Patient Navigation Can Improve Breast Cancer Outcomes among African American Women in Chicago: Insights from a Modeling Study

Abstract

African American (AA) women experience much greater mortality due to breast cancer (BC) than non-Latino Whites (NLW). Clinical patient navigation is an evidence-based strategy used by healthcare institutions to improve AA women’s breast cancer outcomes. While empirical research has demonstrated the potential effect of navigation interventions for individuals, the population-level impact of navigation on screening, diagnostic completion, and stage at diagnosis has not been assessed. An agent-based model (ABM), representing 50–74-year-old AA women and parameterized with locally sourced data from Chicago, is developed to simulate screening mammography, diagnostic resolution, and stage at diagnosis of cancer. The ABM simulated three counterfactual scenarios: (1) a control setting without any navigation that represents the “standard of care”; (2) a clinical navigation scenario, where agents receive navigation from hospital-affiliated staff; and (3) a setting with network navigation, where agents receive clinical navigation and/or social network navigation (i.e., receiving support from clinically navigated agents for breast cancer care). In the control setting, the mean population-level screening mammography rate was 46.3% (95% CI: 46.2%, 46.4%), the diagnostic completion rate was 80.2% (95% CI: 79.9%, 80.5%), and the mean early cancer diagnosis rate was 65.9% (95% CI: 65.1%, 66.7%). Simulation results suggest that network navigation may lead up to a 13% increase in screening completion rate, 7.8% increase in diagnostic resolution rate, and a 4.9% increase in early-stage diagnoses at the population-level. Results suggest that systems science methods can be useful in the adoption of clinical and network navigation policies to reduce breast cancer disparities.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11524-022-00669-9.

Introduction

African American (AA) women were 40% more likely to die from breast cancer relative to non-Latino Whites (NLW) from 2013 to 2017 [1]. This difference in mortality has been consistently found when controlling for age groups [2] and is reflected in lower 5-year survival rates [3]. Mortality disparities are, in part, attributed to racial disparities in breast cancer screening, diagnostic care completion, and stage at diagnosis [4].

Given concerns around over false positive results associated with increased screening, and the consequent financial costs and increased stress associated with overdiagnosis of non-progressive cancers, there has been an ongoing, thoughtful debate on the utility of screening to facilitate early diagnosis of cancers [5–10]. Most of the discrepancy in guidelines is the subject of discussion for women between 40 and 50 years of age, and the utility of screening is known to improve with age [11]. Cancer detections among AA women also tend to occur in more advanced stages than among white women, and AA women are less likely than other groups to receive follow-up care in the event of abnormal screenings [12–15]. Additionally, AA women are less represented in landmark breast cancer screening trials that have raised questions against the utility of increased screening [11]. Combined with the fact that screening remains, to date, the most effective evidence-based tool for early-stage detection of breast cancer [16], increased screening to improve early diagnosis rates among AA women is recommended [17].

Numerous interventions have been developed to improve screening and diagnostic rates for breast cancer among AA women—including strategies that support AA within healthcare institutions and in community settings. Clinical patient navigation is an increasingly widespread, evidence-based strategy used by healthcare institutions to improve AA women’s breast cancer outcomes. In general, navigators are clinic/hospital employees (e.g., nurses, social workers, lay individuals) who are trained to address patients’ individual/unique barriers to care (e.g., costs, childcare, transportation) and often share racial/ethnic/cultural identities with priority populations [18–22]. To increase access to these clinical services, programs often integrate clinical navigation with different forms of community outreach (e.g., partnering with community-based organizations) [23, 24]. Here, we focus on network navigation (e.g., citizen scientists, empowerment approaches), wherein community members who receive clinical navigation services are trained to become resources that promote breast cancer care uptake and refer network members to breast care services [25–28].

A growing body of research has demonstrated the utility of clinical and network navigation for addressing individual-level breast cancer screening and diagnostic completion [29–32]. Relatively less research has assessed these strategies’ effects on stage at diagnosis [33, 34]. While research continues to address population impacts of cancer screening [35], no research of which we are aware has formally quantified the impact of navigation interventions on population-level screening, diagnostic completion, and stage at diagnosis. Such information is crucial for future program planning and resource allocation, yet there are substantial methodological challenges with obtaining such empirical data (e.g., relatively low prevalence of breast cancer/small sample sizes for stage at diagnosis, complexity of co-occurring clinical and community interactions, ethical challenges to track interventions’ spill-over effects beyond participants).

Systems science modeling methods offer an opportunity to characterize and compare the theoretical pathways and population-level consequences that interventions may have on increasing diagnoses during the early stages of breast cancer. In particular, agent-based models (ABMs) are a dynamic systems-modeling technique that provide a platform to co-simulate the evolution of individuals persons (represented as “agents”) and the social network structures that connect them [36–38]. ABMs have been developed to derive practical insights on the implementation of population health interventions, especially in settings where purely empirical studies are infeasible or unethical [39]. Indeed, a number of computational modeling studies have been conducted to provide data to make better policies to improve breast cancer screening, diagnosis and treatment, and provided data on the following: incidence and mortality reduction [35, 40–47], cost-effectiveness of breast cancer screening [17, 48–50], and quantifying the uncertainty in projecting the impact of screening and mammography [51, 52]. Recent work has also developed ABMs to simulate social network influences on screening decisions [53, 54]. Our goal in the current work is to build upon this literature to estimate the potential impacts of clinical and social network navigation on breast cancer diagnosis among AA women in Chicago.

In this paper, we present an ABM to measure the potential impacts that these two intervention strategies may have on population-level screening, diagnostic resolution, and stage at diagnosis. We focus on populations of AA women between 50 and 74 years of age in Chicago, a city with a history of particularly severe breast cancer disparities and significant initiatives to address these disparities [55–57]. Recent reductions in Chicago-based AA-NLW breast cancer disparities have been conceptually attributed due to clinical and network navigation [58, 59]. However, there have been methodological challenges with causally establishing this relationshi Our paper illuminates how different types of navigation may reduce disparities, using available national and local data. Given rigorous data available for clinical navigation, we simulate three scenarios: (1) population-level outcomes when no navigation strategy exists, (2) population-level outcomes when clinical navigation is implemented, and (3) population-level outcomes, when clinical and network navigation are both implemented.

Methods

Initial Population, Social Network Structure, and Temporal Evolution of the Model

A population consisting of 5000 African American women between 50 and 74 years of age was simulated in an ABM, consistent with the recommendation statement from the US Preventive Services Task Force Guidelines, which recommend biennial screening mammography for women aged 50 to 74 years [9]. (The computer code is publicly available, but the citation information redacted for review purposes.) One timestep in the model was defined as being equivalent to one month of calendar time. Each individual was assigned a set of variables describing processes of disease progression, care engagement, social network structure, and patient navigation. The parameters associated with each of these attributes are provided in Table 1. Changes in attributes (e.g., clinical care uptake, disease progression) were tracked over the course of the simulation, which proceeded in monthly time steps, as indicated above.

Table 1.

Simulation modules, attributes, values, and empirical sources

Category	Attribute	Process/parameter values	Source
Demographics	Age limits	50–74 years	US Preventive Services Taskforce Guidelines [78]
	Race/ethnicity	95% are African Americans	Chicago-based cohort study of breast cancer survivors (Breast Cancer Care in Chicago [BCCC] Study) [60]
	All-cause mortality (per person per year)	55–60 years: 9.7% 60–65 years: 13.3% 65–70 years: 17.7% 70–74 years: 25.3%	CDC Wonder Database [79]
Disease risk	Incidence of breast cancer incidence as a function of age (10-year incidence rates)	50–60 years: 2.35% 60–70 years: 3.26% 70–74 years: 3.39%	SEER race-specific data [80]
	Genetic risk/first-degree relatives with breast cancer	12.4%	Hall et al. [63]
	Relative risk of breast cancer incidence as a function of genetic risk/first degree relatives with breast cancer	1.49	Braithwaite et al. [64]
	Breast cancer subtype distribution	Among agents diagnosed with invasive breast cancer: Hormone-negative: 23% Hormone-positive: 77%	Howlader et al. 2014 [81]
	BMI	BMI < 30: 50% BMI ≥ 30: 50%	CDC [82]
	Odds of breast cancer subtype risk as a function of BMI > 30	Post-menopausal, hormone-positive: 1.39 Post-menopausal, hormone-negative: 0.98	Munsell et al. [83]
Disease progression	Cancer stage distribution	Modeled as an ordinal variable with three levels (Localized, Regional, and Distant) Agents are assigned localized stage for 24 months after they develop cancer. They then progress through regional and distant stages at 12-month intervals	Set to be consistent with distribution of known stage data of African American women diagnosed with invasive breast cancer in Cook County [66]
	Symptom severity	Each of the three cancer stages corresponds to a discrete symptom severity level Symptoms become ‘detectable’ and can affect clinical decisions when agents have regional and distant staged cancers	As per discussions with cancer clinicians
Clinical engagement	Probability of a PCP visit	0.034	Set to be consistent with public health data suggesting 84% of African American women have annual PCP visits [84]
	Referral outcome probabilities (No referral, Screening mammogram, Diagnostic test) for symptomatic agents (i.e., distant/regional stage cancers)	Various values	See Appendix for derivation [69, 70]
	Monthly probability of screening mammogram referral completion	0.0476 base probability	Derived from data on yearly mammography rates [84]
	Probability of diagnostic test completion among asymptomatic agents	0.35 base probability up to month 2 of referral, 0.0476 base probability afterwards	See above derivation. First two-month completion rate tuned to match data from Chicago-based navigation study for predominantly AA patients [71]
	Mammogram recall rate	Target range of 5–12%	American College of Radiology [85]
	Probability of false negative screening mammogram	0.00012	US Preventive Services Task Force [78]
	Probability of false positive screening mammogram	0.11	US Preventive Services Task Force [78]
Social network structure	Mean number of social relationships for each agent	1.145	BCCC Study [60]
	Degree distribution (Proportion of nodes with given degree):	Degree 0: 4% Degree 1: 17% Degree 2: 25% Degree 3: 21% Degree 4: 16% Degree 5: 7%	BCCC Study [60]
Clinical navigation (agent receiving navigation from hospital-affiliated staff	Proportion of agents who are randomly assigned to receive clinical navigation	20%	See Sensitivity Analysis, Appendix Section A.7
	Odds for increased screening mammogram completion as a function of being navigated	2.48	Ali Faisal et al. [72]
	Odds for increased diagnostic test completion as a function of being navigated	1.57	Ali Faisal et al. [72]
Network navigation (clinically navigated agent supporting neighbors’ clinical engagement for breast cancer)	Odds of network navigation as a function of clinical navigation	72%	Molina et al. [73]
	Odds for increased screening mammogram/ diagnostic test completion as a function of being socially navigated by neighbor	2.6	Molina et al. [74]

The ABM simulated three counterfactual scenarios: (1) a control setting without any navigation that represents the “standard of care”; (2) a setting with clinical navigation, wherein agents received navigation from hospital-affiliated staff (e.g., hospital-affiliated lay navigators, nurse navigators); and (3) a setting with network navigation, wherein agents received clinical navigation and/or network navigation (i.e., a clinically navigated agent supporting their neighbors’ clinical engagement for breast cancer). Since this model aims to inform a Chicago-based intervention, we used local data—i.e., Chicago, Cook County, Illinois—wherever possible.

Relational ties between these agents were modeled using empirically reported cohort data collected from a past study on Chicago-based African American patients diagnosed with breast cancer [60]. “Ties” in the network denote social relationships between two agents. The network structure was parameterized using the mean and distribution of relationships as reported by patients [60], limiting the degrees that were specified to a maximum of 5 (Table 1). The network structure was modeled using the exponential family random graph models (ERGMs) [61] as implemented in the statnet [62] suite of packages in the R programming language, allowing the simulated network structure to statistically match the simulated empirical parameters.

Model Components

Described below are the four software modules developed to implement the model: Disease Progression, Clinical Engagement, Diagnosis, and Demography.

Disease Progression

The Disease Progression module assessed breast cancer risk factors for individual agents, assigned probability risk scores based on the presence of individual factors, and simulated cancer onset through Bernoulli processes simulated with these probabilities. Agents without breast cancer were assigned a monthly probability of developing it, contingent upon the following risk factors, based on AA-specific data where possible: (1) age, computed in discrete intervals of 50–60 years, 60–70 years, and 70–74 years; (2) genetic risk/presence of first-degree relatives with breast cancer (12.4% of women have first degree relatives with breast cancer [63] and are assigned a relative risk of 1.49 of developing breast cancer themselves [64]); and, (3) BMI ≥ 30 (50% of women are assumed to have BMI ≥ 30). Of the cancers that developed, 77% were classified as hormone-positive, and 23% as hormone-negative [65].

Agents also progressed through cancer stages based on the time since cancer onset. Cancer stages were tracked for agents with breast cancer, defined in terms of localized, regional, and distant cancers, at 24-, 36-, and 48-month intervals, respectively, parameterized to match the distribution of breast cancers among African American women living in Cook County [66]. In line with the standard definitions, localized stage breast cancers were defined as invasive cancers that were confined to the breast of origin [67]. Regional stage breast cancers were invasive cancers that have extended to axillary lymph nodes, tissues, or organs. Distant stage breast cancers were invasive cancers that have metastasized to distant locations in the body, including distant lymph nodes, tissues, or organs. Agents with regional and distant metastasized cancers had discrete symptom severities that influence clinical decisions (i.e., increased per-month likelihood of completing referrals and the possibility of receiving a diagnostic test without screening).

Clinical Engagement

The Clinical Engagement module managed primary care provider (PCP) visits and referrals for breast cancer screening and diagnostic testing. Model scenarios that included patient navigation were also maintained in this module. Each agent without an active referral had a monthly probability to receive a referral. Three referral outcomes, based on shared decision-making practices between agents and PCPs, were modeled: (1) no referral, (2) referral for a screening mammogram, and (3) referral for a diagnostic test. The model assumes that 84% of the AA women in our population visit a PCP annually, consistent with Chicago data [68]. The probability of a screening mammogram referral for asymptomatic agents—i.e., agents without breast cancer and agents with localized breast cancer—was set to be consistent with empirical data in Chicago (and elsewhere) [69, 70]. The probability of a diagnostic test without a screening mammogram was dependent on “detectable symptoms” exhibited by regional and distant staged cancers (see Appendix for detailed derivation).

Diagnosis

For each timestep, the Diagnosis module evaluated whether referrals were completed or expired that month and handled referral completions (i.e., test results) in terms of false positives and false negatives. Referred agents’ completion of screening mammograms and diagnostic tests were determined by per-month probabilities derived to match population-level yearly completion rates (Appendix). False positive and false negative rates for screenings were based on empirical data (see Table 1 and Appendix) [69, 70]. Agents who received an abnormal screening mammogram result (false positive or true positive) were immediately referred to a diagnostic test. Agents’ completion rates of diagnostic tests were based on empirical data regarding non-navigated patients and their symptomatic status. Agents with breast cancer who received a “true positive” from a diagnostic test were marked as “diagnosed” and were excluded from subsequent clinical behavior.

Demography

The Demography module handled the progression of agent age and also managed agents entering and exiting the simulation. The attributes of these agents were set in accordance with the parameter distributions described in Table 1 (e.g., obesity, race, hormone positive and hormone negative cancer risk). Agents exited the model due to aging out of the population at age 75, or through mortality (age-group specific all-cause mortality or stage-specific breast cancer mortality) before then.

Model Calibration

The model was simulated for a 30-year “burnin period,” allowing key population metrics to stabilize and become consistent with the empirically estimated targets. The length of the burnin was set at 30 years because agents in the model survive for no more than 25 years, the length of the burnin allowed for at least one generation of agents to age out of the population, allowing for removal of any artifacts caused by initialization of the model. These calibration targets included: breast cancer prevalence, cancer hormone-sensitivity distribution, screening mammography rates (proportion of screening mammography referrals that are completed), screening mammography recall rates (proportion of screening mammograms that result in diagnostic testing), and diagnostic resolution rates (completed diagnostic tests). Once the burnin was established, the control and navigation scenarios were simulated for another 30 years on top of the burnin. The verification of key outcomes with empirical targets is described in the Appendix.

Simulating Patient Navigation Interventions

Clinical Navigation

Clinical navigation was operationalized in the model as a scenario in which hospital-affiliated staff (e.g., hospital-affiliated lay navigators, nurse navigators) identified and addressed patients’ personal barriers to breast cancer care. The simulated clinical navigation process began with the initial interaction between the agent and the PCP, in line with a recent breast cancer care navigation study for predominantly AA patients in Chicago [71]. Among patients who received a referral for a screening mammogram or diagnostic test, random sample of patients (20%) were immediately assigned to receive navigation (see “Sensitivity Analysis” section). Clinically navigated patients benefit from a higher monthly probability of completing screening or diagnostic testing visits relative to the non-navigated persons, and they continue to receive navigation services for repeat screening until they receive a true positive breast cancer diagnosis (or until they exit the model). Specifically, for screening referrals, the monthly completion probability is improved by a factor of 2.48, while the odds of completing a diagnostic referral improve by a factor of 1.57, consistent with data from a recent meta-analysis [72].

Network Navigation

Network navigation was operationalized in line with network-based cancer interventions (e.g., citizen scientists, empowerment), as an extension of clinical navigation [25–28]. After being clinically navigated, agents were simulated to recommend breast cancer care to individuals in their social networks (e.g., family, friends, acquaintances).

To determine which agents are affected by network navigation the probability that clinically navigated patients will engage network members and recommend breast cancer care was simulated. All network members of navigated women were assumed to have a 72% chance of receiving breast cancer care recommendations [73]. Second, the increased probability of a network member engaging clinical care after receiving this recommendation, was parameterized based on empirical data (odds ratio of 2.6) [74]. Only direct social network contacts who receive breast cancer care recommendations are parameterized to benefit from this factor. Finally, based on network intervention protocols, empirically estimated (51%) proportion of network members who received BC recommendations sought clinical navigation services proactively for themselves at the point of their next referral [74]. These are the agents who received both clinical and network navigation. Other agents who received breast cancer care recommendations, but who were not modeled to seek clinical navigation services, fell into the group who received only network navigation.

Summary of Navigation Protocols

Based on protocols described above, an agent with a navigation referral was classified as belonging to one of four groups: non-navigated agents (all scenarios), agents who received only clinical navigation (clinical and network navigation scenarios), agents who received only network navigation (network navigation scenario), and agents who received both clinical and network navigation (network navigation scenario).

Outcomes

To compare the three scenarios (control, clinical navigation, network navigation), primary population-level outcomes included: (1) screening mammography rates (i.e., proportion of agents who obtained screening mammograms), (2) diagnostic resolution rates (i.e., proportion of persons who completed diagnostic tests), and (3) early stage at diagnosis (i.e., proportion of localized stage cancers) at the time of diagnosis. Overall, results present means of measurements from 30 simulation runs to account for the stochasticity of the model. Unless stated otherwise, the means presented below were calculated by aggregating the means of each individual run. Uncertainty between the model runs is measured using the 95% confidence interval computed using a t-distribution.

Sensitivity Analysis

As stated above, the model assumed that 20% of patients who received a referral for a screening mammogram or diagnostic test were assigned to receive navigation. We varied this proportion of agents receiving navigation across a broad range of values: 0%, 50%, and 80%. Results from this analysis are provided in Appendix Section A.6.

Results

Screening Mammography

Figure 1 (top panel) depicts screening mammography rates at the population level. Across 30 runs, mean population-level screening mammography rates for the control setting was 46.3% (95% CI: 46.2%, 46.4%), consistent with published empirical data [75]. In the clinical navigation and network navigation scenarios, the mean mammography rates were 52.8% (95% CI: 52.4%,53.3%) and 59.4% (95% CI: 59.0%, 59.9%), respectively—see Fig. 2, top panel. Screening mammography rates were comparable across scenarios for non-navigated agents, as expected (mean rate: 46.3–46.6%; 95% CIs: 46.2–46.4%). Increases in screening mammography rates were relatively comparable among agents who only received clinical navigation (mean rate: 72.9–79.1%; 95% CI: 61.1–84.6%) and agents who only received network navigation (mean rate: 73.0%, 95% CI: 61.2–84.7%). Receiving both types of navigation led to a mean rate of 90.8% for screening mammography (95% CIs: 77.8–100.0%) for agents receiving clinical and network navigation.

Fig. 1.

Population rates of screening mammography uptake (top) and diagnostic resolution (bottom) across three scenarios (control, clinical navigation, network navigation)

Fig. 2.

Rates of screening mammography uptake (top) and diagnostic resolution (bottom) among non-navigated agents, agents who have only received clinical navigation, agents who have only received network navigation (NN), and agents who have received both clinical and NN across three scenarios (control, clinical navigation, network navigation)

Diagnostic Resolution

Figure 1 (bottom panel) depicts diagnostic completion rates at the population level. Across 30 runs, mean population-level diagnostic completion rate for the control scenario was 80.2% (95% CI: 79.9%, 80.5%), consistent with published data [70]. The mean diagnostic completion rates for the clinical and network navigation scenarios were 83.7% (95% CI: 83.4%, 84.1%) and 88.0% (95% CI: 87.6%, 88.3%; Fig. 2, bottom panel), respectively. As seen in Fig. 2, across scenarios, there were comparable increases in diagnostic completion rates across agents receiving only clinical navigation, only network navigation, and receiving both clinical and network navigation (mean rates: 86.5–92.9%, 95% CIs: 79.8–100%) relative to non-navigated agents (mean rates: 80.1–80.6, 95% CIs: 79.9–80.9%). Figure 3 shows improvement in the mean diagnostic resolution rates within 60 days across scenarios, with rates increasing across the control setting (80.7%, 95% CIs: 82.1–79.3%), clinical navigation setting (84.1%, 95% CIs: 85.4–82.9%), and network navigation (88.1%, 95% CIs: 89.1–87.0%) setting.

Fig. 3.

Proportion of completed diagnostic tests as a function of time since referral. All 30 simulation trajectories for each of the three counterfactual scenarios are shown

Stage at Diagnosis

Table 2 shows that across 30 runs, mean population-level rates of localized stage at diagnosis in the control scenario was 65.9% (95% CI: 65.1%, 66.7%) for agents who had been diagnosed with breast cancer, consistent with local cancer registry data [66]. Population-level mean rates of localized stage at diagnosis for clinical navigation and network navigation scenarios were respectively 68.6% (95% CI: 67.8%, 69.4%), and 70.8% (95% CI: 70.1%, 71.4%). At the agent level, there were comparable increases in diagnostic completion rates across agents receiving only clinical navigation, only network navigation, and receiving both clinical and network navigation on agents’ stage at diagnosis (mean proportions: 74.0–84.7%; 95% CIs: 72.0–87.2%) relative to non-navigated patients (mean %s: 65.9–68.0%; 95% CIs: 65.1–75.1%).

Table 2.

Location of cancer metastasis at the time of diagnosis across the three scenarios, broken down by the navigation status of agents

Scenario	Localized (%)	Regional (%)	Distant (%)
Control, mean % (95% CI)	65.92% (65.07, 66.77)	24.02% (23.25, 24.79)	10.06% (9.57, 10.55)
Clinical navigation, mean % (95% CI)	68.62% (67.80, 69.45)	22.68% (21.90, 23.46)	8.70% (8.24, 9.16)
Un-navigated agents	66.94% (66.02, 67.85)	23.95% (23.09, 24.81)	9.11% (8.57, 9.65)
Clinically navigated agents only	74.04% (72.93, 75.16)	18.61% (17.4, 19.82)	7.35% (6.56, 8.13)
Network navigation, mean % (95% CI)	70.81% (70.14, 71.49)	20.88% (20.33, 21.42)	8.31% (7.83, 8.79)
Un-navigated agents	69.06% (68.3, 69.83)	22.11% (21.48, 22.74)	8.83% (8.34, 9.31)
Clinically navigated agents only	75.64% (74.49, 76.8)	17.47% (16.42, 18.53)	6.88% (6.08, 7.69)
Network navigated agents only	84.76% (82.12, 87.4)	10.08% (7.69, 12.48)	5.16% (3.47, 6.84)
Clinically and network navigated agents only	81.61% (79.3, 83.91)	11.32% (9.32, 13.31)	7.08% (5.23, 8.92)

Rows sum to 100%, small deviations are due to rounding error

Sensitivity Analysis

Results from the sensitivity analysis (Appendix Section A.7) suggest that even a 20% navigation rate may produce a substantial improvement in early diagnoses over the control setting that assumed no patient navigation. Increasing the navigation rate to 50% or 80% or patients receiving screening or diagnostic testing produces further, though relatively moderate improvements over the 20% navigation rate used as baseline.

Discussion

Our study provided insight into the potential population-level impact of an increasingly popular, widespread tool to address breast cancer disparities—patient navigation [18–22]. Our study suggests that implementation of network navigation may lead up to a 13% increase in screening completion rate, 7.8% increase in diagnostic resolution rates, and a 4.9% increase in early-stage diagnoses at the population-level. These findings offer useful future benchmarks regarding how much improvement we should anticipate in empirical population-level estimates of stage at diagnosis for AA women with the increasing adoption of navigation (e.g., as per the clinical navigation requirement for Commission on Cancer Accreditation [76]).

Our study further offered preliminary data regarding the nuanced, complex ways in which patient navigation is implemented through clinical and/or community settings [23, 24]. Our findings suggest that combining clinical and network navigation may lead to particularly greater rates of screening mammography completion. Yet, navigation strategies (clinical only, network only, both clinical and network) showed relatively comparable improvements in terms of diagnostic completion and stage at diagnosis outcomes. Our findings should be treated cautiously, as there are several systematic reviews and meta-analyses on clinical navigation, but there is comparatively less on network navigation and its impact on screening mammography. As more empirical evidence regarding network navigation emerges, our ABM may be a powerful tool for further comparing the benefits of different navigation strategies. It should further be noted that our ABM did not include more traditional strategies within community settings, wherein lay navigators, community health workers, and other formal change agents are embedded in communities and provide referrals to clinical navigation services. In the future, there will be a need to develop a scenario to model this strategy in order to compare its benefits to the clinical and network navigation scenarios we modeled here.

Our sensitivity analysis was varied the proportion of agents who receive navigation. We chose this parameter for our further analyses for two reasons: (1) in specifying how widespread navigation is, this parameter may be the most important driver of the population-level effects of the impact of navigation; (2) examining the sensitivity of the results to the proportion of women receiving navigation addresses an important implementation question, on assessing the potential benefit of scaling up navigation as an intervention. These questions of scale, effectiveness and cost remain central to our research agenda.

There are several limitations worth noting. The study focuses on an older segment (50–74 years) of the population. Consequently, the impacts of patient navigation among younger segments, and their contributions to it, are unexamined here. Relatedly, the current study incorporated hormonal risk factors to a limited degree, and given the age group, did not examine differences in menopausal onset. For a fuller examination of the impacts of patient navigation, future iterations of this work might also consider the impact of other patient navigation on broader age representations and incorporate other underserved communities (e.g., Latina women, socioeconomically disadvantaged women, sexual and gender minorities). Our study used data from AA women diagnosed with breast cancer to model social networks in the general population. Future work might develop more generalizable models by collecting social network data from women without a cancer diagnosis as well, thus improving generalizability.

This study used a relatively simple model of cancer progression, as expressed by mapping the time of cancer onset to cancer stage. To note, our study focused on stage at diagnosis—yet, cancers detected at an early stage at diagnosis are not necessarily always associated with better outcomes relative to cancers detected at a later stage [77]. Relatedly, our model considered BMI to be stable over the simulated agent life course. Given the importance of BMI as a risk factor for breast cancer, better empirical data and more nuanced future modeling should address this limitation. CISNET microsimulation models address this challenge, by focusing on breast cancer mortality and other long-term outcome, such as life-years gained [35, 42]. Given the dearth of clinical trial data among African American women [11] and because treatment and its implications are not a focus of the current work, we did not calibrate the model to mortality risk due to screening or diagnostic testing, instead focusing on validating the consistency of simulated screening and diagnostic completion rates with empirical data. This pilot study demonstrated the feasibility and utility of modeling interventions and social network effects for breast cancer outcomes. Future studies are warranted that integrate and expand upon our contributions in modeling navigation and social networks with available microsimulation models that incorporate more realistic disease progression frameworks, and might provide more nuanced data on how navigation impacts cancer outcomes, including the impact of treatment on mortality. Such work would likely make a significant impact for ABMs like the one developed here to understand the impact of navigation and other interventions on population cancer disparities.

In conclusion, our study highlighted the benefits that systems science methods, especially ABMs, to provide precise, comprehensive estimates of the population effects that widespread and popular interventions may have on breast cancer disparities in the long term. We also demonstrate the feasibility of modeling different types of intervention strategies to compare their relative benefits throughout the cancer care continuum. Our findings suggest that the increasingly widespread adoption of patient navigation—within clinical settings and throughout patients’ networks—will lead to substantial improvements in early-stage breast cancer diagnoses for AA women.

Supplementary Information

Below is the link to the electronic supplementary material.