Regional Implicit Racial Bias and Rates of Breast Reconstruction, Complications, and Cost Among US Patients With Breast Cancer

Key Points

Question

Are differences in breast reconstruction rates for White patients and patients from minoritized racial and ethnic groups associated with measures of regional implicit racial bias?

Findings

In this cohort study of 52 115 patients, no significant association between regional implicit racial bias and disparities in breast reconstruction utilization for White patients and patients from minoritized racial and ethnic groups was found. Nonetheless, certain regions had higher measures of disparities in reconstruction compared with others.

Meaning

Although implicit racial bias by region was not associated with differences in breast reconstruction utilization for patients from minoritized racial and ethnic groups in this study, there were still regional disparities in care that must be mitigated through coordinated institutional and national policies.

This cohort study evaluates whether variations in implicit racial bias by region are associated with the differences in rates of immediate breast reconstruction, complications, and cost for White patients and patients from minoritized racial and ethnic groups.

Abstract

Importance

Racial disparities influencing breast reconstruction have been well-researched; however, the role of implicit racial bias remains unknown. An analysis of the disparities in care for patients with breast cancer may serve as a policy target to increase the access and quality of care for underserved populations.

Objective

To identify whether variations in implicit racial bias by region are associated with the differences in rates of immediate breast reconstruction, complications, and cost for White patients and patients from minoritized racial and ethnic groups.

Design, Setting, and Participants

This cohort study used data from the National Inpatient Sample (NIS) from 2009 to 2019. Adult female patients with a diagnosis of or genetic predisposition for breast cancer receiving immediate breast reconstruction at the time of mastectomy were included. Patients receiving both autologous free flap and implant-based reconstruction were included in this analysis. US Census Bureau data were extracted to compare rates of reconstruction proportionately. The Implicit Association Test (IAT) was used to classify whether implicit bias was associated with the primary outcome variables. Data were analyzed from April to November 2022.

Exposure

IAT score by US Census Bureau geographic region.

Main Outcomes and Measures

Variables of interest included demographic data, rate of reconstruction, complications (reconstruction-specific and systemic), inpatient cost, and IAT score by region. Spearman correlation was used to determine associations between implicit racial bias and the reconstruction utilization rate for White patients and patients from minoritized racial and ethnic groups. Two-sample t tests were used to analyze differences in utilization, complications, and cost between the 2 groups.

Results

A total of 52 115 patients were included in our sample: 38 487 were identified as White (mean [SD] age, 52.0 [0.7] years) and 13 628 were identified as minoritized race and ethnicity (American Indian, Asian, Black, and Hispanic patients and patients with another race or ethnicity; mean [SD] age, 49.7 [10.5] years). Implicit bias was not associated with disparities in breast reconstruction rates, complications, or cost. Nonetheless, the White-to–minoritized race and ethnicity utilization ratio differed among the regions studied. Specifically, the reconstruction ratio for White patients to patients with minoritized race and ethnicity was highest for the East South Central Division, which includes Alabama, Kentucky, Mississippi, and Tennessee (2.17), and lowest for the West South Central Division, which includes Arkansas, Louisiana, Oklahoma, and Texas (0.75).

Conclusions and Relevance

In this cohort study of patients with breast cancer, regional variation of implicit bias was not associated with differences in breast reconstruction utilization, complications, or cost. Regional disparities in utilization among racial and ethnic groups suggest that collaboration from individual institutions and national organizations is needed to develop robust data collection systems. Such systems could provide surgeons with a comparative view of their care. Additionally, collaboration with high-volume breast centers may help patients in low-resource settings receive the desired reconstruction for their breast cancer care, helping improve the utilization rate and quality of care.

Introduction

Breast reconstruction is underutilized following mastectomy,^1,2 with approximately 40% of patients in the United States undergoing this procedure.^3,4 With overwhelming evidence that breast reconstruction improves the quality of life after breast cancer, efforts to identify disparities are essential to improve breast cancer care quality.^5,6 Research has investigated disparities attributed to race, gender, and insurance status, among other factors.⁷ Understanding drivers of disparities, such as biases among physicians, may serve as the basis for more equitable care.

Implicit bias has been studied in the surgical literature,^8,9,10 but there is minimal evidence on the impact of care for patients receiving breast reconstruction. Implicit biases are unintentional attitudes toward an individual or group that influence one’s behavior or clinical decision-making.¹¹ In an analysis of the National Inpatient Sample (NIS) and US Census Bureau data, Cohen-Levy et al¹² investigated whether regional implicit bias variation was associated with regional variation in total knee and hip arthroplasty. The authors found that although implicit bias by region was not associated with differences in arthroplasty among White and Black patients, the racial disparity significantly varied between the census divisions examined in the study.¹² As regional implicit racial biases may influence health care policy, delivery, and outcomes, identifying these effects in breast cancer care may provide an avenue to provide more equitable care through policy.

Evidence suggests that numerous factors influence the use of breast reconstruction, including age, income, geographic location, and public policy.^13,14,15 This study aims to determine whether differences in breast reconstruction rates for White patients and patients from minoritized racial and ethnic groups are associated with measures of implicit racial bias by region. Secondly, this study aims to identify whether differences in complications and cost of care are associated with regional implicit racial bias. To our knowledge, this project is the first of its kind to assess the influence of implicit racial bias in breast reconstruction care. We hypothesize that patients from geographic regions with higher pro-White implicit bias will experience increased racial disparities pertaining to breast reconstruction rates.

Methods

Database

In this cohort study, NIS data from 2009 to 2019 were used to identify utilization rates among White patients and patients from minoritized racial and ethnic groups. The NIS originates from the Healthcare Cost and Utilization Project (HCUP) and is considered one of the largest all-payer inpatient administrative databases. Specifically, the NIS contains data from approximately 35 million hospital stays a year.¹⁶ This study was considered nonregulated by the institutional review board at the University of Michigan, and therefore, the requirement for informed consent was waived. Methods of data collection and analysis used for this study were referred from Cohen-Levy et al¹² and adjusted for the relevance of this study. These methods are found in eFigure 1 in Supplement 1. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement.¹⁷

Patient Selection

The patient sample included adult female patients with a diagnosis of or genetic predisposition for breast cancer. This ensured specificity of patients undergoing mastectomy for the indication of breast cancer. Patients were included in the final cohort if they received a mastectomy on the same day as the reconstruction. Patients who received delayed reconstruction were not included, as the NIS database does not permit longitudinal analysis following the index operation.¹⁸ Patients with missing data or who died before discharge were excluded.

Patients were identified using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) and International Statistical Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes (eTable 1 in Supplement 1). Breast reconstruction–specific and systemic complications were defined using the diagnostic codes in eTable 2 in Supplement 1. Data on race and ethnicity were included in the databases used to conduct this analysis. Specifically, race and ethnicity data in the NIS are reported to HCUP as documented in administrative claims records. These data reflect the race and ethnicity as documented in patients’ medical records. We decided to study this phenomenon more broadly to provide higher-quality evidence because small sample sizes of minoritized racial and ethnic groups are an inherent bias of using large databases. Black and Hispanic patients and patients belonging to another racial group (Asian, American Indian, other) were categorized as having minoritized race and ethnicity. This categorization is based on the definition of racial minority groups used by the National Institute of Minority Health and Health Disparities.¹⁹

Measuring Implicit Bias

Project Implicit provides publicly accessible Implicit Association Test (IAT) data, a tool used to quantify implicit racial bias. The IAT has been used in the literature to study implicit racial bias among physicians.^20,21 Geographical data from the IAT and NIS were linked to evaluate whether implicit bias was associated with regional variations in breast reconstruction utilization for White patients and patients from minoritized racial and ethnic groups.²² Mean weighted average IAT scores (wa-IAT) were calculated by summing the implicit association test score for every state and weighting this score based on the population size in each region. Regions included in the study were based on the categorization from the US Census Bureau (eFigure 2 in Supplement 1). The IAT score can range from −2 to 2, with a more positive number indicating an implicit preference for White individuals and a more negative number indicating an implicit preference for individuals from minoritized racial and ethnic groups.²³

Adjusting for Population Size

Annual population estimates from the US Census Bureau data were collected to determine a per capita utilization rate by race and ethnicity for the various geographical regions studied. These data are publicly available and can be downloaded at no cost. Specifically, the 5-year American Community Survey population estimates for female individuals were collected.²⁴ These population estimates were then divided into White and minoritized racial and ethnic groups. Volume estimates were obtained by using the stratified sampling method provided by HCUP.¹⁶

Statistical Analysis

Breast reconstruction rates for each geographic region were determined by using the NIS to identify the volume of reconstruction for White and minoritized racial and ethnic groups. The estimated volumes were then divided by the population of female residents in that geographic division, as defined by the US Census Bureau, and multiplied by 100 000 to determine the reconstruction rate per 100 000. Additionally, cost of index reconstruction for patients was studied to determine whether implicit bias by region was associated with higher costs of reconstruction for patients from minoritized racial and ethnic groups. The cost-to-charge ratio file for each year was used to estimate cost from charge data.²⁵ This method has been used by various health science researchers to convert the cost (expenses incurred during index operation) to charges (amount billed by hospital for index operation).^26,27 Although out-of-pocket costs were not attainable using these data, the overall cost provides a meaningful measure of expenditures related to treatment for patients included in this study. Therefore, the cost data reported in the study reflects amounts incurred to the primary payer. All costs were converted to 2022 US dollars.

Spearman correlation was used to evaluate associations between implicit bias and utilization rate. Two-sample t tests were used to evaluate differences in utilization rate, complication rate, and cost. The significance level was set at P < .05 for all analyses, and all tests were 2-tailed. All statistical analyses were performed in R Core Team, 2022 release, version 4.2.1 (R Project for Statistical Computing).

Results

Patient Characteristics

Data from 52 115 patients were included: 38 487 were White and 13 628 belonged to minoritized racial and ethnic groups (Table 1). Most were privately insured (39 627 [76%]) and received implant-based reconstruction (37 354 [72%]). White patients were more commonly privately insured (30 245 [79%]) than patients from minoritized racial and ethnic groups (9382 [69%]). There were no differences in comorbidities between White patients and patients from minoritized racial and ethnic groups.

Table 1. Patient Characteristics.

Characteristic	Patients, No. (%)		P value
	Minoritized racial and ethnic group (n = 13 628)	White (n = 38 487)
Region
New England Division	603 (4.4)	3461 (9.0)	<.001
Middle Atlantic Division	3289 (24.1)	8228 (21.4)
East North Central Division	1219 (8.9)	5233 (13.6)
West North Central Division	142 (1.0)	1427 (3.7)
South Atlantic Division	2923 (21.4)	6862 (17.8)
East South Central Division	296 (2.2)	1844 (4.8)
West South Central Division	2048 (15.0)	3904 (10.1)
Mountain Division	379 (2.8)	1827 (4.7)
Pacific Division	2729 (20.0)	5701 (14.8)
Teaching hospital
No	3278 (24.1)	11 232 (29.2)	<.001
Yes	10 350 (75.9)	27 255 (70.8)
Urban vs rural
Rural	65 (<0.1)	700 (1.8)	<.001
Urban	13 563 (100)	37 787 (98.2)
Reconstruction type
Autologous	3431 (25.2)	7217 (18.8)	<.001
Both	973 (7.1)	3140 (8.2)
Implant-based	9224 (67.7)	28 130 (73.1)
Length of stay, d
Mean (SD)	2.56 (1.93)	2.31 (1.65)	<.001
Median (IQR)	2 (1-3)	2 (1-3)
>5	674 (4.9)	1273 (3.3)	<.001
0-1	3917 (28.7)	12 704 (33.0)
2-3	6727 (49.4)	19 496 (50.7)
4-5	2310 (17.0)	5014 (13.0)
Age at reconstruction, y
Mean (SD)	49.7 (10.5)	52.0 (0.7)	<.001
Median (IQR)	49 (42-57)	52 (45-60)
18-34	881 (6.5)	1846 (4.8)	<.001
35-44	3561 (26.1)	7674 (19.9)
45-54	4943 (36.3)	13 642 (35.4)
55-64	3040 (22.3)	10 236 (26.6)
≥65	1203 (8.8)	5089 (13.2)
Insurance plan
Medicaid	2124 (15.6)	1972 (5.1)	<.001
Medicare	1401 (10.3)	5132 (13.3)
Private	9382 (68.8)	30 245 (78.6)
Other	721 (5.3)	1138 (3.0)
Median income quartile
1	3160 (23.2)	4038 (10.5)	<.001
2	2580 (18.9)	6683 (17.4)
3	3339 (24.5)	10 094 (26.2)
4	4549 (33.4)	17 672 (45.9)
Elixhauser Comorbidity
>8	218 (1.6)	557 (1.4)	.28
0	715 (5.2)	2136 (5.5)
1-3	6692 (49.1)	19 003 (49.4)
4-8	6003 (44.0)	16 791 (43.6)

Implicit Bias Scores by Region

The average IAT score by region is presented in Figure 1. The wa-IAT score for all regions decreased over time for all regions. The mean (SD) wa-IAT for all regions included in our study period was 0.31 (0.03). The West North Central Division had the highest wa-IAT score (0.33), and the South Atlantic Division had the lowest wa-IAT score (0.29) for the period examined.

Figure 1. Weighted Average Implicit Association Test Score by US Bureau Division Over Time.

Utilization Rate by Region

The marginal differences between the White and minoritized racial and ethnic groups were analyzed by comparing White-to–minoritized racial and ethnic group utilization ratios of breast reconstruction. The White-to–minoritized racial and ethnic group utilization ratio for breast reconstruction was 1.03 for all divisions. The ratio of utilization was highest for the East South Central Division (2.17) and West North Central Division (1.66). The utilization ratio was lowest for the West South Central Division (0.75). The White-to–minoritized racial and ethnic group utilization ratio for breast reconstruction is displayed in Figure 2. Differences in utilization were statistically significant for all individual divisions except the Middle Atlantic. Spearman rank correlation between wa-IAT score and breast reconstruction utilization rate for White patients to patients from minoritized racial and ethnic groups was 0.1 for the divisions examined (P = .74), indicating the annual utilization rates did not correlate with changes in IAT at the division level.

Figure 2. White and Minoritized Race and Ethnicity Immediate Breast Reconstruction Utilization in Various Regions.

Complication Rate by Region

The difference in complication ratio between White patients and patients from racial and ethnic minority groups was 0.96 (Table 2). Differences in complication ratio by region were found. Specifically, the complication ratio was highest for the East South Central Division (1.73) and lowest for the West South Central Division (0.73). This result was statistically significant for both divisions. However, Spearman rank correlation between the wa-IAT score and breast reconstruction complication rate ratio for all divisions was 0.1 (P = .81), indicating annual complication rates did not correlate with changes in IAT at the division level.

Table 2. Complications of Care for White Patients and Patients From Minoritized Racial and Ethnic Groups in Various Regions^a.

Region	Ratio of complication, White-to–minoritized group	White complication rate per 100 000	Minoritized group complication rate per 100 0000	P value
New England Division	1.1525	3.5345	3.0667	.44
Middle Atlantic Division	0.9156	3.9715	4.3376	.24
East North Central Division	0.9520	2.3179	2.4347	.66
West North Central Division	1.4215	1.3059	0.9187	.24
South Atlantic Division	0.9841	2.6734	2.7165	.85
East South Central Division	1.7304	1.6678	0.9638	.008
West South Central Division	0.7283	2.5486	3.4996	<.001
Mountain Division	1.3918	1.6988	1.2206	.10
Pacific Division	1.2369	2.8688	2.3193	.007
Total	0.9639	2.5785	2.6751	.28

^a Complications include breast reconstruction–specific complications (wound complications, infection, implant complication, or flap complications) and general complications (transfusion, urinary, stroke, deep venous thrombosis or pulmonary embolism, digestive, cardiovascular, respiratory). Specific diagnoses and corresponding codes are included in eTable 2 in Supplement 1.

Cost by Region

The cost of care differed between White patients and patients from minoritized racial and ethnic groups receiving care (Table 3). Overall, patients from minoritized racial and ethnic groups had a significantly higher mean (SD) cost of care compared with White patients ($25 256 [$14 997] vs $24 934 [$14 030]) on unadjusted analysis. This result was specifically observed in the South Atlantic and West North Central Divisions. The cost was higher for White patients in the New England Division. The Spearman rank correlation between the wa-IAT score and mean cost of care for all divisions was −0.1 (P = .81), indicating that the annual mean cost of care did not correlate with changes in IAT at the division level.

Table 3. Cost of Care for White Patients and Patients From Minoritized Racial and Ethnic Groups in Various Regions.

Region	Cost, mean (SD), US $		P value
	White	Minoritized group
New England Division	30 769 (16 035)	29 093 (15 559)	.02
Middle Atlantic Division	19 861 (10 798)	20 028 (11 492)	.47
East North Central Division	25 197 (13 097)	26 012 (14 103)	.07
West North Central Division	21 962 (11 149)	26 506 (24 933)	.03
South Atlantic Division	24 124 (11 539)	25 483 (13 123)	<.001
East South Central Division	21 598 (11 316)	21 368 (10 522)	.73
West South Central Division	24 113 (13 434)	24 342 (14 597)	.56
Mountain Division	25 956 (13 947)	25 717 (13 986)	.76
Pacific Division	31 507 (17 889)	31 108 (18 259)	.34
National	24 934 (14 030)	25 256 (14 997)	.03

Discussion

This study leveraged a national database with publicly available implicit bias data to evaluate the association of implicit racial bias with the care received by patients undergoing breast reconstruction. No significant association between implicit bias and the utilization ratio for breast reconstruction between White patients and patients from minoritized racial and ethnic groups was found. Regionally, disparities in utilization rate, complication rate, and cost were found.

Evidence suggests that although some physicians may exhibit an implicit bias toward White patients, the bias does not commonly affect clinical care.²⁸ For example, Haider et al⁸ surveyed trauma and acute care surgeons to determine the impact of implicit bias on decision-making and found no association with clinical decision-making.⁸ These findings align with our results that implicit racial bias by region did not correlate with differences in breast reconstruction utilization or complication rates between White patients and patients from minoritized racial and ethnic groups. Nonetheless, to the best of our knowledge, there is no data source that health care disparity researchers can use to examine the effect of implicit physician bias on health care delivery and outcomes. In this study, we applied Project Implicit, which provides data on a sample of the general population that includes, but is not specific to, physicians. The collection of data on the implicit bias of physicians may help facilitate additional research to explore the association of implicit bias with surgical outcomes.

Structural issues of the US health care system pertaining to disparities in access to care are recognized. The findings in this current study suggest that policy targeting physicians may be insufficient to combat disparities. Prior research from our group evaluated whether mandated physician-patient communication was associated with reduced racial disparities in immediate breast reconstruction rates, and found a reduction in disparities of immediate breast reconstruction rates only between Hispanic and White patients.¹⁴ These findings suggest that structural issues in the US health care system merit greater scrutiny, particularly as prior research has highlighted the impact of insurance status on the delivery of health care.^29,30 Additional initiatives, such as community outreach and culturally concordant care, can help facilitate a similar effect. Butler et al³¹ conducted a review of the literature on disparities in breast reconstruction after mastectomy and highlighted how culturally competent outreach to minoritized communities might be a strategy to mitigate these observed disparities. Individual institutions providing breast reconstruction care can identify a diverse group of surgeons and patients to conduct community outreach activities and help improve patient education.

Regional variation in reconstructive surgery access, resource use, and outcomes have been well studied.^32,33,34 For example, in a population-based analysis of women who underwent autologous reconstruction in the United States, the authors found significant variation in resource utilization and length of stay despite similar risk factors and complication rates.³⁴ Our study showed regional variation in breast reconstruction rate, cost, and complication rate. Therefore, a regional cause, particularly region-specific policy or regional patient factors, may contribute to the differences in breast reconstruction care for White patients and patients from minoritized racial and ethnic groups. Standardized methods to collect data from surgeons in various regions will provide researchers and policy makers with more robust data to identify structural barriers to care. Quality collaboratives may help establish standardized methods of data collection to gain a granular view of the differences in the care given to patients from different minoritized racial and ethnic groups. Moreover, these data can also help provide surgeons across the United States with a benchmark to help inform them of inequities in their practice, permitting them to conduct the appropriate outreach or practice modifications to provide more equitable care. Collaboration among surgeons around the nation will be a necessary step in combating regional variation.

Guidelines created by national organizations can serve as a tool to provide more standardized, evidence-based care and reduce variation.^35,36 In our study, we found that patients from minoritized racial and ethnic groups incurred a higher cost of care than White patients. Efforts to standardize care can serve as an opportunity to reduce unnecessary costs. In a systematic review of the effect of guidelines on the regional variation of surgery, the authors found that 3 of 5 studies examining the effect of guidelines on regional variation reported a reduction in the variation after the development of these guidelines.³⁷ Furthermore, Ginzberg et al³⁶ conducted an analysis of the National Cancer Database to identify trends in thyroid cancer care before and after release of the 2015 American Thyroid Association guidelines. The authors found that patients from minoritized racial and ethnic groups were less likely than White patients to receive the appropriate care, but this disparity improved after the guidelines were set in 2015.³⁶ National surgery organizations may develop similar robust guidelines using high-level evidence for the management of breast reconstruction complications, which can help reduce variation in care and consequently reduce costs. Given enhanced recovery protocols have been adopted for breast reconstruction, validated enhanced recovery protocols can help mitigate some disparities in reconstructive surgery. In an analysis of the National Surgical Quality Improvement Project, researchers examined outcomes for Black and White patients with or without an enhanced recovery protocol and found that patients who received enhanced recovery after colorectal surgery had reduced length of stay without any increased risk for readmission, mortality, or common postoperative complications.³⁸ Nonetheless, additional evidence is needed to identify the effects of guidelines prior to their implementation as a strategy to combat racial disparities given that some guidelines may have the potential to exacerbate racial disparities. Given some guidelines can worsen disparities, it is essential that standardized methods for evaluating guidelines are established prior to their implementation.

Patients are diverse, each with a specific set of values that may influence the choice of their reconstruction.³⁹ In a systematic review of barriers to breast reconstruction, Retrouvey et al⁴⁰ identified barriers based on various institution-, clinician-, and patient-level factors. The authors proposed partnerships with large breast reconstruction centers for patients in certain regions with reduced access to breast reconstruction. Specifically, they suggested that small breast cancer centers could collaborate with high-volume breast reconstruction centers to send their more complex patients or patients interested in specific options not offered by smaller centers. This can serve as an important tool to combat observed disparities in the reconstruction rate for patients from minoritized racial and ethnic groups, as higher-volume breast centers may be better equipped to provide culturally competent care, given the volume and diversity of their population. Additionally, the high-volume center may offer more reconstructive options for patients with breast cancer, helping maximize their care.

Limitations

This study has limitations. Our findings are dependent on the accuracy of the data coded in the database; however, HCUP is one of the largest longitudinal hospital databases in the United States and is commonly used in the surgical literature to study health care outcomes.^41,42,43 Additionally, the implicit bias data used in this study do not provide any information on potential confounding variables, ie, education, which may impact measures of implicit bias. Although we categorized patients as White or minoritized race and ethnicity, a more detailed categorization of race may help provide more specific evidence regarding racial disparities in surgical care to guide policy. Future researchers can focus on using prospective data to conduct analyses comparing the specific racial and ethnic groups. Furthermore, we only included patients who underwent immediate reconstruction given the difficulty of studying delayed breast reconstruction using an administrative database. Further analyses can study implicit bias using a prospective design to better capture the outcomes among this patient population. An additional limitation is that insurance type influences cost and access to health care. This may suggest racial disparities in care, reflecting a limitation to our study. Additionally, of note, the cost data in our study represent the costs that hospitals incur by providing care to the patients during their inpatient stay and do not represent actual cost to the patient.

Conclusions

In this cohort study of patients with breast cancer, we found that regional variation in implicit bias was not associated with differences in breast reconstruction utilization, complications, or cost. Nonetheless, regional variation in the utilization, complication rate, and cost of breast reconstruction was observed. Disparities in women’s health care exist and are well described. Despite evidence that some of these disparities are decreasing, efforts to reduce the observed inequities should remain a national priority.⁴⁴ Efforts from individual institutions and national surgical organizations are needed to provide culturally competent, evidence-based care to individuals of all racial and ethnic backgrounds. Qualitative research is an important next step in identifying all the barriers contributing to the observed variation, as it will help guide policy focused on tackling both patient and system-based factors contributing to variation in breast reconstruction care.

Supplement 1.

eFigure 1. Overview of Study Methodology

eTable 1. Codes Used to Identify Patient Sample

eTable 2. Codes Used to Identify Complications

eFigure 2. United States Census Bureau Divisions

Supplement 2.

Data Sharing Statement