Printed statewide intervention (PSI)

Focus groups were held with organizations serving Latinas to review published information about barriers Latinas face in obtaining mammograms. The focus groups identified and confirmed the following barriers faced by Latinas: lack of access to care, modesty, and lack of time because of primary role as family caretaker. Previously identified barriers in the literature, such as strong sense of family (13,16), were also confirmed. Materials for the intervention printed by the National Cancer Institute (NCI) reflected the pre-identified theme of a sense of family with the chosen message, "Do it for you. Do it for your family." The materials were enhanced by adding contact information for the local Colorado Women's Cancer Control Initiative (CWCCI), a program administered by the Colorado Department of Public Health and Environment and funded by the Centers for Disease Control and Prevention (CDC). The CWCCI provides referrals and free mammograms to eligible women.

After an initial contact by the archdioceses, the churches were mailed an intervention package containing a letter describing the Tepeyac Project; the NCI educational materials about breast cancer screening; a display unit; short camera-ready messages in English and Spanish to be delivered from the pulpit, published in the church bulletins, or both; and a fax–back form asking at which level they would agree to participate (display materials, publish messages, deliver messages from pulpit). The first mailing to all churches in the state occurred in March 2000, a second in October 2000, and a third in February 2001. The second and third mailings included issues of the Tepeyac Project newsletter (available from www.cfmc.org) (15).

Information about the level of church participation was obtained by phone call, personal visit, or fax after the first mailing and was available for 154 (72%) of the 213 participating churches (209 in the PSI and 4 in the PI). Of the 154 churches, 61 (40%) displayed the printed materials, 8 (5%) published messages in the bulletin, and 85 (55%) did both. In addition to these activities, 18 (12%) made pulpit announcements. The level of participation was undetermined at 47 churches, and 12 churches declined to participate.

Promotora intervention (PI)

The four churches that received the PI are all in the Denver area. The promotoras selected these churches because the parishioners and priests had expressed interest in having them come and educate women. The churches, located in largely Hispanic areas, are central reference points to Latinos in Colorado. The PI was an expansion of a pre-existing community-based intervention initiated by La Clinica Tepeyac that provides health care to the local underserved Hispanic population. Each promotora was trained using a standardized curriculum developed by La Clinica Tepeyac and assigned a specific church to visit monthly. Promotoras approached their peers after Sunday masses and during church fairs and other church-related activities. La Clinica Tepeyac coordinated the promotoras' work and paid their salaries. The promotoras also facilitated health groups, or platicas, where a group of women met at one another's houses to discuss breast health. This intervention started in 2000 and is ongoing.

Reach of interventions

According to parish register data from the Archdiocese of Denver, the size of the congregations in the four parishes that received the PI varied from 1950 to 5000 total parishioners, of whom 32% to 84% were Latinos, for a potential of 9427 Latino parishioners reached by the PI. Also based on these parish register data, we estimate that the PSI reached a minimum of 349,340 parishioners, of whom 34,419 were Latinos (with an average church size of 3235 parishioners). Latinos are less likely to register than whites; therefore, these numbers are likely to underestimate the number of Latinos potentially reached by these church-based interventions (17).

The eligibility criteria for this study were: women aged 50 to 64 years, enrolled in the Colorado Medicaid fee-for-service health insurance program (not enrolled in an HMO), and enrolled for more than 18 months (continuously or as a sum of individual enrollment periods) during the baseline period January 1998 through December 1999 and similarly during the follow-up period January 2000 through December 2001. Subjects enrolled in a primary care case management (PCCM) program reimbursed by fee-for-service were included in the database for analysis.

Exposure to the PI or PSI among study subjects was determined by zip codes. Women in the study living in the three zip codes of the four churches visited by the promotoras were assumed to be exposed to the PI, whereas women living in remaining zip codes were assumed to be exposed to the PSI. While using zip codes to assess the intervention effects may be a methodological limitation, many churches do not release individual-level parish membership data, as it is potentially damaging to the trusting relationship required to implement this intervention. According to the Archdiocese of Denver's Hispanic Ministry, a large proportion of Latinos attend their neighborhood church (i.e., in the same zip code) because of a recent increase in the number of churches offering masses in Spanish.

Mammogram claims obtained from Medicaid fee-for-service administrative data were used for the analysis. We compared the rates obtained during the baseline period before the intervention (January 1998–December 1999) with those obtained during a follow-up period (January 2000–December 2001) for Medicaid-enrolled women in each of the intervention groups.

Mammogram use was determined by having the claims with any of the following codes: International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) procedure codes 87.36, 87.37, or diagnostic code V76.1X; Healthcare Common Procedure Coding System (HCPCS) codes GO202, GO203, GO204, GO205, GO206, or GO207; Current Procedural Terminology (CPT) codes 76085, 76090, 76091, or 76092; and revenue center codes 0401, 0403, 0320, or 0400 in conjunction with breast-related ICD-9-CM diagnostic codes of 174.x, 198.81, 217, 233.0, 238.3, 239.3, 610.0, 610.1, 611.72, 793.8, V10.3, V76.1x.

The outcome variable was mammography screening status as determined by the above codes. The main predictors were ethnicity as determined by the Passel-Word Spanish surname algorithm (18), time (baseline and follow-up), and the interventions. The covariates collected from Medicaid administrative data were date of birth (to determine age); total length of time on Medicaid (determined by summing lengths of time spent within dates of enrollment); length of time on Medicaid during the study periods (determined by summing only the lengths of time spent within dates of enrollment corresponding to study periods); number of spans of Medicaid enrollment (a span defined as a period of time spent within one enrollment date to its corresponding disenrollment date); Medicare–Medicaid dual eligibility status; and reason for enrollment in Medicaid. Reasons for enrollment in Medicaid were grouped by categories of aid, which were: 1) old age pension, for persons aged 60 to 64; 2) disabled or blind, representing those with disabilities, along with a small number of refugees combined into this group because of similar mammogram screening rates; and 3) those receiving Aid to Families with Dependent Children (AFDC).  

The chi-square test or Fisher exact test (for cells with expected values less than 5) was used for categorical variables, and ANOVA testing was used on continuous variables with the Welch modification when the assumption of similar variances did not hold. An analysis with generalized estimating equations (GEE) was conducted to determine intervention effects on mammogram screening before and after intervention while adjusting for differences in demographic characteristics, dual Medicare–Medicaid eligibility, total length of time on Medicaid, length of time on Medicaid during the study periods, and number of Medicaid spans enrolled. GEE analysis accounted for clustering by enrollees who were present in both baseline and follow-up time periods. About 69% of the PI enrollees and about 67% of the PSI enrollees were present in both time periods.

GEE models were used to directly compare PI and PSI areas on trends in mammogram screening among each ethnic group. The hypothesis for this model was that for each ethnic group, the PI was associated with a larger increase in mammogram rates over time than the PSI. To test this hypothesis, the following two statistical models were used (one for Latinas, one for NLWs):

Logit P = α + β1time (follow-up vs baseline) + β2intervention (PI vs PSI) + β3 (time*intervention) + β4…n (covariates),

where "P" is the probability of having a mammogram, "α" is the intercept, "β1" is the parameter estimate for time, "β2" is the parameter estimate for the intervention, and "β3" is the parameter estimate for the interaction between time and intervention. A positive significant interaction term suggests that the PI had a greater impact on mammogram screening over time than the PSI among that ethnic group. 

An analysis was also conducted to measure the effect of each of the interventions on reducing the disparity of mammogram screenings between ethnic groups. This analysis involved creating two separate models for each of the interventions (PI and PSI)An analysis was also conducted to measure the effect of each of the interventions on reducing the disparity of mammogram screenings between ethnic groups. This analysis involved creating two separate models for each of the interventions (PI and PSI) to test two hypotheses: 1) Among women exposed to the PI, screening disparity between Latinas and NLWs is smaller at follow-up than at baseline; and 2) Among women exposed to the PSI, screening disparity between Latinas and NLWs is smaller at follow-up than at baseline. The two statistical models used (one for the PI, one for the PSI) were:

Logit P = α + β1time (follow-up vs baseline) + β2ethnicity (Latina vs NLW) + β3 (time*ethnicity) + β4…n (covariates),

where "P" is the probability of having a mammogram, "α" is the intercept, "β1" is the parameter estimate for time, "β2" is the parameter estimate for ethnicity, and "β3" is the parameter estimate for the interaction between time and ethnicity. A significant, positive two-way interaction would indicate that for each intervention, mammogram screening improvement (before and after) was significantly greater in Latinas than in NLWs.

The baseline period included 16,277 women aged 50 to 64, of whom 5865 (36%) were enrolled in Medicaid HMO and subsequently removed, leaving 10,412 with fee-for-service reimbursements for analysis. Analyses were restricted to the 6696 (64%) women enrolled in Medicaid fee-for-service longer than 18 months (Table 1). Latinas represented 22% of this Medicaid population, whereas NLWs represented 57%. More than half of the enrollees in this database (59%) were disabled, a small minority (2%) received AFDC, with the remaining receiving old age pensions (39%). The disabled category consisted of enrollees with disabilities and enrollees who are blind. The majority of enrollees receiving old age pensions and the majority of disabled enrollees were dually eligible for Medicare and Medicaid, in contrast to those receiving AFDC, where the majority were eligible for Medicaid only. Sixty percent did not have any mammogram procedure during the baseline study period. Similar characteristics were observed in the follow-up study population (data not shown).

The baseline demographic characteristics of the study population by intervention region and ethnicity are shown in Table 2. Latinas were significantly older than NLWs  in both intervention areas (P = .002 for the PI, P <.001 for the PSI) and significantly less likely to be dually eligible for Medicare and Medicaid in the PSI (P < .001). In the follow-up time period, similar observations were made (data not shown). 

The crude biennial mammogram rates for Latinas and NLWs enrolled in Medicaid during the baseline and follow-up periods by intervention are shown in Figure 1 and Table 3. In the PI region, follow-up mammogram rates for both Latinas and NLWs slightly increased over time, from 25% to 30% for Latinas (unadjusted GEE, P = .30) and from 32% to 38% for NLWs (unadjusted GEE, P = .40); however, this difference was not statistically significant compared with baseline rates. No significant disparities in mammogram rates were observed in either time period, regardless of intervention group, although Latinas had slightly lower rates in the PI area. In the PSI area, follow-up mammogram rates for Latinas remained unchanged when compared with baseline rates (45% in baseline compared with 43% in follow-up [unadjusted GEE, P = .27]). NLWs in the PSI demonstrated a significant increase in mammogram rates over time (41% in baseline compared with 44% in follow-up [unadjusted GEE, P = .02]). Latinas in the PSI had significantly higher mammogram rates than NLWs in the baseline period only (chi-square test, P = .02).

The GEE analysis directly compared the effects of the interventions on mammogram screening rates for each ethnic group. There was a marginally significant positive interaction term between time and intervention (adjusted GEE, P = .07), suggesting that the PI was more effective than the PSI in increasing mammogram screening among Latinas (Figure 2). Among NLWs, the PI was associated with increases in mammogram screening over time (adjusted GEE, P = .10) (Figure 3). These results suggest that the PI was the only intervention in which Latinas demonstrated modest increases in mammogram screening rates. 

GEE was also used to determine the effect of each intervention on mammogram disparity between Latinas and NLWs. No significant ethnic disparities in screening were observed in the PI regions. Although Latinas residing in the PI regions had less screening than NLWs, the difference was not statistically detectable (adjusted GEE, P = .19). The improvement in the PI region among Latinas did not entail a statistically significant difference (adjusted GEE, P = .90). Interestingly, Latinas in the PSI initially had higher mammogram screening rates than NLWs (adjusted GEE, P = .03). Only the PSI group had a negative significant two-way interaction between time and ethnicity due to an increase in mammogram rates for NLWs compared with Latinas over time (adjusted GEE, P = .04 for baseline and adjusted GEE, P = .24 for follow-up). 

An additional analysis excluding the nonparticipating churches in the PSI showed no significant differences in mammogram rates between nonparticipating and participating PSI groups in either ethnic group.