- Email: [email protected]
The Risk of Invasive Cervical Cancer among Hispanics: Evidence for Targeted Preventive Interventions
27, 674–680 (1998) PM980343
PREVENTIVE MEDICINE ARTICLE NO.
The Risk of Invasive Cervical Cancer among Hispanics: Evidence for Targeted Preventive Interventions1 Susan L. Howe, M.P.H.,* Ralph J. Delfino, M.D., Ph.D.,*,†,2 Thomas H. Taylor, Ph.D.,† and Hoda Anton-Culver, Ph.D.† *Graduate School of Public Health, San Diego State University, San Diego, California 92182; †Epidemiology Division, Department of Medicine, College of Medicine, University of California at Irvine, Irvine, California 92697
Background. Mass screening for cervical cancer has considerably reduced invasive cancer rates; however, Hispanic Americans have not fully benefited. This study examined the relationship of demographic, socioeconomic, and acculturation factors to diagnostic stage prior to preventive interventions in a large Hispanic community (San Diego, CA, 1988–1991). Methods. Age-adjusted incidence rates (IR) were estimated for three subareas: Hispanic community targeted with interventions, comparison Hispanic community, and remainder of county. Incident cases of cervical cancer in the California Cancer Registry were used to calculate incidence rate ratios (IRR) of in situ to invasive stage cancer and estimate differences in early detection between Hispanics and white non-Hispanics. Multiple logistic regression was used to examine the relationship of cancer stage to census-linked income, education, poverty status, foreign birth, and language spoken. Results. The in situ cervical cancer IR is significantly lower for Hispanics than for white non-Hispanics in all three regions (P , 0.05). The overall IRR is 3.65 (95% CI 3.19, 4.18) for white non-Hispanics compared with 1.09 (95% CI 0.86, 1.38) for Hispanics. Hispanics in census tracts in the highest versus remaining quartiles for percentage of language-isolated households are at greater risk for invasive stage (OR 5 2.10, 95% CI 1.27, 3.40). White non-Hispanic women residing in census tracts in the highest versus lowest quartile of adults
1 This research was supported by the Cancer Surveillance Program of Orange County and the San Diego/Imperial Organization for Cancer Control. Regional cancer incidence data have been collected by the Epidemiology Division, University of California, Irvine, under Subcontract 050K8707-8710. The subcontract is supported by the California Department of Health Services and its agent, the California Public Helath Foundation, as part of its statewide cancer reporting program, which is mandated by Health and Safety code Section 103875 and 103885. 2 To whom reprint requests should be addressed. Fax: (714) 8244773. E-mail:[email protected].
with less than 12 years education are at greater risk for invasive stage (OR 5 2.10, 95% CI 1.31, 3.38). Conclusions. Results show efforts aimed at cervical cancer early detection for Hispanic women are needed and well targeted. White non-Hispanic women in loweducation communities are also at risk. Populationbased methods in this study can focus efforts in communities needing preventive interventions. q1998 American Health Foundation and Academic Press
Key Words: cervical cancer; Hispanic American; cancer staging; mass screening.
INTRODUCTION
When cervical neoplasms are detected early and treated promptly, the cure rate is high. Consequently, in developed countries in which mass screening has been implemented, mortality rates due to cervical cancer have dropped dramatically [1,2]. However, the success of mass screening has not been uniform in the United States. Within certain subpopulations, such as racial or ethnic minority, low-income, or less educated women, incidence and mortality for invasive cervical cancer notably exceed those for white, higher income, and better educated women [3–5]. Four programs responding to U.S. federal initiatives for improved breast and cervical cancer screening among Hispanics and low-income women [5] were initiated in San Diego County beginning in 1992: Por la Vida, Companeras en la Salud, En Accion, and the Breast and Cervical Cancer Control Program of the State of California (BCCCP) [6,7]. These health-promotion programs were designed to overcome socioeconomic and behavioral barriers to screening among Hispanic and low-income women [6–8]. The BCCCP also provided clinic sites where Pap smears were performed free to low-income women. Since the goal of programs such as these is to reach specific underscreened populations, methods are needed to target intervention sites and evaluate progress. Although, the authors of this paper
674
0091-7435/98 $25.00 Copyright q 1998 by American Health Foundation and Academic Press All rights of reproduction in any form reserved.
STAGE AT DIAGNOSIS OF CERVICAL CANCER AMONG HISPANICS
are not involved in the above programs, we had access to the necessary data needed to apply such methods of identifying populations at risk. We examined the degree to which the populations targeted in San Diego were in need of screening and whether other populations should be targeted. The period examined was the 4 years (1988–1991) prior to the commencement of the preventive interventions. We examined both the population distribution and the demographic determinants of stage at cervical cancer diagnosis for Hispanic and white non-Hispanic women. Few other population-based studies have examined this within a large Hispanic population in the United States [9–11]. This is the first of such studies in San Diego County examining the relationship of ethnic, socioeconomic, and acculturation factors to stage at cervical cancer diagnosis. METHODS
Overview The present study consists of two complementary analyses. The first analysis stratifies subjects by cancer stage, ethnicity, and geographic area within San Diego County, in order to estimate the incidence of cervical cancer and the level of early detection in the subpopulation of Hispanics targeted for preventive intervention as well as Hispanics in a comparison area and in the remainder of the county (described below). Stage-specific incidence rates for Hispanics are compared with rates for white non-Hispanics in all three geographic areas. The second analysis uses logistic regression to examine the relationship of socioeconomic and acculturation factors to the probability of invasive stage compared with in situ stage at diagnosis for Hispanics and white non-Hispanics. Using the results of these analyses, census tracts with higher proportions of risk factors for invasive stage diagnosis are located using computerized mapping tools. Data Sources The two databases used for this study were the 1990 U.S. Census [12] and the San Diego/Imperial County Organization for Cancer Control (SANDIOCC), which is part of the California Cancer Registry (CCR) [13]. Population count data required for denominators were taken from the 1990 100% count of persons, using the U.S. Census Summary Tape Files 3A (STF3A), which report counts at the census tract level stratified by race/ ethnicity, gender, and age. The SANDIOCC data are collected at the Epidemiology Division, Department of Medicine, University of California, Irvine. The CCR sets uniform recording standards and quality control for the state [14]. Case ascertainment is estimated to
675
exceed 97%. Histopathological data sufficient to code the cases as in situ or invasive were available for all reported cases. All cases of neoplasms of the cervix uteri occurring in San Diego County between 1988 and 1991 were extracted from the SANDIOCC data base. Information on patients included the histopathology, stage at diagnosis, census track of residence at time of diagnosis, ethnicity, and place of birth. This study was granted exempt status by the University of California, Irvine Human Subjects Review Board, approval HS 95-005. Cervical cancer cases included both squamous cell carcinoma and adenocarcinoma and were divided into in situ and invasive stage at diagnosis. In situ stage cancers included neoplasms of the cervix uteri that met all microscopic criteria for malignancy, except invasion of the basement membrane, including cervical intraepithelial neoplasia, grade III. Invasive cases included malignant neoplasms of the cervix uteri that displayed any degree of invasion, including local microinvasion of the basement membrane, as well as regional and distant spread. Analysis of the Incidence of Cervical Cancer by Stage, Ethnicity, and Geographic Area Stratification by ethnicity. The two largest ethnic subpopulations in San Diego are Hispanics, comprising 20%, and white non-Hispanics, comprising 67% of San Diego’s 2.5 million population [12]. The majority of Hispanics in San Diego County are of Mexican descent (87%) [8,12]. The Census uses self-identification to classify race/ ethnicity. The CCR classifies a cancer case Hispanic if so indicated by the medical record or if the person’s last name appeared on the 1980 U.S. Census list of Spanish surnames, excluding persons classified as Asian/other (i.e., Filipino or American Indian). White non-Hispanics include those in the white race category excluding Hispanics [15]. Geographic stratification. Using census data, census tracts with populations of over 20, 40, 60, and 80% Hispanic ethnicity were identified. The locations of all intervention sites were then overlaid on these census tract percentages. This two-step process provided the basis for dividing San Diego County into three geographic areas. We assumed that the impact of the four intervention programs is largely limited to the geographic locations of intervention sites. This is justified since the programs focused on the use of local clinic outreach and local social networks within the Hispanic communities to promote both cervical cancer and breast cancer screening among women. One program (En Accion) targeted Hispanic men and women for five screenable cancers including cervical cancer.
676
HOWE ET AL.
The three geographic areas identified were: (1) a “target area” [high concentration of Hispanics ($20% in many census tracts) and the majority of the preventive intervention activities], (2) a “comparison area” (high concentration of Hispanics, but little intervention activity planned), and (3) a “remainder area” (the rest of the county throughout which Hispanics tend to be more widely dispersed). Census tracts were included in clusters called subregional areas (SRAs) since the San Diego Department of Health Services uses SRAs for planning their public health programs annually. The target area was composed of six SRAs extending to the Mexican border and containing approximately 50% of the Hispanics living in San Diego County. There were 32 health promotion sites and 5 BCCCP clinic sites contained in this region, compared with only 1 health promotion site and 3 BCCCP clinic sites located elsewhere in the county. The SRAs comprising the comparison area were selected based on three criteria: (1) a high concentration of Hispanics, (2) Hispanics with similar incomes to Hispanics in the target area, and (3) a minimum of intervention activity. The selected comparison area comprised nine northwestern SRAs containing approximately 25% of the Hispanic population of San Diego County. The per capita income of Hispanics living in the selected comparison area was comparable with Hispanics in the target area (mean per capita income of $9,075, SD 2,237 vs $8,705, SD 2,908) [13]. Intervention activity in this region consisted of one BCCCP clinic site on the Pauma Indian reservation. The remaining contiguous 24 SRAs were grouped together. Another 25% of the Hispanic population resided in this area but was dispersed across a wider geographic area and income distribution (mean per capita income of $11,218, SD $6,250) [13]. Two BCCCP clinics and one health promotion site were located in this area. Calculation of incidence rates and incidence rate ratios. Four-year average annual, age-adjusted incidence rates (IR) were calculated for each ethnic and geographic strata. Cases were assigned to geographic area by census tract of residence at time of diagnosis. The 1990 Census population count was used as an approximate midperiod count for the study period (1988– 1991). Age adjustment was done using 5-year age groups by the direct method, with the 1970 U.S. Census serving as the standard population. Exact 95% confidence intervals were calculated based on the Poisson distribution [16]. As an approach to estimating and comparing the level of early detection between groups, an incident rate ratio (IRR) was calculated by dividing the in situ incidence rate by the invasive incidence rate. Ninety-five percent confidence limits were calculated using the antilogs of the limits on the log of the IRR [17,18].
Analysis of the Relationship of Socioeconomic and Acculturation Factors to Stage at Diagnosis Measures of socioeconomic status (income, poverty, and education status) and acculturation (foreign birth and language spoken in the household) were extracted from U.S. Census data (STF3A). Variables were then derived for the two ethnic groups as follows: (1) mean per capita income; (2) poverty status, percentage of persons with incomes below 200% of the poverty level; (3) low education, percentage of persons over 25 years of age who have not finished high school; (4) foreign birth, percentage of persons born outside the United States (Hispanics only); and (5) percentage of Hispanic households characterized by Spanish language isolation. The language isolation variables is a summary of census questions and is defined on the STF3A as follows. A household is classified as “linguistically isolated” if no person age 14 years or over speaks only English and if no person age 14 years or over both speaks a language other than English (Spanish in our case) and speaks English “very well.” The answers to a question of how well a person speaks English are precoded as “very well,” “well,” “not well,” and “not at all.” All of the members of a linguistically isolated household are tabulated as linguistically isolated, including members under age 14 years who may speak only English. These censusderived variables were assigned to cases of cervical cancer by census tract of residence at time of diagnosis. Poverty status was used in addition to per capita income for its relevance to the fact that the BCCCP targets screening of women who are at #200% poverty level. Principle components analysis examining the distribution and interrelationship of these variables revealed the following: (1) income and education were distributed differently for Hispanics and white non-Hispanics and (2) socioeconomic status and acculturation variables were highly correlated with ethnicity (multiple R2 for Hispanic ethnicity 5 0.99) and intercorrelated with themselves (multiple R2 ranged from 0.55 for per capita income to 0.87 for language isolation). Therefore, Hispanics and white non-Hispanics were analyzed separately in order to clearly estimate the relationships of socioeconomic status and acculturation to stage at diagnosis. Logistic regression modeling (SAS, Procedure Logistic) was used to examine the relationship of socioeconomic and acculturation factors to the probability of invasive versus in situ stage at diagnosis. Bivariate analysis was performed first to: (1) test for significant associations with stage at diagnosis and (2) check for linearity in the logit. Nonlinear variables were categorized into quartiles or dichotomized to create the appropriate parametric relationship in the model [19]. Variables that showed significant associations at the P , 0.05 level were kept for multiple logistic modeling. Age
677
STAGE AT DIAGNOSIS OF CERVICAL CANCER AMONG HISPANICS
adjustment was used in all multivariate modeling. Tenyear age categories were used, with the reference category being women less than 25 years of age. The results from the multiple logistic modeling were mapped to locate “high-risk census tracts,” using computerized mapping tools (Arc View version 2.1).
women in the target region (0.87; 95% CI 0.62, 1.21), indicating that as many or more of their tumors tended to be diagnosed at the invasive stage compared with the in situ stage. The IRR for Hispanics in the comparison area was slightly higher at 1.13 (95% CI 0.66, 1.94), but the difference was not significant (P , 0.4). Effects of Socioeconomic and Acculturation Factors on Stage at Diagnosis
RESULTS
Cervical Cancer Incidence and Stage at Diagnosis Table 1 presents a comparison of the average annual age-adjusted in situ and invasive incidence rates and their rate ratios for Hispanics and white non-Hispanics for all three regions, including a summary for the entire county. The invasive cervical cancer IR is more than twice as high for Hispanics compared with white nonHispanics for the whole county (P , 0.0001) and in all three regions (all P values , 0.05). The in situ incidence rate was significantly higher for white non-Hispanics than for Hispanics in the entire county (P , 0.001), in the target area (P , 0.02), and the comparison area (P , 0.05). However, the difference in in situ incidence rates for Hispanics and white non-Hispanics was not significant in the remainder of the county (P , 0.4). The rates of invasive cervical cancer for Hispanics in the target and the comparison areas were similar. The in situ incidence rate for Hispanics in the comparison area was slightly higher, but the difference was not significant (P , 0.4). Consistent with the above incidence rate measures, the IRR of in situ to invasive cervical cancer was roughly three times higher for white non-Hispanics than for Hispanics in the entire county and in all three geographic regions. The IRR was lowest for Hispanic
After controlling for age, Spanish language isolation was the variable most significantly associated with invasive vs in situ stage at diagnosis among Hispanics (P , 0.05). With Spanish language isolation already in the logistic regression model, the other variables (income, low education, poverty, or foreign birth) did not contribute significantly to the predictability of the model. Hispanic women residing in census tracts in the top 25% for proportion of households characterized by Spanish language isolation were 2.1 times more likely to be diagnosed at the invasive stage than Hispanic women residing in census tracts in the bottom 75% for Spanish language isolation (95% CI 1.27, 3.40). Among white non-Hispanics, low education was the only variable significantly associated with invasive vs in situ stage at diagnosis, after controlling for age (P , 0.05). Again, income or poverty status did not add significantly to the predictability of the model after low education was in the model. The higher the percentage of non-high-school graduates in a census tract, the more likely white non-Hispanics were to be diagnosed at the invasive stage than at the in situ stage (P , 0.05 for the continuous variable). When education is categorized into quartiles, a dose–response relationship is shown: with the reference category being the bottom quartile
TABLE 1 Age-Adjusted Incidence Rates and Rate Ratios for Cervical Cancer by Stage (in Situ and Invasive) for Three Geographical Areas, Including Totals, San Diego County (1988–1991) Female population count (1990) Target area Hispanic White non-Hispanic Comparison area Hispanic White non-Hispanic Remainder area Hispanic White non-Hispanic Total Hispanic White non-Hispanic a
In situ a
Cases
Invasive
In situ/invasive
IR (95%CI)
Cases
IR (95%CI)
IRR (95%CI)
127,266 134,520
86 152
15.70 (12.56–19.39) 21.90 (18.56–25.67)*
74 55
18.10 (14.21–22.72) 8.52 (6.42–11.09)*
0.87 (0.62–1.21) 2.57 (1.86–3.56)*
56,302 215,728
45 289
19.59 (14.29–26.21) 27.17 (24.13–30.49)*
27 84
17.30 (11.40–25.17) 8.13 (6.48–10.07)*
1.13 (0.66–1.94) 3.34 (2.58–4.32)*
63,549 454,300
75 652
23.70 (18.64–29.71) 26.84 (24.82–28.98)
35 151
15.27 (10.64–21.24) 6.46 (5.47–7.58)*
1.55 (1.00–2.42) 4.15 (3.46–4.98)*
247,117 804,548
206 1,093
18.66 (16.20–21.39) 26.11 (24.58–27.71)**
136 290
17.15 (14.39–20.29) 7.15 (6.35–8.02)**
1.09 (0.86–1.38) 3.65 (3.19–4.18)**
Standardized to 1970 U.S. Census. * P , 0.05 comparing white non-Hispanics with Hispanics. ** P , 0.001.
678
HOWE ET AL.
for low education (0– 7.93% without high school diplomas), the risk of being diagnosed at invasive stages was 2.10 (95% CI 1.31–3.38) for the top quartile, 1.78 (95% CI 1.63–2.72) for the third quartile, and 1.45 (95% CI 0.95–2.22) for the second quartile. DISCUSSION
Incidence of Cervical Cancer The age-adjusted ethnic-specific incidence rates per 100,000 for invasive cervical cancer (18.15 for Hispanics and 7.15 for white non-Hispanics in San Diego County) are comparable with IRs for these groups in the state of California during the same 4-year period (16.9 for Hispanics and 7.6 for white non-Hispanics) [13]. The national rate for whites was 7.8 in 1988–1991 [20]. Rates for Hispanic populations elsewhere in the United States include Colorado, 14.8 (1979–81) [21] and New York City, 19.4 (1982–1985) [22]. However, the rate for Hispanics in New Mexico was lower at 10.2 in 1988– 1992. This lower rate for New Mexican Hispanics represents a 61% decrease over a 24-year period and is attributed to increased screening efforts targeting indigent Hispanics [10]. Florida also reported a lower rate for Hispanics of 9.6 in 1990. Florida’s Hispanic population is 67% Cuban, who as a group tend to have higher incomes and be better educated than other Hispanic groups in the United States [23]. Comparing findings for in situ rates is more difficult. The SEER definition for in situ stage used in the present study has been stable only since 1988 [14]. Reporting of the in situ classification may not be uniform from lab to lab or registry to registry, making comparisons across time and in different locations problematic [10,24–27]. The in situ rate and its comparison with invasive incidence rates is best used as a relative measure between groups or across time in the same area and reporting system, provided there have been no major changes in reporting or diagnostic practice over time. With the above restrictions, the in situ to invasive rate ratio can be a useful indicator of the level and trend of early detection in a population. In the 24-year period that rates for invasive cervical cancer for Hispanic women in New Mexico were declining, the IRR of in situ to invasive stage cancer increased from 1.4 in 1969–1972 to 4.7 in 1988–1992 [10]. A study in Newark, New Jersey, following the ratio of in situ to invasive tumors during the period of a government-sponsored preventive intervention observed the ratio increase significantly during the time the program was active and fall again after the program was terminated [24]. The rise in in situ rates as invasive rates fall after implementation of mass screening has been demonstrated in other cancer registry studies as well [2,10,25].
Relationship of Socioeconomic Status and Acculturation to Stage The finding that Spanish language isolation for Hispanics is associated with higher risk for diagnosis at invasive stage cervical cancer is supported by other studies. Speaking English is one of several factors used in measuring acculturation. Lower acculturation has been shown to be a factor associated with less frequent Pap screening in several behavioral risk factor surveys [28,29], including two telephone surveys done in San Diego [8,30]. The present study confirms the association of invasive stage with language isolation using a population-based analysis, with data aggregated at the census tract level. The association of low education among white nonHispanics with invasive stage diagnosis of cervical cancer supports findings reported elsewhere. Studies using both individually linked [31] and aggregate measurements similar to ones used here [32] have shown that women with less education are at higher risk for cervical cancer. Related measures of socioeconomic status have also been associated with higher risk of cervical cancer. Breen and Figueroa [9], using national SEER data and methods similar to the present study, showed that measurement of poverty and a composite measurement of “disadvantaged neighborhood” are associated with higher likelihood of invasive vs in situ stage cervical cancer diagnosis for all women. Limitations Several areas of potential error could have introduced biases in the incidence rate estimates. As discussed above, the classification of in situ stage tumors might be subject to variations in reporting from different pathology labs, which could vary by geographic location or type of medical coverage. If differences in reporting were associated with ethnic group, this could result in a differential misclassification bias. Bias could also arise from misclassification or undercounting of ethnicity. The CCR method of including Spanish surname in addition to medical records for the coding of Hispanic ethnicity was introduced following a SEER study that compared methods to increase sensitivity and minimize bias due to the different classification methods used by the Census and the CCR [15,33]. Although the Census attempts to correct for undercounting, undercounting of one group more than another could introduce bias [34]. Also, since the 1990 U.S. Census count was used as a 4-year midpoint, populations for the different subgroups in different areas of the county could have changed at different rates during the study period. While any of these areas of misclassification could have affected the accuracy of the incidence rate estimates, the differences in the estimates between ethnic groups are large, internally consistent, and similar to
STAGE AT DIAGNOSIS OF CERVICAL CANCER AMONG HISPANICS
findings elsewhere. The ability to control for such errors would not likely change the overall impression of differences. In the case of the logistic regression analysis, the census-derived variables were measurements of central tendency, or probability, for the group residing in the same census tract. These measurements may not apply well to the individual case resulting in misclassification. In addition, the preliminary analyses for this study showed that ethnicity, socioeconomic status, and acculturation are highly intercorrelated and confounded each other. Many of these variables conveyed much the same information. The present study chose the one variable that contributed most to the predictability of the model as a method to pinpoint geographic areas in most need of intervention. Not speaking English or low education probably are causal factors, but they may also represent other, more complex, behavioral determinants of screening [35] as well as barriers to health care. Implications The combined results from the two analyses presented here indicate that the four preventive intervention programs focusing on Hispanics living in a particular area of San Diego County are both needed and well targeted. The results add further evidence that targeting specific groups and developing methods to overcome language, cultural, and educational barriers have the best chance of decreasing the high relative rates of invasive cervical cancer among susceptible groups. The methods used in this study will be repeated again using San Diego County data collected later in the decade to examine whether there is any change in the disparities in cervical cancer stage at diagnosis, possibly resulting from intervention activities in the targeted areas. In addition, the results indicate that the populationbased methods used in this study can aid in the planning, implementation, and monitoring of communitybased promotion programs. Linking cancer registry data with census data and geographic information systems can locate communities with a higher prevalence of particular risk factors and with the highest need of improved cancer screening. Preventive interventions can then be planned in specific locations using knowledge of the types of barriers likely to be present in the local population and using local resources such as family planning clinics or community organizations. Existing data that could be used for this purpose are exemplified by a recent report produced by the CCR, which pinpoints counties, particular zip codes, and Medical Service Study Areas in California where the proportional incidence ratio of late-stage breast cancer is significantly higher than expected [36]. Where population-based registries do exist, they can serve as a resource to provide the foundation for such planning and coordination.
679
ACKNOWLEDGMENTS The ideas and opinions expressed herein are those of the authors, and no endorsement by the State of California Department of Health Services or the California Public Health Foundation is intended or should be inferred. REFERENCES 1. Muir C, Waterhouse J, Powell J, Whelan S, editors. Cancer incidence in five continents. Lyon (France): IARC, 1987;5:No. 88. 2. Parkin DM, Day NE. Evaluating and planning screening programmes. In: Parkin DM, Wagner G, Muir C, editors. The role of the registry in cancer control. Lyon (France): IARC, 1985; No. 66:155. 3. Parker SL, Tong T, Bolden S, Wingo PA. Cancer statistics, 1997. CA 1997;47:5–27. 4. Kosary CL, Ries LAG, Miller BA, Hankey BF, Harras A, Edwards BK, editors. SEER cancer statistics review, 1973–1992: tables and graphs. Bethesda (MD): National Cancer Institute, 1995. [NIH Publication No. 95-2789] 5. U.S. Department of Health and Human Services and Institute of Medicine. Healthy People 2000: national health promotion disease prevention objectives. Washington U.S. Govt. Printing Office, 1990. [DHHS Publication No. (PHS) 91-50212] 6. Castro FG, Elder J, Coe K, Tafoya-Barraza HM, Moratto S, Campbell N, Talavera G. Mobilizing churches for health promotion in ˜ Latino communities: Companeros en la Salud. J Natl Cancer Inst Monogr 1995;18:127–35. 7. Navarro AM, Senn KL, Kaplan RM, McNicholask L, Campo MC, Roppe B. Por la Vida: intervention model for cancer prevention in Latinas. J Natl Cancer Inst Monogr 1995;18:137–45. 8. Elder JP, Castro FG, de Moor C, Mayer J, Candelaria JI, Campbell N, Talavera G, Ware LM. Differences in cancer-risk-related behaviors in Latino and Anglo adults. Prev Med 1991;20:751–63. 9. Breen N, Figueroa JB. Stage of breast and cervical cancer diagnosis in disadvantaged neighborhoods: a preventive policy perspective. Am J Prev Med 1996;12:319–26. 10. Chao A, Becker TM, Jordan SW, Darling R, Gilliland FD, Key CR. Decreasing rates of cervical cancer among American Indians and Hispanics in New Mexico (United States). Cancer Causes Control 1996;7:205–13. 11. Becker T, Wheeler CM, Key CR, Samet JM. Cervical cancer incidence and mortality in New Mexico’s Hispanics, American Indians, and non-Hispanic whites. West J Med 1992;156:376–9. 12. U.S. Bureau of the Census. 1990 census of population and housing (summary tape files 3A). Washington: U.S. Department of Communication, Bureau of Census, 1990. 13. Snipes KP, Perkins CI, Wright WE, Young, JL. Cancer incidence and mortality by race/ethnicity in California, 1988–1991. Sacramento: California Department of Health Services, Cancer Surveillance Section, 1994 May. 14. Seiffert JE, Bragg D, Gordon B. Cancer reporting in California: abstracting and coding procedures for hospitals, 2nd ed., Vol. 1. Emeryville (CA): State of California Department of Health Services, Cancer Surveillance Section, 1991. 15. Young JL, West D. Change in defining race/ethnicity, California cancer registry. [Letter to regional registries on redefinition of Hispanic ethnicity, 1994 Apr] 16. Esteve J, Benhamou E, Raymond L. Statistical methods in cancer research, Vol. iv, Descriptive epidemiology. Lyon (France): IARC, 1994;No. 128:52–53. 17. Breslow NE, Day NE. Statistical methods in cancer research, Vol. II, The design and analysis of cohort studies. Lyon (France): IARC, 1987;No. 82:59.
680
HOWE ET AL.
18. Perkins CI, Morris CR, Wright WE. Cancer incidence and mortality in California by race/ethnicity, 1988–1993. Sacramento: California Department of Health Services, Cancer Surveillance Section, 1996;Mar:359. 19. Hosmer DW, Lemeshow S. Applied logistic regression. New York: Wiley, 1989. 20. Ries LAG, Miller BA, Hankey BF, Kosary CL, Harras A, Edwards BK, editors. SEER cancer statistics in review, 1973–1991: tables and graphs. Bethesda (MD): National Cancer Institute, 1994. [NIH Publication No. 94-2789] 21. Savitz DA. Changes in Spanish surname cancer rates relative to other whites, Denver area, 1969–71 to 1979–81. Am J Public Health 1986;76:1210–5. 22. Wolfgang PE, Semeiks PA, Burnett WS. Cancer incidence in New York City Hispanics, 1982 to 1985. Ethnic Dis 1991;3:263–72. 23. Trapido EJ, McCoy CB, Stein NS, Engel S, McCoy HV, Olejniczak S. The epidemiology of cancer among Hispanic women: the experience in Florida. Cancer 1990;66:2435–41. 24. Holland BK, Foster JD, Louria DB. Cervical cancer and health care resources in Newark, New Jersey, 1970 to 1988. Am J Public Health 1993;83:45–8. 25. Weiss L., Kau TK, Sparks BT, Swanson GM. Trends in cervical cancer incidence among young black and white non-Hispanic women in metropolitan Detroit. Cancer 1994;73:1849–54. 26. Howe HL, Lehnherr M, Qualls RY. Using central-registry data to monitor progress in early detection of breast and cervical cancer (Illinois, United States). Cancer Causes Control 1995;6:155–63. 27. American Association of Central Cancer Registries. Working Group on Pre-invasive Cervical Neoplasia and Population-Based Cancer Registries. Final sub-committee report. Rockville, MD, April 5–6, 1993. Bethesda (MD): National Cancer Institute, 1993;Apr:1–5.
28. Harlan LC, Bernstein AB, Kessler LG. Cervical cancer screening: who is not screened and why? Am J Public Health 1991; 81:885–91. 29. Suarez L, Roche RA, Nichols D, Simpson DM. Knowledge, behavior, and fears concerning breast and cervical cancer among older low-income Mexican-American women. Am J Prev Med 1997; 13:137–42. 30. Swanson, LL. Cervical cancer screening among women in South Bay communities, San Diego. Unpublished M.P.H. thesis, San Diego State University, 1991. 31. Morrison EAB, Ho G, Vermund ST, Goldberg GL, Kadish S, Kelley KF, Burk RD. Human papillomavirus infection and other risk factors for cervical neoplasia: a case–control study. Int J Cancer 1991;49:6–13. 32. Devesa SS, Diamond EI. Association of breast cancer and cervical cancer incidences with income and education among whites and blacks. J Natl Cancer Inst 1980;65:515–28. 33. Stewart SL, Glaser SL, Horn-Ross P L, West DW. SEER study of methods to classify Hispanic cancer patients. Final report, 1993. [Contract No. 1-CN-05224] 34. Hahn RA, Stroup DF. Race and ethnicity in public health surveillance: criteria for the scientific use of social categories. Public Health Rep 1994;109:7–15. 35. Hubbell FA, Chavez LR, Mishra SI, Valdez RB. Beliefs about sexual behavior and other predictors of Papanicolaou smear screening among Latinas and Anglo women. Arch Intern Med 1996;156:2353–8 36. Perkins CI, Wright WE, Snipes KP. Targeting breast cancer interventions: geographic variation in late-stage breast cancer in California, 1988–1992. Sacramento: California Department of Health Services, Cancer Surveillance Section, 1997, Feb.
PREVENTIVE MEDICINE ARTICLE NO.
The Risk of Invasive Cervical Cancer among Hispanics: Evidence for Targeted Preventive Interventions1 Susan L. Howe, M.P.H.,* Ralph J. Delfino, M.D., Ph.D.,*,†,2 Thomas H. Taylor, Ph.D.,† and Hoda Anton-Culver, Ph.D.† *Graduate School of Public Health, San Diego State University, San Diego, California 92182; †Epidemiology Division, Department of Medicine, College of Medicine, University of California at Irvine, Irvine, California 92697
Background. Mass screening for cervical cancer has considerably reduced invasive cancer rates; however, Hispanic Americans have not fully benefited. This study examined the relationship of demographic, socioeconomic, and acculturation factors to diagnostic stage prior to preventive interventions in a large Hispanic community (San Diego, CA, 1988–1991). Methods. Age-adjusted incidence rates (IR) were estimated for three subareas: Hispanic community targeted with interventions, comparison Hispanic community, and remainder of county. Incident cases of cervical cancer in the California Cancer Registry were used to calculate incidence rate ratios (IRR) of in situ to invasive stage cancer and estimate differences in early detection between Hispanics and white non-Hispanics. Multiple logistic regression was used to examine the relationship of cancer stage to census-linked income, education, poverty status, foreign birth, and language spoken. Results. The in situ cervical cancer IR is significantly lower for Hispanics than for white non-Hispanics in all three regions (P , 0.05). The overall IRR is 3.65 (95% CI 3.19, 4.18) for white non-Hispanics compared with 1.09 (95% CI 0.86, 1.38) for Hispanics. Hispanics in census tracts in the highest versus remaining quartiles for percentage of language-isolated households are at greater risk for invasive stage (OR 5 2.10, 95% CI 1.27, 3.40). White non-Hispanic women residing in census tracts in the highest versus lowest quartile of adults
1 This research was supported by the Cancer Surveillance Program of Orange County and the San Diego/Imperial Organization for Cancer Control. Regional cancer incidence data have been collected by the Epidemiology Division, University of California, Irvine, under Subcontract 050K8707-8710. The subcontract is supported by the California Department of Health Services and its agent, the California Public Helath Foundation, as part of its statewide cancer reporting program, which is mandated by Health and Safety code Section 103875 and 103885. 2 To whom reprint requests should be addressed. Fax: (714) 8244773. E-mail:[email protected].
with less than 12 years education are at greater risk for invasive stage (OR 5 2.10, 95% CI 1.31, 3.38). Conclusions. Results show efforts aimed at cervical cancer early detection for Hispanic women are needed and well targeted. White non-Hispanic women in loweducation communities are also at risk. Populationbased methods in this study can focus efforts in communities needing preventive interventions. q1998 American Health Foundation and Academic Press
Key Words: cervical cancer; Hispanic American; cancer staging; mass screening.
INTRODUCTION
When cervical neoplasms are detected early and treated promptly, the cure rate is high. Consequently, in developed countries in which mass screening has been implemented, mortality rates due to cervical cancer have dropped dramatically [1,2]. However, the success of mass screening has not been uniform in the United States. Within certain subpopulations, such as racial or ethnic minority, low-income, or less educated women, incidence and mortality for invasive cervical cancer notably exceed those for white, higher income, and better educated women [3–5]. Four programs responding to U.S. federal initiatives for improved breast and cervical cancer screening among Hispanics and low-income women [5] were initiated in San Diego County beginning in 1992: Por la Vida, Companeras en la Salud, En Accion, and the Breast and Cervical Cancer Control Program of the State of California (BCCCP) [6,7]. These health-promotion programs were designed to overcome socioeconomic and behavioral barriers to screening among Hispanic and low-income women [6–8]. The BCCCP also provided clinic sites where Pap smears were performed free to low-income women. Since the goal of programs such as these is to reach specific underscreened populations, methods are needed to target intervention sites and evaluate progress. Although, the authors of this paper
674
0091-7435/98 $25.00 Copyright q 1998 by American Health Foundation and Academic Press All rights of reproduction in any form reserved.
STAGE AT DIAGNOSIS OF CERVICAL CANCER AMONG HISPANICS
are not involved in the above programs, we had access to the necessary data needed to apply such methods of identifying populations at risk. We examined the degree to which the populations targeted in San Diego were in need of screening and whether other populations should be targeted. The period examined was the 4 years (1988–1991) prior to the commencement of the preventive interventions. We examined both the population distribution and the demographic determinants of stage at cervical cancer diagnosis for Hispanic and white non-Hispanic women. Few other population-based studies have examined this within a large Hispanic population in the United States [9–11]. This is the first of such studies in San Diego County examining the relationship of ethnic, socioeconomic, and acculturation factors to stage at cervical cancer diagnosis. METHODS
Overview The present study consists of two complementary analyses. The first analysis stratifies subjects by cancer stage, ethnicity, and geographic area within San Diego County, in order to estimate the incidence of cervical cancer and the level of early detection in the subpopulation of Hispanics targeted for preventive intervention as well as Hispanics in a comparison area and in the remainder of the county (described below). Stage-specific incidence rates for Hispanics are compared with rates for white non-Hispanics in all three geographic areas. The second analysis uses logistic regression to examine the relationship of socioeconomic and acculturation factors to the probability of invasive stage compared with in situ stage at diagnosis for Hispanics and white non-Hispanics. Using the results of these analyses, census tracts with higher proportions of risk factors for invasive stage diagnosis are located using computerized mapping tools. Data Sources The two databases used for this study were the 1990 U.S. Census [12] and the San Diego/Imperial County Organization for Cancer Control (SANDIOCC), which is part of the California Cancer Registry (CCR) [13]. Population count data required for denominators were taken from the 1990 100% count of persons, using the U.S. Census Summary Tape Files 3A (STF3A), which report counts at the census tract level stratified by race/ ethnicity, gender, and age. The SANDIOCC data are collected at the Epidemiology Division, Department of Medicine, University of California, Irvine. The CCR sets uniform recording standards and quality control for the state [14]. Case ascertainment is estimated to
675
exceed 97%. Histopathological data sufficient to code the cases as in situ or invasive were available for all reported cases. All cases of neoplasms of the cervix uteri occurring in San Diego County between 1988 and 1991 were extracted from the SANDIOCC data base. Information on patients included the histopathology, stage at diagnosis, census track of residence at time of diagnosis, ethnicity, and place of birth. This study was granted exempt status by the University of California, Irvine Human Subjects Review Board, approval HS 95-005. Cervical cancer cases included both squamous cell carcinoma and adenocarcinoma and were divided into in situ and invasive stage at diagnosis. In situ stage cancers included neoplasms of the cervix uteri that met all microscopic criteria for malignancy, except invasion of the basement membrane, including cervical intraepithelial neoplasia, grade III. Invasive cases included malignant neoplasms of the cervix uteri that displayed any degree of invasion, including local microinvasion of the basement membrane, as well as regional and distant spread. Analysis of the Incidence of Cervical Cancer by Stage, Ethnicity, and Geographic Area Stratification by ethnicity. The two largest ethnic subpopulations in San Diego are Hispanics, comprising 20%, and white non-Hispanics, comprising 67% of San Diego’s 2.5 million population [12]. The majority of Hispanics in San Diego County are of Mexican descent (87%) [8,12]. The Census uses self-identification to classify race/ ethnicity. The CCR classifies a cancer case Hispanic if so indicated by the medical record or if the person’s last name appeared on the 1980 U.S. Census list of Spanish surnames, excluding persons classified as Asian/other (i.e., Filipino or American Indian). White non-Hispanics include those in the white race category excluding Hispanics [15]. Geographic stratification. Using census data, census tracts with populations of over 20, 40, 60, and 80% Hispanic ethnicity were identified. The locations of all intervention sites were then overlaid on these census tract percentages. This two-step process provided the basis for dividing San Diego County into three geographic areas. We assumed that the impact of the four intervention programs is largely limited to the geographic locations of intervention sites. This is justified since the programs focused on the use of local clinic outreach and local social networks within the Hispanic communities to promote both cervical cancer and breast cancer screening among women. One program (En Accion) targeted Hispanic men and women for five screenable cancers including cervical cancer.
676
HOWE ET AL.
The three geographic areas identified were: (1) a “target area” [high concentration of Hispanics ($20% in many census tracts) and the majority of the preventive intervention activities], (2) a “comparison area” (high concentration of Hispanics, but little intervention activity planned), and (3) a “remainder area” (the rest of the county throughout which Hispanics tend to be more widely dispersed). Census tracts were included in clusters called subregional areas (SRAs) since the San Diego Department of Health Services uses SRAs for planning their public health programs annually. The target area was composed of six SRAs extending to the Mexican border and containing approximately 50% of the Hispanics living in San Diego County. There were 32 health promotion sites and 5 BCCCP clinic sites contained in this region, compared with only 1 health promotion site and 3 BCCCP clinic sites located elsewhere in the county. The SRAs comprising the comparison area were selected based on three criteria: (1) a high concentration of Hispanics, (2) Hispanics with similar incomes to Hispanics in the target area, and (3) a minimum of intervention activity. The selected comparison area comprised nine northwestern SRAs containing approximately 25% of the Hispanic population of San Diego County. The per capita income of Hispanics living in the selected comparison area was comparable with Hispanics in the target area (mean per capita income of $9,075, SD 2,237 vs $8,705, SD 2,908) [13]. Intervention activity in this region consisted of one BCCCP clinic site on the Pauma Indian reservation. The remaining contiguous 24 SRAs were grouped together. Another 25% of the Hispanic population resided in this area but was dispersed across a wider geographic area and income distribution (mean per capita income of $11,218, SD $6,250) [13]. Two BCCCP clinics and one health promotion site were located in this area. Calculation of incidence rates and incidence rate ratios. Four-year average annual, age-adjusted incidence rates (IR) were calculated for each ethnic and geographic strata. Cases were assigned to geographic area by census tract of residence at time of diagnosis. The 1990 Census population count was used as an approximate midperiod count for the study period (1988– 1991). Age adjustment was done using 5-year age groups by the direct method, with the 1970 U.S. Census serving as the standard population. Exact 95% confidence intervals were calculated based on the Poisson distribution [16]. As an approach to estimating and comparing the level of early detection between groups, an incident rate ratio (IRR) was calculated by dividing the in situ incidence rate by the invasive incidence rate. Ninety-five percent confidence limits were calculated using the antilogs of the limits on the log of the IRR [17,18].
Analysis of the Relationship of Socioeconomic and Acculturation Factors to Stage at Diagnosis Measures of socioeconomic status (income, poverty, and education status) and acculturation (foreign birth and language spoken in the household) were extracted from U.S. Census data (STF3A). Variables were then derived for the two ethnic groups as follows: (1) mean per capita income; (2) poverty status, percentage of persons with incomes below 200% of the poverty level; (3) low education, percentage of persons over 25 years of age who have not finished high school; (4) foreign birth, percentage of persons born outside the United States (Hispanics only); and (5) percentage of Hispanic households characterized by Spanish language isolation. The language isolation variables is a summary of census questions and is defined on the STF3A as follows. A household is classified as “linguistically isolated” if no person age 14 years or over speaks only English and if no person age 14 years or over both speaks a language other than English (Spanish in our case) and speaks English “very well.” The answers to a question of how well a person speaks English are precoded as “very well,” “well,” “not well,” and “not at all.” All of the members of a linguistically isolated household are tabulated as linguistically isolated, including members under age 14 years who may speak only English. These censusderived variables were assigned to cases of cervical cancer by census tract of residence at time of diagnosis. Poverty status was used in addition to per capita income for its relevance to the fact that the BCCCP targets screening of women who are at #200% poverty level. Principle components analysis examining the distribution and interrelationship of these variables revealed the following: (1) income and education were distributed differently for Hispanics and white non-Hispanics and (2) socioeconomic status and acculturation variables were highly correlated with ethnicity (multiple R2 for Hispanic ethnicity 5 0.99) and intercorrelated with themselves (multiple R2 ranged from 0.55 for per capita income to 0.87 for language isolation). Therefore, Hispanics and white non-Hispanics were analyzed separately in order to clearly estimate the relationships of socioeconomic status and acculturation to stage at diagnosis. Logistic regression modeling (SAS, Procedure Logistic) was used to examine the relationship of socioeconomic and acculturation factors to the probability of invasive versus in situ stage at diagnosis. Bivariate analysis was performed first to: (1) test for significant associations with stage at diagnosis and (2) check for linearity in the logit. Nonlinear variables were categorized into quartiles or dichotomized to create the appropriate parametric relationship in the model [19]. Variables that showed significant associations at the P , 0.05 level were kept for multiple logistic modeling. Age
677
STAGE AT DIAGNOSIS OF CERVICAL CANCER AMONG HISPANICS
adjustment was used in all multivariate modeling. Tenyear age categories were used, with the reference category being women less than 25 years of age. The results from the multiple logistic modeling were mapped to locate “high-risk census tracts,” using computerized mapping tools (Arc View version 2.1).
women in the target region (0.87; 95% CI 0.62, 1.21), indicating that as many or more of their tumors tended to be diagnosed at the invasive stage compared with the in situ stage. The IRR for Hispanics in the comparison area was slightly higher at 1.13 (95% CI 0.66, 1.94), but the difference was not significant (P , 0.4). Effects of Socioeconomic and Acculturation Factors on Stage at Diagnosis
RESULTS
Cervical Cancer Incidence and Stage at Diagnosis Table 1 presents a comparison of the average annual age-adjusted in situ and invasive incidence rates and their rate ratios for Hispanics and white non-Hispanics for all three regions, including a summary for the entire county. The invasive cervical cancer IR is more than twice as high for Hispanics compared with white nonHispanics for the whole county (P , 0.0001) and in all three regions (all P values , 0.05). The in situ incidence rate was significantly higher for white non-Hispanics than for Hispanics in the entire county (P , 0.001), in the target area (P , 0.02), and the comparison area (P , 0.05). However, the difference in in situ incidence rates for Hispanics and white non-Hispanics was not significant in the remainder of the county (P , 0.4). The rates of invasive cervical cancer for Hispanics in the target and the comparison areas were similar. The in situ incidence rate for Hispanics in the comparison area was slightly higher, but the difference was not significant (P , 0.4). Consistent with the above incidence rate measures, the IRR of in situ to invasive cervical cancer was roughly three times higher for white non-Hispanics than for Hispanics in the entire county and in all three geographic regions. The IRR was lowest for Hispanic
After controlling for age, Spanish language isolation was the variable most significantly associated with invasive vs in situ stage at diagnosis among Hispanics (P , 0.05). With Spanish language isolation already in the logistic regression model, the other variables (income, low education, poverty, or foreign birth) did not contribute significantly to the predictability of the model. Hispanic women residing in census tracts in the top 25% for proportion of households characterized by Spanish language isolation were 2.1 times more likely to be diagnosed at the invasive stage than Hispanic women residing in census tracts in the bottom 75% for Spanish language isolation (95% CI 1.27, 3.40). Among white non-Hispanics, low education was the only variable significantly associated with invasive vs in situ stage at diagnosis, after controlling for age (P , 0.05). Again, income or poverty status did not add significantly to the predictability of the model after low education was in the model. The higher the percentage of non-high-school graduates in a census tract, the more likely white non-Hispanics were to be diagnosed at the invasive stage than at the in situ stage (P , 0.05 for the continuous variable). When education is categorized into quartiles, a dose–response relationship is shown: with the reference category being the bottom quartile
TABLE 1 Age-Adjusted Incidence Rates and Rate Ratios for Cervical Cancer by Stage (in Situ and Invasive) for Three Geographical Areas, Including Totals, San Diego County (1988–1991) Female population count (1990) Target area Hispanic White non-Hispanic Comparison area Hispanic White non-Hispanic Remainder area Hispanic White non-Hispanic Total Hispanic White non-Hispanic a
In situ a
Cases
Invasive
In situ/invasive
IR (95%CI)
Cases
IR (95%CI)
IRR (95%CI)
127,266 134,520
86 152
15.70 (12.56–19.39) 21.90 (18.56–25.67)*
74 55
18.10 (14.21–22.72) 8.52 (6.42–11.09)*
0.87 (0.62–1.21) 2.57 (1.86–3.56)*
56,302 215,728
45 289
19.59 (14.29–26.21) 27.17 (24.13–30.49)*
27 84
17.30 (11.40–25.17) 8.13 (6.48–10.07)*
1.13 (0.66–1.94) 3.34 (2.58–4.32)*
63,549 454,300
75 652
23.70 (18.64–29.71) 26.84 (24.82–28.98)
35 151
15.27 (10.64–21.24) 6.46 (5.47–7.58)*
1.55 (1.00–2.42) 4.15 (3.46–4.98)*
247,117 804,548
206 1,093
18.66 (16.20–21.39) 26.11 (24.58–27.71)**
136 290
17.15 (14.39–20.29) 7.15 (6.35–8.02)**
1.09 (0.86–1.38) 3.65 (3.19–4.18)**
Standardized to 1970 U.S. Census. * P , 0.05 comparing white non-Hispanics with Hispanics. ** P , 0.001.
678
HOWE ET AL.
for low education (0– 7.93% without high school diplomas), the risk of being diagnosed at invasive stages was 2.10 (95% CI 1.31–3.38) for the top quartile, 1.78 (95% CI 1.63–2.72) for the third quartile, and 1.45 (95% CI 0.95–2.22) for the second quartile. DISCUSSION
Incidence of Cervical Cancer The age-adjusted ethnic-specific incidence rates per 100,000 for invasive cervical cancer (18.15 for Hispanics and 7.15 for white non-Hispanics in San Diego County) are comparable with IRs for these groups in the state of California during the same 4-year period (16.9 for Hispanics and 7.6 for white non-Hispanics) [13]. The national rate for whites was 7.8 in 1988–1991 [20]. Rates for Hispanic populations elsewhere in the United States include Colorado, 14.8 (1979–81) [21] and New York City, 19.4 (1982–1985) [22]. However, the rate for Hispanics in New Mexico was lower at 10.2 in 1988– 1992. This lower rate for New Mexican Hispanics represents a 61% decrease over a 24-year period and is attributed to increased screening efforts targeting indigent Hispanics [10]. Florida also reported a lower rate for Hispanics of 9.6 in 1990. Florida’s Hispanic population is 67% Cuban, who as a group tend to have higher incomes and be better educated than other Hispanic groups in the United States [23]. Comparing findings for in situ rates is more difficult. The SEER definition for in situ stage used in the present study has been stable only since 1988 [14]. Reporting of the in situ classification may not be uniform from lab to lab or registry to registry, making comparisons across time and in different locations problematic [10,24–27]. The in situ rate and its comparison with invasive incidence rates is best used as a relative measure between groups or across time in the same area and reporting system, provided there have been no major changes in reporting or diagnostic practice over time. With the above restrictions, the in situ to invasive rate ratio can be a useful indicator of the level and trend of early detection in a population. In the 24-year period that rates for invasive cervical cancer for Hispanic women in New Mexico were declining, the IRR of in situ to invasive stage cancer increased from 1.4 in 1969–1972 to 4.7 in 1988–1992 [10]. A study in Newark, New Jersey, following the ratio of in situ to invasive tumors during the period of a government-sponsored preventive intervention observed the ratio increase significantly during the time the program was active and fall again after the program was terminated [24]. The rise in in situ rates as invasive rates fall after implementation of mass screening has been demonstrated in other cancer registry studies as well [2,10,25].
Relationship of Socioeconomic Status and Acculturation to Stage The finding that Spanish language isolation for Hispanics is associated with higher risk for diagnosis at invasive stage cervical cancer is supported by other studies. Speaking English is one of several factors used in measuring acculturation. Lower acculturation has been shown to be a factor associated with less frequent Pap screening in several behavioral risk factor surveys [28,29], including two telephone surveys done in San Diego [8,30]. The present study confirms the association of invasive stage with language isolation using a population-based analysis, with data aggregated at the census tract level. The association of low education among white nonHispanics with invasive stage diagnosis of cervical cancer supports findings reported elsewhere. Studies using both individually linked [31] and aggregate measurements similar to ones used here [32] have shown that women with less education are at higher risk for cervical cancer. Related measures of socioeconomic status have also been associated with higher risk of cervical cancer. Breen and Figueroa [9], using national SEER data and methods similar to the present study, showed that measurement of poverty and a composite measurement of “disadvantaged neighborhood” are associated with higher likelihood of invasive vs in situ stage cervical cancer diagnosis for all women. Limitations Several areas of potential error could have introduced biases in the incidence rate estimates. As discussed above, the classification of in situ stage tumors might be subject to variations in reporting from different pathology labs, which could vary by geographic location or type of medical coverage. If differences in reporting were associated with ethnic group, this could result in a differential misclassification bias. Bias could also arise from misclassification or undercounting of ethnicity. The CCR method of including Spanish surname in addition to medical records for the coding of Hispanic ethnicity was introduced following a SEER study that compared methods to increase sensitivity and minimize bias due to the different classification methods used by the Census and the CCR [15,33]. Although the Census attempts to correct for undercounting, undercounting of one group more than another could introduce bias [34]. Also, since the 1990 U.S. Census count was used as a 4-year midpoint, populations for the different subgroups in different areas of the county could have changed at different rates during the study period. While any of these areas of misclassification could have affected the accuracy of the incidence rate estimates, the differences in the estimates between ethnic groups are large, internally consistent, and similar to
STAGE AT DIAGNOSIS OF CERVICAL CANCER AMONG HISPANICS
findings elsewhere. The ability to control for such errors would not likely change the overall impression of differences. In the case of the logistic regression analysis, the census-derived variables were measurements of central tendency, or probability, for the group residing in the same census tract. These measurements may not apply well to the individual case resulting in misclassification. In addition, the preliminary analyses for this study showed that ethnicity, socioeconomic status, and acculturation are highly intercorrelated and confounded each other. Many of these variables conveyed much the same information. The present study chose the one variable that contributed most to the predictability of the model as a method to pinpoint geographic areas in most need of intervention. Not speaking English or low education probably are causal factors, but they may also represent other, more complex, behavioral determinants of screening [35] as well as barriers to health care. Implications The combined results from the two analyses presented here indicate that the four preventive intervention programs focusing on Hispanics living in a particular area of San Diego County are both needed and well targeted. The results add further evidence that targeting specific groups and developing methods to overcome language, cultural, and educational barriers have the best chance of decreasing the high relative rates of invasive cervical cancer among susceptible groups. The methods used in this study will be repeated again using San Diego County data collected later in the decade to examine whether there is any change in the disparities in cervical cancer stage at diagnosis, possibly resulting from intervention activities in the targeted areas. In addition, the results indicate that the populationbased methods used in this study can aid in the planning, implementation, and monitoring of communitybased promotion programs. Linking cancer registry data with census data and geographic information systems can locate communities with a higher prevalence of particular risk factors and with the highest need of improved cancer screening. Preventive interventions can then be planned in specific locations using knowledge of the types of barriers likely to be present in the local population and using local resources such as family planning clinics or community organizations. Existing data that could be used for this purpose are exemplified by a recent report produced by the CCR, which pinpoints counties, particular zip codes, and Medical Service Study Areas in California where the proportional incidence ratio of late-stage breast cancer is significantly higher than expected [36]. Where population-based registries do exist, they can serve as a resource to provide the foundation for such planning and coordination.
679
ACKNOWLEDGMENTS The ideas and opinions expressed herein are those of the authors, and no endorsement by the State of California Department of Health Services or the California Public Health Foundation is intended or should be inferred. REFERENCES 1. Muir C, Waterhouse J, Powell J, Whelan S, editors. Cancer incidence in five continents. Lyon (France): IARC, 1987;5:No. 88. 2. Parkin DM, Day NE. Evaluating and planning screening programmes. In: Parkin DM, Wagner G, Muir C, editors. The role of the registry in cancer control. Lyon (France): IARC, 1985; No. 66:155. 3. Parker SL, Tong T, Bolden S, Wingo PA. Cancer statistics, 1997. CA 1997;47:5–27. 4. Kosary CL, Ries LAG, Miller BA, Hankey BF, Harras A, Edwards BK, editors. SEER cancer statistics review, 1973–1992: tables and graphs. Bethesda (MD): National Cancer Institute, 1995. [NIH Publication No. 95-2789] 5. U.S. Department of Health and Human Services and Institute of Medicine. Healthy People 2000: national health promotion disease prevention objectives. Washington U.S. Govt. Printing Office, 1990. [DHHS Publication No. (PHS) 91-50212] 6. Castro FG, Elder J, Coe K, Tafoya-Barraza HM, Moratto S, Campbell N, Talavera G. Mobilizing churches for health promotion in ˜ Latino communities: Companeros en la Salud. J Natl Cancer Inst Monogr 1995;18:127–35. 7. Navarro AM, Senn KL, Kaplan RM, McNicholask L, Campo MC, Roppe B. Por la Vida: intervention model for cancer prevention in Latinas. J Natl Cancer Inst Monogr 1995;18:137–45. 8. Elder JP, Castro FG, de Moor C, Mayer J, Candelaria JI, Campbell N, Talavera G, Ware LM. Differences in cancer-risk-related behaviors in Latino and Anglo adults. Prev Med 1991;20:751–63. 9. Breen N, Figueroa JB. Stage of breast and cervical cancer diagnosis in disadvantaged neighborhoods: a preventive policy perspective. Am J Prev Med 1996;12:319–26. 10. Chao A, Becker TM, Jordan SW, Darling R, Gilliland FD, Key CR. Decreasing rates of cervical cancer among American Indians and Hispanics in New Mexico (United States). Cancer Causes Control 1996;7:205–13. 11. Becker T, Wheeler CM, Key CR, Samet JM. Cervical cancer incidence and mortality in New Mexico’s Hispanics, American Indians, and non-Hispanic whites. West J Med 1992;156:376–9. 12. U.S. Bureau of the Census. 1990 census of population and housing (summary tape files 3A). Washington: U.S. Department of Communication, Bureau of Census, 1990. 13. Snipes KP, Perkins CI, Wright WE, Young, JL. Cancer incidence and mortality by race/ethnicity in California, 1988–1991. Sacramento: California Department of Health Services, Cancer Surveillance Section, 1994 May. 14. Seiffert JE, Bragg D, Gordon B. Cancer reporting in California: abstracting and coding procedures for hospitals, 2nd ed., Vol. 1. Emeryville (CA): State of California Department of Health Services, Cancer Surveillance Section, 1991. 15. Young JL, West D. Change in defining race/ethnicity, California cancer registry. [Letter to regional registries on redefinition of Hispanic ethnicity, 1994 Apr] 16. Esteve J, Benhamou E, Raymond L. Statistical methods in cancer research, Vol. iv, Descriptive epidemiology. Lyon (France): IARC, 1994;No. 128:52–53. 17. Breslow NE, Day NE. Statistical methods in cancer research, Vol. II, The design and analysis of cohort studies. Lyon (France): IARC, 1987;No. 82:59.
680
HOWE ET AL.
18. Perkins CI, Morris CR, Wright WE. Cancer incidence and mortality in California by race/ethnicity, 1988–1993. Sacramento: California Department of Health Services, Cancer Surveillance Section, 1996;Mar:359. 19. Hosmer DW, Lemeshow S. Applied logistic regression. New York: Wiley, 1989. 20. Ries LAG, Miller BA, Hankey BF, Kosary CL, Harras A, Edwards BK, editors. SEER cancer statistics in review, 1973–1991: tables and graphs. Bethesda (MD): National Cancer Institute, 1994. [NIH Publication No. 94-2789] 21. Savitz DA. Changes in Spanish surname cancer rates relative to other whites, Denver area, 1969–71 to 1979–81. Am J Public Health 1986;76:1210–5. 22. Wolfgang PE, Semeiks PA, Burnett WS. Cancer incidence in New York City Hispanics, 1982 to 1985. Ethnic Dis 1991;3:263–72. 23. Trapido EJ, McCoy CB, Stein NS, Engel S, McCoy HV, Olejniczak S. The epidemiology of cancer among Hispanic women: the experience in Florida. Cancer 1990;66:2435–41. 24. Holland BK, Foster JD, Louria DB. Cervical cancer and health care resources in Newark, New Jersey, 1970 to 1988. Am J Public Health 1993;83:45–8. 25. Weiss L., Kau TK, Sparks BT, Swanson GM. Trends in cervical cancer incidence among young black and white non-Hispanic women in metropolitan Detroit. Cancer 1994;73:1849–54. 26. Howe HL, Lehnherr M, Qualls RY. Using central-registry data to monitor progress in early detection of breast and cervical cancer (Illinois, United States). Cancer Causes Control 1995;6:155–63. 27. American Association of Central Cancer Registries. Working Group on Pre-invasive Cervical Neoplasia and Population-Based Cancer Registries. Final sub-committee report. Rockville, MD, April 5–6, 1993. Bethesda (MD): National Cancer Institute, 1993;Apr:1–5.
28. Harlan LC, Bernstein AB, Kessler LG. Cervical cancer screening: who is not screened and why? Am J Public Health 1991; 81:885–91. 29. Suarez L, Roche RA, Nichols D, Simpson DM. Knowledge, behavior, and fears concerning breast and cervical cancer among older low-income Mexican-American women. Am J Prev Med 1997; 13:137–42. 30. Swanson, LL. Cervical cancer screening among women in South Bay communities, San Diego. Unpublished M.P.H. thesis, San Diego State University, 1991. 31. Morrison EAB, Ho G, Vermund ST, Goldberg GL, Kadish S, Kelley KF, Burk RD. Human papillomavirus infection and other risk factors for cervical neoplasia: a case–control study. Int J Cancer 1991;49:6–13. 32. Devesa SS, Diamond EI. Association of breast cancer and cervical cancer incidences with income and education among whites and blacks. J Natl Cancer Inst 1980;65:515–28. 33. Stewart SL, Glaser SL, Horn-Ross P L, West DW. SEER study of methods to classify Hispanic cancer patients. Final report, 1993. [Contract No. 1-CN-05224] 34. Hahn RA, Stroup DF. Race and ethnicity in public health surveillance: criteria for the scientific use of social categories. Public Health Rep 1994;109:7–15. 35. Hubbell FA, Chavez LR, Mishra SI, Valdez RB. Beliefs about sexual behavior and other predictors of Papanicolaou smear screening among Latinas and Anglo women. Arch Intern Med 1996;156:2353–8 36. Perkins CI, Wright WE, Snipes KP. Targeting breast cancer interventions: geographic variation in late-stage breast cancer in California, 1988–1992. Sacramento: California Department of Health Services, Cancer Surveillance Section, 1997, Feb.