Why Are Thyroid Cancer Rates So High in Southeast Asian Women Living in the United States? The Bay Area Thyroid Cancer Study¹

Variation in thyroid cancer incidence by gender and racial/ethnic group is striking, with rates being particularly high in women and in Filipinos and SAs³ living in the United States (1, 2, 3, 4). In contrast, Chinese and Japanese populations in the United States overall have rates comparable with the rates for United States Caucasians (Fig. 1). However, the pattern is complex, with thyroid cancer incidence rates among Asian migrants to the United States often being higher than those in their native or host countries (4, 5, 6). In the San Francisco Bay Area, rates of thyroid cancer in Filipino migrants are twice as high as those observed either among women in the Philippines or among Caucasian women (1, 4).

Established risk factors for thyroid cancer include ionizing radiation and a history of benign proliferative thyroid diseases, including goiter and thyroid nodules (7). Although iodine and seafood consumption (as a proxy for iodine exposure) have been investigated as potential risk factors for thyroid cancer, several recent studies have suggested that iodine intake may be, at most, weakly related to thyroid cancer risk (8, 9). The higher incidence of thyroid cancer in women than in men has prompted research into the role of hormonal and reproductive factors in thyroid carcinogenesis. Although much of the research investigating these associations provides only weak evidence for a relationship between standard menstrual and reproductive factors and thyroid cancer risk, recent studies suggest that a recent pregnancy may increase risk among premenopausal women (10, 11, 12, 13). Protective factors against thyroid cancer may include the consumption of cruciferous vegetables, antioxidant vitamins, and phytoestrogens (14, 15, 16, 17).

The reasons for the high incidence rates of thyroid cancer in Filipino and other SA women residing in the United States are not understood. The purpose of this analysis was to estimate, within a single geographic area, what proportion of the differences in incidence rates between SA women and women of ethnic groups with lower incidence rates (NA and Caucasian women) can be attributed to differences in the prevalence of thyroid cancer risk factors between the populations.

The current analysis was conducted within the framework of the Bay Area Thyroid Cancer Study, a population-based, case-control study of women aged 20–74 years, whose methods have been described elsewhere (8). Briefly, cases were identified through the Greater Bay Area Cancer Registry of the Northern California Cancer Center, which is part of the Surveillance, Epidemiology, and End Results program and the California Cancer Registry. Women with an incident thyroid cancer diagnosed between June 1, 1995 and May 31, 1998 were eligible for the study. In addition, Asian women diagnosed between June 1, 1992 and May 31, 1995 were included because of particular interest in this group. Of 817 cases identified, 608 (74%) were interviewed, 106 (13%) declined to participate, and 103 (13%) were not interviewed for other reasons.

Controls were identified through random digit dialing conducted between 1995 and 1998. Of 5002 known residences contacted, 3928 (79%) were successfully enumerated. From these enumerations, controls were frequency matched to cases on age (5-year age groups) and racial/ethnic group (non-Latina Caucasian, African-American, Latina, Asian-American, or Native-American). Of 793 controls selected, 558 (70%) were interviewed, 154 (19%) declined to participate, and 81 (10%) were not interviewed for other reasons.

A standardized structured questionnaire was used to collect information, including dietary intake, migration and ethnic identification, benign thyroid disease, reproductive and menstrual history, and medical history (8). Interviews were conducted in six languages (i.e., English, Spanish, Tagalog, Vietnamese, Cantonese, and Mandarin) by trained bilingual, bicultural interviewers.

The present analysis was limited to Caucasian and Asian women (n = 515 cases and 486 controls). Asian participants were classified as SA (i.e., high risk) or NA (i.e., low risk) based on self-identification. For the few women whose self-reported ethnicity was ambiguous or comprised two or more ethnic groups, the birthplace of the subject and those of her parents and grandparents were used to classify her into one of the two Asian subgroups. The SA group (n = 214) consisted primarily of Filipino (n = 143; 67%) and Vietnamese (n = 55; 26%) women, as well as Thai (n = 3), Indonesian (n = 2), and Pacific Islander (n = 11) women. The NA group (n = 196) consisted primarily of Chinese (n = 127; 65%) and Japanese women (n = 37; 19%) and also included Korean (n = 16) and Asian-Indian (n = 16) women. Non-Latino Caucasian women (n = 592) were included in the analysis as a second comparison group (low risk).

Attributable risk analyses focused on factors that were found to be associated with thyroid cancer risk in this case-control study (8, 12, 16). These factors included radiation treatment to the head or neck ≥5 years before the cancer diagnosis (for cases) or the date of selection for the study (for controls); a personal history of goiter or thyroid nodules diagnosed ≥2 years before cancer diagnosis/selection; a family history of proliferative thyroid disease (i.e., thyroid cancer, goiter, or thyroid nodules); education; dietary factors, including consumption of phytoestrogens (isoflavones and lignans), fiber, and antioxidant vitamins (vitamin C, vitamin E, and carotenoids); hormonal factors, including age at menarche, recency of last pregnancy, and OC use; and for Asian women, their percentage of lifetime spent residing in the United States

Continuous variables were categorized into percentiles or a priori groups based on the distribution among all Asian and Caucasian controls combined. Unconditional logistic regression, adjusted for age, was used to estimate ethnic-specific ORs for each factor of interest. Breslow-Day tests for homogeneity of ORs across ethnic groups were performed (18). Because none of these tests showed departure from homogeneity, common (i.e., ethnicity adjusted rather than ethnic specific) ORs were used in RAR calculations. Unconditional logistic regression was used to estimate the common OR and 95% CIs for two age groups (<50 and ≥50 years), adjusting for age as a continuous variable and race/ethnicity. For dietary variables, age, race/ethnicity, and average daily caloric intake were adjusted for. RAR analyses for dietary variables compared women in the lowest two tertiles of consumption with those in the highest tertile.

RARs were calculated using methods described by Breslow and Day (18), and CIs were calculated using the bootstrap approach with 2000 replications (19, 20). The RAR indicates how much of the difference in incidence rates between two populations can be attributed to the difference in patterns of exposure to a particular risk factor or combination of risk factors, hence, the proportion of disease excess in a high-risk population that is attributable to greater exposure to certain risk factors. High-risk women (i.e., SAs) were compared with two reference groups (i.e., NA women and Caucasian women). Each comparison was made for younger, largely premenopausal women (20–49 years old) and older, largely postmenopausal women (50–74 years old) to account for the change in slope around age 50 in the overall age-specific incidence rates of thyroid cancer.

To select the variables to include in the composite RAR estimates, a priori criteria were set: (a) for each comparison, variables with an OR > 2, a Wald test P < 0.1, and a positive individual RAR were identified; (b) logistic regression was used to determine which variables, in combination, were significantly associated with thyroid cancer risk at a P < 0.05 level; (c) a composite RAR was calculated based on these variables; and (d) a “reduced” composite RAR was calculated, which represented the most parsimonious combination of variables that still explained a maximum of the variation in incidence rates.

For some variables, the RARs were negative. When using common ORs, negative RARs occur when the prevalence of the risk factor is higher among the low-risk group than the high-risk group. Thus, these factors do not explain the higher incidence of disease in the high-risk group but suggest that incidence differences would be greater if the distribution of the factor of interest among the high-risk group more closely resembled that of the low-risk group. In the present analysis, these factors were not considered further.

Table 1 presents the common ORs for the primary risk factors of interest by age group, and Table 2 presents the prevalence of these factors among controls by age and ethnic groups. A personal history of goiter or thyroid nodules and a family history of proliferative thyroid disease were most strongly associated with thyroid cancer risk (OR = 4.3, 95% CI 2.6–7.2 and OR = 3.5, 95% CI 1.9–6.2, respectively). The prevalence of goiter/nodules was higher in SA women (11%) than among NA (6%) and Caucasian women (5%), whereas prevalence differences for a family history of proliferative thyroid disease were small: (a) SA (6%); (b) NA (6%); and (c) Caucasians (4%). Among parous women, two or more pregnancies in the previous 5 years were associated with an ∼2-fold increase in risk of thyroid cancer (OR = 1.8; 95% CI 1.1–3); however, this event was more common among Caucasian (19%) and NA women (14%) than SA women (10%). Lower consumption of phytoestrogens, antioxidant vitamins, and fiber were all associated with a 1.5–2-fold increase in risk compared with the highest tertile of consumption. Caucasian women consumed lower levels of isoflavones and higher levels of antioxidant vitamins and fiber than both NA and SA women.

Table 3 presents the RAR for each factor of interest taken individually and for a composite of selected variables. Because of the small size of the subgroups examined, many of the RAR estimates are very imprecise, as indicated by wide CIs, including the value of 0% (Table 3). A history of radiation treatment to head or neck was not included in these calculations because of the very small numbers of exposed women (Table 2). Considered independently, the higher prevalence of goiter/nodules among SAs accounted for 27% of the difference in thyroid cancer incidence rates compared with NA women and 40% of the difference compared with Caucasian women. A family history of proliferative thyroid disease accounted for only 1% of the difference in incidence between SAs and NAs but explained 13% of the difference when comparing young SA with Caucasian women. Lower consumption of isoflavones among SA women explained 36% of the difference in incidence between this group and NAs. The lower consumption of antioxidant vitamins also explained a portion of the difference in rates when Caucasian women were the comparison group.

Composite RARs were determined taking joint distributions of variables into account. Goiter or thyroid nodules, a family history of proliferative thyroid disease, and low consumption of isoflavones and vitamin E accounted for 74% of the incidence rate difference between young SA and NA women. Goiter/nodules and isoflavones account for the majority (66%) of this difference. Goiter/nodules and lower consumption of carotenoids accounted for the majority (67%) of the difference between SA and Caucasian women.

As observed in younger women, goiter or thyroid nodules were most strongly associated with thyroid cancer risk (OR = 4.6, 95% CI 2.2–9.5; Table 1). The prevalence of this condition among controls was higher among SAs (13%) than among either NAs (4%) or Caucasians (7%; Table 2). There was a >2.5-fold increase in thyroid cancer risk in women with less than a high school education and among recent immigrants, with the prevalence of these factors being substantially higher among SA women compared with NA and Caucasian women. There was an increased risk of thyroid cancer in women with either an early or late age at menarche; a larger proportion of SA women had a late age at menarche (42%) compared with Caucasian women (19%). Women who never used OCs were at an increased risk for thyroid cancer; however, no dose-response relationship was observed for OC use (data not shown). Most SA women had never used OCs (74%), whereas most Caucasian women had (73%).

For older women, 37% and 23% of the difference in thyroid cancer incidence between SA women and NA and Caucasian women, respectively, were explained by the higher prevalence of goiter/nodules among SA women (Table 3). Education and OC use accounted for a substantial amount of the difference in both comparisons. About 50% of the difference in incidence between SA and NA women were associated with recent immigration. Age at menarche explained a minimal amount of the difference in incidence in either comparison. With the exception of isoflavones and carotenoids, the RARs for dietary factors in the older women for the NA and Caucasian comparisons, respectively, were either small or negative.

Taking their joint relationships into account, goiter/nodules, low consumption of isoflavones, and recent migration accounted for 95% of the difference in thyroid cancer incidence between SA and NA women. When comparing SA and Caucasian women, goiter/nodules, never using OCs, and lower education accounted for 81% of the difference in incidence.

To our knowledge, this is the first study that has directly examined which factors might explain the excess incidence of thyroid cancer in SA women. Unlike the standard population attributable risk, which represents the proportion of cases occurring within a single population that can be explained by a given risk factor, the RAR estimates how much of the difference in incidence between two populations is explained by the difference in patterns of exposure to a particular risk factor (18). Given the consistent and striking differences in thyroid cancer incidence rates between SA women living in the United States and other United States women, the RAR is a potentially informative tool for explaining the notable excess risk seen in this group. Indeed, the data from this study indicate that although a number of factors are independently associated with thyroid cancer risk, the higher prevalence of benign proliferative thyroid disease (i.e., goiter or thyroid nodules) among SA women compared with other women explains a substantial proportion of the higher incidence of thyroid cancer in this group, regardless of age.

Dietary differences also play a role, with isoflavone and carotenoid consumption perhaps most important. Isoflavones are weak estrogenic compounds found in plants or derived from plant precursors. In addition to antioxidant effects, they have been shown to exhibit antiestrogenic effects and inhibit the growth and proliferation of estrogen-dependent cancers (21, 22). Isoflavones are found primarily in soy-based foods, such as tofu and soy milk, but are also present in smaller amounts in Western-style foods containing added soy flour or soy protein, including some brands of doughnuts, white bread, and canned tuna (23). Carotenoids are found primarily in orange or yellow and dark green vegetables and have strong antioxidant effects (24).

Among older Asian women, in addition to goiter/nodules and isoflavone consumption, recent migration, OC use, and education are explanatory. It is likely that these last two factors are measures of SES. Several studies in different populations have found decreased risks of thyroid cancer associated with OC use but consistently have not observed a dose-response relationship (11, 25, 26). Thus, the ever/never use of OCs might simply be a reflection of SES or lifestyle differences between cases and controls rather than an indicator of the impact of hormones on thyroid cancer risk. In the Southeast and NA comparison, the contribution of recent migration may reflect the same SES influences as OC use and education in the SA and Caucasian comparison, because these latter two factors had larger individual RARs but dropped out of multivariate models. In addition, the recent migration variable may reflect the influence of some aspect of early lifestyle or exposure in southeastern Asia affecting the risk of thyroid cancer.

The RAR approach helps delineate areas where interventions might be most effectively targeted to have the greatest impact on reducing the elevated thyroid cancer rates in SA women. Increasing the consumption of isoflavone- and carotenoid-rich foods may be particularly beneficial. Similarly, it demonstrates that other risk factors for thyroid cancer, such as a family history of proliferative thyroid disease, although a consistent predictor of risk in studies of thyroid cancer, actually account for very little of the ethnic differences in incidence rates. This analysis also suggests that early life exposures that are more common in Asia than the United States warrant further exploration.

Our results, however, should be interpreted with caution. The statistical power of this study is limited by the relatively few women in each ethnic group, particularly for the analyses of women aged ≥50. This limited power impacted the ability to detect departures from homogeneity when applying the Breslow-Day tests for homogeneity of the OR. Not surprisingly, no significant departures were observed. As a result, common ORs were assumed in all RAR calculations, thereby limiting RAR interpretation to the influence of prevalence differences and masking any true relative risk differences that may have existed between groups. However, visual inspection of ethnic-specific ORs, suggested only minimal differences between them. Thus, this limitation may have only minimal impact on our understanding of the incidence differences.

The limited statistical power also made compromises necessary in the computation of the composite RAR, e.g., we used the most parsimonious model to minimize the variance in its calculation. Fortunately, the parsimonious model should reflect the stronger risk factors for thyroid cancer. Other compromises included combining tertiles of dietary intake into dichotomous variables to avoid zero cell counts and obtain more stable RAR estimates, thus possibly decreasing explanatory potential. Limited power also resulted in substantially wide CIs for the RAR estimates, reducing the precision of some of our findings.

As in all retrospective studies, exposure information from self-reported medical history, family medical history, and dietary recall is subject to error and recall bias. In the original case-control study, measures were taken to minimize potential biases, including the following: (a) the identification of cases within several months after diagnosis minimizing the time between diagnosis and interview; (b) the use of visual aids in estimating portion size of foods; (c) trained bilingual, bicultural interviewers; and (d) when collecting information on benign thyroid disease, for each diagnosis reported, associated symptoms were inquired about, and these reports were reviewed by a clinician to verify consistency of symptoms and diagnosis.

Finally, the RAR assumes controls are reasonably representative of the population. Although the functional response rate for controls in this study was 55% (i.e., 79% random digit dialing enumeration × 70% interview response rate), interviewed controls were similar to the 79% identified on 5-year age group and ethnicity, including the distribution of specific Asian subgroups. Thus, some concerns about selection biases are lessened.

Despite these limitations, the RAR approach has provided some insights into thyroid cancer risk over and above what has been concluded from standard case-control analyses of risk factors. In summary, the high thyroid cancer incidence rates in SA women in the United States can be largely attributed to the higher prevalence of goiter or thyroid nodules in this group compared with others. This was observed consistently in both younger and older women and whether the comparison group was NA or Caucasian women. Dietary factors also seem to have a significant impact on the observed difference in thyroid cancer risk and are areas that might be most effectively targeted in prevention efforts. The impact of recent migration among older women may reflect the influence of yet unidentified early life exposures more common in Southeast Asia than the United States and warrants further exploration, as does the influence of socioeconomic factors.