Stalled at the intersection: insurance status and disparities in post-mastectomy breast reconstruction

Post-mastectomy breast reconstruction (PMBR) is an important component of breast cancer treatment, but disparities relative to insurance status persist despite legislation targeting the issue. We aimed to study this relationship in a large health system combining a safety-net hospital and a private academic center. Data were collected on all patients who underwent mastectomy for breast cancer from 2011 to 2019 in a private academic center and an adjacent public safety-net hospital served by the same surgical teams. Multivariable logistic regression was used to assess the effect of insurance status on PMBR, controlling for covariates that included socioeconomic, demographic, and clinical factors. Of 1554 patients undergoing mastectomy for breast cancer, 753 (48.5%) underwent PMBR, of which 592 (79.9%) were privately insured, 50 (6.7%) Medicare, 68 (9.2%) Medicaid, and 31 (4.2%) uninsured. Multivariable logistic regression showed a significantly higher likelihood of not undergoing PMBR for uninsured (OR 6.0, 95% CI 3.7–9.8; p < 0.0001), Medicare (OR 1.9, (95% CI 1.2–3.0; p = 0.006), and Medicaid (OR 1.5, 95% CI 1.0–2.3; p = 0.04) patients compared with privately insured patients. Age, stage, race and ethnicity, and hospital type confounded this relationship. Patients without health insurance have dramatically reduced access to PMBR compared to those with private insurance. Expanding access to this important procedure is essential to achieve greater health equity for breast cancer patients.


Introduction
It is estimated that over 275,000 new cases of breast cancer were diagnosed in 2020 in the USA [1]. Of these, around one third will undergo mastectomy [2,3]. Post-mastectomy Orli Friedman-Eldar and Jonathan Burke are co-first authors.
* Orli Friedman-Eldar eldaror@gmail.com breast reconstruction (PMBR) is an essential aspect of care for breast cancer, conferring a variety of benefits for patients. These include the restoration of breast appearance [4], alleviation of psychological distress [4,5], improvement of self-esteem [6], and enhancement of quality of life [4,7]. While around half or more of women undergoing mastectomy may prefer PMBR [8], in the USA actual rates have historically been considerably lower, ranging from around 10-15% in the late 1990s to early 2000s [2,9]. In 1997, Medicare began to cover PMBR for breast cancer, and the Women's Health and Cancer Rights Act (WHCRA) of 1998 mandated coverage of PMBR by private insurers. Many states, including Florida, later mandated Medicaid coverage of PMBR as well [10][11][12]. While the WHCRA helped increased PMBR rates, albeit mostly in states without prior legislation requiring coverage for the procedure [13], many patients are still unable to access PMBR. The most recent rates are estimated to be around 35-40% [2,3], but stark disparities in PMBR exist, particularly surrounding socioeconomic factors such as race and ethnicity [14,15], educational level [16], and insurance status [14,17]. In fact, these disparities have persisted or even increased following the implementation of the WHCRA [18], leaving many patients without the option to undergo this important procedure. Previous research examining the effect of insurance status on PMBR rates has utilized state and national databases. While these other investigations give a more global view, a real-world perspective of how insurance status affects access to care in a local population is lacking. This study aims to fill this gap by examining the local effect of insurance status on PMBR rates in a public, safety-net hospital, and an adjacent private academic center.

Study design
This is an observational cohort study of patients undergoing mastectomy using a multi-hospital institutional database. The study was approved by the institutional review board. We retrospectively collected data from the charts of all patients undergoing mastectomy at the local public safety-net hospital and private university hospital between 2011 and 2019. The same multidisciplinary team composed of surgical oncologists, plastic surgeons, medical oncologist, and radiation oncologist cover and evaluate the patients at both institutions. The same team of surgical oncologists and plastic surgeons operate at both hospitals and get incentive based on volume but with no distinction or difference based on insurance status or type of hospital (private vs. public) for offering or performing PMBR. This article adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist for cohort studies [19].

Patient selection
All patients undergoing mastectomy from 2011 to 2019 were identified in the institutional medical records using the primary Current Procedural Terminology (CPT) codes for simple mastectomy (19303), subcutaneous mastectomy (19304), radical mastectomy (19305)(19306), and modified radical mastectomy (19307). Male patients and patients undergoing mastectomy for indications unrelated to cancer, such as gynecomastia or gender-affirming surgery, were excluded. Stage IV patients were included, despite the fact that the majority of them are not being offered reconstruction, to represent the entire spectrum of patients referred to mastectomy for breast cancer.

Exposure and potential confounding variables
Inclusion criterion for all variables was the presence of 80% or more data for all records in the patient sample to prevent the need for extensive imputation. We categorized the primary exposure variable, insurance status, as uninsured/ self-pay, Medicare, Medicaid, and privately insured. We identified potential confounding variables based on demonstrated relationships with PMBR rates in the literature and our institution's clinical experience. We then developed a Directed Acyclic Graph (DAG) to determine which variables should be assessed as potential confounding variables in the multivariable model. These included demographic variables (age, race and ethnicity, income) [14,18], comorbidities (Body Mass Index [BMI], Diabetes) [20,21], perioperative factors (stage, functional status, neoadjuvant chemotherapy) [22], hospital type, and behavioral risk factors (smoking) for breast implant complications and failure [23].
We categorized race and ethnicity as non-Hispanic Black, non-Hispanic White, Hispanic, and Other, according to self-identified race and ethnicity in the medical record. As done in previous studies [18], income was inferred from the patients' self-listed zip code associated with their address based on the median income for each zip code as recorded in the publicly accessible 2017 Florida Census [24]. Functional status was assessed using the recorded American Society of Anesthesiologists (ASA) physical status classification system [25]. We determined stage using the American Joint Committee on Cancer (AJCC) 8 th Edition criteria [26] and limited categorizations to the five numbered stages along with prophylactic and other (for cancer types other than adenocarcinoma such as Phyllodes tumors where the staging does not apply). We likewise determined nodal status and utilized N0, N1 (including microscopic N1), and N2-3, as has been done in previous studies assessing the effect of nodal status on radiation therapy [27], which also influences decision making with PMBR [28]. We categorized smoking status as current, former, and never. Continuous variables, including age, income, and BMI were grouped into deciles and plotted against the log odds for PMBR to assess for linearity. Annual income was categorized empirically into < $50,000, $50,000-$80,000, > $80,000; BMI was categorized according to standard measures [29] and age was categorized empirically into < 50, 50-70, and ≥ 70 years-old. For technical reasons, all flap reconstructions were done physically at the public hospital, however, for the purpose of this study, a patient that was seen in the private hospital and underwent a flap reconstruction was still considered as a private facility patient. Finally, a large sample of records were evaluated for documentation of the reason patients did not undergo reconstruction (categorized to advanced disease, lack of insurance, high risk for complications, patient's preference), and whether it was offered to them during the clinic visit.

Clinical end point
The primary outcome was the occurrence of PMBR, utilizing immediate or delayed reconstruction with either tissue expanders, implants, or flaps. The time frame to look for reconstruction was the day of chart review and not less than 1 year after diagnosis.

Addressing potential sources of bias
In order to reduce selection bias, we included data from two hospitals with the same surgeons, minimizing practice variations. We also included only female patients undergoing mastectomy for an indication related to breast cancer. We utilized objective insurance data from the charts and a careful review of post-mastectomy care for patients not undergoing PMBR to identify delayed reconstruction and prevent misclassification bias. We minimized missing data by adding an "Unknown" category to all covariables with > 5 missing data points.

Statistical analysis
Exposure variables were characterized using frequency and percentage of categorical variables and mean and standard deviation for continuous variables. Associations with the main outcome were determined in univariable analyses using the Chi-squared test or Student's t test as appropriate. Potential confounding variables with p < 0.10 were included in multivariable logistic regression, and those variables that did not meet that criteria but were determined to have sufficient potential for confounding based on past literature and/or clinical expertise were also included.
A multivariable logistic regression model accounting for covariates was used to assess the effect of insurance status on not undergoing PMBR, and the effect was considered statistically significant if p < 0.05. Confounding was assessed by removing a potential confounding variable from the model and examining any change in effect size of the primary exposure variable, with a 10% difference considered to be significant. Sensitivity analyses were performed to see the effect of excluding patients with Stage IV disease given the uncommon indication for mastectomy with or without reconstruction in that setting, as well as to assess the impact of the creation of "Unknown" categories for all missing data to have complete case analysis. Post hoc secondary analysis was performed by adding an interaction term between insurance status and race and ethnicity to the full multivariable model to assess if race and ethnicity modifies the relationship between insurance and PMBR. Data were analyzed using SAS 9.4 (SAS Institute, Cary, NC).

Results
Of 1802 total patients undergoing mastectomy, 1554 met inclusion criteria for analysis as specified above (Fig. 1). The overall patient population had a mean age of 53.9 years, and the majority were Hispanic (59.5%) with substantial proportions of non-Hispanic Black (17.9%) and non-Hispanic White (20.9%) patients. The majority of patients, 68.3% (n = 1061), had early-stage disease (Stage 0-II), while 22.5% (n = 350) had Stage III, and 4.3% (n = 67) had stage IV disease. Patients without insurance were more likely to be Hispanic and have a lower income (Table 1).
Overall, 48.5% (n = 753) of patients underwent PMBR, 741 of them had insurance type recorded, with 79.9% (n = 592) privately insured, 6.7% (n = 50) Medicare, 9.2% (n = 68) Medicaid, and 4.2% (n = 31) uninsured. Type of reconstruction was known for 737 patients in the PMBR group, 664 (90%) of them underwent immediate reconstruction (446 received tissue expanders, 178 direct implants, and 40 free flaps) and 73 (10%) underwent delayed reconstruction. Patients that underwent PMBR were more likely to be of non-Hispanic White origin, to have earlier stage disease, and to have fewer comorbidities than those that did not. Univariable analyses demonstrated statistically significant associations between insurance status and PMBR (p < 0.0001), as well as for all co-variates except for smoking (p = 0.32) ( Table 2).
Multivariable logistic regression controlling for age, race and ethnicity, income, comorbidities, functional status, stage, nodal status, and neoadjuvant chemotherapy demonstrated a persistent effect of insurance status on PMBR, with uninsured patients having 6.0 times the odds (95% Confidence Interval [CI] 3.7-9.8, p < 0.0001) of not undergoing PMBR as privately insured patients (Table 3). This effect was smaller, yet still statistically significant greater odds of not undergoing PMBR for both Medicare (Odds Ratio [OR] 1.9, 95% CI 1.2-3.0; p = 0.006) and Medicaid (OR 1.5, 95% CI 1.0-2.3; p = 0.04) compared to privately insured patients. Age, stage, race and ethnicity, and hospital type confounded the relationship, while diabetes, BMI, ASA status, smoking, income, nodal status, and neoadjuvant chemotherapy did not. Sensitivity analysis revealed little effect (< 10%) on the primary effect measure after creation of "Unknown" categories for all missing data to have complete case analysis. Post hoc analysis showed that neither race and ethnicity (p = 0.98) nor nodal status (p = 0.23) modified the relationship between insurance status and PMBR.

Discussion
This study demonstrates the dramatic effects of insurance status on PMBR utilization. This affirms the global relationship demonstrated in previous studies [14,18,30] in a local, real-world context in urban teaching hospitals, with lack of insurance resulting in far less access to PMBR. Based on prior literature showing substantial effect of PMBR on various quality of life measures [4][5][6][7], lack of access to PMBR for uninsured patients may translate to poorer health outcomes after mastectomy.
While Medicare and Medicaid both fully cover the procedure in our state, patients with government insurance were significantly more likely to not undergo PMBR compared to patients with private insurance (OR 1.9 for Medicare, 95% CI 1.2-3.0; p = 0.006 and OR 1.5 for Medicaid, 95% CI 1.0-2.3; p = 0.04). The effect was much   more dramatic for uninsured patients (OR = 6.0, 95% CI 3.7-9.8; p < 0.0001). Like other health disparities that exist for uninsured patients, such as in cancer survival [31] and major complications after general surgery procedures [32], this may be partially attributed to lack of access to consistent, high-quality primary and preventive care, leading to more advanced disease. In this study, the difference of effect for government-insured patients with inclusion of stage in the model supports this notion. Indeed, more advanced stage may be viewed as a mediator in the relationship between insurance status and PMBR, as lack of insurance can prevent access to screening mammography, leading to a later stage breast cancer at presentation that may be less eligible for reconstruction. Medicare patients had significantly higher rates of comorbidities including diabetes and increased ASA score ( Table 1) that often accompany advanced age, yet even after stratifying by age in the raw data, about half the proportion of patients ≥ 70 years old with Medicare underwent PMBR compared to those with private insurance (16.3% vs 32.7%). After controlling for age and the other confounders in multivariable analysis, these patients still had significantly lower odds of undergoing PMBR.
Age, race and ethnicity, and hospital type substantially confounded the relationship between insurance status and PMBR, demonstrating the important intersection between insurance status and these other factors. Specifically, Non-Hispanic Black patients were significantly more likely to not undergo PMBR (Table 3), a fact that has impact on insurance status as part of the wellestablished association between Black race and lower socioeconomic status [33]. The large effect of age on PMBR (OR 15.0, 95% CI 8.4-26.9 for age ≥ 70 compared to < 50) is reflected in its considerable confounding on the relationship between insurance and PMBR, particularly with Medicare patients (121% decrease in effect size when including age). Interestingly, income did not confound the effect of insurance on PMBR, likely because income and insurance status are often intertwined with socioeconomic status.
The distribution of uninsured and insured patients at each hospital highlights the differences in access to care. A private academic center, serving primarily privately insured patients, allows for PMBR to most patients who may desire it, while the need to provide a large proportion of care as charity care in a public, safety-net hospital strains financial resources [34,35] and may restrict funding for the procedure. Indeed, the safety-net hospital in this study previously provided PMBR to all patients regardless of insurance status, but lack of adequate funding around 2008 necessitated cutting services, including PMBR. Still, after inclusion of hospital type, there is a 127% decrease in effect size for Medicaid patients and 37% decrease for uninsured patients, indicating that the public hospital mitigates the disparities for both these groups. This is the first study to our knowledge to demonstrate the disparity in undergoing PMBR for uninsured patients adjusting for a broad range of salient clinical and social factors in a local, real-world setting. Previous research utilizing statewide and national administrative and clinical databases [14,18,30,36] lacked the perspective of showing the on-the-ground effects of insurance status in a particular community. Further, as previously mentioned, in this study the same surgeons work at both institutions with financial incentive based on volume, but not based on insurance status or type of hospital, which minimizes provider dependent differences in practice that may exist in research utilizing state or national databases. Taken together with the research based on broader data, our data support the persisting disparities in PMBR surrounding insurance status. Still, generalizability from this study may be limited for other localities given the impact of local policy, such as state or local funding for PMBR that may exist elsewhere.
While state and federal legislation over the past decades has significantly expanded access to PMBR through mandated coverage by private [11] and government insurance [10], past literature as well as this study show that considerable disparities persist [14,18,37]. Lack of health insurance in particular has repeatedly been shown to be a major contributing factor [14,18,36]. Coverage of the procedure increases access and leads to substantially higher PMBR rates. This can be seen following the aforementioned US legislation [13] and perhaps most clearly in countries with nationalized health insurance such as Korea, where there was a dramatic increase in PMBR rates (from 4 to 52%) following the institutional coverage of the procedure [38].
Our study has several limitations. First, medical records did not consistently document patients' preference. In a large sample of records, the reason for not undergoing PMBR was present in less than 20%, and so no conclusions may be drawn from these data. The most common reasons cited in these records were advanced disease and lack of insurance. Not surprisingly, the latter was more frequent in the safetynet hospital which serves most of the uninsured patients in the area. Unfortunately, it is not clear in most records of patients that did not undergo reconstruction if it was actually offered or not. Still, given historical PMBR rates and disparities as well as the drastically different rates between the insurance statuses, it is unlikely that this would significantly confound the analysis. Second, there is a potential for bias with the missing data (< 2%, Fig. 1); however, the impact on the primary effect size was minimal with sensitivity analysis. Loss to follow-up may have occurred if patients underwent reconstruction at an outside hospital, died before a planned reconstruction, or were scheduled for delayed reconstruction after the time of data collection, but based on review of the records is presumed to be minimal. Another possible selection bias is the fact that stage IV patients were included in the cohort (4.3% of the cohort). However, it is unlikely that this represents an important source of selection bias given the lack of meaningful change in primary effect size (< 15%) on sensitivity analysis. Finally, the analysis lacks immigration status, which could be an important confounding variable but cannot be determined from the records.
Future directions should include state by state variations in PMBR rates to assess the effect of Medicaid expansion through the Affordable Care Act. Additionally, data examining the disparities in PMBR faced by undocumented patients would be useful, particularly in the setting of local interventions to allow access to PMBR for this population. Finally, Community Based Participatory Research involving uninsured patients that lack access to PMBR would be helpful to assess their perception of the problem and what priorities they would like to see targeted through policy and/ or interventions.

Conclusion
In a large public and private hospital system, served by the same surgeons, patients without health insurance have dramatically reduced access to PMBR compared to those with private insurance. Further expanding access to this important procedure is essential to achieve greater PMBR equity for breast cancer patients.