Organizational Factors Affect Safety‐Net Hospitals’ Breast Cancer Treatment Rates

Abstract

Objective

To identify key organizational approaches associated with underuse of breast cancer care.

Setting

Nine New York City area safety‐net hospitals.

Study Design

Mixed qualitative–quantitative, cross‐sectional cohort.

Methods

We used qualitative comparative analysis (QCA) of key stakeholder interviews, defined organizational “conditions,” calibrated conditions, and identified solution pathways. We defined underuse as no radiation after lumpectomy in women <75 years or mastectomy in women with ≥4 positive nodes, or no systemic therapy in women with tumors ≥1 cm. We used hierarchical models to assess organizational and patient factors’ impact on underuse.

Principal Findings

Underuse varied by hospital (8–29 percent). QCA found lower underuse sites designated individuals to track and follow‐up no‐shows; shared clinical information during handoffs; had fully integrated electronic medical records enabling transfer of responsibility across specialties; had strong system support; allocated resources to cancer clinics; had a patient‐centered culture paying close organizational attention to clinic patients. High underuse sites lacked these characteristics. Multivariate modeling found that hospitals with strong approaches to follow‐up had low underuse rates (RR = 0.28; 0.08–0.95); individual patient characteristics were not significant.

Conclusions

At safety‐net hospitals, underuse of needed cancer therapies is associated with organizational approaches to track and follow‐up treatment. Findings provide varying approaches to safety nets to improve cancer care delivery.

Minority patients tend to receive care from hospitals serving predominantly minority patient populations (Hasnain‐Wynia et al. 2007; Jha, Orav, and Epstein 2011). These sites are typically challenged by limited resources, with uninsured and Medicaid patients often receiving poorer quality care and having higher mortality rates (Hewitt and Simone 1999; Bradley et al. 2005, 2012a,b; Bickell et al. 2006; Keating et al. 2009; Joynt, Orav, and Jha 2011; Chatterjee et al. 2012; Dimick et al. 2013; Rangrass, Ghaferi, and Dimick 2014). There are different possible approaches to improve the quality of cancer care of vulnerable populations. One approach ties Medicaid reimbursements to centers of excellence, paying for procedures only if they are done at high‐performing institutions. This approach, however, has the potential to limit access, particularly for vulnerable populations and, thus, may not be feasible. Another approach to improve quality care identifies best practices at high‐performing sites (Bradley et al. 2012a,b) that can be adopted at low‐performing sites. Some safety‐net hospitals, despite limited resources, are able to provide excellent cancer care quality (Fiscella et al. 2012; Bickell et al. 2014). Such sites could provide models to guide quality improvement efforts at poorer performing sites. Because many safety‐net hospitals face limited resources to provide care for challenging populations and it is unlikely these institutions will receive substantially increased funding under the Affordable Care Act (Sabik and Gandhi 2013), it is critical to understand how some safety‐net hospitals are able to deliver high‐quality cancer care. The majority of studies assessing hospital organizational characteristics’ association with quality cancer care focus on constructs such as volume, insurance status, market share, and teaching status (Hébert‐Croteau, Brisson, and Pineault 2000). These variables do not help institutions improve their care. Others focus on multilevel contextual factors and multidisciplinary teams, but the best ways to ensure optimal quality remain uncertain (Fennell et al. 2010; Taplin et al. 2012, 2015).

We sought to identify organizational factors associated with high‐performing safety‐net hospitals and determine their impact on treatment underuse, key components of breast cancer quality measures (National Quality Forum 2012). We chose to look at breast cancer as it shares key characteristics with other cancers—(1) it is relatively common; (2) appropriate and timely treatment can improve survival; and (3) it requires evaluation, treatment, and coordination among numerous different specialists. We looked at underuse of adjuvant therapies as a quintessential quality measure as these treatments improve survival (Early Breast Cancer Trialists’ Collaborative Group [EBCTCG] et al. 2011, 2012, 2014) and their underuse may contribute to racial disparities in mortality.

Methods

We recruited nine inner‐city hospitals, four municipal, and five community hospitals, in the New York City metropolitan area to participate in a trial testing a tool for closing the referral loop between surgeons and oncologists. Hospitals were chosen if they served a large proportion of black and Hispanic women with breast cancer as this population has higher rates of underuse due to system failures than white women (Bickell et al. 2006). We focused on the quality problem of underuse because black women historically had higher mortality rates from breast cancer than white women (American Cancer Society 2015) and were more likely to have treatment omitted due to a system failure, cases in which oncology referrals are requested, not refused, yet care does not ensue. Between 42 and 100 percent of breast cancer patients treated at the participating hospitals were black or Hispanic and all institutions are designated as Disproportionate Share Hospitals (CMS). Institutional Review Board approval was obtained at all sites. To obtain baseline rates of underuse prior to implementing the intervention, we used hospital discharge data and tumor registries to identify all women with a new primary stage I‐IIIa breast cancer who had been treated surgically from 2009 to 2012. We tracked patients’ treatment using previously described methods (Bickell et al. 2006) and abstracted their inpatient and outpatient records to measure adjuvant treatment rates, abstracting records on 493 women. Despite tracking, we could not ascertain treatment for 34 women (8 percent), the majority of whom were treated at a community hospital that closed during the study. Patients missing treatment data were more likely to have commercial insurance than those for whom treatment data were available; there were no differences in cancer stage, patient age, or race between these groups. Our final sample size was 389 women surgically treated at nine hospitals.

Site principal investigators identified key informants who could speak about how breast cancer care and organizational change occur in their institutions. At each hospital, interviewees included hospital leadership of cancer care and quality, breast surgical, medical and radiation oncologists, and their staff including nurse managers, nurses, clerks, and patient navigators. We interviewed 90 key informants (59 clinical, 16 administrative, 12 clerical, and 3 other‐IT and social work) across the nine sites using an interview guide that focused on how breast cancer care is coordinated and how change happens at that institution. Semistructured interviews were recorded and transcribed verbatim. In an iterative and interactive process, three investigators (N. A. B., A. S. M., A. D. M.) coded the transcripts to identify themes. To identify key organizational characteristics associated with low underuse cancer care, we used qualitative comparative analysis (QCA), a methodology created to investigate causally complex outcomes (Ragin, Drass, and Davey 2006).

Measures and Data Analysis

QCA requires five steps: (1) identify the outcome of interest and possibly related conditions; (2) create calibration structures for each measure; (3) calibrate and weight the data; (4) create and minimize a truth table using Boolean logic; and (5) assess solutions based on parameters of fit—consistency and coverage. Consistency measures how often the solution occurs with the outcome (Ragin 2008). Coverage is the proportion of the outcome explained by the solution (Ragin 2008; Schneider and Wagemann 2010; Schneider and Wagemann 2012).

We identified 13 themes important to underuse (see Table 1). Similar themes were grouped together into overarching conditions; for example, the condition “information sharing across specialties” includes themes of integrated electronic medical record (EMR), effective handoffs, and interdepartmental communication. Within each theme are measures or detailed descriptions of a theme's elements. For example, the integrated EMR theme includes two measures assessing (1) what data were electronically available; and (2) which physicians could view the EMR. Each measure was calibrated for each hospital after reviewing transcript quotes describing that measure. Each measure within themes was weighted. For example, because it was more important that surgical and medical oncologists see pathology data to inform adjuvant decisions as compared to the radiation oncologist, who could see information was weighted more heavily than whether the content was available electronically.

Table 1.

Patient Characteristics Comparison by Treatment Groups

	Total (N = 389)	Treated (N = 331) (85%)	Underuse (N = 58) (15%)	p‐value
Age	59.0 ± 12.7	58.3 ± 12.2	62.8 ± 14.9	.03
Race/ethnicity (missing 45)
Black	130 (38%)	111 (37%)	19 (42%)	NS
Asian	31 (9%)	28 (9%)	3 (7%)
White	33 (10%)	29 (10%)	4 (9%)
Hispanic	96 (28%)	85 (28%)	11 (24%)
Other	54 (16%)	46 (15%)	82 (18%)
Insurance
Third‐party commercial	168 (43%)	144 (43%)	24 (46%)	.04
Medicaid	171 (44%)	154 (46%)	17 (33%)
Medicare	32 (8%)	27 (8%)	5 (10%)
Self‐pay	6 (2%)	5 (1%)	1 (2%)
Unknown	12 (3%)	7 (2%)	5 (10%)
Comorbidity (>0)	100 (26%)	84 (25%)	16 (28%)	.783
Tumor size (cm)	2.25 ± 1.38	2.26 ± 1.34	2.22 ± 1.60	.87
Stage
I	153 (40%)	131 (39%)	22 (45%)	.02
IIA	121 (32%)	112 (34%)	9 (18%)
IIB	60 (16%)	54 (16%)	6 (12%)
III	48 (13%)	36 (11%)	12 (24%)
Triple negative	66 (17%)	54 (16%)	12 (23%)	.23
Mastectomy	188 (48%)	165 (49%)	23 (44%)	.25

After all measures were calibrated and weighted for each hospital (see Appendix Calibration) (Basurto and Speer 2012), we created a data matrix (Appendix Data Matrix). Applying Boolean algebra to the matrix, coding all values ≥0.5 as 1, and those <0.5 as 0, we created a truth table indicating the various combinations of conditions among hospitals ranked by their underuse rates. The truth table contains all possible combinations of conditions with 2^k rows, where k is the number of conditions in the analysis (2⁶ = 64). The truth table we present includes only 7 rows because there are no hospitals with the distribution of conditions represented in the remaining 57 rows. For the various combinations of conditions, also called solution formulas, we calculated consistency and coverage scores (Ragin, Drass, and Davey 2006). Finally, to address the common difficulty encountered in QCA research of limited diversity of combinations of conditions, we followed Ragin's approach to deal with logical remainders (Ragin 2008).

We defined underuse as any of the following: no radiation following breast‐conserving surgery (excluding women ≥75 years), no radiation after mastectomy with ≥4 positive nodes, no chemotherapy for hormone receptor negative tumors, and no hormonal therapy for hormone receptor positive tumors ≥1 cm. We assessed comorbidity with the Charlson comorbidity index (Charlson et al. 1987).

To obtain measures of association of underuse with the organizational conditions after controlling for patient‐level variables within each institution, we fit hierarchical models using both patient characteristics (age, insurance, comorbidity, stage) and organizational conditions identified in the QCA (e.g., information sharing, system support, follow‐up, patient‐centeredness). Each organizational condition score was centered around the mean by subtracting the condition's grand mean from the individual hospital score creating a centered measure of 0; hospitals with lower than average measures were negatively scored, whereas hospitals with higher than average measures were positively scored. We used SAS Procedure GLIMMIX (SAS 9.3, Cary, NC, USA) to fit the model. Odds ratios were converted to relative risks as the outcome was not a rare event (McNutt et al. 2003).

Results

Patient characteristics and treatment rates of women with a new primary early‐stage breast cancer surgically treated at the nine safety‐net study hospitals are listed in Table 1. Ten percent of women were white, 44 percent had Medicaid insurance. Fifty‐two percent underwent breast‐conserving surgery, of whom 86 percent received radiation. Of 88 women with hormone receptor negative tumors ≥1 cm, 83 percent received chemotherapy. Of 265 women with hormone receptor positive tumors ≥1 cm, 92 percent received hormonal therapy as per NCCN guidelines (NCCN 2013). Older women were more likely to experience underuse; race and insurance did not affect treatment rates. Hospital rates of adjuvant treatment underuse varied from 7.7 to 29 percent (Table 2). There was no association between volume of breast cancer cases and underuse (r = 0.07; p = .9).

Table 2.

Hospital Characteristics and Underuse Rate

Hospital	Hospital Type	Breast Cancer Cases/Year	Underuse Rate
1	Municipal	17	7.7%
2	Municipal	16	8.6%
3	Community	52	9.7%
4	Community	27	11.3%
5	Municipal	16	12.0%
6	Municipal	16	15.5%
7	Community	30	19.7%
8	Community	12	23.1%
9	Community	25	28.9%

Table 3 lists and defines the thirteen organizational themes we identified as relevant to the delivery of low‐underuse cancer care, grouped into six overarching conditions: information sharing, follow‐up, system support, a patient‐centered culture, private practice, and flexibility.

Table 3.

Organizational Themes Grouped by Condition

Condition	Theme	Measures/Definitions
Information sharing	Integrated EMR	Includes labs, imaging, notes, and order entry Shared between treatment sites/providers
	Effective handoffs	A handoff occurs in which a clinical person from one specialty exchanges information with and transfers responsibility to a clinical person in another specialty An appointment is made for the other specialty without the patient having to call and make it
	Interdepartmental communications	Providers send treatment summaries/consult notes OR notes are available in the EMR and providers look at them Willingness to get in touch with other providers about a patient Specialties treating cancer are in the same place Effective tumor board for discussing breast cancer cases
Follow‐up	No‐show tracking and follow‐up	“Extra” measures are taken when “baseline” measures have failed Clinicians are actively involved in the no‐show follow‐up process Tracking no‐shows is someone's designated responsibility
	Tracking across specialties	Tracking patient care is someone's designated responsibility
System support	Clinical leadership	The leader worked to affect positive change/improvements in the past year
	Organizational attention to clinic patients	Everyone gets the same caliber of treatment and follow‐up regardless of ability to pay There are enough staff/providers with sufficient resources to handle the patient flow in the clinic; intolerance of understaffed, chaotic clinics
	Resource support	The organization allocates adequate resources to support the cancer program
	Supported doctors	Doctors are not overwhelmed and able to handle the volume of work they are expected to do
Patient‐centered culture	System, providers, and staff are responsive to patient needs	Providers and staff work for the patients and not the other way around The system makes it easy for patients to get what they need
Private practice	Organizational attention to private patients	The system is responsive to private patients
	High proportion of private patients	Private patient population of 70% or greater
Flexibility	Development of creative work‐arounds	Providers or staff developed work‐arounds to solve specific and/or system problems

EMR, electronic medical record.

Table 4 shows a modified Truth Table which displays the number of hospitals with their varying organizational characteristics. All five high‐quality hospitals had strong information sharing, follow‐up, and patient‐centeredness. All high‐underuse hospitals had poor system support and most had weak follow‐up strategies.

Table 4.

Organizational Characteristics in High‐ and Low‐Quality Hospitals (Truth Table)

Information Sharing	Follow‐Up	System Support	Patient‐Centered	Private Practice	Flexibility	No. of Hospitals with Organizational Characteristics	High‐Quality Hospitala
+	+	−	+	+	−	1	Y
+	+	+	+	−	−	2	Y
+	+	−	+	−	+	2	Y
+	+	−	+	−	−	1	N
+	−	−	+	−	+	1	N
+	−	−	−	−	+	1	N
−	−	−	−	−	+	1	N

^aY = Yes, high quality (underuse rate <15%); N = No high quality (underuse rate ≥15%).

Number of hospitals with this row's distribution of organizational characteristics (total no. of hospitals is 9).

QCA finds three distinct pathways to low rates of underuse (Figure 1): all hospitals with low underuse had high levels of information sharing, approaches to follow‐up, a patient‐centered culture, and one additional condition. The fourth condition varied by hospital. Two hospitals had strong system support, including effective clinical leadership, resources for the cancer program, and organizational attention to clinic patients. Their physicians were able to handle the volume of work. Providers and staff appreciated a system that allowed them to provide high‐quality care and propose changes and improvements. For instance, a medical oncologist at one of these sites observed, “We are given a lot of time to see patients. So we see one patient maybe every 20–30 minutes as a follow‐up, and for a new patient it's a whole hour. This is not a lot of patients compared to what might be going on in an outside practice. We are not seeing 40 patients. All of our patients are getting a lot of time.”

Figure 1.

Pathways to Low Underuse

Two hospitals did not have strong system support but their clinical staff were flexible and creative, devising work‐arounds to ensure their patients got the care they needed. However, staff and providers at these hospitals were frustrated with the broken system and exhausted by constantly needing to work around it. One medical oncologist remarked, “It's a lot of doing favors for people in a time of need and that's sort of how this system works, right? You help people out and then they'll help you out. It's very, almost third world‐ish. [For example], when I was a fellow, I used to bring chocolate and a Christmas gift to the person who did the CT scan scheduling because there was a 3‐month wait just to get a regular CT scan over here. And my patients would get to the front of the line just through that.” This pathway, containing flexibility and work‐arounds, had low rates of underuse but takes a toll on providers and staff.

Follow‐Up Care

Follow‐up care includes approaches to track patients who do not show up for their surgical appointments and to ascertain patients’ connections with multispecialty consults. While nearly all physicians used follow‐up appointments as a method to check their patients’ status, few had systems in place to check on no‐shows. Some EMRs enabled providers to see specialty appointments and notes, but in our study sample, none had “ticklers” to prompt intervention for lack of connection with requested consults. High‐performing hospitals assigned staff to follow up with those patients who do not show for appointments and close this potential gap in care. Personnel assigned this task varied by hospital and included schedulers, navigators, and nurses.

Information Sharing

Information sharing included use of an integrated EMR and effective handoffs across specialists. While electronic records do allow multiple disciplines to view information important to treatment decision making, personal handoffs, including phone calls between specialists, appear to add value as patients were more likely to follow‐through with requested consults when personal hand‐offs occurred (Jeffs et al. 2013). All participating sites had EMRs, but these systems had varying capabilities and were not always integrated across the institutions. In fact, a few had more than one EMR with ensuing challenges in access to information from different systems. Several clinicians spoke with consultants while patients were in the office—in fact, some walked their patients to the medical oncologist and introduced them. We were unable to determine how often these personalized approaches were implemented, but at sites in which personal handoffs were described, surgeons, medical, and radiation oncologists spoke of this shared approach to care. We were unable to ascertain from our data whether personalized transitions in care were effective because patients’ oncology appointments were made before they left the surgeons’ office, or because there was a clear transfer of responsibility from one clinical entity to another.

Patient‐Centeredness

Patient‐centeredness (AHRQ 2010) reflects a commitment to work for and with patients, to make the system easy for patients to get what they need. For example, a radiation oncology nurse noted, “I think it was more of just allowing the patient to know that somebody is there for them because when you give them a card and that card has your oncology nurse on it, you can call me here between this time and this time and they know that they have somebody to call and they know that somebody cares for them, and I've seen an overall improvement since I've been here. I [also] think [the radiation oncologist] got on them with a personal level. He gave them a cell phone number, he said, ‘okay, well this is how you can contact me if you have any problems.’ Because here, like after 4 o'clock, you'll have nobody to talk to basically. You will leave a message—the message is going to stay on until on Monday when somebody comes in to take the message. So he really got on a personal level with them and they're able to contact someone. They know that we're out looking for them to come in for treatment.” This core value is highly correlated with strong follow‐up procedures and sharing information.

System Support

System support is composed of clinical leaders who work to affect positive change and improvements; an intolerance of understaffed, chaotic clinics; and adequate resources to support clinical care. Some hospitals with poor system support achieved high quality. They were able to do so because their providers and staff were flexible, creating work‐arounds to solve specific or systemic problems. The cost of flexibility, though, was evident in staff burnout and dissatisfaction, consistent with findings from prior research (Humphries et al. 2014), and is not a sustainable approach to care delivery.

Private Practice

One high‐performing site had parallel systems for surgeons seeing patients in their private practice and another for surgeons seeing patients in the clinic. Independent, private practice community physicians created their own systems to track and respond to patients which likely contributed to the overall high performance rates attributed to the hospital. At the same time, however, the clinic system was variable in its follow‐up. This sort of dual system with less accountability and follow‐through for clinic patients does little to address disparities in care.

Assessing the Solution Formula

Each pathway solution had high consistency (range: 0.82–1.0). Coverage rates varied from 0.16 to 0.49 (Figure 1). The low coverage rate for the pathway with private practice reflects the small proportion of safety‐net hospitals in our sample that had substantial proportions of private practice patients. Together, these three pathways characterizing organizations’ approaches to ensuring delivery of needed adjuvant therapies explain a large proportion of the high treatment rates these safety‐net hospitals experience (0.72).

The hierarchical model assessing patient and organizational factors and controlling for clustering within the hospital found that hospitals using strong follow‐up approaches had lower rates of underuse (RR = 0.28; 95 percent CI: 0.08–0.95). None of the patient factors such as stage ≥2b (RR = 0.99; 95 percent CI: 0.5–1.9), comorbidity (RR = 1.04; 95 percent CI: 0.56–1.92), Medicaid insurance (RR = 1.01; 95 percent CI: 0.53–1.95), black race or Hispanic ethnicity (RR = 1.05; 95 percent CI: 0.61–1.8), and older age (70 + yrs) (RR = 1.20; 95 percent CI: 0.6–2.2) was associated with treatment underuse. Hierarchical models with the other organizational factors—information sharing, patient‐centeredness, system support, flexibility—were not statistically significant.

Discussion

Despite challenging resource limitations at safety‐net hospitals, we found key organizational characteristics associated with low underuse. Hospitals with strong approaches to follow‐up care, information sharing, patient‐centeredness, and system support had lower rates of underuse and provided excellent quality treatment to their breast cancer patients. Hospitals without these approaches had higher rates of adjuvant therapy underuse. Assessing and addressing organizational approaches to cancer care may offer concrete steps to improve the quality of cancer care delivery as it does for chronic care (Bickell and Young 2001; Wagner et al. 2001).

Where patients get their care impacts the quality of care received (Rhoads et al. 2008; Chen et al. 2011; López and Jha 2013). Patients receiving care at hospitals serving predominantly minority populations tend to have worse outcomes than those treated at nonminority serving institutions (López and Jha 2013), and vulnerable populations tend to get their get care at poorer performing hospitals (Hasnain‐Wynia et al. 2010; Jha, Orav, and Epstein 2011). All safety‐net hospitals participating in this study receive Disproportionate Share Hospital (DSH) Medicaid payments. Given the key role these facilities play in providing care to the poor, steps are needed to ensure high quality. While restricting Medicaid funding for breast cancer surgery to higher volume hospitals is one approach to improve vulnerable populations’ cancer care quality, it runs the risk of impeding easy access to cancer treatment.

Our QCA findings offer such hospitals different pathways to achieve excellent cancer care quality, particularly by ensuring follow‐up with requested consults and treatment. Assigning an individual to follow‐up “no‐shows” offers a potentially low‐cost intervention to safety‐net hospitals as it relies on current employees and scheduling systems but creates a process with accountability. Additionally, warm handoffs of patients across specialists and personal calls to physicians helped combat “alert fatigue” and ensure transitions across care. All study hospitals had an EMR, yet this tool did not guarantee sharing of information. Rather, test and referral tracking, coordination, and communication beyond the EMR are needed (Kern, Edwards, and Kaushal 2014). While all institutions had multidisciplinary tumor boards, these meetings cannot be relied on to ensure communication across specialists (Keating et al. 2013). Hospitals with strong patient‐centered approaches to care achieved high treatment rates. Sites without such an emphasis will likely require a shift in organizational culture, a challenge to even the strongest organizations (Shortell and Kaluzny 1999).

Given the planned reductions in DSH reimbursement scheduled under the Affordable Care Act (KFF 2013), it is anticipated that safety‐net hospitals will experience greater financial constraints despite a boost in Medicaid recipients, particularly in areas with high rates of uninsured (Coughlin et al. 2014). Currently, these hospitals experience greater financial penalties for readmissions than hospitals serving the wealthy (Medicare 2014; Gilman et al. 2015). There may be important lessons the cancer community can learn from Medicaid Health Homes and Patient‐Centered Medical Homes that focus on improving care coordination and pay for such services (Medicaid 2015; AHRQ). However, safety‐net hospitals lag in the digital divide (Jha et al. 2009) and so face significant challenges attaining higher levels of meaningful use required for PCMH designation. Our patient‐level analysis suggests that providing strategies that focus on follow‐up and handoffs across specialties may offer the greatest yield for safety‐net hospital cancer care quality improvement efforts.

Our work is limited by geographic region and institution type. The lack of insurance effect on treatment differs from others (Bradley et al. 2005, 2012a,b) and likely reflects generous Medicaid reimbursement in New York and a sample limited to safety‐net institutions. The time and energy required for in‐depth case studies creates significant challenges for large‐scale, multiregion research, and it limits our ability to generalize results across hospitals with different organizational structures and patient populations. At present, many hospitals lack the ability to assess levels of effective follow‐up and transitions in care, let alone patient‐centeredness and adequate system support (Shaller 2007; Groene 2011; Davis et al. 2012). The calibration matrix developed in this project may offer process performance measures to aid with this endeavor but should be tested in other sites to further validate their accuracy and usefulness. The high correlation between patient‐centeredness, follow‐up, information sharing, and system support suggests the multiple dimensions associated with high‐quality organizations. That follow‐up was the single organizational approach achieving statistical significance through quantitative modeling may reflect the key cause of underuse in the hospitals we studied, namely, gaps in care transitions (Bickell et al. 2008). Our study focused on breast cancer treatment underuse, but the organizational issues identified may be relevant to other cancers and should be evaluated.

In summary, in safety‐net hospitals, underuse of needed breast cancer adjuvant treatment is affected by patient age and remediable hospital organizational factors. Safety‐net hospitals with high treatment rates shared information, tracked their follow‐ups, and fostered a patient‐centered culture. System support is important, but in its absence, a flexible staff creative in their work‐arounds is able to achieve low underuse, a fundamental component of high‐quality breast cancer care.

Supporting information

Appendix SA1: Author Matrix.

Appendix SA2: Calibration.