Improving follow-up to abnormal breast cancer screening in an urban population

A patient navigation intervention

Authors

  • Tracy A. Battaglia MD, MPH,

    Corresponding author
    1. Women's Health Unit, Boston University School of Medicine, Boston, Massachusetts
    2. Women's Health Interdisciplinary Research Center, Boston University School of Medicine, Boston, Massachusetts
    • Women's Health Unit, Section of General Internal Medicine, Evans Department of Medicine, Boston University School of Medicine, Doctors Office Building, 720 Harrison Avenue, Suite 1108, Boston, MA 02118, USA
    Search for more papers by this author
    • Fax: (617) 638-8026

  • Kathryn Roloff BS,

    1. Department of Organizational Behavior, Harvard Business School, Boston, Massachusetts
    Search for more papers by this author
  • Michael A. Posner PhD,

    1. Department of Mathematical Sciences, Villanova University, Villanova, Pennsylvania
    Search for more papers by this author
  • Karen M. Freund MD, MPH

    1. Women's Health Unit, Boston University School of Medicine, Boston, Massachusetts
    2. Women's Health Interdisciplinary Research Center, Boston University School of Medicine, Boston, Massachusetts
    Search for more papers by this author

  • Presented at Exploring Models to Eliminate Cancer Disparities Among African American and Latino Populations: Research and Community Solutions, Atlanta, GA, April 21–22, 2005.

Abstract

Delays in follow-up after cancer screening contribute to racial/ethnic disparities in cancer outcomes. We evaluated a patient navigator intervention among inner-city women with breast abnormalities. A full-time patient navigator supported patients using the care management model. Female patients 18 years and above, referred to an urban, hospital-based, diagnostic breast health practice from January to June 2000 (preintervention) and November 2001 to February 2003 (intervention), were studied. Timely follow-up was defined as arrival to diagnostic evaluation within 120 days from the date the original appointment was scheduled. Data were collected via computerized registration, medical records, and patient interview. Bivariate and multivariate logistic regression analyses were conducted, comparing preintervention and intervention groups, with propensity score analysis and time trend analysis to address the limitations of the pre–post design. 314 patients were scheduled preintervention; 1018, during the intervention. Overall, mean age was 44 years; 40% black, 36% non-Hispanic white, 14% Hispanic, 4% Asian, 5% other; 15% required an interpreter; 68% had no or only public insurance. Forty-four percent of referrals originated from a community health center, 34% from a hospital-based practice. During the intervention, 78% had timely follow-up versus 64% preintervention (P < .0001). In adjusted analyses, women in the intervention group had 39% greater odds of having timely follow-up (95% CI, 1.01–1.9). Timely follow-up in the adjusted model was associated with older age (P = .0003), having private insurance (P = .006), having an abnormal mammogram (P = .0001), and being referred from a hospital-based practice, as compared to a community health center (P = .003). Our data suggest a benefit of patient navigators in reducing delay in breast cancer care for poor and minority populations. Cancer 2007. © 2006 American Cancer Society.

The unequal burden of cancer is highlighted among lower socioeconomic status and racial/ethnic minority women who suffer higher mortality from breast cancer compared with their more affluent non-Hispanic white counterparts. Age standardized death rates among African American women with breast cancer exceed those of non-Hispanic white women. Similarly, residents of poorer counties have higher death rates from breast cancer than do residents of more affluent counties. 1 Advanced stage at diagnosis, which contributes to poorer outcomes, is found more frequently in these populations. Recent SEER statistics show that the proportion of women diagnosed with regional- and distant-stage breast cancer continues to be higher among African Americans and Hispanics than among non-Hispanic whites. The same is true when one compares high- versus low-poverty census tracts; the proportion diagnosed with distant stages is higher in high-poverty census tracts.1

Improvements in access and equity in mammography screening alone have not translated into survival improvements for these disadvantaged women. 1 Although racial/ethnic differences in mammography utilization rates have been found to explain some of the observed differences in stage at diagnosis,2 several observational studies have not found this to be a significant explanatory factor.3–5 Potential reduction in morbidity and mortality through breast cancer screening will never be realized without timely and efficient follow-up care once an abnormality has been detected. Investigating the disparities in follow-up after abnormal breast screening serves as the next step in addressing breast cancer health disparities. Although measures of follow-up lack precision,6 numerous studies have documented that racial/ethnic minority women often suffer from the longest delays at this point.5, 7–11

Patient navigation, a type of care management that encompasses a wide-range of advocacy and coordination activities, has been proposed to address known barriers to the delivery of high-quality cancer care in underserved populations. 12–14 A recent national case study review of emerging programs to reduce racial/ethnic disparities in cancer found that almost all adopted a variant of patient navigation.15 Despite increasing interest along with both federal and private resource allocation,16, 17 published evidence of the benefit of navigation programs is limited.18

We evaluated a hospital-based patient navigator intervention among inner-city minority women with breast abnormalities. The main objective of the intervention is to improve the rate of timely diagnostic follow-up in comparison to a group of historical controls, and to identify characteristics of patients who are most at risk for loss to follow-up.

MATERIALS AND METHODS

This pre–post intervention study was conducted at a hospital-based diagnostic breast health practice at a major academic medical center in Boston, Massachusetts. The specialty practice accepts referrals for evaluation of any breast health issue, including screening abnormalities, suspicious symptoms, or elevated cancer risk. Patients served include those who receive their primary care at the academic medical center and those from over 20 affiliated community health centers throughout Boston. Together with its affiliated health centers, the academic institution serves as the major safety-net hospital in the region. All income eligible, uninsured women can receive services without charge. This study was approved by our Institutional Review Board. Written informed consent was waived since the intervention was implemented as a new standard of care for all patients. The final dataset used in these analyses was devoid of all patient identifiers.

Study Sample

The sample for this study included all women with scheduled visits at the diagnostic breast health practice from January through June 2000 (preintervention, N = 314) and November 2001 through February 2003 (intervention, N = 1018). We included consecutive female patients >18 years referred for evaluation during the study period, as referral criteria to the practice were consistent throughout this time period. Individual women were only included once throughout the study, so that preintervention and intervention subjects are unique women. For women seen more than once during this timeframe, we evaluated their initial scheduled visit only, using subsequent visits for the sole purpose of evaluating follow-up within 120 days.

Preintervention

As a result of initial observations of patient care data showing high rates of failure to arrive for scheduled evaluation, a formal program evaluation was conceived. At the time of preintervention data collection, the standard protocol for these patients used existing secretarial staff to attempt telephone contact without clinical oversight. In the ensuing 16 months prior to program implementation, this protocol continued without other institutional or practice changes while a needs assessment was undertaken to inform the intervention. Findings from structured interviews with patients, breast health providers/support staff, and referring providers identified lack of coordination of care and communication as major barriers to diagnostic evaluation. Program planning, hiring, and navigator training occurred during this time period.

Intervention

The patient navigator intervention was guided by the principles of Care Management.19 Services provided by the navigator focused around 4 key activities: 1) case identification, 2) identification of individual barriers to care, 3) implementation of a care plan, and 4) tracking through completion. Criteria for hiring of the patient navigator included experience caring for a diverse patient population and knowledge of the existing local health systems. The patient navigator was trained to coordinate care for each patient referred for diagnostic evaluation. Training included written triage and follow-up protocols as well as monthly barriers-focused cultural competence training at the local department of public health. Initial contact with patients began after appointments were scheduled, 1 week before their scheduled visit. Telephone outreach with all patients, using interpreters for non-English speaking women, was attempted to confirm appointments, provide information about the visit, and learn about any individual barriers to arriving for that appointment. The navigator utilized available resources to address those barriers. A key function of the navigator included advocating for the patient through rapid communication with breast health providers, referring providers, and with other specialty sites, including radiology, surgery, and pathology. In accordance with the 4th key activity, the patient navigator, with daily assistance and weekly oversight by the study coordinator, tracked information on patient demographics, reason for referral, site and provider who referred the patient, diagnostic evaluation conducted, and outcomes of evaluation. Dates from time of initial referral through arrival to first scheduled visit were collected.

Data Collection

Preintervention data collection was conducted via retrospective chart review for women referred to the diagnostic breast evaluation practice from January through June 2000. The administrative registration database was reviewed for the date that the first diagnostic appointment was scheduled, demographic data, and the date at which the patient first arrived for evaluation. If the patient's race was not listed in the registration database, but their birthplace and native language were listed, race was extrapolated from birthplace and native language. For example, if a woman was born in Haiti and spoke Haitian-Creole, she was categorized as black. The clinical record was abstracted for reason for visit based on the following categories: abnormal mammography (BI-RADS category 0, 3, 4, 5), abnormal clinical breast exam, or other (including suspicious symptoms or elevated cancer risk). To ensure accuracy of data, all preintervention data were reviewed by a second chart abstractor.

Intervention data were collected prospectively by the navigator beginning 1 week before the scheduled diagnostic visit at the time schedules were reviewed for case identification and telephone outreach. Following preintervention data collection protocol, the administrative registration database was reviewed for the date that the diagnostic appointment was scheduled and demographic data while the clinical record was abstracted for reason for visit. The date the subjects first arrived to a diagnostic evaluation was documented prospectively by the navigator. To ensure completeness and accuracy of data, all intervention data were reviewed by a one of the authors on a weekly basis (TB, KF). The final analytic file was a limited dataset devoid of key patient identifiers.

The main outcome in this intervention study was the dichotomous variable timely follow-up (yes, no). Subjects were considered to have timely follow-up if they arrived to a diagnostic evaluation visit within 120 days from the date the original appointment was scheduled. Although an ideal outcome would be time to actual diagnosis or resolution of the abnormality, limited data collected during the preintervention period prohibited this comparison. Since no gold standard exists to determine what constitutes timely diagnostic follow-up, 6 we operationalized the concept based on: 1) literature that suggests diagnostic and treatment delays of 3–6 months may impact survival20; and 2) review of our data, which indicated that beyond 120 days no appreciable additional follow-up was achieved. Furthermore, repeating our analyses using 90 days as the definition of timely follow-up did not change our findings. To determine 120 day follow-up rates for patients who were seen within the last 120 days for the specified study time period, records beyond the study end date were examined.

Statistical Analyses

Baseline demographic data, including age, race, reason for consultation, site of referral, need for interpreter services, and insurance status, were compared between subjects presenting for care during the preintervention and intervention period. Bivariate comparisons using χ2 tests of independence tested for differences in timely follow-up between the 2 groups overall, and within specific demographic variables. Logistic regression was conducted to assess the proportion of benefit that can be attributed to the navigator intervention, once demographic variables were taken into account, and 95% confidence intervals were calculated. All analyses were done using PC-SAS version 8.02. 21

Because baseline differences were noted between the preintervention and intervention group, 2 subsequent analyses were performed to control for differences in case-mix in the 2 study groups. First, a propensity score analysis was conducted to adjust for differences between the study periods. This methodology selects a random subset of subjects in each study period, so that baseline characteristics of the 2 groups are similar. One intervention subject was chosen per preintervention subject. After demonstrating that the groups were similar in the their baseline characteristics (age, race/ethnicity, insurance, reason for referral, and source of referral), we conducted a multiple logistic regression on the reduced dataset predicting timely follow-up, again adjusting for age, race/ethnicity, insurance, reason for referral, and referral source. Second, time was considered as a covariate in the model to control for longitudinal effects not related to the intervention. Time was calculated as the month of the study when the visit was scheduled. For example, a women who scheduled her appointment during the first month of the study (January 2000) would have a value of time of 1, while a women who was in the intervention period might have a value of time of 25, representing that she scheduled her appointment in January 2002. Three models were considered: including time to the main model including covariates; including an interaction effect between time and intervention in the main model; and considering only the effects of time and intervention on timely follow-up. 22

RESULTS

During the study period, 2044 scheduled patient visits were identified, which corresponded to 1381 individual patients. Of those patients, 21 (1.5%) men were excluded, 10 (0.7%) visits for female teens under age 18 were excluded, and 14 (1.0%) were excluded due to missing dates. The final analytic sample consisted of 1332 individual women with scheduled visits (314 preintervention, 1018 during the intervention period).

Baseline characteristics by study group are shown in Table 1. The majority of women scheduled for evaluation were under age 65 (91%), with over half (52%) between 40 and 64 years of age. The majority of women were of minority race (40% black, 14% Hispanic, 4% Asian), while 36% were non-Hispanic white. Fifteen percent required a language interpreter during their visit, and most (68%) had no insurance or some type of public health insurance (Medicaid, Medicare only, or uncompensated care coverage). Over half the women were referred for evaluation of a screening abnormality (47% abnormal exam, 12% abnormal mammogram). Forty-one percent of women in the “other” category which included breast pain, increased breast cancer risk based on family or personal history of breast cancer, or nipple discharge, respectively. The majority of women were referred from a Community Health Center (CHC; 44%) or a hospital-based practice site (34%).

Table 1. Characteristics of Patients Before and After Patient Navigation Intervention
Subject characteristicTotal (N = 1332)Preintervention, (N = 314)Intervention (N = 1018)P*
  • CHC indicates community health center.

  • All values given are in number (percentages).

  • *

    Comparing preintervention versus intervention periods.

Age, y
 18–39522 (39)149 (47)373 (37).0008
 40–64689 (52)147 (47)542 (53)
 ≥65121 (9)18 (6)103 (10)
Race
 White480 (36)107 (34)373 (37).03
 Black534 (40)143 (46)391 (38)
 Hispanic190 (14)44 (14)146 (14)
 Asian59 (4)13 (4)46 (5)
 Other69 (5)7 (2)62 (6)
Insurance
 Private423 (32)82 (26)341 (34).01
 Public/none909 (68)232 (74)677 (67)
Interpreter
 Needed197 (15)25 (8)172 (17)<.0001
Reason for visit
 Breast mass630 (47)117 (37)513 (50)<.0001
 Abnormal mammogram158 (12)29 (9)129 (13)
 Other544 (41)168 (54)376 (37)
Source of referral
 CHC582 (44)96 (31)486 (48)<.0001
 Hospital448 (34)78 (25)370 (36)
 Private105 (8)13 (4)92 (9)
 Unknown/other197 (15)127 (40)70 (7)

Table 1 also demonstrates that subjects who presented for evaluation during the intervention period differed significantly from preintervention subjects in most demographic characteristics. Intervention subjects were more likely to be older, white, to have private health insurance coverage, to require an interpreter, to be referred for a screening abnormality and to have been referred from an affiliated CHC.

Overall, 64% of subjects referred for diagnostic evaluation had timely follow-up during the preintervention period (N = 314) compared with 78% of women during the intervention period (N = 1018) (P < .0001, unadjusted OR: 2.0 [95% CI, 1.5–2.6]). Table 2 presents the results of our adjusted analysis. Controlling for age, race, insurance status, reason for referral, and source of referral, women in the intervention group had a 39% greater odds of having timely follow-up (OR = 1.39 [95% CI, 1.01–1.91]). Compared with women aged 40–64, women over 65 years of age were more likely to have timely follow-up (OR of 1.9; 95% CI, 1.1–3.4), while those aged 18–39 were less likely to have timely follow-up (OR of 0.7; 95% CI, 0.5–0.9). Women with private health insurance were more likely to have timely follow-up, compared with those with only public or no health insurance (OR of 1.5; 95% CI, 1.1–2.1). Compared with women referred for evaluation of an abnormal screening mammogram, timely follow-up was less likely among women referred for evaluation of a breast mass (OR of 0.5; 95% CI, 0.3–0.9) or other breast abnormality (OR of 0.3; 95% CI 0.2–0.5). Compared with women referred from community health centers, timely follow-up was more likely among those referred from hospital-based practice sites (OR of 1.4; 95% CI, 1.0–2.0).

Table 2. Factors Associated with Timely Follow-Up by Logistic Regression and Propensity Score Analysis: Patient Navigation Intervention
VariableLogistic Regression Odds Ratio*95% CIPropensity Score Odds Ratio*95% CI
  • CHC indicates community health center.

  • *

    Analyses adjusted for all variables listed in table.

Patient navigation intervention1.41.01–1.91.71.2–2.6
Age, y
 18–390.70.5–0.90.40.3–0.6
 40–641.0xxx1.0xxx
 ≥651.91.1–3.42.30.8–6.4
Insurance
 Private (vs. public)1.51.1–2.12.21.3–3.7
Race
 White1.0xxx1.0xxx
 Black0.80.6–1.00.90.6–1.5
 Hispanic1.20.8–1.82.01.0–3.9
 Asian1.50.7–3.02.10.6–7.8
 Other0.60.3–1.10.40.1–1.2
Reason for visit
 Breast mass0.50.3–0.90.90.4–1.8
 Abnormal mammogram1.0xxx1.0xxx
 Other0.30.2–0.50.60.3–1.1
Source of referral
 CHC1.0Xxx1.0xxx
 Hospital1.41.0–2.00.80.5–1.3
 Private1.50.8–2.71.60.4–5.7
 Unknown/other0.70.5–1.00.50.3–0.8

Table 2 also shows the results of our propensity score analysis to address the differences in case-mix. Our propensity score analysis selected a subset of 284 preintervention and 284 intervention subjects with similar baseline characteristics, using the same 5 covariates as the main analytic model to calculate a propensity of being in the intervention group. Within each quintile of propensity score, an even number of preintervention and intervention women were chosen, so that the resulting subsample would be matched on baseline covariates. All covariates were associated with the intervention before sampling, while only source of referral remained associated with the intervention postsampling (P-values were .79 for age, .34 for insurance, .41 for race, .0004 for source, and .13 for source). Comparison of these 2 matched subgroups did not change the direction or significance of the intervention effect. Using these reduced data, 65% of preintervention versus 76% of intervention subjects had timely follow-up (P = .008). The odds ratio (OR) for the postsampling intervention was 1.7 (95% CI, 1.2–2.6). The only difference noted in the propensity score analysis is a change in effect due to source of referral, specifically from hospital versus CHC, where the effect size changed direction from 1.4 to 0.8. These results can be found in Table 2.

As a second sensitivity analysis, we considered the trend over time. Three models were considered using the time variable. We first added time to the final multiple logistic regression reported in Table 2. In this model, controlling for other covariates, time was not significant (P = .57). Second, a time-intervention interaction was considered. Once again, this term was not significant (P = .41). Lastly, in recognition of the fact that the effect of the time variable might be confounded within the other covariates, which have changed over time, a model including only time and intervention was evaluated to predict timely follow-up. In this model, once again, the trend over time being associated with timely follow-up was not significant (P = .34). These results lead us to conclude that any change over time has been successfully addressed by the modeling on the other covariates.

DISCUSSION

Patient, provider, cultural, and system level factors have been identified as barriers to the provision of effective cancer diagnostic and treatment services in an equitable and timely manner, 23–37 thus leading to health disparities.38 This is the first publication, to our knowledge, to provide evaluation of a barrier-focused, patient navigation program within an academic medical center, using outcomes data on the group of patients receiving service for breast screening abnormalities. Our data suggest that patient navigation improved rates of timely diagnostic follow-up for abnormal breast cancer screening among a racially diverse group of urban women. We determined that nearly all subgroups of women benefited from the intervention even after adjusting for selection bias. The vast majority of our study population were women from either an ethnic minority population, and/or from a low income group, thus representing the women at greatest risk of poor outcomes from cancer.

Our navigation intervention is modeled after the care management model, with 4 key components to navigation. 19 These include a mechanism for case identification, a process of identifying barriers to follow-up for each individual case, a plan for addressing barriers, and a system of tracking. Each of these components is critical to both accomplish effective navigation, but also to track the outcomes of individual women included in our system. Our system utilized hand entered systems for all aspects of care and tracking, thus demonstrating a method which is easily generalizable to many systems that care for underserved and minority populations.

Care management, used in nursing and social work since the mid 1800s, has been employed for those at increased risk for adverse outcomes and excessive healthcare utilization. 39 The concept has evolved over the years to target various chronic diseases, most notably diabetes,40 where evidence has led to clinical recommendations for such interventions by medical organizations. There has been growing interest in the use of patient navigators as a mechanism to reduce cancer disparities through care management since the concept was first introduced in the early 1990s.18 Since then, navigator programs have been implemented across the country with support from private foundations13 and more recently the federal government16, 41 after 2 bills in Congress42, 43 proposed support for navigation programs to address cancer disparity in underserved populations. To date, however, the data on the effectiveness of patient navigation interventions has been anecdotal or uncontrolled.18 Although evidence does exist to demonstrate the effectiveness of individual components of our model, such as telephone outreach or tracking,44–46 intervention studies to improve diagnostic follow-up using patient navigation encompassing the full care management model only exist for women with abnormal Papanicolaou (Pap) tests.47,48 This marks the first report of an evaluation to investigate the benefit of a patient navigator in an academic medical center serving an urban, minority patient population with screening breast abnormalities.

We evaluated timely follow-up, defined as the time from when the first diagnostic breast appointment is scheduled until actual follow-up evaluation was initiated, in a racially diverse group of inner-city women referred for diagnostic breast evaluation at an urban academic medical center. Our study did not permit an evaluation of timeliness to actual diagnosis. Rather, the intermediate outcome of time to initiation of diagnostic evaluation was chosen based on the limited data collected in the preintervention period. Currently, there lacks clear and widely accepted definitions of what constitutes timely follow-up after a breast abnormality is detected, 6 which limits our ability to compare our findings with existing literature. We demonstrated a 15% improvement in timely follow-up after program implementation. This increase is consistent with a previously published report of a similar care management intervention for inner-city minority women with abnormal Pap tests.48

We identified characteristics of those patients who are most at risk for loss to follow-up. Significant predictors of timely follow-up include older age, having a referral, which originated on-site, having private health insurance, or being referred for evaluation of an abnormal screening mammogram. Women referred from community health centers benefited from the intervention, but not to the same degree as other women. Although transportation or convenience factors could have been operant, our system of care provides regularly scheduled shuttle rides to and from all our affiliated health centers. Rather, the etiology may lie in other patient characteristics not measured in our study, such as fear, mistrust, lack of appreciation of the potential seriousness of the problem, inability to get time off from work, child care issues, etc.

Our model employed a medical assistant with some clinical experience recruited to the navigator position. Although no standard guidelines exist, patient navigation typically utilizes a person who is not the provider of direct healthcare for coordination and implementing care. The literature to date has described navigation across a broad spectrum of experience, including trained lay navigators, to those with advance nursing degrees. 12 Lack of clarity exists for what constitutes patient navigation, as it has multiple components, although a several studies sought to evaluate programs with some components of patient navigation.46, 47 Our data cannot address which level of training is most effective or most cost effective or which components of navigation are critical to successful outcomes. Our data precluded an analysis of the specific intervention components. Future studies should focus on developing measurements of the 4 principles of care management utilized in this study to determine which components have the greatest impact on target populations.

Limitations of our study include the evaluation of the intervention in only one health care system. However, the study had a large sample size and a diverse population referred from multiple sites. Nonetheless, our system has some unique features that may encompass a more responsive system of care for underserved urban communities, limiting generalizability to other health care systems, even with similar underserved communities. Unique features include a managed Medicaid plan, which covers the greater proportion of the otherwise uninsured, and an uncompensated care pool, to support those without any health insurance access otherwise. Furthermore, extensive interpreter services and transportation strategies are in place to allow patients to travel easily from community health centers to the medical center.

Another limitation of this study is the pre–post design, which is vulnerable to secular trends as the explanation of the findings. Although our pre–post design introduces the possibility of historical bias, the above-mentioned resources were in place throughout the study duration. Furthermore, there were no major institutional, city- or statewide breast cancer diagnostic programs implemented during the study period to introduce secular trends as a potential explanation for our findings. A propensity score analysis to adjust for known baseline differences in the 2 groups found the same magnitude of effect of the intervention. A time trend analysis was also conducted and identified no significant time trend. For these reasons, it is less likely that the observed differences in case-mix in the 2 study periods are attributable to changes in the healthcare system that are being confused with a navigator effect.

Our data suggest a benefit of patient navigators for addressing follow-up after abnormal cancer screening. Patient navigation shows promise as the missing link between available cancer care services and delivery to vulnerable populations. Future funding of patient navigation is critical to continue more rigorous evaluation efforts, especially to address which type of navigator and what components of navigation are most effective. Furthermore, understanding the work design issues and how to coordinate patient navigator efforts across specialists and disease conditions is especially timely, as multiple venues of the health care system begin to employ navigators. These data will be critical in policy decisions on incorporating support through insurers to accomplish cancer care management for vulnerable populations.

Acknowledgements

We acknowledge the daily efforts of our Patient Navigator, Wanda Turner, for her steadfast dedication and commitment to the women we serve. In addition, we thank Emily Looney and Shreya Patel for their contributions to the final dataset and manuscript preparation

Ancillary