Studying Cancer Incidence and Outcomes in Immigrants: Methodological Concerns

The study of disease patterns in immigrant populations has become an increasingly important public health activity in the United States. For example, in 2000, 28.4 million foreign-born individuals resided in the United States, representing 10.4% of the population.1

The population-based Surveillance, Epidemiology, and End Results (SEER) program represents an important resource for the study of cancer incidence and outcomes in immigrant and minority racial/ethnic populations. Information on birthplace, an indicator of migrant status, is routinely collected by SEER registries and is generally derived from a patient’s medical records or death certificate.2 Unfortunately, these data are problematic for 2 reasons. First, to our knowledge, birthplace information in these registries is not validated, and thus its accuracy is unknown. Second, birthplace information is missing for a substantial proportion of patients, and its level of completeness continually decreases.3

Despite these methodological problems, this variable continues to be used in analyses of cancer incidence among Asian subgroups and their descendants,4–11 examinations of the relationship of birthplace to tumor size and stage among Asians12 and Hispanics,13 and investigations of the effect of birthplace on breast cancer survival among Asians.14 Hoping to inform and improve the quality of future analyses of birthplace as it relates to cancer patterns, we outline here some of our ongoing research on the utility of the birthplace variable in the San Francisco/Oakland and San Jose/Monterey SEER registries and discuss the results of previously published studies in the light of our findings involving this variable.

We found that 67% of patients diagnosed in the period 1988 to 1996 had complete birthplace information in the registries.2 This percentage varied by race/ethnicity, with Vietnamese and other Southeast Asians having the highest level of complete data (88.5%) and non-Hispanic Whites having the lowest (65.3%). The most significant factor associated with birthplace completeness was vital status; information from death certificates can supplement existing information by supplying missing birthplace data.

In addition, completeness of the birthplace variable in the registries examined was not random but varied according to characteristics such as patient age, year of diagnosis, and hospital teaching status, ownership status, and size. Preliminary results from an ongoing study comparing registry data and selfreported birthplace revealed that the sensitivity and positive predictive value of birthplace data among specific Asian subgroups were quite high, indicating generally good validity (Table 1 ▶). Overall, however, Asians with no birthplace information were less likely to be foreign born than Asians with birthplace information (Table 2 ▶). This pattern was seen across all Asian subgroups except for Vietnamese, of whom the vast majority were foreign born.

TABLE 1.

—Validity of Registry Data Birthplace Information, by Asian Subgroup: Greater Bay Area Cancer Registry, 1988–2000

	All Asians Combined	Chinese	Japanese	Filipino	Vietnamese	Other
Measure of Validity	(n = 1183)	(n = 690)	(n = 52)	(n = 96)	(n = 215)	(n = 130)
PV+: foreign borna	98.8	98.5	84.6	100.0	100.0	98.2
PV+: US bornb	92.3	90.8	100.0	100.0	. . .c	90.0
Sensitivity: foreign bornd	98.9	98.9	100.0	100.0	99.1	98.2
Sensitivity: US borne	91.0	88.5	95.1	100.0	. . .	90.0

Note. PV+ = positive predictive value.

^aAmong those classified as foreign born, the percentage who self-identified as such.

^bAmong those classified as US born, the percentage who self-identified as such.

^cNot reported for fewer than 5 US-born persons.

^dAmong those self-identified as foreign born, the percentage who were classified as such.

^eAmong those self-identified as US born, the percentage who were classified as such.

TABLE 2.

—Birthplace Distribution (Percent Foreign–Born) Among Participants With Known and Unknown Registry Birthplace Information, by Asian Subgroup: Greater Bay Area Cancer Registry, 1988–2000

	All Asians Combined	Chinese	Japanese	Filipino	Vietnamese	Other
	(n = 1830)	(n = 1078)	(n = 104)	(n = 196)	(n = 241)	(n = 211)
Known birthplace, no.	1183	690	52	96	215	130
Registry classification, % (Pa)	87.7 ( < .001)	88.7 ( < .001)	21.2 (.807)	93.8 (.238)	100.0 (1.0)	84.6 (.002)
Self-report, % (Pa)	87.9 ( < .001)	89.0 ( < .001)	25.0 (.478)	93.8 (.238)	99.1 (.621)	84.6 (.002)
Unknown birthplace, no.	647	388	52	100	26	81
Self-report, %	74.3	77.8	19.2	89.0	100	66.7

^aFor χ² test (vs unknown birthplace).

Therefore, incidence rates computed under the assumption that patients with no birthplace data have the same birthplace distribution as those with complete data4–11,15 will represent overestimates for foreign-born Asians and underestimates for US-born Asians. For example, in a survival analysis conducted by Pineda et al.,14 inflation of the hazard rate of death due to breast cancer was greater for US-born Asians than for their foreign-born counterparts, because patients with missing birthplace information were excluded from the analysis (and because complete birthplace data are more likely to be available for deceased patients than for patients who are alive).

Similarly, in 2 recently published studies examining birthplace in relation to breast cancer tumor size and stage among Asians12 and Hispanics,13 24.2% of Asians and 37.3% of Hispanics with unknown birthplace data were excluded. As a result, the findings of these studies are subject to selection bias in that completeness of birthplace information was likely to be associated with the outcomes under investigation, given that breast cancer patients at advanced stages and with larger tumor sizes generally have worse survival rates than patients at earlier stages and with smaller tumor sizes.

Inclusion of birthplace information in research designs can result in the identification of vulnerable subpopulations to be targeted for public health services, and such data can provide etiological insights as well. However, if this increasingly important information is to be useful, hospitals need to adopt policies aimed at improving its collection; the quality and completeness of birthplace data can be improved only at the level of the reporting facilities.