Impact of Patient Navigation on Population-Based Breast Screening: a Systematic Review and Meta-analysis of Randomized Clinical Trials

Background Unsatisfactory cancer screening results are often associated with poor prognosis. This study synthesized the literatures addressing the impact of patient navigation (PN) interventions on population-based breast cancer screening promotion to identify characteristics of the model for addressing breast cancer disparities. Methods We searched Pubmed, Embase, Web of Science, and the Cochrane Central Registry from inception to 31 December 2020 for randomized controlled trials (PROSPERO: CRD42021246890). We double blindly abstracted data and assessed study quality. We assessed screening completion rates and diagnostic resolution using random-effects models between those receiving navigation and controls. Results Of 236 abstracts identified, 15 studies met inclusion criteria. Nine of the papers evaluated the impact of PN on breast screening, while the other six were on the resolution of abnormal screening results. Compared to the non-PN group, PN improved screening completion (OR: 2.0, 95% CI: 1.4–2.8]) and shortened the time to diagnosis (WMD: − 9.90 days, 95% CI: − 19.09 to − 0.71). Conclusions Patient navigation improves breast cancer screening rates but does not improve resolution of abnormal tests. Supplementary Information The online version contains supplementary material available at 10.1007/s11606-022-07641-y.

BACKGROUND: Unsatisfactory cancer screening results are often associated with poor prognosis. This study synthesized the literatures addressing the impact of patient navigation (PN) interventions on population-based breast cancer screening promotion to identify characteristics of the model for addressing breast cancer disparities. METHODS: We searched Pubmed, Embase, Web of Science, and the Cochrane Central Registry from inception to 31 December 2020 for randomized controlled trials (PROSPERO: CRD42021246890). We double blindly abstracted data and assessed study quality. We assessed screening completion rates and diagnostic resolution using random-effects models between those receiving navigation and controls. RESULTS: Of 236 abstracts identified, 15 studies met inclusion criteria. Nine of the papers evaluated the impact of PN on breast screening, while the other six were on the resolution of abnormal screening results. Compared to the non-PN group, PN improved screening completion (OR: 2.0, 95% CI: 1.4-2.8]) and shortened the time to diagnosis (WMD: − 9.90 days, 95% CI: − 19.09 to − 0.71). CONCLUSIONS: Patient navigation improves breast cancer screening rates but does not improve resolution of abnormal tests.

INTRODUCTION
Mortality due to breast cancer remains high globally. Two causes include low participation in screening and delays in diagnosis. 1 Early detection and treatment reduce breast cancer death. 2 However, in some specific populations, socioeconomic factors may be an obstacle to participation in population-based breast screening.
Patient navigation (PN) has emerged as one partial solution to reduce disparities in cancer care delivery. Trained navigators can promote cancer screening, follow-up of abnormal tests, and timely treatment. 3,4 However, the literature on navigator effectiveness is mixed [5][6][7] and interpreting these studies is difficult, owing to considerable heterogeneity. Two recent systematic reviews including both RCTs and observational studies concluded that PN improves the screening rate of many tumors, including breast cancer. 8,9 However, these studies did not analyze the impact of PN on the diagnosis rate of abnormal screening findings. Reduction in breast cancer mortality requires that abnormal mammograms be followed by a process that leads to timely definitive diagnosis and treatment. 10 Unfortunately, this follow-up process was shown to be incomplete or delayed in some vulnerable populations. Our systematic review aims to assess the impact of patient navigation on screening and resolution of abnormal findings for breast cancer.

PATIENTS AND METHODS
We followed PRISMA guidelines in conducting our review 11 a n d r e g i s t e r e d o u r p r o t o c o l w i t h P R O S P E R O (CRD42021246890).
Supplementary data. Both unpublished and published studies were eligible for inclusion. After removing the duplicate literature, three authors (Lu T., L. H., and Y. Y.) screened the titles and abstracts of the search results. The final inclusion decision was based on the independent review of the full text by the three authors. Any discrepancies arising from the process should be settled by consensus.

Eligibility Criteria
Our review was limited to randomized trials of PN among female participants over 18 years old and not pregnant. Studies were excluded if any of the following conditions occurred: (1) participant(s) had a history of cancer or were receiving antitumor treatment; (2) participant(s) had received PN intervention in the past; (3) participant(s) were/are living in a pension institution; (4) participant(s) had a history of mastectomy; (5) the screening mode was not community-based; (6) interventions did not meet the definition of PN, such as only phone-call or email reminder; (7) data were insufficient to obtain OR and 95% CI for outcome; (8) full texts were unavailable; and (9) the types of literature comprised abstract, letter, review, protocol, conference presentations, editorials, and/or expert opinions. When the same or partially identical cohorts were reported in different published studies, the most comprehensive study was selected.

Data Extraction
For all studies eventually included in the meta-analysis, study characteristics were independently extracted using a standard data extraction form by each of the three authors (Table 1). Again, discrepancies were resolved by consensus. For studies that provided graphs of time to DR by days from initial screening instead of mean and SD, we used the Engauge-Digitizer software (version 11.2) to obtain the approximate number of people diagnosed at a specific time point, so as to reasonably estimate the diagnostic time. For studies that provided data by BI-RADS group or race, we pooled the data to get overall effect measures.

Risk-of-Bias Appraisal
We assessed literature quality using the Cochrane Risk of Bias tool. 25 Review Manager (v 5.3.5) was used to generate "risk of bias" graph and summary.

Statistical Analysis
We performed statistical analyses using Stata/SE (College Station, TX, v13.0) and the meta package in R (version 3.4.3). We calculated ORs and their associated 95% CIs to assess outcomes and considered a p value less than 0.05 to be statistically significant. A significant degree of heterogeneity between studies was defined as both the I 2 statistic with a cut off of ≥ 50%, and the χ 2 test with a p value < 0.10. 26 Effect size was calculated using random-effects models. 27 We assessed heterogeneity using subgroup analyses (Stata), sensitivity analysis (Stata), and meta regression (R). The criteria of grouping in subgroup analysis was established based on clinical significance and overall data distribution. Publication bias was assessed by Egger's test with visual inspection of funnel plots (Stata). 12 We assessed the potential impact of publication bias using the Trim and Fill approach (Stata). 12

Study Characteristics
We included 9 papers 13-21 on the impact of PN on screening and six papers 5, 6, 22-24, 28 on diagnostic resolution (Fig. 1). Study characteristics are presented in Table 1. A considerable number of trials had unclear risk of bias on some measures, suggesting only modest study quality (Fig. 2). Three trials on screening 15,19,20 were judged to be at high risk of bias. All trials enrolled were partially or fully sponsored by the government.
In our sensitivity analysis, we found that no single study was overly influential. Our results were not significantly changed by excluding trials at high risk of bias (OR: 1.95, 95% CI: 1.3-2.9, I 2 = 96.3%).
For diagnostic resolution, a sample size > 500 and less than 6-month follow-up length explained 94.5% of the heterogeneity. Excluding any single trial did not change our effect estimate. There was no evidence of publication bias (Egger's test, p = 0.08). For time until diagnostic resolution, sample size and length of follow-up explained 86.1% and 91.5% of the heterogeneity. We found no evidence of publication bias (Egger's test, p = 0.87).

DISCUSSION
We found that patient navigators doubled the likelihood of screening for breast cancer. The impact of PN on diagnostic Recent studies have shown that PN improved mammography screening rates in medically underserved settings, and this effect may be especially pronounced for women who have not been previously screened. 3,29 We found that PN is effective in increasing mammography utilization among minority or underserved communities similar to other recent meta-analyses. 8,9 However, there was significant qualitative and quantitative heterogeneity with a great deal of variability in the design and implementation of the programs assessed. Additional studies would help clarify the significance and identify patients most likely to benefit from PN. To facilitate the aforementioned task, exploring the sources of heterogeneity and its significance is important.
Our findings suggest that follow-up time is one of the important sources of heterogeneity, which could be explained by the results of several previously published studies, demonstrating that longer trials demonstrated greater benefits, and shorter trials may underestimate the effectiveness of PN. 30,31 Trialists and policy makers should anticipate this lag when studying and implementing PN, as PN evidently takes time to achieve maximum benefit. 6,30 Other sources of heterogeneity, such as education and race, suggest that PN may be useful in helping certain patients navigate the health care system.
Reducing mortality rate through breast cancer screening will be incompletely realized if timely diagnostic follow-up for abnormal screening does not occur. We found a trend toward improved diagnostic resolution with a reduction in the number of days between the abnormal mammogram and final disposition. The paucity of trials limited our ability to provide definitive recommendations.
Our study has a number of limitations: first, most of the studies were from the USA; second, we identified evidence of publication bias as it is likely that small studies with negative results could remain unpublished; third, we found that PN doubled diagnostic resolution, an important clinical difference. The lack of statistical significance could be explained  by the few included studies; our analysis lacked power. That we reduced the time but not the odds may also reflect differences in parametric and nonparametric tests in demonstrating statistical significance; finally, some studies combined data on the impact of PN on breast, cervical, and rectal screening, making it impossible to extract the data of the breast screening population.
Women needing breast cancer screening are a heterogeneous group, varying by menopausal status, age, education, and race. PN may be particularly useful in assuring screening and follow-up for vulnerable women. PN improves screening rates. Additional studies are required to assess the impact on diagnostic resolution, to determine which subgroups may