Boston Patient Navigation Research Program: The Impact of Navigation on Time to Diagnostic Resolution after Abnormal Cancer Screening

There is widespread recognition that increasing complexity of cancer care contributes to outcome disparities (1, 2) as evidenced by the impact that lack of access to timely, quality cancer services has on outcomes among vulnerable populations defined by low income, inadequate insurance coverage, and minority race/ethnicity (3–7). To address this failure to equally deliver the spectrum of cancer services to all Americans, there is growing emphasis on patient navigation programs (8–10).

Patient navigation was designed to identify and eliminate barriers to accessing cancer services to achieve timely completion of care (11, 12). The goal of patient navigation is to facilitate timely care by meeting cultural needs and addressing barriers to accessing care at the individual level. Examples of navigation services include identifying patients at risk for delays, facilitating appointment scheduling, coordinating care among providers, organizing interpreter services, ensuring access to prior medical records, and providing logistic support for transportation, childcare, or linkage to community resources. Navigators are trained to help patients advocate for themselves in the health care system and provide emotional support during this stressful period.

Several demonstration projects and small clinical trials report the potential benefit of patient navigation at certain points in the cancer care continuum and in specific populations (10, 13–15) resulting in several organizations recommending patient navigation as standard care (16, 17). However, large-scale clinical trials across diverse populations are still lacking. Implementation research is needed to assess the ability of navigation to be adopted in usual clinical practice, especially in safety net institutions that care for low-income patients (18). To provide controlled clinical trial data on the efficacy and effectiveness of patient navigation for follow-up of screening abnormalities, the Center to Reduce Cancer Health Disparities (CRCHD) and the American Cancer Society funded 10 intervention sites through 9 grants (11) to conduct a national Patient Navigation Research Program (PNRP). The Boston PNRP, one of the funded programs, was designed as an effectiveness study to assess the impact of patient navigation under usual clinical practice rather than the ideal circumstances of a randomized clinical trial.

We conducted a quasi-experimental intervention across 6 federally qualified inner-city community health centers targeting women with breast and cervical cancer screening abnormalities. The Boston program was designed as a clinical effectiveness study of patient navigation as a new standard of care, where all patients with screening abnormalities were included into the trial and thus the Boston University Institutional Review Board approved the study with a waiver of written informed consent. Using the community based participatory research process (19, 20) guided by a Community Advisory Panel, each health center partner agreed to collaborate in developing and evaluating the patient navigation program at their site. Each health center was assigned a navigation site for either abnormal breast or cervical cancer screening and a control for the other condition. To adjust for temporal trends and observe for contamination between the intervention and control conditions, we collected baseline data from the health centers on a random sample of all subjects with abnormal screening in 2004 to 2005, and collected data on all subjects with abnormal screening during the intervention period from January 2007 to December 2008. We followed all subjects for 365 days, censoring those who had not achieved diagnostic resolution by the end of the study.

Because there were only 6 sites, each serving different racial/ethnic communities, rather than randomly assigning each to the navigation condition, the study team developed a strategic allocation protocol to reduce imbalance across the study by race/ethnicity.

We used electronic medical records to identify eligible women with an abnormal breast or cervical cancer screening test conducted at one of the participating health centers during the study periods. Breast abnormalities include abnormal mammograms, ultrasound or magnetic resonance imaging (MRI) of the breasts, coded as BI-RADS 0, 3, 4, or 5 using the Breast Imaging Reporting and Data System (21, 22), or abnormal clinical breast examinations with a mass or other lesion suspicious for cancer. Cervical cancer screening abnormalities included Pap test findings of low- and high-grade squamous intraepithelial lesions (LGSIL and HGSIL), atypical cells of unknown significance (ASCUS), when reflex testing for high risk human papilloma virus (HPV) was positive, or atypical glandular cells of unknown significance (AGUS). Women were excluded if they were pregnant at the time of the screening abnormality, as pregnancy may alter the evaluation protocol and its time course, or had a cognitive impairment that prevented their participation with a patient navigator. During the baseline period, we sampled all high-grade (BI-RADS 4, 5, and HGSIL) lesions and a random sample of all other mammographic and cytologic lesions with up to 100 subjects per health center (23). We included all women with abnormalities during the intervention period. When guidelines for management of ASCUS HPV–positive and LGSIL changed for women ages 18 to 21 years in 2007 (24), recommending annual Pap tests and not diagnostic colposcopy, we excluded, thereafter, those subjects where the clinician choose repeat Pap tests as the management, retaining any subjects if the clinician referred the patient for colposcopy.

Patient navigators were hired as employees of the health centers, not part of the research staff. With personnel turnover, there were a total of 23 female navigators. They all had a high school education and some health care experience, such as medical assistant training. None had advanced training in social work or nursing. Each health center attempted to hire navigators who could address the language needs of their communities; five navigators spoke another language, including Spanish, Vietnamese, Albanian, and Portuguese Creole; all other navigators had access to robust interpreter services. They were integrated into the clinical team at the health center, receiving direct supervision from health center staff. All received standard training to ensure minimal competency. The research team conducted bi-monthly local training for all navigators and their supervisors throughout the study period. All navigators participated in annual face-to-face national PNRP trainings and several webinars to standardize the training across the national sites (25). A protocol was developed at each health center for identifying women with abnormal cancer screening. Navigators contacted patients by telephone to initiate navigation after a clinician had informed the patient of the test results. Navigators used the care management model (26) to identify barriers to recommended care, and developed strategies to address these barriers, with the focus on timely completion of the diagnostic evaluation. Follow-up occurred by telephone, mail, and in face-to-face meetings, usually at the health center. Navigators documented their activities in a structured template in the electronic medical record, which allowed providers to view the activities in real time and allowed electronic capture of the navigator activities.

The research team conducted annual core competency assessments on all navigators, and provided feedback to the local supervisors. The research team also provided each site with a monthly spreadsheet of patients eligible for navigation at the health center and their progress in reaching diagnostic resolution.

All data were captured through the electronic medical records at each of the health centers regardless of control or navigation status. Demographic information was captured from registration and billing systems. Electronic charts recorded all screening studies and their results, so we were able to capture abnormalities electronically. All subsequent information on the clinical diagnostic evaluation, including tests ordered, completed, and their results, were collected through manual abstraction by trained abstractors, with 10% of charts reviewed by a second abstractor for accuracy of abstraction with 3% errors identified.

Race/ethnicity was collected from registration data, as our prior work indicated a high correlation of this administrative data with patient self-report (27). We imputed the n = 229 (8%) missing race/ethnicity data from the following sources, in order of: providers' notes, country of origin (28), last name (29), or based on the majority race/ethnicity group of their community health center site. Missing language (n = 174, 6%) was imputed as other than English if there was documentation in the medical record of the primary language, of need for interpreter, or correspondence with the patient in another language.

The primary outcomes of interest in our study were whether and when diagnostic resolution of the screening abnormality was achieved, as defined by the Design and Analysis Committee for the National PNRP (11). Time to diagnostic resolution was defined as the time from the date of the initial screening abnormality to the date when the final definitive diagnostic test or evaluation was completed, a timepoint determined by the clinicians caring for the patient. For BI-RADS 4 and 5 lesions, it was usually biopsy. For BI-RADS 3 imaging, where serial 6 month imaging for up to 2 years is recommended, we observed only to confirm when the next imaging was completed after 6 months. We subtracted 180 days from the diagnostic evaluation time, which allowed us to compare all abnormalities in a similar timeframe. If the BI-RADS 3 was addressed in less than 180 days, we used 0.5 days for the time to diagnostic resolution. For BI-RADS 0 lesions, resolution was defined as either follow-up imaging designated BI-RADS 1 or 2, or completion of whatever diagnostic testing through biopsy had been recommended. For cervical abnormalities, diagnostic resolution was usually colposcopy.

We followed our a priori analysis plan to conduct a difference in differences survival analysis, comparing the differences across the baseline and intervention time periods between the navigated and control study sites. We calculated separate Kaplan–Meier survival curves and proportional hazards regression models for breast and cervical subjects to examine the effect of patient navigation on time to definitive diagnosis for each clinical abnormality. Proportional hazards regression models examined the effect of patient navigation, controlling for differences between the baseline and intervention periods, grade of abnormality, and demographic covariates. We included as potential confounders the covariates of age (breast categories: 18–40, 41–64, >65 years and cervical categories: 18–20, 21–30, >30 years), high (BIRADS 4 and 5, HGSIL) versus lower (BIRADS 0 and 3, LGSIL) grade screening abnormality, race/ethnicity, primary language and insurance (private, public or no insurance), and percentage of zip code under federal poverty level as proxy for socioeconomic status. These models accounted for clustering by health center. We included interactions terms with the time to diagnostic resolution variable where proportional hazards assumptions were violated.

All analyses were conducted in SAS, and the proportional hazards regression procedure in SAS used a robust sandwich estimator for the covariance matrix in Wald tests for the global null hypothesis and individual parameters to account for clustering (30). The effect of patient navigation met assumptions of proportionality for the cervical cancer screening arm of the trial, but did not for the breast cancer screening arm. We noted that the survival curves for the baseline and intervention period for the breast navigated sites crossed at approximately 60 days, and therefore included an interaction term between intervention and time allowing differential effects of patient navigation over days 0 to 59 and days 60 to 365. All observations were censored at 365 days. A limitation of the sandwich estimator to account for clustering is that it is biased downward when there are few clusters (31). Therefore we conducted a secondary analysis using a shared gamma frailty model with a proportional hazards model fit through a penalized likelihood approach (using the frailtypack procedure in R; 32) as a second approach to account for clustering. This method does not allow one to account for the nonproportional hazards for navigation in the breast cancer screening arm in a single model. Therefore, we ran separate proportional hazards for diagnostic resolution within 59 days and for diagnostic resolution between 60 and 365 days in this secondary analysis.

Because we were specifically interested in knowing whether navigation had a greater impact on vulnerable groups of women, described here by race/ethnicity, primary language, and insurance coverage, we developed models including interaction terms for these variables with the navigation variable. Our primary analysis was conducted as an intention to treat analysis, including all women whether or not they received patient navigation. We conducted a secondary analysis, excluding those navigated subjects where the navigator did not initiate patient navigation within 30 days for breast screening abnormalities and 60 days for cervical screening abnormalities.

There were a total of 997 subjects in the baseline period and 3,041 subjects during the intervention period, of which 1,497 received navigation and 1,544 were controls. Tables 1 and 2 display demographics separately for breast and cervical subjects, looking at the overall group divided by both study period (baseline versus intervention) and by study site (navigation versus control sites). The majority of subjects were from racial or ethnic minority groups, including 30% from African American communities and 29% from Hispanic populations. The subjects spoke 18 primary languages, with Spanish (19%), Vietnamese (3.5%), Albanian (2%), and Portuguese (1%) being the most common. The differences in the proportion of women from each racial and ethnic group in the navigated versus control sites reflects the different communities served by each health center, and different sizes of the health centers. There are also differences in race/ethnicity over time within the health center populations. The proportion uninsured dropped from 40% to 29% overall between the baseline and intervention time periods, reflecting the impact of Massachusetts Health Insurance Reform, initiated in 2006 (33, 34).

Figure 1 represents the unadjusted proportion of subjects who ever reach a diagnostic resolution by 365 days, comparing differences across the 2 study periods (baseline and intervention) within navigation and control sites separated for breast and cervical cancer subjects. For the breast cancer screening arm, at baseline, the control sites had similar rates of diagnostic resolution compared with the navigated sites (89.6% vs. 89.7%). The control sites showed no improvement in the proportion of women reaching diagnostic resolution from the baseline to the intervention period (89.6% vs. 90.2%, P = 0.7809), whereas the percentage of women at navigated sites who resolved increased from 89.7% to 92.6% from the baseline to intervention period, although this difference did not reach statistical significance (P = 0.12). For the cervical cancer screening arm, the control sites had slightly higher rates of diagnostic resolution during the baseline period compared with the navigated sites (87.9% vs. 79.1%). The control sites again showed no change temporally at 365 days after abnormal screening (80.0% versus 78.6%, P = 0.64). However, the navigated sites showed significant improvement in diagnostic resolution comparing the baseline with the intervention period (79.1% vs. 87.9%; P = 0.0008).

Figures 2 and 3 display the Kaplan–Meier curves for time to diagnostic resolution (where the y-axis represents the proportion achieving diagnostic resolution) comparing the baseline and intervention periods across both the navigation and control sites. For breast cancer screening subjects in the navigation sites, there is a superimposition of the baseline and intervention period curves through about 60 days, when 73% of subjects in the baseline period have achieved resolution compared with 71% of subjects in the intervention period. Median days to resolution are also similar in the 2 periods, with median days to resolution of 29 days in the baseline period and 32 days in the intervention period. After 60 days, there is a consistently higher rate of resolution among subjects during the intervention than the baseline period. For patients without resolution by 60 days, median days to resolution decreased from 157 days to 70 days from the baseline to intervention period. The control sites had no change between the baseline and intervention time periods, also reflected at 60 days when 73% of subjects in the baseline period achieved resolution compared with 72% of subjects in the intervention period, and with median days to resolution of 27 days in the baseline period and 34 days in the intervention period. For cervical cancer screening subjects in the navigation sites, the intervention period showed more subjects achieved timely resolution almost immediately. For example, at 60 days, only 27% of subjects in the baseline period achieved diagnostic resolution compared with 39% of subjects in the intervention period, and median days to resolution decreased from 110 in the baseline period to 76 in the intervention period. These improvements among intervention subjects in the cervical navigation sites persisted throughout the 365 day follow-up, compared with the baseline period, and compared with the control sites in either time period where the proportion of subjects achieving diagnostic resolution at 60 days was similar for baseline and intervention subjects (37% vs. 31%, respectively), and median time to resolution was also similar (84 days vs. 90 days).

Table 3 presents the adjusted HRs for the proportional hazards models, for breast and cervical cancer subjects, respectively. We found a significant benefit of patient navigation in reaching diagnostic resolution after adjusting for differences in differences between the navigation and control sites of the study and the intervention and baseline time periods. The model also adjusted for covariates and for clustering within each of the health centers. In the breast model, the adjusted HR for completing diagnostic evaluation in the first 59 days is 1.04 (95% CI, 0.83–1.30), indicating no difference between the navigated and control sites after adjustment. From days 60 through 365, there was a significant impact of patient navigation (aHR 1.40; 95% CI, 1.06–1.86), where HR more than 1.0 indicates that patient navigation is associated with more timely diagnostic resolution. For the cervical cancer screening subjects of the study, the adjusted HR was 1.46 (95% CI, 1.14–1.88), indicating a benefit of navigation over the control sites in reaching diagnostic resolution throughout the course of the study. Secondary analyses using a shared gamma frailty model to account for clustering gave similar effect size for breast cancer screening subjects, but were not significant in separate models for the first 59 days (aHR 1.03; 95% CI, 0.8–1.3) or for days 60–365 (aHR 1.5; 95% CI, 0.9–2.4), whereas significance remained among the cervical cancer screening subjects (aHR 1.4; 95% CI, 1.1–1.8).

In the breast model, several covariates were associated with delays in achieving diagnostic resolution. Compared with patients who had private insurance, those with public insurance (aHR 0.8; 95% CI, 0.8–0.9), but not uninsured (aHR 0.9; 95% CI, 0.8–1.0; P = 0.07), were less likely to have timely resolution. Patients with a primary language other than English or Spanish were also less likely to have timely resolution (aHR 0.8; 95% CI, 0.7–0.9). Index screening abnormalities of BI-RADS 3 (aHR 12.7; 95% CI, 8.7–18.6) or BI-RADS 4 or 5 (aHR 4.5; 95% CI, 2.0–10.0) were initially associated with more timely resolution, which decayed over time, compared with those with a BI-RADS 0 screening abnormality. Age, race/ethnicity, and percentage of zip code below federal poverty level were not predictors of timely resolution in the model.

In the cervical model, public insurance status and younger age were associated with delays in achieving diagnostic resolution. Compared with private insurance, publically insured (aHR 0.8; 95% CI, 0.7–0.9) but not uninsured women (aHR 1.0; 95% CI, 0.9–1.2) were less likely to have timely resolution. Women ages 18 to 40 (aHR 0.7; 95% CI, 0.6–0.9) were less likely to have timely resolution than older women. Race/ethnicity, language, screening abnormality, and socioeconomic status were not predictors of the outcome.

We were specifically interested in examining whether vulnerable populations defined by minority race/ethnicity, language other than English, or public or no insurance had a benefit from the intervention. None of the interaction terms of these variables by the navigation variable were significant in either the breast or cervical model, suggesting that there was no differential benefit of navigation for any specific group of women defined by these demographic variables.

Our data indicated that 124 (16.1%) women in the navigated arm did not receive navigation services within 30 days of the breast screening abnormality, and 103 (14.2%) of women did not receive navigation within 60 days of the cervical cancer screening abnormality. These findings varied by health center, and ranged from 4.0% to 29.6%. Removing women enrolled later into navigation did not change our findings (breast navigation effect ≥ 60 days: aHR, 1.4; 95% CI, 1.0–1.8; cervical navigation effect: aHR, 1.6; 95% CI, 1.3–2.1). Median time that a navigator spent per case varied by disease status; breast navigators spent a median of 60 minutes per subject with navigation activities [75% interquartile range (IQR), 30–127.5], with a median of 1 encounter per subject (range 0–15); whereas cervical navigators spent a median of 75 minutes (75% IQR, 30–120) in navigation activities, with a median of 2 encounters per subject (range 0–18).

The Boston Patient Navigation Research Program showed a benefit of patient navigation for vulnerable communities of women with both breast and cervical cancer screening abnormalities with an adjusted HR between 1.4 and 1.5. Observing the navigated arm over 365 days, this translated into an additional 9% of patients with cervical abnormalities and 3% of women with breast screening abnormalities completing diagnostic resolution to determine whether or not they have a premalignant or invasive cancer. By including all individuals with a screening abnormality, this study represents an assessment of the effectiveness of the patient navigation intervention for an entire population cared for in a safety net system of care, leading to greater external validity. Individual randomized clinical trial methodologies are often unable to randomize the most difficult to reach patients who may be in greatest need of navigation (14), and thus limits their external validity. Our effectiveness design includes this difficult to reach group of women while ensuring identical methods in the navigated and control arms of the study, and thus provides an estimate of the effect of navigation when applied to an entire practice population. Our data provide scientifically rigorous evidence of how navigation can be implemented.

The generalizability of our findings extends to a diverse group of women from multiple communities and cultures: more than 55% of our sample were from minority communities, and more than 60% were either publicly insured or uninsured. Each community health center that participated in the trial provided care to unique immigrant and minority communities in the greater Boston area. Our work shows that navigation is broadly beneficial to all groups. We did not find that our navigation intervention preferentially supported any group of subjects, defined by race/ethnicity, language, or insurance status. The increased rates of insurance seen in our intervention period of 2007 to 08 compared with the baseline 2004 to 05 periods reflect the impact of Massachusetts Health Insurance reform. The persistent uninsured rates of 8% to 47% by health center, all higher than the overall 5% uninsured in the state in 2008 (35), reflect that women who are uninsured despite insurance reform disproportionately seek their care at community health centers. Our finding that publically insured subjects had less timely care compared with uninsured subjects is consistent with other studies and suggests that insurance status may be a marker for other social determinants of health (36, 37).

The benefit of patient navigation does not seem to be only from shifting the diagnostic completion curve to an earlier time for all, but also in increasing the proportion completing diagnostic resolution. Our data in the breast cancer screening abnormalities suggests that navigation does not show an immediate benefit, but may be of greatest benefit for those with initial delays. This may be because many patients who complete in a timely manner will do so without the help of a navigator. This has implications for institutions with limited resources that wish to develop navigation programs with the most effective impact. Several navigation programs addressing annual mammography screening, for example, began at 18 months, providing patients a 6-month window to complete screening on their own before activating navigation (38).

The main limitation of our trial is the lack of randomization of subjects into the intervention and control groups. This could have resulted in potential imbalance of both measured and unmeasured differences between the groups. We attempted to address this design limitation by collecting baseline data from all sites. This allowed us to adjust for secular trends and baseline differences in the health centers. The lack of any differences over time in the control groups suggest that there was no secular trend to account for our findings. It also suggests that there was little contamination of the navigation intervention into the control arms. A related limitation is that our study was conducted at 6 urban, East Coast community health centers. While our analyses do account for clustering of patients within health centers, available methods of accounting for clustering in the context of proportional hazards regression for time-to-event data do not perform well with a small numbers of clusters. Our results may not generalize to less urban settings, or other regions of the country, and we acknowledge that our data collection methods did not account for care received outside of the affiliated health care system, though our previous work has shown that the majority of this population have documented longevity within this safety-net health system (39, 40).

To successfully disseminate patient navigation programs, there are several components critical to ensure success. Safety-net institutions, including community health centers and hospital ambulatory centers, will require an initial and ongoing training program to ensure the competence of navigators. Care sites will require the ability to track patients with screening abnormalities, and link this tracking to the work of the navigators. The methods developed for the health centers in this project fulfill these components while also meeting many of the principles of the Patient Centered Medical Home, including criteria for care coordination, registry/tracking, and meaningful use of electronic medical records (41). Because this study was not powered to provide clarity on all aspects of navigation, including its impact among specific sub-populations or on patient-reported outcomes or cost-effectiveness, it lays the groundwork for the future studies necessary to inform best practices.

This study highlights the benefits of a community based participatory research process (20) where the design of the research protocol as a quality improvement effectiveness research project was developed with leadership of the health centers and under the guidance of an active Community Advisory Panel. The findings provide controlled clinical trial data on the effectiveness of patient navigation to promote equal access to timely diagnostic cancer care for a racial/ethnically diverse low-income population. Perhaps the strongest indicator of the success of this intervention is the fact that the health centers continued to implement the intervention with internal funds following completion of the trial and saw this as a key component of their Patient Centered Medical Home implementation.

K.M. Freund has a commercial research grant from Komen for the Cure and Avon Foundation and is a consultant/advisory board member of Komen for the Cure and Avon Foundation. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T.A. Battaglia, R. Kalish, K.M. Freund

Development of methodology: T.A. Battaglia, S.M. Bak, S. Tringale, K.M. Freund

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.A. Battaglia, S.M. Bak, S. Tringale, J.O. Taylor, A.P. Egan, K.M. Freund

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.A. Battaglia, S.M. Bak, T.C. Heeren, C.A. Chen, R. Kalish, K.M. Freund

Writing, review, and/or revision of the manuscript: T.A. Battaglia, S.M. Bak, T.C. Heeren, C.A. Chen, R. Kalish, B. Lottero, A.P. Egan, K.M. Freund

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.A. Battaglia, S.M. Bak, K.M. Freund

Study supervision: T.A. Battaglia, S.M. Bak, K.M. Freund

Other (study and staff supervision at community health center): N. Thakrar

This work was funded by the National Cancer Institute and the Center to Reduce Cancer Health Disparities, National Cancer Institute (U01 CA116892-01).