Randomized Trial of an Intervention to Improve Mammography Utilization Among a Triracial Rural Population of Women

Abstract

Introduction

Mammography is underused by certain groups of women, in particular poor and minority women. We developed a lay health advisor (LHA) intervention based on behavioral theories and tested whether it improved mammography attendance in Robeson County, NC, a rural, low-income, triracial (white, Native American, African American) population.

Methods

A total of 851 women who had not had a mammogram within the past year were randomly assigned to the LHA intervention (n = 433) or to a comparison arm (n = 418) during 1998–2002. Rates of mammography use after 12–14 months (as verified by medical record review) were compared using a chi-square test. Baseline and follow-up (at 12–14 months) surveys were used to obtain information on demographics, risk factors, and barriers, beliefs, and knowledge about mammography. Linear regression, Mantel–Haenszel statistics, and logistic regression were used to compare barriers, beliefs, and knowledge from baseline to follow-up and to identify baseline factors associated with mammography.

Results

At follow-up, 42.5% of the women in the LHA group and 27.3% of those in the comparison group had had a mammogram in the previous 12 months (relative risk = 1.56, 95% confidence interval [CI] = 1.29 to 1.87). Compared with those in the comparison group, women in the LHA group displayed statistically significantly better belief scores (difference = 0.46 points on a 0–10 scale, 95% CI = 0.15 to 0.77) and reduced barriers at follow-up (difference = −0.77 points, 95% CI = −1.02 to −0.53), after adjusting for baseline scores.

Conclusions

LHA interventions can improve mammography utilization. Future studies are needed to assess strategies to disseminate effective LHA interventions to under-served populations.

It is estimated that, during 2006, 212 920 women in the United States will be diagnosed with breast cancer and 40 970 will die of the disease (1). Techniques for early detection of breast cancer, including mammography and clinical breast examination, have been shown to reduce breast cancer mortality by 30% in women aged 50 and older (2–5), although the benefit of mammography screening among women aged 40–49 is still under debate (5–7). Despite the fact that rates of screening mammography have increased over the last several years, many women do not routinely obtain mammograms, and it is estimated that only 70% of women aged 40 and older had a mammogram within the past 2 years (8).

Minority women, older women, women of low socioeconomic status, and those who live in rural areas have low rates of mammography (8, 9). Compared with white women, African American women have lower rates of breast cancer screening examinations (8, 10) and higher rates of breast cancer mortality. Racial differences in rates of screening and mortality are thought to be even more pronounced among rural women because these women are more likely than residents of more urban areas to be poor (9, 11, 12). For example, data from a study conducted in two rural counties in eastern North Carolina indicated that white women aged 50–74 were twice as likely to have had a recent mammogram as African American women of the same age group (54% and 27%, respectively) (9). Native Americans have the poorest rates of survival from cancer of any racial group in the United States (13). Some studies suggest that this racial/ethnic group has higher rates of cancer risk factors (e.g., smoking, alcohol use, and poor dietary habits) and lower rates of cancer screening than other groups (14–18). However, few cancer prevention and control studies have been conducted among Native American populations.

One strategy that has been shown to improve cancer screening in minority and rural populations is the use of lay health advisors (LHAs), community residents who have been trained to deliver health information. LHAs have been used to improve breast cancer screening in rural, mostly African American women in community settings (19, 20) and to improve breast and cervical cancer screening among mainly minority, urban women in a community trial (21). A recent meta-analysis found a small but positive effect of LHAs in promoting breast cancer screening (22). Although the LHA approach has been shown to be effective for various racial and ethnic groups of women (23–31), no randomized trial, to our knowledge, of LHA interventions for mammography utilization has been conducted among minority, rural women.

The goal of the Robeson County Outreach Screening and Education (ROSE) Project (32) is to use LHAs to deliver an individualized health education intervention to improve rates of mammography screening in a population of women who are traditionally underserved by cancer control efforts—specifically, low-income, rural white, African American, and Native American women. The ROSE project includes 897 women aged older than 40 years who had not had a mammogram within the 12 months before enrollment. Each participant completed a survey that asked about her knowledge and behaviors regarding breast and cervical cancer screening. We previously reported that these low-income, rural women, especially Native American and African American women, experienced statistically significantly higher barriers than other women, maintained inaccurate beliefs, and had reduced knowledge regarding cancer screening (32).

These same women were randomly assigned to receive an LHA intervention or to a comparison group for 12–14 months following their baseline survey. Here we report the results of this randomized trial in terms of the overall impact of the LHA intervention. Our primary hypothesis was that women who received the LHA intervention would have higher mammography screening rates than the comparison group at follow-up and that the effect would be similar within each racial group. Our secondary hypotheses were that the LHA intervention would result in improved knowledge and beliefs about mammography screening and in reduced barriers to obtaining this screening. Other goals of the study were to compare self-report and medical record–verified screening to assess the need for medical record review in intervention studies of this type and to assess the cost of the intervention for future replication.

Subjects and Methods

Setting

The study was conducted from February, 1998, through January, 2002, in Robeson County, NC, a rural county whose population consists of three main racial groups: Native Americans (predominantly Lumbee) (38%), African Americans (25%), and Caucasians (33%), with a few Hispanics and Asian/Pacific Islanders (4%). During the study, Robeson County ranked as the eighth poorest of the 100 counties in North Carolina, and less than half of the adults were high school graduates. The lack of a countywide public transportation system contributed to limited access to health care. Age-adjusted breast cancer incidence rates during the study period were higher in Robeson County than in North Carolina as a whole (30.5 and 25.4 per 100 000 women, respectively). Breast cancer mortality rates were 140.2 and 147.1 per 100 000 for the county and the state, respectively (33, 34).

All of the women in the study received health care from the Robeson Health Care Corporation (RHCC), a federally funded consortium of four community health centers. In 1996, RHCC served approximately 12 778 clients from the three largest racial groups in Robeson County; 15% of its clients were white, 26% were African American, and 59% were Native American. The majority (63%) of RHCC users had incomes below the poverty line, 33% had no health insurance, and 32% were covered by Medicaid. Uninsured patients received services on a sliding fee scale.

Subject Selection

Subject selection has been described in detail elsewhere (32). In brief, 2948 charts of women over age 40 who had visited RHCC within the last 2 years were selected at random from a list of female RHCC patients. After review and determination of mammogram receipt in the past 12 months, 1503 women (51%) were identified as eligible to participate—i.e., they had not had a mammogram within that time period. Stratified selection and blocked randomization by race and clinic (two of the four clinics were combined into one stratum) were performed by the statistician (JR) such that approximately one-third of the women were recruited from each racial group and one-third were recruited from each clinic stratum to assess the effect of the intervention within each racial group and to control for any clinic effect.

Each woman's physician was contacted, informed of the patient's eligibility, and asked to contact the study office if he or she did not want the patient to participate. A trained study interviewer then contacted each woman and conducted a telephone screening interview to further confirm eligibility (i.e., that the woman had not had a mammogram since chart review). The study interviewer scheduled an in-person visit with all interested eligible women to obtain written informed consent, sign a medical release form, and complete a baseline interview (see Supplementary Data, available at http://jncicancerspectrum.oxfordjournals.org/jnci/content/vol98/issue17). At the time of consent women were told that they would receive a $10 gift certificate for completing the baseline survey and another $10 certificate for completing a follow-up survey in 12–14 months. The baseline interview inquired about history of breast cancer screening, health care visits, family history of cancer, social support, stage of readiness to change, and knowledge, beliefs, and barriers related to breast cancer screening. After completion of the baseline survey and verification that each woman had not had a mammogram in the past 12 months, the participating women were randomly assigned using stratification by race and clinic stratum to either the LHA intervention or the comparison group by the study statistician (JR). Interviewers, medical records abstractors, and primary care physicians were blinded to intervention assignment. After completing the survey, each woman was mailed a $10 grocery store gift certificate in appreciation of her participation.

As a quality control measure, 10% of the women were recontacted by the assistant project manager to verify that the interviewer had visited and that the survey had been administered. This study was approved by the Institutional Review Boards at Wake Forest University School of Medicine and The Ohio State University.

Intervention Design

Theoretical framework

Several theories were used to guide the design of the LHA intervention. The PRECEDE–PROCEED Model (35,36) provided a framework to identify important screening barriers among participants, e.g., transportation and scheduling problems. Social learning theory (37,38) guided the educational program design and how messages were presented, i.e., the use of LHAs to increase self-efficacy. The Communication–Behavior Change model provided an organizing framework for choosing specific, culturally appropriate messages for the LHAs to deliver (39). The Minority Health Communication Model (Alcalay R: unpublished data) was used to focus on issues related to developing health education for African Americans and Native Americans by incorporating five phases of developing effective channels and methods of communication. The Transtheoretical Model (TTM) (40) was used to judge each woman's stage of readiness to obtain a mammogram and to deliver tailored intervention materials, e.g., messages on postcards.

Intervention development

The educational intervention was designed to address specific barriers experienced by rural women in a manner that was culturally acceptable to all racial groups. The goal of the intervention was to increase awareness of the benefits of early detection of breast cancer and to encourage women to reduce their own risk of breast cancer death by identifying and reducing important barriers to obtaining mammography screening and by providing basic knowledge and education about the breast, breast abnormalities, and breast cancer screening. The intervention was developed through several steps: community analysis, development of prototype materials, focus group review, pretesting, and revision (41).

Two Native American and one African American women who lived in the community were hired as the LHAs. These women— a former nurse, a social worker, and a research study interviewer— were selected because they had good social skills; were organized, professional, and courteous; and could work flexible hours. The women received extensive LHA training over a 1-week period at both Wake Forest University and local settings in Robeson County, with additional follow-up sessions throughout the study period. The training was conducted by the Wake Forest ROSE Project Manager (CT) and other study personnel.

The LHA training included general project information and extensive training on breast development, breast abnormalities, and breast cancer screening, diagnosis, treatment, and risk factors. The project protocol, procedures (visits and telephone and mail contacts), and documentation (study forms) were reviewed. The training included practice and role playing, a review of the LHA resource manual, and discussions of how to handle problems (such as scheduling and protocol adherence). At the end of the training, LHAs took a comprehensive written examination, conducted practice intervention sessions, and performed a breast self-examination demonstration on breast models. LHA supervision consisted of weekly phone or in-person meetings with the LHA supervisor, during which assigned cases were reviewed and any problems or concerns were discussed with the LHA. The LHA supervisor periodically attended patient visits with each LHA to ensure standardization and protocol adherence in delivering the intervention.

Intervention delivery

The LHA intervention was an individualized health education program that was culturally acceptable and tailored to the needs of each woman. It started 2–4 weeks following completion of the baseline survey and consisted of an intensive, face-to-face interactive educational program. The intervention program was administered over a 9- to 12-month period and consisted of three in-person visits, with educational materials provided at each visit and follow-up phone calls and mailings after each visit. During or after each contact, the LHA completed an encounter form to document interactions with each woman, any barriers to obtaining a mammogram cited by the woman, any materials she gave to the woman, and whether the woman had obtained a mammogram since the last contact.

First home visit (45–60 minutes)

The LHA described the project and objectives of the visit, provided educational materials about individual cancer risk and ways to overcome barriers to mammography, and discussed mammography, breast cancer, breast self-examination, and scheduling a mammogram.

Second home visit (30–45 minutes, 2–3 weeks after first visit)

The main focus was to teach and reinforce breast self-examination and to provide feedback on barriers and cancer risk information discussed at the first visit. The need for regular breast self-examination and setting up mammogram appointments, which the LHA offered assistance in scheduling, was also discussed with the patient.

Phone calls and mailings (months 3–9)

Phone calls were made twice—during months 2 and 6, respectively, following the second visit—to assist participants in making mammography appointments, discuss any remaining barriers to obtaining a mammogram, provide information on other important health topics, determine stage of readiness to change, and encourage women to discuss their mammogram experiences. Mailings addressed women's readiness to obtain a mammogram and were tailored to her specific stage of change (as based on the TTM) in relation to getting a mammogram. Each woman received two postcard reminders (at months 4 and 8) that addressed her specific stage of change in relation to obtaining a mammogram.

Final visit (months 10–14)

The LHA asked about screening received and the participant's ability to do breast self-examination and underscored the importance of good breast care. At this last visit, small gifts (such as calendars and cups) were given to participants in appreciation of their time.

Comparison group

Six months after random assignment, women in the comparison group were sent a letter and a National Cancer Institute (NCI) brochure from RHCC calling attention to the need for regular cervical cancer screening. Three months after completing the follow-up assessment (see below), women in the comparison group were sent a letter from RHCC inviting them to obtain a mammogram and an NCI brochure (designed for low-literate women) about mammography.

Follow-up assessment

Twelve to 14 months after completing the baseline survey, all participants were asked to complete a follow-up survey (see Supplementary Data, available at http://jncicancerspectrum.oxfordjournals.org/jnci/content/vol98/issue17). This survey was similar to the baseline survey and administered by the same interviewers except that it asked about the intervention components to assess the value of the intervention and any contamination in the comparison group. Each woman's medical record from RHCC or any other appropriate health/mammogram facility that was reported by the participant during the survey was checked to determine whether she had received a mammogram since the baseline survey was completed in the previous 12 months. Women who refused to complete the entire follow-up survey were asked to complete a shorter version. As a quality control measure, 10% of women were recontacted by the assistant project manager to verify that the follow-up survey had been administered. All participants reported that they had received the follow-up survey.

Statistical Methods

The sample size for this study was based on estimates of an intervention effect from a previous project (21) and was designed to achieve 80% power to detect a difference in mammogram use of at least 10% between treatment groups and 80% power to detect a difference in mammogram use of at least 20% between treatment groups within each racial group. This required a sample size of 800 women (400 per treatment group). On the basis of a predicted attrition rate of 10% for 12-month medical record review data (21), our goal was to randomly assign approximately 900 women. With an attrition rate of 20% for women completing the follow-up survey (as was seen in a previous study) (21), 720 women (360 per intervention group) would be available to analyze the secondary hypotheses assessing changes in barriers to obtaining a mammogram, beliefs about mammography, and knowledge about breast cancer.

Composite scores were calculated for barriers, beliefs, and knowledge using the individual items listed in Table 3 (42). Responses to each item contributed to the score by adding or subtracting one point, resulting in observed ranges of −12 to 6 for the 12 barrier items, −4 to 4 for the four belief items, and −10 to 12 for the 12 knowledge items. The observed scores were then transformed into a 0–10 range. High scores indicated many barriers to receiving mammography, positive beliefs about mammography, and good knowledge of breast cancer, respectively. Baseline differences between treatment groups in demographics, health behaviors, health care access, barriers, beliefs, and knowledge about mammograms were analyzed using chi-square tests and t tests. Health care access variables were important modifiable factors that did not fall specifically under the barriers, beliefs, and knowledge categories but were believed to contribute to receiving mammography screening.

Table 3.

Changes in modifiable factors, by treatment group (N = 815)^*

	LHA intervention group (n = 411)		Comparison group (n = 404)
Factor	Baseline %	Follow-up %	Baseline %	Follow-up %	Treatment effect,† RR (95% CI)
Barriers
Too hard to find time	23.2	16.0 (P = .002)	23.7	20.9	0.77 (0.59 to 1.02)
Do not like to get mammogram at a different place	29.7	22.1 (P = .003)	34.4	29.5	0.79 (0.63 to 0.99) (P = .037)
Do not know where to go to get a mammogram	11.1	3.2 (P<.001)	12.7	7.6 (P = .006)	0.44 (0.24 to 0.82) (P = .008)
No one has ever encouraged me to get a mammogram	44.8	19.0 (P<.001)	45.6	31.4 (P<.001)	0.62 (0.51 to 0.75) (P<.001)
No insurance	26.1	23.8	30.7	28.7	0.96 (0.83 to 1.10)
I have to sit and wait too long	41.6	35.0 (P = .033)	38.2	35.0	0.96 (0.78 to 1.19)
Too hard to get to the doctor's office	15.5	13.9	12.2	11.8	1.01 (0.71 to 1.45)
Doctors and nurses do not treat me with respect	3.3	1.0	3.5	2.4	0.43 (0.11 to 1.65)
Technicians are rude	21.6	9.1 (P<.001)	26.0	16.5 (P<.001)	0.63 (0.45 to 0.88) (P = .005)
Doctor recommended a mammogram but I did not get it	30.7	32.1	34.9	35.3	0.95 (0.75 to 1.21)
I believe my regular doctor wants me to get a mammogram	87.2	92.1 (P = .005)	85.9	88.5	1.04 (0.99 to 1.08)
Cost is a barrier‡	52.8	34.1 (P<.001)	55.0	42.1 (P<.001)	0.83 (0.70 to 0.97) (P = .022)
Mean score (95% CI)	3.41 (3.22 to 3.61)	2.92 (2.73 to 3.12) (P<.001)	3.61 (3.42 to 3.81)	3.78 (3.58 to 3.98)	b = −0.77 (−1.02 to −0.53) (P<001)
Beliefs
It is embarrassing to get a mammogram	23.8	16.0 (P<.001)	27.0	22.1 (P = .046)	0.77 (0.59 to 1.00) (P = .048)
Radiation from mammography causes cancer	39.7	35.5	42.5	43.0	0.85 (0.72 to 1.00) (P = .043)
A mammogram hurts	40.8	31.7 (P = .003)	40.1	33.8 (P = .027)	0.93 (0.76 to 1.14)
I feel OK so why bother getting a mammogram	27.6	15.0 (P<.001)	34.6	23.4 (P<.001)	0.72 (0.55 to 0.94) (P = .016)
Belief score (0–10), Mean (95% CI)	6.65 (6.40 to 6.90)	7.55 (7.30 to 7.80) (P<.001)	6.33 (6.07 to 6.58)	6.95 (6.70 to 7.21) (P<.001)	b = 0.46 (0.15 to 0.77) (P =.004)
Knowledge§
Women aged 40 and older should get mammograms regularly	96.8	95.0	97.9	96.9	0.98 (0.95 to 1.01)
Women should not stop getting screening mammograms at any age∥	74.0	83.5 (P<.001)	72.8	81.9 (P<.001)	1.02 (0.96 to 1.08)
Women should get a screening mammogram once a year	73.0	82.8 (P<.001)	71.2	79.6 (P = .002)	1.04 (0.97 to 1.11)
A woman can tell if she has breast cancer¶	82.0	83.5	77.9	79.1	1.04 (0.98 to 1.11)
Black women are more likely to get breast cancer¶	41.2	45.4	37.9	51.7 (P<.001)	0.86 (0.75 to 0.98) (P = .026)
Breast removal is the only good treatment for breast cancer¶	56.5	61.5	57.5	57.3	1.08 (0.97 to 1.20)
Women with no children are less likely to get breast cancer¶	65.3	66.0	56.7	68.4 (P<.001)	0.93 (0.85 to 1.02)
Older women are more likely to get breast cancer	52.6	52.6	53.9	53.4	0.99 (0.88 to 1.12)
Women should begin regular mammography at age 40#	32.4	36.9	30.0	30.0	1.20 (0.99 to 1.44) (P = .056)
Women over age 50 should have a CBE by a physician every year	98.5	96.8 (P = .035)	98.0	97.7	0.99 (0.97 to 1.01)
After two mammogram results are OK, I do not need another mammogram¶	89.4	93.1 (P = . 019)	88.0	89.8	1.03 (0.99 to 1.08)
Breast cancer runs in families	82.8	73.5 (P<001)	78.6	75.3	0.96 (0.89 to 1.04)
Knowledge score (0–10), mean (95% CI)	6.75 (6.59 to 6.91)	7.0 (6.84 to 7.15) (P = .002)	6.56 (6.40 to 6.72)	6.92 (6.76 to 7.09) (P<.001)	b = −0.02 (−0.21 to 0.17)
Health care access
Ever had mammogram without doctor ordering it	22.2	54.6 (P<.001)	21.6	32.1 (P<.001)	1.68 (1.47 to 1.93) (P<.001)
Ever asked doctor to order mammogram	12.8	25.3 (P<001)	14.0	24.4 (P<.001)	1.08 (0.91 to 1.29)
Ever encouraged by doctor to get mammogram	34.1	42.7 (P<.001)	32.5	49.6 (P<.001)	0.84 (0.76 to 0.92) (P<.001)
Never heard of BCCDP	91.1	74.9 (P<.001)	87.7	91.1 (P<.001)	0.89 (0.84 to 0.94) (P<.001)

^*A total of 815 out of 851 women completed the follow-up survey. Comparisons between baseline and follow-up for each intervention group were performed with McNemar's test for agreement (for categorical factors) and paired t tests (for comparison of means). P values less than .05 were reported. LHA = lay health advisor; CI = confidence interval; CBE = clinical breast examination; BCCDP = Breast and Cervical Cancer Detection Program.

^†Risk Ratios (with 95% CIs) represent the multiple of risk in the LHA intervention group compared with the comparison group for agreement with each barrier/belief/knowledge item. P values are calculated from a Mantel–Haenszel test for the association of LHA intervention group with each barrier/belief/knowledge item at follow-up, after adjusting for baseline values. Linear regression was performed on each continuous barrier, belief, and knowledge score with treatment group as the main predictor and adjusting for baseline score as a covariate, and parameter estimates (with P values) are reported for the difference between treatment and comparison groups; P values less than .05 were reported.

^‡Answered yes to one of five questions related to cost.

^§Percentage of women who answered each item correctly.

^∥Percentage of women who chose “never stop” when asked to select an age when they should stop getting a mammogram.

^¶Percentage and risk ratio for women who disagreed with this false statement.

^#Percentage and risk ratio for women who chose the category “age 40–44.”

The primary outcome was medical record–verified mammography in the past 12 months (yes/no) at the follow-up assessment. The relative risk (RR), with 95% confidence interval (CI), of mammography for the LHA group compared with the comparison group was computed, and mammography rates in each treatment group were compared with a chi-square test. Relative risks and chi-square tests also were used to compare mammography rates between treatment groups within each racial/ethnic group. Mantel–Haenszel tests were used to test for an association between mammography use at the follow-up assessment and baseline factors after adjustment for treatment group. The baseline factors included smoking status (current/former/never); socioeconomic status (higher [private health insurance, high school graduate, and annual household income of $20 000 or more]/lower [no private health insurance, not high school graduate, or annual income less than $20 000]); personal history of colon, breast, or ovarian cancer (no/yes); family history of breast cancer (no/yes); reproductive history, i.e., nulliparity or age over 30 years at birth of first child (no/yes); had a regular checkup in the last 12 months (disagree/agree); last mammogram (within last 3 years/more than 3 years ago/never); last clinical breast examination (within last year/more than 1 year ago/never); ever had a mammogram without a doctor ordering it (no/yes); ever asked doctor to order a mammogram (no/yes); ever encouraged by a doctor to get a mammogram (yes/no); never heard of the Breast and Cervical Cancer Detection Program (BCCDP) (disagree/agree); clinic (three strata); race (white/Native American/African American); age group (40–49/50–59/60–69/70–79/80+); education (less than high school graduate/high school graduate/some college or college graduate); work status (full time or part time/retired/homemaker/disabled/other); marital status (married or living together/divorced or separated/widowed/never married); and any health insurance (no/yes). Logistic regression was used to test the relationship of continuous scores on the baseline bar riers, belief, and knowledge scales with mammography at the follow-up assessment after adjustment for treatment group. Relative risks and 95% confidence intervals were calculated for the categorical variables. Odds ratios (ORs) for a one-unit increase in each barrier, belief, and knowledge score with 95% confidence intervals were calculated from the logistic regression models.

Medical record–verified mammography reports were compared with self-reports for women who completed the follow-up survey using κ statistics and McNemar's test to examine accuracy and false-positive rates, respectively. Logistic regression modeling (43) was performed with purposeful forward selection to identify a set of baseline factors (from those listed in Fig. 2) that were most strongly associated with receipt of mammography during the 12 months before the follow-up assessment, after simultaneously adjusting for treatment group and race. Resulting odds ratios were converted to risk ratios according to Zhang and Yu (44).

Fig. 2.

Relative risks of mammography at follow-up for each baseline factor among 851 women available for analysis at follow-up, after adjustment for treatment group. Relative risks (points) and 95% confidence intervals (bars) were calculated with Mantel–Haenszel statistics.

*Odds ratios from logistic regression models are presented for continuous barrier, belief, and knowledge scores instead of relative risks; † P = .051; ‡ P = .017. All other P values are >.10.

Changes in barriers, beliefs, and knowledge were analyzed by comparing the baseline survey responses with those at the follow-up assessment separately by treatment group, using McNemar's test for paired responses. These analyses included the 815 women who completed the follow-up survey and who were not of “mixed” race/ethnicity (n = 5). Risk ratios were calculated to evaluate the relative risk in the LHA intervention group compared to the comparison group for agreement with each barrier, belief, or knowledge item (i.e., the items listed in Table 3) at follow-up, after adjustment for baseline values. These risk ratios, along with their 95% confidence intervals and P values, were calculated from a Mantel–Haenszel test for the association of treatment group with each item at the follow-up assessment, after adjusting for baseline values. For continuous composite scores, paired t tests compared means between time points within treatment group, and linear regression models tested the intervention effect on the follow-up assessment responses after adjustment for baseline values.

Analyses were conducted using SAS System for Windows version 9.1 (45). All statistical tests used a two-sided α = .05 level of significance. No adjustments for multiplicity were made because the primary analysis was of only one comparison.

The costs of providing the intervention were estimated based on the salaries of the LHAs and the supervisor and all direct travel and supply costs associated with training and delivery of the intervention. The costs of delivering the intervention for each additional mammogram received in the LHA group were estimated by 1) adding the costs associated with the intervention delivery, 2) calculating the number of additional mammograms received in the LHA group versus the comparison group, and 3) dividing the total costs by the number of additional mammograms received (46).

Results

Study Population

Of 1503 potentially eligible women, 481 were excluded from the study after the initial telephone screening interview (Fig. 1). Reasons for exclusions included not meeting the eligibility criteria (n = 206 women had already had a mammogram within the last year or were scheduled to have one and n = 3 were under age 40 years), having moved (n = 70), being deceased (n = 20), being mentally or physically unable to participate (n = 40), and being unreachable (n = 128). An additional 14 women were excluded for miscellaneous reasons, such as language/hearing impairment or duplicate contact. Of the remaining 1022 women, 901 answered the baseline survey, for an overall participation rate of 88%. During the baseline interviews, it was discovered that four women had had a recent mammogram, and these women were excluded from the study, resulting in the random assignment of 897 women, 453 (51.5%) to the LHA group and 444 (49.5%) to the comparison group.

Fig. 1.

Flow of participants through the trial.

A total of 41 women were lost to follow-up (23 in the LHA group and 18 in the comparison group) (Fig. 1). Thus, 856 members of the original cohort (95%) were eligible for the follow-up survey. Among these women, 820 completed the follow-up survey, for a response rate of 96% (95% and 97%, respectively, for the treatment and comparison women). The rates of completion of the first, second, and third intervention visits were 96%, 95%, and 89%, respectively. For the 4% of women who refused or could not be reached to complete any version of the follow-up survey, medical records were used to examine mammography use. Outcome data (i.e., receipt of a mammogram within the last 12 months) were available from medical records for all 856 women who were eligible for follow-up.

Baseline Survey

Women who defined themselves as “mixed” race/ethnicity (n = 5) were removed from the analysis for consistency with our previous publication (32), leaving 851 women to be analyzed based on intention-to-treat principles (433 in the LHA group and 418 in the usual-care group). Thirty-three percent of the participants were African American, 42% were Native American, and 25% were white (Table 1). The average age was 55.08 years (95% CI = 54.33 to 55.83). Overall, 29% of the women did not have any health insurance, and 83% were in the lower socioeconomic status group (i.e., they did not have private health insurance, they had not graduated from high school, or their household income was less than $20 000 per year). Almost half (44%) of the women had not finished high school, and 43% worked outside the home. Regarding breast cancer risk factors, 21% of women had a family history of breast cancer and 50% had a history of cigarette smoking. A total of 74% of the women reported having had a checkup from a physician in the last year. There were no statistically significant differences in any of these characteristics by treatment group, although there were some differences in these characteristics by racial group (32).

Table 1.

Demographics and risk factors of participants, by treatment group^*

Factor	Percentage of LHA intervention group (n = 433)	Percentage of comparison group (n = 418)	Percentage of total group (n = 851)
Race
African American	33	33	33
Native American	42	42	42
White	24	25	25
Age, y
40–49	45	41	43
50–59	28	27	28
60–69	15	18	16
70–79	10	10	10
80+	2	4	3
Mean (95% CI)	54.52 (53.50 to 55.53)	55.67 (54.56 to 56.78)	55.08 (54.33 to 55.83)
Educational level
<High school	42	45	44
High school graduate	30	33	31
Some college/college graduate	28	22	25
Work status
Work full time/part time	45	41	43
Retired	13	12	13
Homemaker	12	16	14
Disabled	24	26	25
Other	5	6	6
Socioeconomic status
Higher†	20	15	17
Lower	80	85	83
Marital status
Married/living together	45	47	46
Divorced/separated	23	23	23
Widowed	24	22	23
Never married	8	8	8
Health insurance
Yes	74	69	71
No	26	31	29
Personal history of colon, breast, or ovarian cancer	3	2	2
Family history of breast cancer	18	23	21
Nulliparity or >30 y at first birth	11	11	11
Smoking status
Current	30	30	30
Former	20	19	20
Never	49	51	50
Regular checkup in the last 12 months	75	73	74
Last mammogram
Within last 3 y	60	56	58
>3 y ago	18	19	19
Never	22	24	23
Clinical breast examination
Within last year	51	50	50
>1 y ago	41	43	42
Never	8	7	8

^*Statistical significance of differences among groups was tested using chi-square tests for each categorical factor and t tests to compare means. No differences were statistically significant. LHA = lay health advisor; CI = confidence interval.

^†Higher socioeconomic status is defined as having private health insurance, being at least a high school graduate, and having a household income of ≥$20 000 per year.

At baseline (Table 2), the most often reported barriers to mammography receipt were cost (54%) and lack of encouragement (45%). Nevertheless, 86% of women believed that their doctor wanted them to get a mammogram. Again, no statistically significant differences were evident between the treatment groups in terms of barriers.

Table 2.

Modifiable factors at baseline by treatment group: barriers, beliefs, knowledge, and health care access^*

	Percentage of LHA intervention group (n = 433)	Percentage of comparison group (n = 418)	Percentage of total group (n = 851)
Barriers
Too hard to find time	24	24	24
Do not like to get a mammogram at different place	30	33	32
Do not know where to go to get a mammogram	11	12	12
No one ever encouraged me to get a mammogram	45	45	45
Do not have insurance	26	31	29
I have to sit and wait too long	41	39	40
It is too hard to get to MD's office	14	12	13
MDs/nurses do not treat me with respect	3	4	3
Technicians are rude	21	26	24
MD recommended mammogram but I did not get it	26	27	26
You believe your regular doctor wants you to get a mammogram	87	86	86
Cost is a barrier†	53	54	54
Barrier score
Mean (95% CI)	3.44 (3.25 to 3.63)	3.58 (3.38 to 3.78)	3.51 (3.37 to 3.65)
Beliefs
Too embarrassed to get a mammogram	24	26	25
Radiation from mammogram can cause cancer	40	42	41
Mammogram hurts	41	40	41
I feel okay, so why bother getting mammogram?	28	34 (P = .034)	31
Beliefs score
Mean (95% CI)	6.63 (6.38 to 6.88)	6.37 (6.11 to 6.63)	6.50 (6.32 to 6.68)
Knowledge‡
Women should begin getting screening mammograms at age 40§	34	30	32
Women aged 40+ should get mammograms regularly	96	98	97
Women should not stop getting screening mammograms at any age∥	74	73	73
Women should get a screening mammogram every year	73	72	73
A woman can tell if she has breast cancer¶	81	78	80
Black women are more likely to get breast cancer¶	42	38	40
Breast removal is the only good treatment for breast cancer¶	56	57	56
Women who had no children are less likely to get breast cancer¶	65	57 (P = .015)	61
Older women are more likely to get breast cancer than younger women	54	55	54
Women who are more than 50 years old need a CBE each year	98	98	98
After two mammogram results are OK, do not need another mammogram¶	90	89	89
Breast cancer runs in families	83	79	81
Knowledge score
Mean (95% CI)	6.76 (6.60 to 6.92)	6.56 (6.40 to 6.72)	6.67 (6.55 to 6.78)
Health care access
Ever encouraged by doctor to get mammogram	34	33	33
Ever had mammogram without doctor ordering it	23	21	22
Ever asked MD to order mammogram	12	14	13
Never heard of BCCDP	91	88	90

^*Chi-square tests were performed for each categorical factor, and t tests were used to compare means; P values less than .05 are reported. LHA = lay health advisor; CI = confidence interval; CBE = clinical breast examination; BCCDP = Breast and Cervical Cancer Detection Program.

^†Answered yes to one of five questions related to cost.

^‡Percentage of women who answered each item correctly.

^§Percentage values reflect women who chose the category “Age 40–44.”

^∥Percentage values reflect women who chose “never stop” when asked to select an age when they should stop getting a mammogram.

^¶Percentage values reflect women who disagreed with this false statement.

The most commonly reported negative beliefs about mammography at baseline were that radiation from a mammogram can cause cancer (41%), that mammograms hurt (41%), and that “I feel okay, so why bother getting a mammogram” (31%). Only the last belief showed a statistically significant difference between treatment groups at baseline (28% of the LHA group versus 34% of the comparison group; P = .034).

In terms of baseline knowledge, 44% of women believed that “the only good treatment for breast cancer is an operation to remove the breast.” Only approximately one-third of women knew of the recommendation to begin getting mammograms at age 40, and 61% knew that women who had children were less likely to get breast cancer. Responses to the latter item were statistically significantly different between groups (with the correct answer provided by 65% of women in the LHA intervention group and 57% of those in the comparison group; P = .015).

Health care access factors that were most often reported included lack of encouragement from a doctor (67%) and no knowledge of BCCDP in Robeson County (90%), which provides free mammograms to low-income women. Responses were not statistically significantly different between groups.

Mammogram Receipt

Our analysis of medical record data for all 851 women (not shown) revealed that 42.5% of women in the LHA intervention group and 27.3% of those in the comparison group received a mammogram in the 12 months before the follow-up assessment (RR = 1.56, 95% CI = 1.29 to 1.87, P <.001). The intervention showed a statistically significant association with mammography receipt within each racial group: African Americans (RR = 1.54, 95% CI = 1.11 to 2.14, P = .008), Native Americans (RR = 1.58, 95% CI = 1.18 to 2.13, P = .002), and whites (RR = 1.54, 95% CI = 1.05 to 2.25, P = .024). No statistically significant differences in screening rates were observed between racial groups or clinics (data not shown).

To assess the validity of women's self-report of mammography, we compared medical record–validated mammography receipt reports and self-reports of mammogram receipt in the past 12 months. These reports showed a 76.9% overall agreement (κ = 0.55, McNemar's test; P <.001). Self-reports were statistically significantly less reliable (P <.001, chi-square test) in the LHA group (71.2% overall agreement, κ = 0.45) than in the comparison group (82.8% overall agreement, κ = 0.63). The difference was related to the difference in the rates of erroneous reports of receiving screening (48.2% in the LHA group and 22.1% in the comparison group, P <.001, chi-square test). There were no differences in the proportion of missing medical records by treatment group (data not shown).

The cost of delivering the intervention over the course of the study was estimated to be $329 054 (for LHA and LHA supervisor salaries and benefits plus supply and travel costs). The difference in mammography rates between the two groups was 15.2%, which translates into 66 additional mammograms in the LHA group; therefore, each additional mammogram in the LHA group cost $4986.

Association Between Baseline Factors and Receiving a Mammogram at Follow-up

The associations between baseline factors and whether a mammogram was received during the 12 months before the follow-up assessment are shown in Fig. 2. A one-unit increase on the barrier scale (0–10) was associated with a borderline statistically significantly lower odds of mammogram receipt (OR = 0.93; 95% CI = 0.87 to 1.00, P = .051). Women who currently smoked had 0.77 times the probability of receiving a mammogram at follow-up than those who never smoked (95% CI = 0.62 to 0.96, P = .016). No other baseline factors were statistically significantly associated with receiving a mammogram. There were also no statistically significant interactions between baseline factors and treatment group at the α = .05 level.

Logistic regression modeling revealed the baseline factors most strongly associated with mammography receipt during the 12 months before the follow-up assessment to be smoking status and time since last clinical breast examination, after adjustment for treatment and racial group (data not shown). Women who never smoked were 1.25 (95% CI = 1.03 to 1.52) times as likely to obtain a mammogram as former or current smokers, and women who had a clinical breast examination in the past year were 1.38 (95% CI = 1.09 to 1.69) times as likely to have a mammogram as women with no clinical breast examination or who had had a clinical breast examination more than a year ago after adjustment for treatment and racial group. In addition, after adjusting for racial group, smoking, and clinical breast examination history, the probability of receiving a mammogram during the 12 months before the follow-up assessment in the LHA intervention group was 1.57 (95% CI = 1.31 to 1.84) times that in the comparison group.

Changes From Baseline Survey to Follow-up Survey

Data on changes in modifiable factors (barriers, beliefs, and knowledge) from the baseline to follow-up survey are presented in Table 3, along with risk ratios for the difference between treatment groups in each barrier, belief, or knowledge item at follow-up, after adjusting for baseline values. The average barrier score was statistically significantly reduced in the LHA intervention group (P<.001), with no statistically significant change in the comparison group. At follow-up, the average barrier score was significantly smaller in the LHA intervention group than the comparison group (P<.001), after adjusting for baseline. Women in the LHA intervention group also showed statistically significantly reduced risks of reporting many barriers at follow-up, such as not knowing where to get a mammogram (P = .008) and nobody having encouraged them to get a mammogram (P<.001). The probability of ever having obtained a mammogram without a doctor ordering it was 1.68 times as high for the women in the LHA intervention group (P<.001) and their relative risk of never having heard of BCCDP was 0.89 compared with the comparison group (P<.001).

The proportion of women reporting inaccurate beliefs was statistically significantly reduced between the baseline and follow-up surveys for three of the four items in both the treatment and comparison groups (Table 3). Composite belief scores (higher is better) were statistically significantly increased, from 6.65 to 7.55 in the LHA intervention group (P<.001) and 6.33 to 6.95 in the comparison group (P<.001). The average belief score was 0.46 points higher (P = .004) in the LHA intervention group at follow-up, after adjustment for baseline belief score, than in the comparison group.

For six of the 12 knowledge items, the proportion of women in the LHA intervention group who provided a correct answer underwent a statistically significant increase between the baseline and follow-up surveys. The comparison group showed a statistically significant increase for only four of the items. Composite knowledge scores showed a statistically significant increase in the LHA intervention group (P = .002) and in the comparison group (P<.001). The knowledge scores at follow-up were not statistically significantly higher in the LHA intervention group after adjusting for baseline. Women in the LHA intervention group did not have statistically significantly higher chances of giving correct answers at the follow-up survey than the comparison group for any of the knowledge items (P>.05).

DISCUSSION

The ROSE study was designed to improve mammography screening among low-income women from three racial groups in a rural county. An LHA personally delivered the intervention, which was based on multiple behavioral theories, because not all women have similar barriers or learning styles (47). The population included in the ROSE study was medically underserved in that almost 30% of the participants did not have any health insurance. They also had lower socioeconomic status and lower educational levels than North Carolina averages, as well as relatively poor knowledge of breast screening. Although most women reported having regular checkups, 67% of the women reported not getting a recommendation for a mammogram from their physician. Similar studies have reported this finding (i.e., women see a doctor but do not receive a recommendation for screening) and characterized it as a missed opportunity for conducting or recommending screening (21,32,48,49).

The women assigned to the LHA intervention had higher mammography rates at the follow-up assessment than the comparison group (42.5% versus 27.3%), and this effect was found for all three racial groups. Results of further analyses that examined changes in knowledge, beliefs, and barriers suggest that the LHA intervention was effective at improving beliefs, reducing barriers, and increasing knowledge regarding mammography utilization. In addition, the LHA was an agent for behavior change related to mammogram use in that, at the follow-up survey, women in the LHA group were 0.6 times as likely to report that nobody had encouraged them to get a mammogram and twice as likely to obtain a mammogram without a doctor's order compared to the comparison group. Thus, this trial demonstrates that LHA interventions can promote mammography screening in low-income, rural, minority women.

Many barriers to receiving mammograms, such as the lack of encouragement and time and the cost of obtaining the test, were statistically significantly reduced in women who received the LHA intervention. In addition, statistically significantly more women in the LHA intervention group than the comparison group knew at follow-up that free or low-cost mammograms were available through the county BCCDP.

Smoking and time since last clinical breast examination were the factors that were most strongly associated with receiving a mammogram during the 12 months before the follow-up assessment and after adjustment for treatment group and race. This result is consistent with findings from past studies (50,51). We did not find associations between mammography screening and commonly studied factors, such as insurance status, age, and recommendation by a physician.

Many interventions have been developed and tested to improve mammography rates in underserved populations. Mandelblatt and Yabroff (52,53) conducted a meta-analysis to identify strategies and interventions that were most effective at enhancing clinician- and patient-targeted strategies to increase mammography utilization. They found that the most effective interventions were patient-targeted behavioral interventions (e.g., telephone reminders or letters of invitation), theory-based cognitive interventions (e.g., health education), and sociologic interventions (e.g., use of peer or lay support). The present study used a combination of these three types of interventions: postcards, health education by an LHA, and a sociologic intervention in the form of the LHA who offered support for mammography.

The study demonstrates that the LHA intervention strategy was successful in a triracial, rural population. LHAs provide a personalized intervention, as well as navigation through the health care system, social networking, and social support (54), and serve as a link between community members and the medical care system through outreach, education, information dissemination, and, in some cases, system navigation (55). With the findings of this study, this type of intervention has now been reported to be successful in studies of community outreach targeting breast and/or cervical screening for women of many racial and ethnic groups throughout the United States (20,21,23–31).

Secondary outcomes from this study included the accuracy of patient self-report of mammography and the cost of the intervention. Assessment of the accuracy of self-reports indicated poor overall agreement with validated reports, especially in the LHA group. Women in the LHA intervention group were asked by the LHAs about their receipt of mammography during the intervention period, but these responses were not used as outcome data. However, constant reminding about mammography might have biased self-reports of mammography to interviewers on the follow-up survey. Because interviewers asked women where the reported mammogram was received and then searched records at the reported facilities, this result is not likely to have been influenced by missing medical records because there were no differences in the proportion of missing medical records by treatment group. The low accuracy of self-report of mammography, which has also been seen in other studies (56–58), suggests that future studies should not rely only on self-reports of screening for outcome assessment, especially among intervention groups, due to possible differences in the accuracy of reporting by treatment group. Failure to correct for this reporting bias by treatment arm could overestimate the efficacy of an intervention.

The cost of the intervention per additional mammogram received was $4986 in direct costs. This estimate is higher than the cost per additional mammogram of the most costly intervention in the study population reported by Andersen et al. (46) ($2451 per additional regular user) but closer to the cost per additional mammogram for interventions in women who were not in compliance with screening guidelines ($2267–$4650 per additional new user in their population) (46). Thus, the cost of this intervention is comparable to other interventions tested among women who needed a mammogram.

The present study has several strengths, including the use of a randomized controlled design, a fairly large sample size, complete follow-up by medical records for outcomes, high participation rates, and a focus on a multiracial, underserved population. Few prior studies examining LHA interventions have been designed in this manner. Limitations of the study include the limited generalizability of the results to other populations, because our population was rural, low income, and of three racial groups, and the cost of delivering such an in-person intervention for physician offices. The use of medical record verification reduced reporting bias; however, some data on mammography use could have been missing. We know of no reason to believe that there would be any difference in the amount of missing information on screening test receipt by treatment arm. The high response rates to the study and follow-up survey also reduce respondent bias. These results should be replicated in other settings to assess the transferability of the intervention. Other ways to deliver the intervention, e.g., using trained volunteers, may be feasible and could reduce costs.

In conclusion, LHAs are an effective way to deliver messages about health to underserved populations. Future research should examine cost-effective ways to disseminate such interventions. These strategies are needed to reduce the disparate burden of disease among underserved populations.