2.1 Study design and sample

This retrospective cohort study included women with breast cancer receiving care at either of two large, academic, public and private, Georgia-based health care systems, Grady Health System and Emory Healthcare. Women ≥18 years of age diagnosed with breast cancer between 2004 and 2022 and who received at least one treatment for breast cancer were included (Figure 1). Data for this study were abstracted from electronic patient records by hospital administrators. Patients with missing data for age, race, ethnicity, treatment start dates, and address or nine-digit zip codes were excluded. Unknown categories were created for American Joint Committee on Cancer (AJCC) stage and breast cancer subtype variables to address high rates of missing data for these variables. This study was approved by the Georgia State University Institutional Review Board (IRB) and by the hospitals providing data. The IRB waived the need for consent as medical record data were collected retrospectively.

2.2 Delayed treatment

The outcome for our study was delayed treatment. Patients were considered to have received delayed treatment if the number of days from breast cancer diagnosis to initiation of breast cancer treatment was higher than that recommended in prior studies and the National Accreditation Program for Breast Centers guidelines.^{6, 20, 21} Breast cancer treatments included surgery, neoadjuvant systemic therapy, and other (systemic therapy only, radiation therapy only, systemic and radiation therapy). Surgical treatment refers to time from diagnosis to curative surgery date, and does not include excisional biopsies or non-cancer directed surgeries. Delayed treatment was defined as treatment initiated >90 days after diagnosis for patients receiving surgery as first treatment and >120 days for those receiving systemic therapy first (i.e., chemotherapy, endocrine therapy, and targeted therapy).

2.3 Exposure: Insurance coverage status

Health insurance coverage was captured during the initial oncology clinic visit and classified into one of the following six categories: private insurance, Medicaid, Medicare, other public coverage (Indian Health Service, Military, and Veterans Affairs), none (i.e., uninsured), and unknown.

2.4 Neighborhood deprivation

Neighborhood deprivation was used as a proxy for patient socioeconomic status and was calculated using the Area deprivation index (ADI).²² The ADI is a composite measure comprised of 17 metrics of income, education, employment, and housing quality determined using census block group 12-digit Federal Information Processing Standards (FIPS) codes or nine-digit zip codes. ADI scores were calculated based on the patient's home address at breast cancer diagnosis by geocoding the address to a 12-digit FIPS code using the United States Census Bureau's Census Geocoder²³ and to a nine-digit zip code using the US Postal Service zip code database.²⁴ Nine-digit zip codes or 12-digit FIPS codes were matched to census block ADI scores obtained from the Neighborhood Atlas website.^{25, 26} The ADI is scored from 1 to 100, with higher scores indicating greater neighborhood deprivation. Areas with ADI scores ≥85 were considered neighborhoods of high deprivation, whereas those with scores <85 were considered neighborhoods of low deprivation. These categories of ADI scores were based on previous studies using similar cut points.^27-29

2.5 Covariates

Abstracted patient demographic information included age at diagnosis, hospital, and race and ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, and other race). Abstracted patient clinicopathologic data included AJCC stage (0, I, II, III, IV), breast cancer subtype (hormone receptor [HR]+/human epidermal growth factor receptor 2 [HER2]+, HR−/HER2−, HR+/HER2−, HR−/HER2+), and treatment modality (neoadjuvant [systemic treatment followed by surgery], surgery [surgery as first treatment], and other [radiation as first treatment or systemic treatment only]).

2.6 Statistical analysis

Descriptive statistics were calculated using frequency counts and percentages for categorical variables and medians and interquartile ranges (IQR) for continuous variables. The association of delayed treatment and insurance status was estimated using odds ratios (OR) and 95% confidence intervals (CI) using a logistic regression model. The model was adjusted for patient age at diagnosis, race and ethnicity, hospital, neighborhood deprivation, treatment modality, stage, and breast cancer subtype. To understand the effect of neighborhood deprivation on the relationship between time to treatment initiation and insurance status, an interaction term was added to our model. Interaction significance was assessed using likelihood ratio tests, which did not result in a significantly different fit. Multicollinearity of model covariates was evaluated using the “car” package³⁰ in R, which estimates variance inflation factors. Absence of multicollinearity was confirmed for all covariates included in the final logistic regression model. Data cleaning, analysis, and modeling were performed using R version 4.2 (Vienna, Austria) and SAS software version 9.4.³¹ To qualitatively compare geographic clusters of deprivation by insurance status, we created maps using ArcGIS and US Census Bureau shapefiles.^{32, 33}

3.1 Sample characteristics

Data from 14,195 of the 19,284 (74%) women with breast cancer in our dataset were included in our analysis (Figure 1) after excluding patients with missing data for age, race, ethnicity, and ADI score. Compared to included patients, excluded patients were more often Medicare enrolled (36% vs. 28%) and were less often non-Hispanic Black (35% vs. 54%) and living at low neighborhood deprivation (5% vs. 85%; Supplemental Table S1). Among the included patients, median age at diagnosis was 58 years (IQR 49–67) and 54% of patients were non-Hispanic Black. Patients were most often diagnosed with stage I disease (30%), 37% had HR+/HER2− breast cancer, and 74% underwent surgery as their first treatment.

Over half of the included patients were privately insured (52%), 28% were Medicare enrollees, 11% were Medicaid enrollees, 4% were uninsured, and <1% had other public insurance (Table 1). The majority of privately insured patients were non-Hispanic White (49%) or non-Hispanic Black (49%). Non-Hispanic Black patients were more often Medicaid enrollees (17% vs. 4%) or uninsured (5% vs. 2%), compared to non-Hispanic White patients.

3.2 Delayed treatment and insurance status

Median time to treatment initiation was 43 days for uninsured patients (IQR 28–69), 44 days for Medicaid enrollees (IQR 28–68), 40 days for privately insured patients (IQR 23–60), and 39 days for Medicare enrollees (IQR 24–60; Table 1). A total of 9% (n = 1241) of patients received delayed treatment, which varied by first treatment type. Of patients who underwent surgery as first treatment, 11% experienced delays. Medicaid enrollees and patients with other public insurance had the longest time to surgical treatment initiation (median 45 days, IQR 27–70 and median 57 days, IQR 35–108, respectively; Table 2). Only 4% of patients who received systemic therapy first received delayed treatment. Patients who were uninsured had the longest time to systemic treatment initiation (median 47 days, IQR 30–66; Table 2).

In our multivariable-adjusted model, uninsured patients had 79% higher odds of experiencing treatment delays compared to those privately insured (OR: 1.79, 95% CI: 1.32–2.43; Table 3). Compared to patients with private insurance, Medicaid enrollees had 75% higher odds and Medicare enrollees 27% higher odds of experiencing treatment delays (OR: 1.75, 95% CI: 1.43–2.13; OR: 1.27, 95% CI: 1.06–1.51, respectively).

3.3 Modification by neighborhood deprivation

Patients living in neighborhoods of high deprivation comprised 15% of our sample. Compared with those living in neighborhoods of low deprivation, patients living in neighborhoods of high deprivation were more often non-Hispanic Black (85% vs. 49%). Women living in neighborhoods of high versus low deprivation were less often privately insured (34% vs. 55%) and were more often enrolled in Medicaid, Medicare, or uninsured (22% vs. 9%, 31% vs. 27%, and 7% vs. 4%, respectively). Patients living in neighborhoods of higher vs. lower deprivation also had similar median time to treatment initiation (43 days, IQR 27–65 vs. 40 days, IQR 24–60).

In qualitative geographical comparisons, we observed higher concentrations of patients who were Medicaid or Medicare enrollees living in areas of higher deprivation compared to patients with private insurance (Figure 2). For patients living in neighborhoods of high deprivation, the odds of delayed treatment did not differ between patients covered by different insurance (Table 3). Conversely, patients living in low-deprivation neighborhoods who were uninsured, Medicaid beneficiaries, or Medicare beneficiaries had 100%, 84%, and 26% higher odds of delayed treatment compared with privately insured patients (OR: 2.00, 95% CI: 1.44–2.78; OR: 1.84, 95% CI: 1.48–2.30; OR: 1.26, 95% CI: 1.05–1.53, respectively).

4.1 Limitations

Results from this study should be considered within the context of several limitations. This study assumes that patients did not receive care at other facilities and that treatment delays were not due to patient requests, such as a patient's decision to receive a second opinion on a treatment plan at another hospital. The address was available only for residence at the time of diagnosis. ADI provides information on neighborhood deprivation rather than individual-level socioeconomic status. We also did not have information for other factors affecting access to care due to residence, such as availability of public transportation, walkability, and safety. Few patients were covered by non-Medicare or Medicaid public insurance, which limited our ability to reliably compare associations for this group. Furthermore, this study was limited in the geographic generalizability of results because most patients were located and treated in the Atlanta metropolitan area.