential to conduct ancillary studies using biologic markers. Pilot work for the project was initiated in Hawaii and Los Angeles in 1988; the full-scale study began in 1993. Initially, four ethnic groups (African-Americans, Japanese-Americans, Latinos, and Whites) were selected for the study based on the size of the respective populations in the two areas and on the striking differences among them in the reported rates for several common cancers (table 1) (6, 7). Subsequently, a smaller ethnic group, Native Hawaiians, was added.

Table 1. Incidence (1988–1992)^* for some common cancers in the ethnic groups of the cohort^†.

Ethnicity	Cancer site
	Prostate	Breast	Colorectum	Lung
	Males
Latino	89.0		38.3	41.8
African-American	180.6		60.7	117.0
Japanese-American	88.0		64.1	43.0
White	134.7		56.3	76.0
Native Hawaiian	57.2		42.4	89.0
	Females
Latino		69.8	24.7	19.5
African-American		95.4	45.5	44.2
Japanese-American		82.3	39.5	15.2
White		111.8	38.3	41.5
Native Hawaiian		105.6	30.5	43.1

^*Per 100,000, age-adjusted to 1970 US standard population.

^†Source: Miller et al. (7).

Materials and Methods

Selection of subjects

Our intention was to establish a cohort that would include all socioeconomic strata within each ethnic group, so that generalization of findings to the population at large would be possible. Because drivers' license files in both Hawaii and California include the names of most resident adults, encompass all socioeconomic strata, and contain information on age and sex, we selected this source as the primary sampling frame for the study. An additional source in Hawaii was the voters' registration file, which was used to identify names not on the drivers' license file, especially among older Japanese-American women. An additional source of African-Americans in California was the Health Care Financing Administration (HCFA) files, as described below.

In order to avoid oversampling of Whites in Los Angeles County and to exclude, as much as possible, ethnic groups other than those of primary interest, an ethnic identifier was needed prior to sampling. Because drivers' license files do not contain information on ethnicity, we added this item to the files based on ethnic-specific surname lists, an approach that has been shown to be relatively effective for identifying Latinos in the United States (8). These surname lists were created from several sources, including rosters of participants in our prior epidemiologic studies, the files of the population-based tumor registries in the two areas, and commercial publications. In addition, we used a Japanese first-name list to identify Japanese-American women married to non-Japanese-American men. This effort resulted in a database of separate name lists for Chinese, Filipinos, Japanese-Americans, Native Hawaiians, Koreans, Latinos, and Samoans. Because some names fell into more than one list (e.g., Lee could be Chinese, Korean, or White), these lists were not mutually exclusive. In addition, the voters' registration file in Hawaii has a special identification field for persons of Native Hawaiian ancestry.

In Hawaii, names that do not appear on any of these lists primarily identify Caucasians (Whites), whereas in Los Angeles, names that do not occur on any of these lists primarily identify Caucasians or African-Americans. Because it was not possible to distinguish between the latter two groups by this approach, we categorized the census tracts of Los Angeles County by the proportion of African-American residents (9). In addition, we obtained lists of names of African-Americans in California age 65 years and older from the HCFA files. Thus, by a combination of selective sampling from drivers' license files, census tracts with ≥65 percent African-Americans, and the HCFA files, we were able to access the appropriate ethnic populations for the study. Since the ultimate assignment of ethnicity to the subjects in the cohort was based on their questionnaire responses, any errors in assignment of ethnicity at the time of initial sampling does not affect the quality of the data.

Questionnaire development

Because our primary interest was in diet, we developed a questionnaire that emphasized the assessment of food intake. The diet portion of the questionnaire was designed specifically for use in this study. We began by collecting 3-day measured food records from approximately 60 men and women, 45 to 75 years of age, from each of the four main ethnic groups in the study, as well as from a similar sample of Native Hawaiians. We utilized these records to identify food items for inclusion in the questionnaire. For each ethnic group, we computed the contribution of each food item to the total intake of the nutrients of major interest (e.g., fat, dietary fiber, vitamin A, carotenoids, and vitamin C). For a particular nutrient, the foods were ordered from highest to lowest in their contribution to intake, and the minimum set that yielded at least 85 percent of the intake was determined. Because this was done for each ethnic group and nutrient, the final questionnaire list actually accounts for much more than 85 percent of the intake of the major nutrients. In addition, we used the food records to identify for inclusion specific food items uniquely associated with the traditional diets of a particular group irrespective of their contribution to intake (e.g., tofu and salted fish for Japanese-Americans; tortillas and tamales for Latinos). Food items specific to the diets of other ethnic groups in Hawaii and Los Angeles, such as Filipinos, Chinese, and Koreans, were not included in the questionnaire, unless they were commonly eaten by other ethnic groups (e.g., chow mein noodles).

A single diet questionnaire suitable for all ethnic groups, rather than multiple versions, was chosen because: 1) we developed it from a face-to-face instrument that had been validated in a multiethnic population; 2) many ethnic foods are consumed regularly by several different groups and would have to be included on all versions; and 3) we did not know the precise ethnic identification of each prospective participant at the time the questionnaire was mailed.

The questionnaire was designed for self-administration and for processing by optical scanning. Eight frequency categories for foods and nine for beverages were included, together with three (in a few instances four) choices of portion size. As an additional aid to quantification, photographs showing selected foods in representative portion sizes were provided at the head of several pages of the questionnaire. It was important to collect portion size information not only because of significant variations among individuals, but also because of significant variations in the typical serving sizes consumed by different ethnic groups. We found in preliminary studies, for example, that nearly one-half of the subjects in each ethnic-sex group selected either the small or large portion size when given a choice, and thus would likely have been misclassified if assigned a standard medium portion. The portion size options on the questionnaire were based on typical serving sizes for each single food or grouping of foods as reflected in the original 3-day measured food records.

In addition to the section on diet, the questionnaire included information on demographic factors (including migrant status), personal behaviors (smoking, solar exposure, physical activity), history of prior medical conditions, use of medications, reproductive history and the use of replacement estrogens/oral contraceptives (for women), and a brief family history of cancer. The questionnaire was extensively pretested in order to assure that the instrument could be self-administered by individuals with less than a high school education and would yield reliable information. We began with face-to-face administration of an early draft by trained interviewers who identified ambiguities, awkward wording of questions, and problems in the layout of questions. After appropriate revisions, the questionnaire was mailed to a small (130 subjects) sample of residents in 1989. Following this mailing, a subset of respondents and non-respondents were telephoned to determine reasons for their participation or not, their reaction to the cover letter about the study, and comments on questionnaire content. One result of this activity was a recognition that many Latinos could not complete the form in English. Thus, a Spanish translation of the questionnaire was developed.

The pre-testing activities culminated in the mailing of a draft questionnaire to 1,000 randomly selected subjects of each ethnic group. This gave us some experience with the logistics of large sample mailings. In addition, for 200 subjects in Hawaii, we compared their responses on the pilot questionnaire to their responses to the dietary portion of a past interviewer-administered questionnaire used for other epidemiologic studies in Hawaii and found very comparable reports of consumption. We also tested the reproducibility of the questionnaire on approximately 300 subjects to whom we remailed the form 6 months later; correlation coefficients for nutrient intakes between the two questionnaires were in the range of 0.5–0.7 for most items. Finally, we collected unannounced 24-hour recalls monthly over a 12-month interval from a sample of 337 subjects representing all ethnic-sex groups who had completed the pilot questionnaire. This enabled us to determine where further modifications to the questionnaire were needed (including identification of any omitted important foods) and provided useful data for designing an appropriate calibration study for the final study instrument. These data are discussed in a methodological paper on calibration study design (10).

Data collection, editing, and management

Assembly of the cohort began in spring 1993 and was completed in 1996. Mailings of the self-administered, 26-page questionnaire to persons between ages 45 and 75 years in the entire state of Hawaii and in Los Angeles County were conducted in waves of approximately 100,000 subjects each using the facilities of local mailing houses in both study areas. In Hawaii, these mailings were done on a geographic basis. In Los Angeles, they were primarily done on an ethnic basis. Up to three mailings were made to each potential participant, with the exception of the last mailing waves to African-Americans and Latinos in late 1995 and early 1996, when only one or two mailings were possible because of limited funds. In order to expand the African-American sample, we also mailed questionnaires to African-American residents of additional southern and northern California counties (Riverside, Orange, San Bernardino, San Diego, San Francisco, San Mateo, Contra Costa, and Alameda) who were identified from the HCFA files. Final response rates by initially assigned ethnicity were as follows: Japanese-American men 46.3 percent, women 51.3 percent; White men 39.1 percent, women 47.0 percent; African-American men 20.0 percent, women 25.5 percent; Latino men 18.6 percent, women 21.3 percent; and Native Hawaiian men 36.1 percent, women 42.4 percent.

All questionnaires were processed by optical scanning and the information was stored in a structured database. Editing began with an initial manual review of questionnaires for completeness and suitability for optical scanning, followed by computer-assisted editing for such problems as multiple responses to a single item, inconsistencies between parts of the questionnaire, and other correctable errors. For purposes of the baseline dietary data presented in this paper, we excluded persons whose reported daily intake was <500 or >8,000 calories, or who completed fewer than five diet pages (out of 18) of the questionnaire. (The latter criterion may seem liberal. However, by recontacting a sample of respondents by telephone, we had determined that many subjects provided no response to items they did not eat, rather than marking the “never” category on the form.)

The study is coordinated from the Hawaii center where the computer management and nutritional activities take place. A computerized tracking system and food composition database specific for this study were developed and are being maintained by the coordinating center. The complex tracking database allows for recording of all addresses, as well as tumor, death, and other follow-up information. The staff at each location use this system to keep the addresses and follow-up activities up-to-date for the cohort members in each of the two locations.

Food composition table

An extensive food composition database for use in the multiethnic population of Hawaii has been developed and maintained at the Cancer Research Center of Hawaii for many years. For the present project, additional data on foods consumed by Latinos and African-Americans had to be added. These data were obtained from USDA Handbook 8 (11, 12), with supplementation from special laboratory analyses (Kolonel et al., unpublished data) and other research and commercial publications (13–16). In addition to values for energy, macro- and micro-nutrients, the database includes values for carotenoids, tocopherols, isoflavonoids, non-starch polysaccharides, and individual fiber components. This database includes over 1,500 foods and 700 recipes, the latter developed from food records and 24-hour recalls collected over many years, including the pilot work for this project, as well as from established cookbooks. This extensive database is constantly reviewed and updated.

For the self-administered quantitative food frequency questionnaire (QFFQ), we prepared a customized food composition table, based on the larger database, to use for the analyses. For grouped (composite) food items on the questionnaire, we weighted the individual foods according to their respective frequencies of consumption as determined from the 24-hour recalls collected on the first 1,362 participants in a calibration study (see below). Thus, for a composite item such as “pizza”, the nutrient values were weighted averages of the different types of pizza reported. For a few of these composite items, the values were ethnic-specific, because of sufficient variation in the frequencies of consumption of the individual foods that made up the category (e.g., corn tortillas, corn muffins, or cornbread).

The food composition table has separate records for meats trimmed of all visible fat and for chicken with skin removed. Thus, for a given subject, the value used is based on his/her response as to usual practice regarding fat/skin removal. Similarly, a respondent's usual preference for a specific type of oil (e.g., safflower), butter, or margarine (e.g., whipped) was used to assign the appropriate nutrient values for these items.

Calibration study

Recall information, especially in the assessment of dietary intakes, can entail considerable measurement error (17). To address this concern, we collected three unannounced 24-hour diet recalls about 1 month apart on approximately 260 randomly selected subjects of each ethnic-sex group in the study. Day of the week was randomly selected for the second and third recalls to ensure an adequate representation across days of the week. These data will be used to calibrate the dietary information from the questionnaires so that estimates of risk associated with intake can be made with less misclassification error (10, 18). An accompanying paper describing the design and findings of the calibration study appears in this issue of the Journal (19).

Results

Demographic characteristics

After excluding approximately 1 percent of questionnaires because certain demographic or other essential information was missing, the total number of subjects who comprised the cohort was 215,251. (As noted above, some additional subjects will be excluded from certain analyses because of incomplete information on the dietary portion of the questionnaire.) The distribution of the cohort by age, sex, and ethnicity is given in table 2. Because actual ethnicity was not known at the time the questionnaires were mailed, a number of persons of ethnicities other than the five that were targeted responded. Japanese-Americans comprise the largest subgroup within each sex (27.9 percent of men and 25.3 percent of women), followed by Whites (23.6 percent and 22.4 percent), Latinos (23.6 percent and 20.8 percent), African-Americans (13.3 percent and 18.8 percent), and Native Hawaiians (6.3 percent and 6.6 percent). A majority (about two-thirds) of the “Other” category is comprised of Filipinos. The number of Latinos in the cohort is somewhat lower than originally planned, because of the high cost involved in enrolling this group. This resulted from the combination of a low response rate and the need to mail the questionnaires in both Spanish and English. (We had determined from the pilot study that it was not possible to determine in advance which version a subject would prefer.) In order to include as many respondents as possible in the cohort, persons of mixed ancestry were assigned to one of the categories in table 2 according to the following priority ranking: African-American, Native Hawaiian, Latino, Japanese-American, White, and Other. Within each ethnic-sex group, the sample is well distributed across the three age strata. About 49 percent of the cohort members reside in Hawaii and 51 percent reside in California.

Table 2. Distribution of the cohort by age,^* sex, and ethnicity, Hawaii and Los Angeles, 1993-1996.

Ethnicity	Age group (years)						Total no.
	45–54		55–64		65–75
	No.	%	No.	%	No.	%
Males
Latino	6,376	27.9	9,453	41.4	6,989	30.6	22,818
African-American	2,899	22.6	3,503	27.3	6,449	50.2	12,851
Japanese-American	8,065	29.9	7,667	28.4	11,232	41.7	26,964
White	8,718	38.1	6,672	29.2	7,467	32.7	22,857
Native Hawaiian	2,779	45.4	1,969	32.2	1,372	22.4	6,120
Other	1,744	33.5	1,801	34.6	1,655	31.8	5,200
Total	30,581		31,065		35,164		96,810
Females
Latino	7,767	31.5	10,481	42.6	6,372	25.9	24,620
African-American	6,136	27.6	6,315	28.4	9,805	44.1	22,256
Japanese-American	8,809	29.4	9,299	31.0	11,849	39.6	29,957
White	10,216	38.5	7,937	29.9	8,349	31.5	26,502
Native Hawaiian	3,753	47.8	2,426	30.9	1,672	21.3	7,851
Other	2,949	40.6	2,495	34.4	1,811	25.0	7,255
Total	39,630		38,953		39,858		118,441

^*Age at baseline (completion of the quantitative food frequency questionnaire).

Foreign-born subjects also add to the diversity of the cohort because they are more likely to have dietary habits that diverge from the US mainstream. The overall proportion of foreign-born subjects was 20.4 percent. Excluding the mixed group “Other”, the rate varied from a low of <1 percent for Native Hawaiians to a high of 52.0 percent for Latinos. Although only 8.4 percent of Japanese-Americans were foreign-born, 44.0 percent of them were second generation immigrants. Latinos also included a substantial proportion (22.3 percent) of second generation immigrants. Among the Latinos, 38.6 percent completed the Spanish-language version of the questionnaire, an indicator of degree of acculturation and another reflection of diversity within the cohort.

As intended, the cohort reflects a substantial range in educational level (table 3). Overall, 44.8 percent of the cohort members (42.4 percent of men and 46.8 percent of women) have a high school education or less; 7.7 percent have vocational training; and 47.4 percent have at least some college education. However, as shown in table 3, this varies considerably by ethnicity: for example, 28.7 percent of Latino men compared with 72.7 percent of Caucasian men have ever attended college. Differences by gender within ethnicity are not substantial. The table also compares the cohort with the distribution from the US Census for these groups in Hawaii and Los Angeles County in the same age range (9). As we expected, the cohort members are a somewhat better educated subset of the general population, but all levels of education are comparatively well represented.

Table 3. Distribution (%) of the cohort members by educational level and comparison with the US census,^* Hawaii and Los Angeles, 1993–1996.

Ethnicity	Educational level
	≤8 years		9–12 years		Vocational		At some least college
	Cohort	Census	Cohort	Census	Cohort	Census	Cohort	Census
Males
Latino	33.6	43.2	31.1	36.8	6.6	2.0	28.7	18.0
African-American	9.3	13.0	33.7	47.5	5.5	3.1	51.5	36.4
Japanese-American	3.2	7.9	34.7	40.6	12.6	7.2	49.4	44.3
White	3.4	2.9	19.7	30.8	4.2	3.8	72.7	62.5
Native Hawaiian	5.8	7.0	45.9	61.4	8.5	4.3	39.9	27.4
Females
Latino	36.7	46.8	35.9	39.3	6.4	1.4	21.0	12.6
African-American	6.4	11.5	35.6	49.2	6.5	3.4	51.5	35.9
Japanese-American	4.7	13.6	38.3	39.9	12.9	6.6	44.1	39.9
White	3.0	4.1	27.7	34.2	4.8	4.7	64.5	57.1
Native Hawaiian	4.3	7.1	52.4	60.8	7.6	3.6	35.7	28.5

^*Census data for the corresponding geographic areas of Hawaii and California, 1990 (9).

Power considerations

Table 4 shows the number of cases expected in this cohort by the year 2002 for four common cancer sites in the four larger ethnic groups. (The Native Hawaiian group was an addition to the original cohort, and, because the sample is much smaller, a longer period of follow-up will be required for definitive analyses in this population.) These projections are based on Hawaii and Los Angeles cancer incidence rates from the respective population-based tumor registries (6). The number of lung cancer cases was adjusted downward to reflect a projected lower incidence in the cohort because of reduced smoking levels (see below). The table also shows minimal detectable relative risks (MDRs), comparing the highest to the lowest quartile of exposure, assuming 80 percent power and a type I error of 0.05. These MDRs were computed assuming a steady trend in the relative risk over the four quartiles of exposure (20). Relative risks of 1.5 or lower will be detectable in all four groups for prostate and breast cancers, while risks of 1.9 or lower will be detectable for colorectal cancer and 2.0 or lower for lung cancer.

Table 4. Expected number of cases by the year 2002 and minimal detectable risks (MDRs) from the lowest to the highest quartile of exposure in the cohort, Hawaii and Los Angeles, 1993–1996.

Site and ethnicity	Males		Females
	No. of cases	MDR	No. of cases	MDR
Colorectum
Latino	255	1.6	160	1.9
African-American	221	1.7	310	1.6
Japanese-American	579	1.4	356	1.5
White	411	1.5	319	1.6
Total	1,533	1.2	1,204	1.3
Lung
Latino	201	1.7	120	2.0
African-American	264	1.6	225	1.7
Japanese-American	255	1.6	115	2.1
White	354	1.5	326	1.6
Total	1,171	1.3	863	1.3
Prostate
Latino	680	1.4
African-American	733	1.3
Japanese-American	882	1.3
White	1,044	1.3
Total	3,426	1.1
Breast
Latino			406	1.5
African-American			484	1.4
Japanese-American			657	1.4
White			806	1.3
Total			2,542	1.2

Non-dietary exposure variables

Table 5 shows the prevalence of smoking, drinking, obesity, and vigorous physical activity among subjects in the cohort. Current smoking levels show a wider range in men (from 13.3 percent in Japanese-Americans to 28.8 percent in African-Americans) than in women (from 11.1 percent in Japanese-Americans to 21.2 percent in Native Hawaiians). Surprisingly, there is little effect of migrant status on smoking behavior: for example, 18.4 percent of first generation Latino men are current smokers compared with 18.1 percent of second or later generation men (data not shown).

Table 5. Smoking, drinking, obesity, and physical activity by sex and ethnicity^* in the cohort, Hawaii and Los Angeles, 1993-1996.

Ethnicity	Current smokers (%)	Current drinkers† (%)	Obesity‡ (%)	Vigorous physical activity§ (%)
	Males
Latino	18.3	40.5	21.6	36.1
African-American	28.8	37.9	22.8	27.5
Japanese-American	13.3	38.6	7.3	28.2
White	19.8	56.0	14.6	40.1
Native Hawaiian	21.8	38.5	35.1	47.1
	Females
Latino	13.6	13.7	29.5	12.0
African-American	20.9	17.6	37.7	9.3
Japanese-American	11.1	9.9	6.7	9.6
White	17.6	35.3	19.6	19.5
Native Hawaiian	21.2	16.2	34.5	23.1

^*Standardized to the overall cohort age distribution.

^†At least one alcoholic beverage per week.

^‡Defined as body mass index (BMl) ≥30, where BMI = weight (kg)/[height(m)]²×100.

^§At least 3 hours of strenuous exercise or vigorous work per week.

Current consumers of alcohol were defined as persons who drank any type of alcoholic beverage at least once a week. The rate of consumption is much higher in Whites than in the other groups. Consumption rates are much lower among women than men. Only 9.9 percent of Japanese-American women are current consumers, compared with 35.3 percent of White women.

Using current weight and height information, we computed a body mass index (BMI) for each subject. Obesity among men, i.e., BMI ≥30, is highest among Native Hawaiians (35.1 percent). Obesity among women is highest in African-Americans (37.7 percent), followed closely by Native Hawaiians (34.5 percent). African-Americans show the most striking difference between men and women in the percent of obese individuals.

The questionnaire included a section on physical activity. Table 5 shows the proportion of subjects who engage in more than 3 hours per week of vigorous physical activity (strenuous sports or vigorous work). Among men, the highest rate is seen in Native Hawaiians, and the lowest in Japanese-Americans and African-Americans. Among women, the rates are much lower than in men, but the ethnic pattern is the same.

Dietary exposure variables

Table 6 shows mean intake values for calories and several nutrients by ethnicity and sex. These results are based on food intake only, and do not include supplements. Mean intake of calories ranges from 2,242 to 2,728 per day among men, and from 1,865 to 2,349 per day among women.

Table 6. Energy and nutrient intakes^* from the quantitative food frequency questionnaire among the main ethnic groups in the cohort, Hawaii and Los Angeles, 1993–1996.

Nutrient	Males					Females
	Latino	African-American	Japanese-American	White	Native Hawaiian	Latino	African-American	Japanese-American	White	Native Hawaiian
Calories	2,679.0	2,278.1	2,241.8	2,340.4	2,727.6	2,285.5	1,966.1	1,864.9	1,899.4	2,348.7
Protein (g)	102.2	85.6	82.0	83.3	95.7	88.3	77.6	67.7	71.3	85.1
Total fat (g)	98.1	85.7	70.1	80.9	91.6	82.8	72.5	59.6	65.3	79.6
P/S† ratio	0.77	0.83	1.01	0.84	0.89	0.80	0.86	1.03	0.83	0.91
Cholesterol (mg)	329.4	298.4	238.5	246.1	309.7	261.9	240.2	176.7	193.9	244.2
Carbohydrate (g)	335.6	276.0	306.8	290.5	362.9	304.9	254.5	269.9	254.2	326.6
Dietary fiber (g)	33.5	25.5	21.0	26.4	25.9	30.6	25.6	22.2	24.5	27.4
Vitamin C (mg)	194.9	167.8	152.5	182.9	192.0	211.7	187.3	180.5	180.0	218.4
Vitamin E (mg α-TE‡)	15.1	12.7	11.2	12.5	13.9	13.6	11.7	10.6	10.8	13.3
B-Carotene (meg)	5,226.3	4,532.3	4,455.8	4,878.4	5,441.3	6,218.7	5,427.9	5,503.2	5,437.7	6,486.4
Lycopene (meg)	4,514.2	3,622.8	2,791.6	4,562.3	3,915.3	5,565.7	3,499.7	2,554.0	3,552.8	3,673.8
Lutein (meg)	3,317.5	3,406.8	3,039.8	2,801.4	3,819.0	3,418.1	3,878.4	3,292.4	3,083.2	4,113.9
Folate (meg)	493.7	382.8	317.7	403.5	395.1	440.7	365.2	322.4	361.1	402.3
Calcium (mg)	1,056.2	771.0	634.6	882.1	834.8	965.3	745.8	633.8	843.8	847.4

^*Mean daily intake, standardized to the overall cohort age distribution.

^†P/S, polyunsaturated fat (g)/saturated fat (g).

^‡α-TE, a-tocopherol equivalents.

Several interesting ethnic variations are apparent in the table. Latinos and Native Hawaiians of both sexes have higher average intake of calories than the other ethnic groups; fat and protein intake are highest among the Latinos, whereas carbohydrate intake is highest among the Native Hawaiians. Latinos also have the highest intakes of cholesterol, dietary fiber, vitamin E, calcium, and folate compared with the other groups. In contrast, Japanese-Americans have notably lower intakes of cholesterol, fiber, lycopene, calcium, and folate relative to the other groups, and are the only group with a polyunsaturated/saturated fatty acid (P/S) ratio >1.0. African-Americans and Whites show the fewest extremes among the groups. Nevertheless, African-American men have the lowest intake of carbohydrate among males; White men have the highest intake of lycopene and the lowest intake of lutein among males; and White women have the lowest intake of carbohydrate, vitamin C, and lutein among females. The distribution of percent calories from fat also differs among the ethnic groups in the cohort (data not shown). Overall, Japanese-Americans have the lowest proportion of calories from fat and African-Americans the highest.

As shown in table 7, there are also notable differences among these groups in their intakes of particular foods and food groups that contribute to the intake of particular nutrients. For example, intake of red meat is higher among Latinos and Native Hawaiians compared with the other three groups. Processed meat is consumed in greater amounts by African-Americans and Native Hawaiians. Fish consumption is highest among the Japanese-Americans and Native Hawaiians. Whites and Latinos have high intakes of dairy products, especially compared with the Japanese-Americans. Intake of legumes is much higher among the Latinos than any other group, although tofu is consumed most by the Japanese-Americans. Latinos are the lowest consumers of green and cruciferous vegetables, whereas they (together with Native Hawaiians) are the highest consumers of yellow-orange and allium vegetables.

Table 7. Intake of selected foods^* from the quantitative food frequency questionnaire among the main ethnic groups in the cohort, Hawaii and Los Angeles, 1993–1996.

Food	Males					Females
	Latino	African-American	Japanese-American	White	Native Hawaiian	Latino	African-American	Japanese-American	White	Native Hawaiian
Red meat	72.9	54.4	49.0	51.2	68.4	54.0	38.5	36.9	35.0	52.6
Poultry	54.0	60.5	50.1	38.2	42.5	50.3	63.3	33.9	35.9	39.0
Processed meat	16.4	22.9	14.5	15.7	21.7	11.0	14.7	9.7	9.5	15.5
Fish	17.2	20.6	32.9	25.1	39.4	14.0	18.6	23.5	17.6	32.0
Dairy products	277.8	202.6	135.2	257.9	214.6	273.1	201.8	153.8	265.7	237.1
Rice	66.1	64.6	364.7	102.2	337.0	75.0	47.6	257.9	73.8	216.2
Potatoes/tubers	38.9	43.0	30.4	49.0	80.8	33.4	35.3	26.7	40.7	82.7
Legumes†	96.4	38.9	24.2	34.4	27.4	67.4	31.6	20.4	26.5	24.5
Tofu	2.9	1.9	22.0	7.4	14.7	3.0	1.7	23.9	5.4	16.1
Green vegetables	99.1	103.8	120.1	113.2	133.6	113.5	123.2	135.9	127.5	149.8
Yellow-orange vegetables	29.4	24.4	26.8	29.1	30.6	36.0	29.3	34.5	32.6	38.1
Cruciferous vegetables	34.2	35.6	44.9	40.4	46.4	38.3	47.0	49.2	43.2	51.4
Allium vegetables	11.8	7.2	8.0	7.9	8.9	11.3	7.7	8.4	7.4	9.1
Citrus fruit	155.8	146.2	102.3	129.0	123.4	170.4	157.1	130.3	119.4	141.3
Other fruit	214.5	170.0	148.4	197.6	188.1	236.4	212.0	196.6	202.2	241.6

^*Mean daily intake (g), standardized to the overall cohort age distribution.

^†Legumes excluding tofu.

Vitamin and mineral supplement use by members of the cohort shows some variation (data not shown). Usage is generally higher for the oldest age group (65-75 years) compared with younger ages. Overall, age-adjusted usage of any supplements (multivitamin preparations and/or single vitamins or minerals) among the cohort subjects is high: Latino men 55 percent, women 67 percent; African-American men 58 percent, women 69 percent; Japanese-American men 59 percent, women 76 percent; White men 62 percent, women 74 percent; and Native Hawaiian men 44 percent, women 57 percent.

Surveillance and follow-up

Long-term surveillance for cancer incidence and mortality in this cohort will be accomplished in several ways. Both Hawaii and California have state-wide population-based cancer registries and experience low out-migration rates in the adult population. Thus, incidence cases of cancer can readily be identified by linkage of the cohort to the respective registries in the two states. For cohort members known to have moved to another state, we can also check with the corresponding state registry for incident cancers. Mortality from cancer and other causes will be determined from routine linkages to the death certificate files in Hawaii and California; periodically, we will also use the National Death Index to identify deaths among cohort members who have moved to other parts of the United States.

In addition to these passive approaches, we will also conduct active follow-up on the cohort. We have already begun to send the participants annual newsletters which not only inform them about the progress of the study, but also enable us to update our address files. Less frequently, we plan to recontact the participants with follow-up questionnaires. These will seek information on intercurrent illnesses, as well as additional and updated exposure information, including diet. Planning for the first of these follow-up questionnaires is already underway.

Comment

This large prospective cohort for the study of diet and cancer is unique in several aspects, but particularly with respect to its ethnic composition. No other cohort of African-American men, Japanese-American women, or Latinos and Native Hawaiians of either sex yet exists for the study of chronic disease risk. An added advantage of this cohort is its broad representation of social strata. By using drivers' license files as the main sampling frame, with supplementation from other sources, we were able to include a diverse range of subjects within each ethnic and sex group. Indeed, despite the fact that the participants in this cohort study are volunteers, the distributions on educational level broadly resemble those reported by the US Census, so that findings from this study can be compared across ethnic and social strata and should be widely generalizable. Nevertheless, some selection bias is inevitable, as reflected in the data on current smoking in table 5. These rates are lower by 30–50 percent when compared with corresponding data from national surveys (21). Thus, we can expect a lower incidence of lung and other smoking-related cancers in the cohort compared with the general population of Hawaii and California.

Another diet cohort in the United States (22) has also used drivers' license files to identify study participants; other cohorts in the United States and elsewhere (23–27) used different sampling frames. Currently, several cohorts are being assembled across different populations and countries of Europe in order to investigate the role of nutrition in cancer (28). Because the methodologies for sampling and data collection vary across the participating centers, a calibration study, based on 24-hour diet recalls, has been included (29). This is analogous to the calibration study included in the present cohort, which is designed both to reduce measurement error and to permit valid comparisons across the different populations (19).

Our scheme for ethnic-specific sampling using surnames (and in some instances first names) or census tracts proved to be quite successful. Of the participants identified a priori as Latino, 78 percent self-identified themselves as Latino on the questionnaire. Corresponding figures were 85 percent for African-Americans, 94 percent for Japanese-Americans, and 83 percent for Native Hawaiians.

Some baseline findings with regard to non-dietary exposure variables are shown in table 5. The well-known association of cigarette smoking with lung cancer risk is demonstrated by an ecologic comparison of the cigarette smoking data in table 5 with the ethnic-sex-specific lung cancer incidence rates in table 1. African-Americans and Native Hawaiians have the highest percent of current cigarette smokers and the highest lung cancer incidence rates in the cohort for men and women. Japanese-American and Latino women have the lowest percent of current smokers and the lowest lung cancer incidence rates.

The substantial variation in alcohol consumption and obesity (table 5) will enable us to examine relations of diet and other risk factors to cancer within strata of these variables. The obesity data show that a substantial proportion of cohort subjects are overweight, similar to the general US population (30). It will be important in future analyses of the relation between obesity and cancer or other outcomes (31–34) to try to separate the effects of obesity per se from those of excessive energy intake.

We have validated some of these non-dietary data. For example, reports of physical activity on the questionnaire agreed closely with those obtained by direct interview in a subset of subjects who also participated in a case-control study of colorectal cancer.

As shown in tables 6 and 7, there is notable variation among the ethnic groups in intake of both foods and nutrients. Ecologic comparisons of these findings with the corresponding ethnic-sex-specific cancer incidence rates in table 1 already suggest possible relations worth testing at the individual level as cases accrue in the cohort over the next several years. For example, many reports have suggested that the intake of fiber may protect against colorectal cancer (35–37). Intake of dietary fiber, and of legumes, an important food source of fiber, is much higher among the Latinos, whose incidence of colorectal cancer is much lower compared with the other ethnic groups. Again, some studies (38–42), but not all (43, 44), have suggested that calcium intake may be inversely related to colon cancer; limited data also suggest that dietary folate may be protective against this type of cancer (45, 46). In our cohort, Latinos have the highest intakes of both calcium and folate and the lowest incidence of colon cancer.

The ecologic data are less consistent with regard to certain other hypotheses. For example, some researchers have suggested that soy products, because of their content of isoflavones such as genistein and daidzein, may be protective against breast and prostate cancers (47, 48); there are limited data to support the hypothesis (49, 50). Similarly, a high intake of the omega-3 polyunsaturated fats found in fish has been postulated to protect against cancer (51). Native Hawaiian men have the highest intake of fish and a relatively high intake of tofu, and their rate of prostate cancer is the lowest of the five ethnic groups. On the other hand, among women, the consumption of tofu is highest in the Japanese-Americans and their consumption of fish is also high, but their rate of breast cancer is higher than that of Latino women, whose tofu and fish consumption is low. Interestingly, though, consumption of other legumes among Latinas is especially high. Thus, these ecologic observations suggest that there may be multiple factors involved in protection against these cancers.

Several carotenoids have been found to be protective against lung cancer in epidemiologic studies (52–54), though intervention trials using beta-carotene in heavy smokers (55, 56) and physicians (57) have shown no benefit, and possible harmful effects. The variations among the ethnic groups in their intake of the different carotenoids might enable us to identify which, if any, of these constituents has a beneficial effect on lung cancer risk, and why intake of fruits and vegetables shows a consistent inverse association with the risk of this cancer (58). There is growing recognition of the importance of physical activity in relation to cancer risk. Interestingly, African-American and Japanese-American men and women have the lowest levels of vigorous physical activity (table 5) and the highest rates of colorectal cancer (table 1). Physical activity has been convincingly related to a lower colon cancer risk (59), although no particular type of activity has yet been established as most beneficial. It will be important in future prospective analyses of the cohort to distinguish between activities with high energy expenditure of relatively short duration and moderate activities of longer duration (60).

Future work

Over the next few years, sufficient numbers of cases of the common cancer sites (especially lung, colorectum, breast, and prostate) will accrue for prospective analysis. We plan to investigate the role of nutritive constituents such as fat, carotenoids, tocopherols, fiber, and phytoestrogens on risk, as well as the effects of cooking methods, physical activity, obesity, and other potential risk factors. The multiethnic make-up of the cohort offers some unique advantages for these investigations. It will enable us to confirm in four minority ethnic groups dietary (as well as other) relations that have previously been reported in largely White populations. The multiethnic composition also contributes to study power: each ethnic subcohort has substantial within-group variability and also different ranges in exposures, introducing between-group variability in dietary and non-dietary exposures. This combination of within-and between-group variability of intake means that, if dietary factors are important in cancer causation, we should expect to see two somewhat distinct forms of evidence (61). First, individual consumption of particular nutrients or foods will be predictive of cancer risk consistently when the subcohorts defined by gender and ethnic group are analyzed separately. Second, the overall (age-specific) rates of cancer for each of the ethnic groups will show a general relation with average nutrient or food intake, after adjusting for other risk factors (an “ecologic analysis” with adjustments for other risk factors). One can then investigate consistency of effect across ethnic groups, and the extent to which differences in risk among ethnic groups are explained by the differences in food or nutrient consumption.

Because we have quantitative data on amounts of foods and dietary components consumed (based on portion-size information), we may be able to determine optimal intakes for any beneficial effects we observe, thereby contributing further to public health guidelines. This cohort also lends itself to interesting opportunities for studying the relation of genetic factors to cancer risk. Of particular interest are differing genetic susceptibilities to dietary and other environmental risk factors. Polymorphisms in metabolizing enzymes, for example, distribute differentially across ethnic groups (62–65) and this may help to explain differing risks of cancer among these groups that are not fully explained by differences in levels of exposure to environmental risk factors (66, 67). Currently, we are collecting blood samples from a cross-section of cohort members and from cases of particular cancers (breast, prostate, colon) as part of a separate study. Findings from this large cohort over the next several years should contribute not only to general knowledge on the relation to cancer of diet, other etiologic factors, and gene-environment interactions, but also to addressing the role of these factors in explaining differences in the risk of cancer among diverse ethnic and minority populations in the United States.

Abbreviations

BMI

body mass index

HCFA

Health Care Financing Administration

MDR

minimal detectable relative risk

QFFQ

quantitative food frequency questionnaire