Decomposing Racial Disparities in Severe Maternal Morbidity within Insurance Groups

Abstract

Rates of severe maternal morbidity (SMM) are related to maternal, hospital, and residential factors, but the contribution of these factors to racial disparities in SMM within Medicaid and private insured groups is largely unknown. Linked Georgia vital records/hospital discharge data for 2016–2020 are used to identify SMM during delivery or within 42 days postpartum for Medicaid and private insured. The Oaxaca-Blinder decomposition is used to describe the percentage of the Black-White SMM gap explained, based on linear probability models without and with hospital fixed-effects. While the rate of SMM is higher for Medicaid than private insured, the Black-White SMM gap is lower within Medicaid than private insured (1.15 vs. 1.40 per 100 deliveries). Including hospital fixed-effects increased the explained gap by 29.1 percentage points (from 13.8% to 42.9%) within Medicaid and by 9.4 percentage points (from 20.0% to 29.4%) within private insured. Residential factors significantly reduced the Black-White gap explained for Medicaid (−19.6%) but were insignificant (~0%) for privately insured. According to the Oaxaca-Blinder algebraic calculation, differences in within-hospital processes by race contribute a large portion of the discriminatory Black-White SMM gap among Georgia deliveries while residential areas with greater provider access tends to reduce the gap among Medicaid insured.

1. Introduction

Severe maternal morbidity (SMM) refers to unexpected complications during labor and delivery that can have serious short- or long-term health consequences (Kilpatrick et al., 2016). SMM is regarded as a ‘near-miss’ outcome, proximate to maternal mortality but nearly 100 times more common (Callaghan et al., 2012). Recognizing and addressing clinical and social drivers of SMM is essential for improving maternal health (Carmichael et al., 2021). Rates of SMM are higher among racial and ethnic minorities, those insured by Medicaid or uninsured, and those residing in low-income communities (Fingar et al., 2018; HCUP, 2022). SMM is costly (CDC, 2022), with per-patient prenatal, delivery, and postpartum care costs found to be 111% higher than those without a SMM among private insured and 175% higher among Medicaid insured (Black et al., 2021). In 2019, excess costs related to SMM inclusive of hospitalization, productivity losses, and social services for mother and infant were estimated at $32.3 billion (O’Neil et al., 2021). To the extent SMM costs are avoidable, both Medicaid and private insurance plans would benefit as over 95% of all US births are paid by these two forms of insurance (CDC, 2023).

Sociodemographic and medical factors account for a portion of racial/ethnic disparities in SMM but much remains unexplained (Callaghan et al., 2008; Geller et al., 2004; Selo-Ojeme et al., 2005). A growing literature cites delivery hospital as contributing to racial disparities in SMM and points to the need to understand reasons for lower rates in ‘high- performing’ hospitals (Agrawal, 2014). Between-hospital differences are estimated to contribute as much as 47.7% to New York City’s Black-White racial gap in SMM, while disparities within the same hospital and insurance stratum (private, Medicaid) are also observed (Howell et al., 2016; Brown et al., 2021). The concentration of Medicaid service provision among providers (Zuckerman et al., 2021; Ranji et al., 2022; Ludomirsky et al., 2022) and ‘sorting’ to hospitals with low operating costs and service offerings (Dranove et al., 1993) resulted in delivery at hospitals with worse outcomes among Medicaid insured (Phibbs et al., 1993).

We use population-based administrative data and the Oaxaca-Blinder algebraic decomposition method to analyze the non-Hispanic Black-White gap in SMM during the delivery hospitalization through 42 days postpartum within Medicaid and within privately insured deliveries in Georgia, a state with a 30% higher rate of SMM compared to the overall national rate (189.2 vs 144 per 10,000 deliveries) (Kramer et al., 2023). In earlier analysis using this approach for Georgia deliveries inclusive of all payers and the uninsured (Adams, et al, 2023) we found the largest proportion of Georgia’s Black-White gap in SMM was related to unmeasured factors within the same delivery hospital. Given the importance of payer in influencing access to care and the comparatively large percentage of Medicaid paid deliveries to Black mothers, we focus here on factors affecting the Black-White SMM gap separately within the Medicaid and private insured groups.

2. Materials and Methods

2.1. Data.

Using deterministically-linked Georgia hospital discharge, birth and fetal death records for 2016 through 2020, we identified 389,828 deliveries to non-Hispanic Black or White individuals with either Medicaid (203,868; 52.3%) or private insurance (185,960; 47.7%) at delivery. Of the deliveries covered by Medicaid, 60.6% were to non-Hispanic Blacks and 39.4% to non-Hispanic Whites. Of the privately insured deliveries, 26.8% were to non-Hispanic Blacks and 73.2% to non-Hispanic Whites (see Table 1, Appendix Table 1).

Table 1:

Characteristics of Deliveries to Non-Hispanic Black and White Individuals by Insurance Group (Medicaid and Private), Georgia, 2016 – 2020

DELIVERY CHARACTERISTICSa	MEDICAID N = 203,868		PRIVATE N = 185,960
	Non-Hispanic Black N = 123,443	Non-Hispanic White N = 80,425	Non-Hispanic Black N = 49,796	Non-Hispanic White N = 136,164
Severe Maternal Morbidity (per 100 deliveries)
Delivery Hospitalization to 42 Days	3.18	2.03***	2.91	1.51***
Delivery Hospitalization	2.58	1.64***	2.32	1.26***
Delivery Discharge to 42 Days	0.69	0.43***	0.69	0.27***
Maternal Sociodemographic Factors
Age (years):      < 20	9.8%	10.3%***	2.2%	1.1%***
20–29	63.1%	64.4%	41.9%	40.5%
30–34	17.7%	17.2%	30.2%	37.8%
35–39	7.8%	6.7%	20.1%	17.4%
40+	1.7%	1.3%	5.6%	3.3%
Education:      < High School	15.7%	20.5%***	2.1%	1.3%***
High School Graduate	46.5%	43.9%	19.4%	13.0%
Some College	37.8%	35.5%	78.5%	85.7%
USA Born	94.5%	96.8%***	86.5%	94.9%***
Married	15.1%	40.2%***	52.6%	89.9%***
Medical Factors b
Diabetes mellitus	2.0%	1.5%***	2.1%	1.1%***
Gestational diabetes	6.6%	7.7%***	8.9%	7.8%***
Chronic hypertension	7.6%	4.1%***	8.4%	4.0%***
Gestational hypertension	15.4%	13.3%***	15.0%	13.5%***
Smoker	11.0%	28.8%***	4.3%	6.5%***
Obstetrical Factors
Prior Births      None	35.8%	37.0%***	43.0%	44.4%***
One	26.2%	30.7%	31.2%	35.6%
Two	18.4%	18.8%	16.0%	14.3%
Three or more	19.6%	13.6%	9.8%	5.7%
Singleton	99.0%	99.3%***	99.0%	99.2%***
Access to Care
No or Third Trimester Prenatal Care	10.8%	7.4%	5.2%	2.5%***
Hospital Characteristics – Time-Invariant
Ownership For-Profit	12.3%	13.6%***	6.8%	6.9%***
Government Hospital	4.6%	8.3%	1.6%	3.1%
Non-Profit Hospital	79.4%	72.9%	87.9%	83.7%
Religious Affiliation	3.6%	5.2%	3.8%	6.3%
Teaching Status None	25.9%	45.9%***	17.8%	28.7%***
Minor Teaching	49.1%	45.6%	62.0%	60.1%
Major Teaching	25.0%	8.5%	20.2%	11.3%
Perinatal Level 0 or 1	5.5%	14.9%***	2.3%	4.9%***
2	26.6%	37.6%	16.3%	24.5%
3 or 4	68.0%	47.6%	81.3%	70.6%
Hospital Characteristics – Time-Varying
Delivery Volume Highest Quartile	65.8%	46.3%***	81.1%	72.4%***
2nd Quartile	25.2%	35.0%	15.8%	22.5%
3rd Quartile	8.0%	16.9%	2.7%	4.7%
Lowest Quartile	1.1%	1.8%	0.4%	0.4%
Deliveries per Full-time Equivalent Registered Nurse, mean	4.06	3.96***	4.43	4.19***
Patient Safety Indicator Score, mean	1.06	1.00***	1.10	1.07***
Doctor Communication Score, mean	78.62	79.27***	79.21	79.53***
Residential Characteristics c
Residential Area; Large Central Metro	12.9%	1.5%***	14.2%	10.1%***
Large Fringe Metro	41.4%	36.5%	57.8%	49.7%
Medium Metro	13.5%	10.1%	10.2%	11.7%
Small Metro	15.9%	19.3%	10.9%	14.4%
Micropolitan	9.9%	18.6%	4.4%	8.7%
Noncore	6.5%	14.0%	2.5%	5.3%
Neighborhood Deprivation Index (NDI)	0.43	−0.17***	−0.19	−0.70***
Uninsured women of reproductive age (WRA)	16.6%	14.7%***	14.6%	11.3%***
Obstetrician/Gynecologists per 10,000 women of reproductive age (Mean)	6.76	4.19***	6.56	5.67***
Federally Qualified Health Centers (FQHC), mean	5.27	1.75***	5.74	4.12***
Medically Underserved Area (Yes)	15.1%	27.9%***	8.1%	12.0%***
Community Mental Health Centers, mean	0.36	0.12***	0.50	0.36***
Year of Delivery 2016	20.3%	21.7%***	18.8%	20.5%***
2017	20.2%	20.5%	19.9%	20.3%
2018	20.2%	20.0%	20.2%	20.1%
2019	20.0%	19.5%	20.5%	19.9%
2020	19.4%	18.3%	20.7%	19.1%

^*P<.05

^**P<.01

^***P<.001 (Significant differences between Non-Hispanic Black and White estimated with Pearson Chi-square for categorical variables and Student t-tests for continuous measures)

^a.Data Sources Hospital discharge: SMM; Vital records: age, education, USA born, married, prior births, singleton birth, payer at delivery, no or 3^rd trimester PNC, delivery volume, year of delivery; Vital records and hospital discharge: pre-pregnancy diabetes and hypertension, gestational diabetes and hypertension, smoker (see codes below); American Hospital Association (AHA): hospital ownership, teaching status, full-time equivalent registered nurses; Area Health Resource File (AHRF): percent uninsured women of reproductive age (WRA), Ob/Gynecologists per 10,000 WRA, number FQHCs, medically underserved area, number community mental health centers; American Community Survey (ACS): NDI (NDI scaled to distribution in Georgia in 2015 with 1 unit equal to 1 standard deviation increase; higher scores represent more deprivation); Center for Medicare and Medicaid Services (CMS): patient safety and adverse event composite score (Hospital Compare-Agency for Healthcare Research and Quality quality indicator), doctor communication (Hospital Consumer Assessment of Healthcare Providers and Systems Survey – percent responding doctors always communicate well); National Center for Health Statistics (NCHS): Residential area 2013 urban–rural classification scheme for counties (https://www.cdc.gov/nchs/data/series/sr_02/sr02_166.pdf)

^bDefined using vital records (VR) and ICD-10 diagnosis codes from hospital discharge (HD)

Pre-pregnancy diabetes: VR – Pre-pregnancy diabetes (yes) HD – E08, E09, E10, E11, O240, O241, O242, O243, O248

Gestational diabetes: VR – Gestational diabetes (yes) HD – O244, O99810, O99814

Chronic hypertension: VR - Chronic hypertension (yes) HD – I10, O10, O11

Gestational hypertension: VR – Gestational hypertension (yes) or hypertension eclampsia (yes) HD – O13, O14, O15

Smoker: VR – Tobacco use during pregnancy (yes) HD – F172, O9933, Z87891

^cAll residential characteristics are measured at county level, except NDI scale is measured at census tract.

Appendix Table 1.

Percentage of Georgia Deliveries Paid by Medicaid or Private Insurance^a -by Individual Hospital, 2016–2020

Hospital	% Medicaid Deliveries	Private Deliveries		Medicaid Deliveries
		N	% NH Black	N	% NH Black
1	62.5%	75	12.0%	125	28.8%
2	81.0%	1,085	44.8%	4,624	69.4%
3	39.6%	559	15.6%	366	30.1%
4	100.0%	0	n/a	938	0.5%
5	86.6%	59	39.0%	382	57.9%
6	58.5%	144	37.5%	203	66.0%
7	76.3%	367	52.9%	1,184	61.4%
8	91.5%	83	12.0%	892	42.6%
9	37.4%	5,797	35.3%	3,464	68.2%
10	71.6%	744	13.7%	1,879	44.8%
11	39.4%	1,496	21.4%	971	57.9%
12	75.7%	335	18.8%	1,043	43.7%
13	21.5%	7,267	6.7%	1,993	11.2%
14	75.7%	155	20.0%	482	51.9%
15	46.5%	5,669	31.0%	4,932	58.5%
16	85.3%	138	53.6%	801	75.8%
17	86.7%	269	5.6%	1,761	28.8%
18	95.1%	64	4.7%	1,241	5.5%
19	71.5%	299	2.0%	750	4.3%
20	57.9%	1,785	1.7%	2,453	4.0%
21	45.8%	2,628	8.4%	2,221	24.3%
22	62.3%	2,596	32.1%	4,293	61.8%
23	48.4%	1,148	9.5%	1,077	37.0%
24	6.5%	833	19.4%	58	46.6%
25	100.0%	0	n/a	683	1.5%
26	89.5%	31	19.4%	265	61.1%
27	65.6%	2,060	20.7%	3,922	54.8%
28	71.8%	1,201	8.2%	3,061	40.9%
29	75.3%	1,064	20.3%	3,249	45.9%
30	79.4%	880	11.8%	3,388	53.3%
31	89.2%	46	19.6%	380	56.1%
32	68.1%	1,650	15.5%	3,522	47.4%
33	85.3%	218	21.6%	1,265	46.9%
34	81.3%	279	12.2%	1,211	47.8%
35	79.6%	436	22.9%	1,699	60.7%
36	47.0%	3,358	48.8%	2,975	71.4%
37	65.4%	1,237	47.3%	2,340	69.2%
38	64.6%	833	10.9%	1,522	29.8%
39	88.3%	156	10.9%	1,177	63.3%
40	36.8%	3,207	7.5%	1,869	11.8%
41	62.7%	1,169	7.0%	1,963	18.4%
42	83.6%	117	4.3%	596	17.3%
43	64.8%	3,572	43.8%	6,582	61.1%
44	94.3%	315	72.4%	5,208	86.1%
45	81.7%	315	47.0%	1,411	73.5%
46	55.7%	3,736	10.7%	4,696	41.3%
47	61.5%	1,303	36.1%	2,078	55.1%
48	62.1%	2,783	7.8%	4,552	20.8%
49	47.8%	4,762	4.2%	4,363	12.6%
50	67.9%	883	13.9%	1,872	59.8%
51	34.4%	11,937	16.6%	6,248	46.9%
52	59.6%	3,812	46.5%	5,631	78.1%
53	2.4%	11,448	24.8%	281	86.5%
54	57.1%	3,298	24.9%	4,392	61.5%
55	64.5%	1,432	21.5%	2,597	61.0%
56	36.4%	4,213	6.6%	2,411	37.9%
57	44.4%	6,116	25.9%	4,892	70.8%
58	75.3%	568	17.3%	1,729	39.6%
59	52.8%	2,450	16.5%	2,740	43.6%
60	19.6%	33,701	33.7%	8,221	75.4%
61	71.5%	1,421	36.6%	3,557	74.5%
62	46.8%	4,666	20.7%	4,107	54.7%
63	68.5%	1,961	21.7%	4,261	77.6%
64	53.8%	8,796	63.3%	10,234	95.4%
65	69.8%	1,658	46.8%	3,841	92.6%
66	95.9%	311	86.8%	7,191	97.1%
67	56.6%	4,198	16.8%	5,467	69.8%
68	49.1%	5,151	25.6%	4,977	74.0%
69	73.0%	1,018	23.3%	2,747	75.3%
70	73.4%	3,421	46.6%	9,451	94.1%
71	59.6%	2,930	29.1%	4,322	60.4%
72	20.8%	2,248	29.8%	589	67.4%
ALL	52.3%	185,960	26.8%	203,868	60.6%

^aData omit uninsured deliveries which account for ~ 5% of all deliveries in the analytic file.

Hospital-level data from the American Hospital Association (AHA) and the Center for Medicare and Medicaid Services (CMS) Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) were merged using a crosswalk from Georgia’s Department of Public Health and used to derive both time-invariant (e.g., ownership) and time-varying hospital characteristics (e.g. delivery volume, patient safety scores). Data from the Area Resource File (ARF) and the American Community Survey (ACS) were used to categorize maternal residence according to the National Center for Health Statistics (NCHS) urban/rural continuum. FIPS codes were used to link data reflective of access to care within the county (e.g. Ob/Gyns and other providers in county) and area-level social and economic conditions as noted in Table 1. Details on the sources and definitions of all variables are included in Appendix Table 2.

Appendix Table 2.

Data Sources and Definitions of Characteristics used in Analysis of SMM Rates among Non-Hispanic Black and White Individuals in Georgia, 2016–2020

Characteristics	Sources and Definitions
Severe Maternal Morbidity (per 100 deliveries) by Timing	SMM codes downloaded from https://www.cdc.gov/maternal-infant-health/php/severe-maternal-morbidity/icd.html?CDC_AAref_Val=https://www.cdc.gov/reproductivehealth/maternalinfanthealth/smm/severe-morbidity-ICD.htm (January 2022 version)
Source: Georgia Hospital Discharge Abstracts
Delivery Hospitalization to 42 Days	SMM diagnosed during delivery hospitalization or through 42 days postpartum
Delivery Hospitalization	SMM diagnosed during delivery hospitalization
Delivery Hospital Discharge to 42 Days	SMM diagnosed post-delivery hospitalization discharge through 42 days postpartum
Maternal Sociodemographic Factors
Source: Georgia Birth and Fetal Death Records
Age (years)	Mother age in years
Education	Mother education - last grade of formal education completed
USA Born	Mother country of birth = “US”
Married	Mother marital status at delivery
Medical Factors
Source: Georgia Hospital Discharge Abstracts and Georgia Birth and Fetal Death Records (Vital Records)
Diabetes mellitus	Vital records: pre-pregnancy diabetes (yes) Hospital discharge: ICD-10 diagnosis codes E08, E09, E10, E11, O240, O241, O242, O243, O248
Gestational diabetes	Vital Records: gestational diabetes (yes) Hospital discharge: ICD-10 diagnosis codes O244, O99810, O99814
Chronic hypertension	Vital Records: cronic hypertension (yes) Hospital discharge: ICD-10 diagnosis codes I10, O10, O11
Gestational hypertension	Vital Records: gestational hypertension (yes) or hypertension eclampsia (yes) Hospital discharge: ICD-10 diagnosis codes O13, O14, O15
Smoker	Vital Records: tobacco use during pregnancy (yes) Hospital discharge: ICD-10 diagnosis codes F172, O9933, Z87891
Obstetrical Factors
Source: Georgia Birth and Fetal Death Records
Prior Births	Mother birth event order
Singleton	Singleton birth
Access to Care
Source: Georgia Birth and Fetal Death Records
No or Third Trimester Prenatal Care	Trimester prenatal care began
Hospital Characteristics – Time Invariant
Source: American Hospital Association Annual Survey (AHA) and Georgia Department of Public Health
Hospital Ownership	AHA: hospital ownership defined using CNTRL variable Government 12–16, 41–48 Not-for-profit 23 For-profit 31 – 33 Religious 21, 22
Teaching Status	AHA: major teaching are all hospitals that have the Council of Teaching Hospitals designation (MAPP8); minor teaching are all hospitals that have any one or more of the other four MAPP codes identified below: MAPP3 – Participating site recognized for one or more Accreditation Council for Graduate Medical Education accredited programs. MAPP5 – Medical school affiliation reported to the American Medical Association (AMA). MAPP8 – Member of the Council of Teaching Hospitals (COTH) of the Association of American Medical Colleges. MAPP12 – Internship approved by American Osteopathic Association. MAPP13 – Residency approved by American Osteopathic Association
Perinatal Level	Georgia Department of Public Health: Levels of Maternal Care 0–4
Hospital Characteristics – Time Varying
Source: Georgia Birth and Fetal Death Records (Vital Records), American Hospital Association Annual Survey (AHA) and Center for Medicare and Medicaid Services (CMS)
Delivery Volume	Vital Records: Quartiles of delivery volume
Deliveries per Full-time Equivalent Registered Nurse	Vital Records and AHA: number of deliveries divided by AHA FTE registered nurses
Patient Safety Indicator Score	CMS: patient safety and adverse event composite score (Hospital Compare-Agency for Healthcare Research and Quality quality indicator)
Doctor Communication Score	CMS: doctor communication (Hospital Consumer Assessment of Healthcare Providers and Systems Survey – percent responding doctors always communicate well)
Residential Characteristics
Source: National Center for Health Statistics (NCHS), American Community Survey (ACS), Area Health Resources Files from the Health Resources and Services Administration (HRSA)
Residential Area	NCHS: residential area 2013 urban–rural classification scheme for counties (https://www.cdc.gov/nchs/data/series/sr_02/sr02_166.pdf )
Neighborhood Deprivation Index (NDI)	ACS: Census tract level measure. NDI scaled to distribution in Georgia in 2015 with 1 unit equal to 1 standard deviation increase; higher scores represent more deprivation.
Uninsured women of reproductive age (WRA)	AHRF: percent uninsured women of reproductive age in county
Obstetrician/Gynecologists per 10,000 women of reproductive age	AHRF: Ob/Gynecologists per 10,000 women of reproductive age in county
Federally Qualified Health Centers (FQHC)	AHRF: number of Federally Qualified Health Centers in county
Medically Underserved Area (Yes)	AHRF: county is medically underserved area
Community Mental Health Centers	AHRF: number of Community Mental Health Centers in county
Payer at Delivery and Year of Delivery	Payer at delivery: Private = BCBS, Tricare, HMO, commercial, PPO, POS Medicaid = Medicaid managed care, Medicaid, Georgia Better Health, Peachcare
Source: Georgia Birth and Fetal Death Records

The covariates in the models include: 1) sociodemographic factors (age, level of education, marital status, nativity); 2) medical conditions (diabetes mellitus, gestational diabetes, chronic hypertension, gestational hypertension or preeclampsia, smoker); 3) obstetrical factors (singleton, prior births) 4) access to care (no or third trimester prenatal care); 5) hospital characteristics that are time-invariant (ownership, profit status, religious affiliation, teaching status, perinatal level of care) and time-varying (deliveries per full-time equivalent registered nurse, doctor communication, patient safety and adverse event composite score), and 6) residential characteristics (urban/rural classification, percent uninsured women of reproductive age (WRA), obstetrician-gynecologists per WRA, number of federally-qualified health centers (FQHCs) and community health centers (CMHCs) in county of maternal residence, medically-underserved area (yes/no), and a validated neighborhood deprivation index (NDI) used in the maternal and child health area at the census tract level (Messer et al., 2006; Glassman et al., 2020).

For each delivery, hospital discharge data were used to identify occurrences of any of the 21 SMM indicator conditions during the delivery hospitalization or any subsequent hospitalization through 42 days postpartum using established ICD codes (CDC, 2022).

2.2. Statistical Analysis.

To analyze maternal and hospital characteristics as determinants of SMM, we first estimated two basic linear probability models as follows:

$$y_{iht}={\beta }_0+{\beta }_1{Maternal}_{it}+{\beta }_2{Hospital}_{iht}+{\beta }_3{CtyACCESS}_{it}{+{\beta }_{4}CenSES}_{it}+{\beta }_{5}t+{\epsilon }_{iht}$$ (1)

$$y_{iht}={\beta }_0+{\beta }_1{Maternal}_{it}+{\beta }_2{Hospital}_{iht}+{\beta }_3{CtyACCESS}_{it}{+{\beta }_{4}CenSES}_{it}+{\beta }_{5}t+{\beta }_6{\delta }_h+{\epsilon }_{iht}$$ (2)

Model 1: $y_{iht}={\beta }_0+{\mathit{M}\mathit{a}\mathit{t}\mathit{e}\mathit{r}\mathit{n}\mathit{a}\mathit{l}}_{\mathit{i}\mathit{t}}{\mathit{\beta }}_1+{\mathit{H}\mathit{o}\mathit{s}\mathit{p}\mathit{i}\mathit{t}\mathit{a}\mathit{l}}_{\mathit{i}\mathit{h}\mathit{t}}{\mathit{\beta }}_2+{\mathit{C}\mathit{t}\mathit{y}\mathit{A}\mathit{C}\mathit{C}\mathit{E}\mathit{S}\mathit{S}}_{\mathit{i}\mathit{t}}{\mathit{\beta }}_3{+\mathit{C}\mathit{e}\mathit{n}\mathit{S}\mathit{E}\mathit{S}}_{\mathit{i}\mathit{t}}{\mathit{\beta }}_4+{\mathit{\lambda }}_{\mathit{t}}+{\epsilon }_{iht}$

Model 2: $y_{iht}={\beta }_0+{\mathit{M}\mathit{a}\mathit{t}\mathit{e}\mathit{r}\mathit{n}\mathit{a}\mathit{l}}_{\mathit{i}\mathit{t}}{\mathit{\gamma }}_1+{\mathit{H}\mathit{o}\mathit{s}\mathit{p}\mathit{i}\mathit{t}\mathit{a}\mathit{l}}_{\mathit{i}\mathit{h}\mathit{t}}{\mathit{\gamma }}_2+{\mathit{C}\mathit{t}\mathit{y}\mathit{A}\mathit{C}\mathit{C}\mathit{E}\mathit{S}\mathit{S}}_{\mathit{i}\mathit{t}}{\mathit{\gamma }}_3{+\mathit{C}\mathit{e}\mathit{n}\mathit{S}\mathit{E}\mathit{S}}_{\mathit{i}\mathit{t}}{\mathit{\gamma }}_4+{\mathit{\lambda }}_{\mathit{t}}+{\mathit{\delta }}_{\mathit{h}}+{\epsilon }_{iht}$

In these equations, y represents the SMM (0/1) observation for the i^th individual at hospital (h) in year (t). The vector Maternal includes variables related to maternal sociodemographic, medical, obstetrical characteristics, and access to care. The Hospital vector includes time-varying (Models 1 and 2) and time-invariant (Model 1 only) delivery hospital characteristics. CtyACCESS and CenSES represent residential factors at the county (Cty) and census tract (Cen) levels that may affect access. Models include a year fixed-effect (${\mathit{\lambda }}_{\mathit{t}}$) while Model 2 includes both a year and hospital fixed-effect. The hospital fixed-effects (δ_h) represent unmeasured time invariant characteristics of hospitals (which may include hospital cultural characteristics or care processes). In the fixed-effects models, all time-invariant hospital characteristics from Model 1 are necessarily excluded. Within each insurance group (Medicaid and privately insured), all models were estimated separately for deliveries to non-Hispanic Black and non-Hispanic White individuals. Models were estimated using linear probability models with robust standard errors.

Assuming a discriminatory effect for deliveries to non-Hispanic Black individuals, we used the Oaxaca-Blinder algebraic decomposition technique (Oaxaca, 1973; Blinder, 1973), similar to other studies (Sen, 2014; Thoma et al., 2019; Zuvekas, 2003; Lhila & Long, 2012) examining Black-White gaps in birth outcomes, to derive: 1) an “explained portion” attributed to differences in the averages of the covariates for non-Hispanic Blacks and Whites (shown by the first term on the right hand side of equation 3); and 2) an “unexplained portion” attributed to differences in the coefficient estimates for non-Hispanic Blacks and Whites (as shown by the second term on the right hand side of equation 3):

$${\overline{Y}}_B‐{\overline{Y}}_W=\left({\overline{X}}_B‐{\overline{X}}_W\right)\mathrm{\text{'}}\widehat{{\beta }_W}+{\overline{X}}_B\left(\widehat{{\beta }_B}‐\widehat{{\beta }_W}\right)$$ (3)

Based on this algebra, the overall gap reflects the combination of both differences in the average characteristics of the racial groups (i.e., compositional differences in the explanatory variables) along with differences in the ‘returns’ (i.e., the coefficients) of those characteristics on the SMM rate.

3. Results

3.1. Descriptive.

Within both insurance groups, we observed a Black-White gap in SMM (Table 1). Within Medicaid, the SMM rate for delivery through 42 days postpartum was 3.18 per 100 deliveries for Black individuals and 2.03 for White individuals (gap = 1.15 pp; p<0.01). Within the privately insured, the SMM rate for delivery through 42 days postpartum was 2.91 per 100 deliveries for Black and 1.51 for White individuals (gap = 1.40 pp; p<0.01).

Within Medicaid insured deliveries in Georgia from 2016–2020, non-Hispanic Black individuals tend to be older (ages 35–39), more likely to be a high school graduate (46.5% vs 43.9%) and to have chronic hypertension (7.6% vs. 4.1%) or gestational hypertension (15.4% vs. 13.3%), far less likely to be married (15.1% vs 40.2%) and to smoke (11% vs. 28.8%) than their White counterparts (Table 1). Other differences include a higher percentage with three or more births and no or third trimester prenatal care (10.8% vs 7.4%). Within the privately insured, non-Hispanic Black individuals were also more likely to be in the 35–39 and 40+ age groups, more likely to be a high school graduate (19.4% vs 13.0%) and far less likely to be married (52.6% vs 89.9%). As within Medicaid insured, non-Hispanic Black individuals within the private insured were more likely to have chronic hypertension (8.4% vs. 4.0%) or gestational hypertension (15.0% vs. 13.5%) as well as pre-pregnancy or gestational diabetes. While the privately insured non-Hispanic Black mothers are also more likely to have no or third trimester prenatal care (5.2% vs 2.5%) these rates are lower than those found for Medicaid insured (10.8% vs. 7.4%).

With respect to delivery hospital, both Medicaid and privately insured, non-Hispanic Black individuals were more likely to deliver at non-profit, major teaching hospitals and those with larger delivery volumes, characteristics of ‘safety-net’ hospitals (Howell et al., 2014). Within both insured groups, a higher percentage of non-Hispanic Black individuals delivered at a regional perinatal center and at hospitals with more deliveries per full-time equivalent registered nurse and higher (worse) scores on the patient safety and adverse event composite score (AHRQ, n.d.). Within both insured groups non-Hispanic Black individuals more likely reside in areas with a higher Neighborhood Deprivation Index (NDI) and with more uninsured women of reproductive age (WRA), but also have higher rates of Ob/Gyns per WRA and other safety-net providers [e.g. Federally Qualified Health Centers (FQHCs)]. Across the 72 hospitals with deliveries in our study period (Appendix Table 1), the percentage covered by Medicaid was 52.3% with the percentage Medicaid across the individual hospitals ranging from a low of 2.4% to 100%.

3.2. Multivariable.

Linear probability model results for SMM at delivery through 42 days for non-Hispanic Black and White individuals within each insurance group for models, with and without hospital fixed-effects are shown in Appendix Tables 3 and 4. Within a given insurance group, the direction of the association of a given covariate with SMM was largely similar for deliveries to non-Hispanic Black and White individuals.

Appendix Table 3:

Regression coefficients of Severe Maternal Morbidity at Delivery through 42 Days Postpartum for Sociodemographic, Medical, Obstetric, Access, Hospital and Residential Factors for MEDICAID Insured Deliveries to Non-Hispanic Blacks and Whites, Georgia, 2016–2020

Delivery Characteristics	Model 1a		Model 2b Hospital Fixed Effects
	Non-Hispanic Black (N = 123,443)	Non-Hispanic White (N = 80,425)	Non-Hispanic Black (N = 123,443)	Non-Hispanic White (N = 80,425)
	Coefficient (Std Err)	Coefficient (Std Err)	Coefficient (Std Err)	Coefficient (Std Err)
Maternal Sociodemographic Factors
Age (years): < 20	0.0040* (0.0019)	0.0003 (0.0019)	0.0041* (0.0019)	0.0003 (0.0019)
20–29	Ref.	Ref.	Ref.	Ref.
30–34	0.0025 (0.0015)	0.0009 (0.0014)	0.0026 (0.0015)	0.0010 (0.0014)
35–39	0.0054* (0.0022)	0.0063** (0.0024)	0.0058** (0.0022)	0.0064** (0.0024)
40+	0.0140** (0.0050)	0.0142* (0.0059)	0.0146** (0.0050)	0.0144* (0.0059)
Education: < High School	Ref.	Ref.	Ref.	Ref.
High School Graduate	−0.0011 (0.0016)	−0.0044** (0.0015)	0.0000 (0.0016)	−0.0042** (0.0015)
Some College	−0.0053** (0.0017)	−0.0068*** (0.0016)	−0.0037* (0.0017)	−0.0066*** (0.0016)
USA Born	0.0030 (0.0022)	0.0047 (0.0025)	0.0010 (0.0022)	0.0039 (0.0026)
Married	−0.0042** (0.0014)	−0.0023* (0.0011)	−0.0041** (0.0014)	−0.0021* (0.0011)
Medical Factors
Diabetes mellitus	0.0182*** (0.0051)	0.0332*** (0.0070)	0.0183*** (0.0051)	0.0339*** (0.0070)
Gestational diabetes	−0.0014 (0.0023)	−0.0024 (0.0021)	−0.0012 (0.0023)	−0.0021 (0.0021)
Chronic hypertension	0.0301*** (0.0027)	0.0103** (0.0035)	0.0287*** (0.0027)	0.0106** (0.0035)
Gestational hypertension	0.0364*** (0.0019)	0.0250*** (0.0021)	0.0342*** (0.0019)	0.0241*** (0.0021)
Smoker	0.0035* (0.0018)	0.0009 (0.0012)	0.0022 (0.0018)	0.0009 (0.0012)
Obstetrical Factors
Prior Births: None	Ref.	Ref.	Ref.	Ref.
One	−0.0010 (0.0013)	−0.0049*** (0.0012)	−0.0012 (0.0013)	−0.0052*** (0.0012)
Two	0.0017 (0.0015)	−0.0004 (0.0015)	0.0017 (0.0015)	−0.0006 (0.0015)
Three or more	0.0070*** (0.0017)	0.0030 (0.0019)	0.0069*** (0.0017)	0.0025 (0.0019)
Singleton	−0.0620*** (0.0085)	−0.0464*** (0.0105)	−0.0623*** (0.0084)	−0.0471*** (0.0105)
Access to Care
No or Third Trimester PNC	0.0048** (0.0017)	0.0047* (0.0021)	0.0039* (0.0017)	0.0038 (0.0021)
Hospital Characteristics – Time Invariant
Ownership:      For-Profit	Ref.	Ref.	Ref.	Ref.
Government Hospital	0.0093** (0.0029)	0.0067** (0.0024)
Non-Profit Hospital	0.0106*** (0.0016)	0.0056*** (0.0016)
Religious Affiliation	−0.0008 (0.0026)	−0.0001 (0.0022)
Teaching Status:      None	Ref.	Ref.	Ref.	Ref.
Minor Teaching	−0.0036* (0.0014)	−0.0039** (0.0012)
Major Teaching	0.0152*** (0.0021)	0.0014 (0.0025)
Perinatal Level:      0 or 1	Ref.	Ref.	Ref.	Ref.
2	−0.0014 (0.0025)	0.0010 (0.0018)
3 or 4	0.0035 (0.0027)	0.0032 (0.0021)
Hospital Characteristics – Time Varying
Delivery Volume: Highest Quartile	Ref.	Ref.	Ref.	Ref.
2nd Quartile	−0.0025 (0.0016)	−0.0070*** (0.0015)	0.0047 (0.0042)	−0.0032 (0.0044)
3rd Quartile	0.0018 (0.0028)	−0.0002 (0.0024)	0.0303* (0.0126)	0.0108 (0.0108)
Lowest Quartile	0.0208** (0.0066)	0.0089 (0.0053)	0.0205 (0.0190)	0.0115 (0.0154)
Deliveries per RN FTE	0.0010* (0.0005)	0.0016*** (0.0005)	−0.0001 (0.0012)	−0.0001 (0.0016)
Patient Safety Indicator score (PSI)_	−0.0184*** (0.0036)	−0.0000 (0.0039)	−0.0062 (0.0048)	0.0079 (0.0048)
Doctor Communication	0.0008*** (0.0002)	0.0002 (0.0002)	0.0008** (0.0003)	−0.0001 (0.0003)
Residential Characteristics
Residential Area:
Large Central Metro	Ref.	Ref.	Ref.	Ref.
Large Fringe Metro	−0.0059* (0.0026)	−0.0058 (0.0050)	0.0055 (0.0030)	0.0067 (0.0053)
Medium Metro	−0.0137*** (0.0026)	−0.0098 (0.0050)	0.0016 (0.0058)	0.0093 (0.0061)
Small Metro	−0.0115*** (0.0027)	−0.0080 (0.0051)	0.0037 (0.0056)	0.0090 (0.0060)
Micropolitan	−0.0117*** (0.0032)	−0.0047 (0.0052)	0.0082 (0.0056)	0.0116 (0.0060)
Noncore	−0.0110** (0.0034)	−0.0042 (0.0053)	0.0044 (0.0058)	0.0120* (0.0061)
Neighborhood Deprivation Index (NDI)	0.0007 (0.0007)	0.0004 (0.0009)	0.0002 (0.0007)	−0.0003 (0.0009)
Percent Uninsured Women of Reproductive Age (WRA)	−0.0005 (0.0083)	0.0016 (0.0091)	−0.0001 (0.0086)	0.0080 (0.0096)
Obstetrician/Gynecologists per 10,000 WRA	0.0000 (0.0002)	0.0001 (0.0002)	−0.0003 (0.0003)	−0.0003 (0.0002)
FQHC	−0.0008*** (0.0002)	−0.0009** (0.0003)	0.0003 (0.0002)	−0.0001 (0.0003)
Medically Underserved Area (Yes)	−0.0022 (0.0020)	−0.0006 (0.0017)	−0.0015 (0.0022)	−0.0017 (0.0018)
Number Community Mental Health Centers	0.0009 (0.0012)	0.0004 (0.0015)	−0.0025 (0.0014)	0.0013 (0.0017)
Year of Delivery: 2016	Ref.	Ref.	Ref.	Ref.
2017	0.0027 (0.0015)	0.0008 (0.0015)	0.0028 (0.0016)	0.0015 (0.0016)
2018	0.0009 (0.0016)	−0.0007 (0.0016)	0.0011 (0.0016)	−0.0012 (0.0016)
2019	0.0013 (0.0017)	0.0008 (0.0016)	0.0023 (0.0018)	0.0003 (0.0018)
2020	0.0102*** (0.0018)	0.0031 (0.0018)	0.0077*** (0.0020)	0.0008 (0.0021)

^*P<.05

^**P<.01

^***P<.001

^aAdjusted for all characteristics presented.

^bAdjusted for all characteristics presented and hospital fixed effects. Hospital fixed effects are not shown. Time invariant hospital characteristics of hospital ownership, teaching status, and perinatal level excluded from fixed effects models.

Appendix Table 4:

Regression coefficients of Severe Maternal Morbidity at Delivery through 42 Days Postpartum for Sociodemographic, Medical, Obstetric, Access, Hospital and Residential Factors for PRIVATE INSURED Deliveries to Non-Hispanic Black and Whites, Georgia, 2016–2020

Delivery Characteristics	Model 1a		Model 2b Hospital Fixed Effects
	Non-Hispanic Black (N =49,796)	Non-Hispanic White (N =136,164)	Non-Hispanic Black (N =49,796)	Non-Hispanic White (N =136,164)
	Coefficient (Std Err)	Coefficient (Std Err)	Coefficient (Std Err)	Coefficient (Std Err)
Maternal Sociodemographic Factors
Age (years): < 20	0.0013 (0.0052)	−0.0019 (0.0035)	0.0016 (0.0052)	−0.0019 (0.0035)
20–29	Ref.	Ref.	Ref.	Ref.
30–34	0.0026 (0.0018)	0.0018* (0.0008)	0.0026 (0.0018)	0.0015 (0.0008)
35–39	0.0056* (0.0023)	0.0063*** (0.0011)	0.0056* (0.0023)	0.0060*** (0.0011)
40+	0.0207*** (0.0044)	0.0132*** (0.0025)	0.0207*** (0.0044)	0.0128*** (0.0025)
Education: < High School	Ref.	Ref.	Ref.	Ref.
High School Graduate	0.0049 (0.0053)	−0.0027 (0.0034)	0.0036 (0.0053)	−0.0026 (0.0034)
Some College	0.0034 (0.0052)	−0.0047 (0.0033)	0.0020 (0.0052)	−0.0048 (0.0033)
USA Born	−0.0041 (0.0024)	−0.0013 (0.0016)	−0.0042 (0.0024)	−0.0010 (0.0016)
Married	0.0004 (0.0017)	−0.0022 (0.0013)	0.0005 (0.0017)	−0.0021 (0.0013)
Medical Factors
Diabetes mellitus	0.0165* (0.0074)	0.0090* (0.0044)	0.0161* (0.0074)	0.0094* (0.0045)
Gestational diabetes	0.0007 (0.0030)	0.0027 (0.0015)	0.0006 (0.0030)	0.0028 (0.0015)
Chronic hypertension	0.0272*** (0.0039)	0.0096*** (0.0023)	0.0266*** (0.0038)	0.0097*** (0.0023)
Gestational hypertension	0.0297*** (0.0028)	0.0189*** (0.0014)	0.0291*** (0.0028)	0.0188*** (0.0014)
Smoker	0.0106* (0.0044)	0.0058*** (0.0016)	0.0104* (0.0044)	0.0057*** (0.0016)
Obstetrical Factors
Prior Births: None	Ref.	Ref.	Ref.	Ref.
One	−0.0022 (0.0018)	−0.0063*** (0.0008)	−0.0021 (0.0018)	−0.0062*** (0.0008)
Two	−0.0027 (0.0022)	−0.0052*** (0.0010)	−0.0025 (0.0023)	−0.0050*** (0.0010)
Three or more	0.0013 (0.0030)	−0.0031 (0.0016)	0.0014 (0.0030)	−0.0030 (0.0016)
Singleton	−0.0561*** (0.0127)	−0.0270*** (0.0062)	−0.0564*** (0.0127)	−0.0270*** (0.0062)
Access to Care
No or Third Trimester PNC	0.0020 (0.0035)	0.0022 (0.0023)	0.0002 (0.0035)	0.0017 (0.0023)
Hospital Characteristics – Time Invariant
Ownership: For-Profit	Ref.	Ref.	Ref.	Ref.
Government Hospital	0.0130 (0.0076)	0.0031 (0.0025)
Non-Profit Hospital	0.0035 (0.0032)	0.0006 (0.0014)
Religious Affiliation	−0.0098* (0.0048)	−0.0070*** (0.0015)
Teaching Status: None	Ref.	Ref.	Ref.	Ref.
Minor Teaching	−0.0004 (0.0025)	−0.0035*** (0.0009)
Major Teaching	0.0109*** (0.0032)	−0.0005 (0.0015)
Perinatal Level: 0 or 1	Ref.	Ref.	Ref.	Ref.
2	0.0037 (0.0053)	−0.0011 (0.0019)
3 or 4	0.0041 (0.0057)	−0.0013 (0.0021)
Hospital Characteristics – Time Varying
Delivery Volume: Highest Quartile	Ref.	Ref.	Ref.	Ref.
2nd Quartile	−0.0030 (0.0028)	−0.0033*** (0.0010)	0.0049 (0.0061)	−0.0017 (0.0030)
3rd Quartile	−0.0070 (0.0060)	−0.0049* (0.0023)	−0.0200 (0.0139)	−0.0068 (0.0076)
Lowest Quartile	−0.0043 (0.0141)	−0.0030 (0.0065)	0.0070 (0.0349)	−0.0061 (0.0078)
Deliveries per RN FTE	0.0011 (0.0006)	0.0005* (0.0003)	−0.0002 (0.0019)	−0.0002 (0.0008)
Patient Safety Indicator score (PSI)_	0.0066 (0.0057)	0.0058* (0.0024)	0.0049 (0.0074)	0.0065* (0.0031)
Doctor Communication	0.0007* (0.0003)	0.0006*** (0.0001)	−0.0006 (0.0005)	0.0005* (0.0002)
Residential Characteristics
Residential Area:
Large Central Metro	Ref.	Ref.	Ref.	Ref.
Large Fringe Metro	−0.0028 (0.0039)	0.0016 (0.0018)	0.0046 (0.0041)	0.0032 (0.0018)
Medium Metro	−0.0064 (0.0043)	−0.0066*** (0.0018)	−0.0045 (0.0106)	0.0003 (0.0027)
Small Metro	−0.0068 (0.0040)	−0.0018 (0.0019)	0.0022 (0.0090)	0.0027 (0.0025)
Micropolitan	0.0010 (0.0057)	−0.0008 (0.0022)	0.0110 (0.0093)	0.0036 (0.0027)
Noncore	0.0020 (0.0068)	−0.0008 (0.0025)	0.0102 (0.0102)	0.0026 (0.0029)
Neighborhood Deprivation Index (NDI)	0.0018 (0.0012)	0.0004 (0.0007)	0.0015 (0.0013)	0.0005 (0.0007)
Percent Uninsured Women of Reproductive Age (WRA)	−0.0164 (0.0131)	−0.0075 (0.0069)	−0.0134 (0.0135)	−0.0061 (0.0071)
Obstetrician/Gynecologists per 10,000 WRA	0.0005 (0.0004)	0.0001 (0.0001)	0.0007 (0.0004)	0.0001 (0.0001)
FQHC	−0.0006* (0.0002)	−0.0002 (0.0001)	−0.0001 (0.0003)	−0.0002 (0.0002)
Medically Underserved Area (Yes)	−0.0012 (0.0036)	0.0006 (0.0014)	0.0007 (0.0038)	0.0006 (0.0015)
Number Community Mental Health Centers	0.0002 (0.0014)	−0.0001 (0.0007)	−0.0029 (0.0016)	−0.0006 (0.0008)
Year of Delivery: 2016	Ref.	Ref.	Ref.	Ref.
2017	−0.0002 (0.0024)	−0.0016 (0.0010)	0.0000 (0.0024)	−0.0018 (0.0010)
2018	0.0002 (0.0025)	−0.0000 (0.0011)	−0.0003 (0.0026)	−0.0008 (0.0011)
2019	−0.0012 (0.0025)	0.0008 (0.0011)	−0.0031 (0.0028)	−0.0002 (0.0011)
2020	0.0047 (0.0027)	0.0029* (0.0012)	−0.0007 (0.0033)	0.0014 (0.0013)

^*P<.05

^**P<.01

^***P<.001

^aAdjusted for all characteristics presented.

^bAdjusted for all characteristics presented and hospital fixed effects. Hospital fixed effects are not shown. Time invariant hospital characteristics of hospital ownership, teaching status, and perinatal level excluded from fixed effects models.

3.3. Decomposition.

Results of the Oaxaca Blinder algebraic decomposition of the Black-White SMM gap for the two insurance groups are shown in Table 2 and Figures 1 and 2. While there are stark differences in the percentage of the Black-White gap explained by specific sets of factors for Medicaid and private insured, for both insurance groups the inclusion of hospital fixed-effects (Model 2) markedly increased the total percentage of the Black-White SMM gap explained. The total percentage of the Black-White SMM gap explained in Model 1 and Model 2 for Medicaid is 13.8% and 42.9%, respectively, and for private insured is 20% and 29.4%, respectively. For both insurance groups, hospital fixed-effects (Model 2) explain the largest percentage of the discriminatory effect on the Black-White SMM gap (47% within Medicaid, 14.1% within private insured).

Table 2.

Oaxaca Blinder Decomposition of the Black-White Gap in SMM at Delivery to 42 Days Postpartum, Models without and with Hospital Fixed-Effects, by Insurance Group (Medicaid and Private), Georgia, 2016 – 2020

	Medicaid		Private
	Model 1a	Model 2b Hospital Fixed Effects	Model 1a	Model 2b Hospital Fixed Effects
	NH White and Black (N = 203,868)	NH White and Black (N = 203,868)	NH White and Black (N = 185,960)	NH White and Black (N = 185,960)
SMM Delivery to 42 Days postpartum: Gap	1.15	1.15	1.40	1.40
Explained gap ${\left({\stackrel{\mathbf{-}}{\mathit{X}}}_{\mathit{b}}\mathbf{-}{\stackrel{\mathbf{-}}{\mathit{X}}}_{\mathit{w}}\right)}^{\mathbf{\prime }}{\widehat{\mathit{\beta }}}_{\mathit{w}}$	0.16 (13.8%)	0.49 (42.9%)***	0.28 (20.0%)***	0.41 (29.4%)***
Maternal Sociodemographic Factors Age Education USA Born Married	0.03 (2.9%) 1.1% −2.3% −0.9% 5.1%	0.03 (2.7%) 1.1% −2.3% −0.8% 4.6%	0.14 (10.3%)** 2.3% 1.2% 0.8% 6.0%	0.14 (9.7%)** 2.3% 1.3% 0.6% 5.5%
Medical Factors  Diabetes mellitus  Gestational diabetes  Chronic hypertension  Gestational hypertension  Smoker	0.09 (8.1%)*** 1.5% 0.2% 3.1% 4.7% −1.4%	0.09 (8.1%)*** 1.5% 0.2% 3.2% 4.5% −1.4%	0.07 (5.0%)*** 0.6% 0.2% 3.0% 2.1% −0.9%	0.07 (5.1%)*** 0.7% 0.2% 3.0% 2.1% −0.9%
Obstetrical Factors  Birth Order  Singleton Birth	0.05 (4.6%)*** 3.5% 1.1%	0.05 (4.5%)*** 3.4% 1.1%	0.01 (0.8%) 0.5% 0.3%	0.01 (0.8%) 0.5% 0.3%
Access to Care  No or 3rd Trimester PNC	0.02 (1.4%)*	0.01 (1.1%)	0.01 (0.4%)	0.00 (0.3%)
Hospital Characteristics – Time Invariant  Ownership  Teaching Status  Perinatal Level	0.08 (6.6%) 1.1% 0.9% 4.6%	n/a	−0.00 (−0.1%) 1.1% −0.8% −0.4%	n/a
Hospital Characteristics – Time Varying  Delivery Volume  Deliveries per RN FTE  Patient Safety Indicator Score  Doctor Communication	0.07 (5.8%) 5.7% 1.4% 0% −1.2%	−0.02 (−1.7%) −6.2% 0% 4.2% 0.4%	0.05 (3.3%)*** 2.3% 0.9% 1.6% −1.4%	0.03 (2.2%) 1.8% −0.3% 1.7% −1.0%
Residential Characteristics  Residential Area  NDI  Percent Uninsured WRA  Ob/Gyn per 10,000 WRA  FQHC  Medically Underserved Area  Number of Comm. Mental Health Centers	−0.18 (−15.8%)* 3.4% 2.3% 0.3% 3.0% −26.3% 0.7% 0.8%	−0.23 (−19.6%)* −13.8% −1.5% 1.3% −6.1% −4.2% 1.9% 2.7%	−0.00 (−0.1%) 2.5% 1.5% −1.8% 0.5% −2.5% −0.2% −0.1%	−0.04 (−3.0%) −0.5% 1.7% −1.5% 0.4% −2.4% −0.2% −0.6%
Year of Delivery	0.00 (0.3%)	0.00 (0%)	0.01 (0.4%)**	0.00 (0.2%)
Hospital Fixed Effects	n/a	0.55 (47.8%)**	n/a	0.20 (14.1%)**
Unexplained gap	0.99 (86.2%)***	0.66 (57.1%)***	1.12 (80.0%)***	0.99 (70.6%)***

^*P<.05

^**P<.01

^***P<.001

^aAdjusted for all characteristics presented.

^bAdjusted for all characteristics presented and hospital fixed effects. Time invariant hospital characteristics of hospital ownership, teaching status, and perinatal level excluded from fixed effects models.

Figure 1. Percent of Black-White Disparity in Severe Maternal Morbidity Explained by Covariate Groups in a Model Without (Model 1) and with (Model 2) Hospital Fixed Effects, Medicaid Insured Deliveries, Georgia 2016–2020.

Figure 2. Percent of Black-White Disparity in Severe Maternal Morbidity Explained by Covariate Groups in a Model Without (Model 1) and with (Model 2) Hospital Fixed Effects, Private Insured Deliveries, Georgia 2016–2020.

In Table 2 we show the total percentage of the explained Black-White SMM gap that each set of factors (e.g. sociodemographic) algebraically contributes and the percentage that each individual factor (e.g. marital status) within the set contributes. In Model 1, for Medicaid the largest contribution to the discriminatory effect on the Black-White gap is medical factors (8.1%). Time-invariant (6.6%) and time-varying (5.8%) hospital characteristics combined have larger, although statistically insignificant effects, with hospital perinatal level being the largest time-invariant contributor (4.6%) and delivery volume being the largest time-varying contributor (5.7%) within these sets. In Model 1, for private insured, time-varying hospital characteristics had a smaller but significant, discriminatory effect (3.3%) with delivery volume contributing 2.3%, while time-invariant hospital characteristics did not have a significant discriminatory effect; the largest contribution to the discriminatory effect is from maternal sociodemographic factors (10.3%) with the largest individual contribution coming from marital status (6.0%). Sociodemographic factors do not have a significant discriminatory effect within Medicaid.

Within Medicaid, in both Models 1 and 2, there was a large negative (protective) effect of residential characteristics on the discriminatory gap (−15.8% and −19.6%, respectively) largely reflecting the individual protective effect of FQHCs in county of residence on SMM (−26.3% and −4.2%, respectively) within the counties in which non-Hispanic Black, versus White, individuals reside. Within private insured, residential characteristics had a non-significant negative discriminatory effect in both Models 1 and 2 (−0.1% and −3.0%, respectively).

Medical risk factors have a significant discriminatory effect of 8.1% for both models within Medicaid and around 5.0% within the private insured. The largest individual contributor within the medical factors is gestational hypertension for Medicaid (4.7% and 4.5%, respectively) and chronic hypertension for privately insured (3.0% and 3.0%, respectively). Obstetrical factors are significant contributors to the discriminatory effect among Medicaid insured in Models 1 and 2 (4.6% and 4.5%, respectively), but are not significant among the privately insured.

It is notable that the unexplained percentage of the Black-White SMM gap, which is algebraically related to differences in the coefficients for the two racial groups, remains large even in Model 2 (57.1% for Medicaid, 70.6% for private insured). In Appendix Figures 1 and 2 we present the contributions of each set of covariates to the unexplained gap. Here, the effects of obstetrical factors and observable time-varying hospital characteristics are the largest contributors to the gap, which differs from the explained portions where observed medical factors, residential characteristics and hospital fixed-effects are the largest contributors.

Appendix Figure 1. Percent of Black-White Disparity in Severe Maternal Morbidity Unexplained by Covariate Groups in a Model Without (Model 1) and with (Model 2) Hospital Fixed Effects, Medicaid Insured Deliveries, Georgia 2016–2020.

Appendix Figure 2. Percent of Black-White Disparity in Severe Maternal Morbidity Unexplained by Covariate Groups in a Model Without (Model 1) and with (Model 2) Hospital Fixed Effects, Private Insured Deliveries, Georgia 2016–2020.

3.4. Sensitivity Analysis

In sensitivity analyses we: 1) randomly sampled a single delivery among individuals with more than one in the study period to assess whether correlated repeat births to the same person could have biased our estimates; and 2) excluded blood transfusion from the SMM definition as prior research has demonstrated the low construct validity for SMM identified by blood transfusion only (Snowden et al, 2021). Results are shown in Table 3. In models without repeat deliveries, the percentages explained by factors are roughly similar to the main models for both insured groups. In models omitting blood transfusions, the Black-White SMM gap is narrower than in the main models but Model 2 (with hospital fixed-effects) still has a greater explanatory proportion than Model 1 (without hospital fixed-effects). The percentage explained by factors are similar to the results for the main model for the private insured but some differences are seen for the Medicaid.

Table 3.

Oaxaca Blinder Decomposition of the Black-White Gap in SMM at Delivery to 42 Days Postpartum, in Models without and with Hospital Fixed Effects, by Insurance Group (Medicaid and Private), Georgia, 2016–2020 – Sensitivity Analyses Omitting Repeat Deliveries and Omitting Blood Transfusions

	MEDICAID				PRIVATE
	Omitting Repeat Deliveries	Omitting Repeat Deliveries	Omitting Blood Transfusion	Omitting Blood Transfusion	Omitting Repeat Deliveries	Omitting Repeat Deliveries	Omitting Blood Transfusion	Omitting Blood Transfusion
	Model 1 N=174,754	Model 2 N=174,754	Model 1 N=203,868	Model 2 N=203,868	Model 1 N=163,187	Model 2 N=163,187	Model 1 N=185,960	Model 2 N=185,960
SMM Delivery to 42 Days postpartum: Gap	1.16	1.16	0.51	0.51	1.42	1.42	0.80	0.80
Explained gap ${\left({\stackrel{\mathbf{-}}{\mathit{X}}}_{\mathit{b}}\mathbf{-}{\stackrel{\mathbf{-}}{\mathit{X}}}_{\mathit{w}}\right)}^{\mathbf{\prime }}{\widehat{\mathit{\beta }}}_{\mathit{w}}$	0.20 (16.8%)	0.56 (48.1%)***	0.12 (24.4%)*	0.24 (47.5%)**	0.28 (20.4%)***	0.42 (29.1%)***	0.16 (20.3%)***	0.22 (28.0%)***
Sociodemographics	1.6%	1.3%	5.7%	5.6%	9.2%**	8.7%*	8.9%*	8.1%
Medical Factors	10.0%***	9.9%***	11.8%***	11.4%**	4.4%***	4.5%***	6.9%***	6.8%***
Obstetrical Factors	4.2%***	4.2%***	4.7%**	4.6%**	0.6%	0.7%	1.4%*	1.4%*
Access to Care	1.5%*	1.2%*	0.9%	0.9%	0.6%	0.5%	0.5%	0.5%
Hospital Characteristics – Time Invariant	7.2%	n/a	19.1%**	n/a	−0.2%	n/a	1.4%	n/a
Hospital Characteristics – Time Varying	5.0%	1.0%	8.9%	−15.7%	3.3%**	2.4%	2.1%	2.2%
Residential Factors	−12.9%	−17.8%	−25.9%*	−33.9%*	2.1%	−1.8%	−0.7%	−1.9%
Year	0.2%	−0.1%	−0.8%*	−0.8%*	0.3%	0.1%	−0.3%	−0.3%
Hospital Fixed Effects	n/a	48.4%**	n/a	75.6%**	n/a	14.0%**	n/a	11.0%
Unexplained gap	0.96 (83.2%)***	0.60 (51.9%)***	0.39 (75.6%)***	0.27 (52.5%)*	1.14 (79.6%)***	1.00 (70.9%)***	0.64 (79.7%)***	0.58 (72.0%)***

^*P<.05

^**P<.01

^***P<.001

Given disruptions in healthcare in 2020, we completed sensitivity analysis omitting 2020 (Appendix Table 5). The percentages of the gap explained by Model 1 and Model 2 when 2020 is are excluded for both Medicaid (13.6% and 42.9%, respectively) and private (19.3% and 29.0%, respectively) are similar to those when 2020 is included (13.8% and 42.9%, respectively, for Medicaid; and 20.0% and 29.4%, respectively for private). The percentages explained by the subsets of factors and their significance are largely similar as well.

Appendix Table 5.

Oaxaca Blinder Decomposition of the Black-White Gap in SMM at Delivery to 42 Days Postpartum, Models without and with Hospital Fixed Effects, by Insurance Group (Medicaid and Private), Georgia, 2016 – 2019 (Omits 2020 Data)

	Medicaid		Private
	Model 1a	Model 2b Hospital Fixed Effects	Model 1a	Model 2b Hospital Fixed Effects
	NH White and Black (N = 165,248)	NH White and Black (N = 165,248)	NH White and Black (N = 149,601)	NH White and Black (N = 149,601)
SMM Delivery to 42 Days postpartum: Gap	1.05	1.05	1.35	1.35
Explained gap ${\left({\stackrel{\mathbf{-}}{\mathit{X}}}_{\mathit{b}}\mathbf{-}{\stackrel{\mathbf{-}}{\mathit{X}}}_{\mathit{w}}\right)}^{\mathbf{\prime }}{\widehat{\mathit{\beta }}}_{\mathit{w}}$	0.14 (13.6%)	0.45 (42.9%)**	0.26 (19.3%)***	0.39 (29.0%)***
Sociodemographic Factors Age Education USA Born Married	0.03 (2.5%) 1.0% −2.7% −1.7% 5.8%	0.03 (2.8%) 1.1% −2.5% −1.5% 5.7%	0.14 (10.0%)* 2.3% 1.4% 1.3% 5.0%	0.13 (9.7%)* 2.3% 1.6% 1.0% 4.7%
Medical Factors Diabetes mellitus Gestational diabetes Chronic hypertension Gestational hypertension Smoker	0.09 (8.9%)*** 1.2% 0.3% 3.0% 5.3% −0.9%	0.09 (8.9%)*** 1.2% 0.3% 3.0% 5.1% −0.8%	0.06 (4.7%)*** 0.6% 0.2% 2.9% 2.0% −1.0%	0.07 (4.8%)*** 0.6% 0.2% 3.0% 2.0% −1.0%
Obstetrical Factors  Birth Order  Singleton Birth	0.04 (4.0%)*** 3.0% 1.0%	0.04 (3.8%)*** 2.9% 1.0%	0.01 (0.5%) 0.2% 0.3%	0.00 (0.5%) 0.2% 0.3%
Access to Care  No or 3rd Trimester PNC	0.01 (0.7%)	0.01 (0.5%)	0.01 (0.4%)	0.01 (0.4%)
Hospital Characteristics – Time Invariant  Ownership  Teaching Status  Perinatal Level	0.10 (9.1%) 0.5% 2.5% 6.1%	n/a	−0.01 (−0.4%) 1.4% −1.9% 0.2%	n/a
Hospital Characteristics – Time Varying  Delivery Volume  Deliveries per RN FTE  Patient Safety Indicator Score  Doctor Communication	0.06 (5.4%) 6.4% 2.1% −1.2% −1.8%	−0.13 (−12.5%) −21.2% 0.9% 5.9% 2.0%	0.06 (4.1%)*** 2.2% 1.4% 1.6% −1.1%	0.00 (0.1%) −1.5% 1.0% 2.2% −1.6%
Residential Characteristics  Residential Area  NDI  Percent Uninsured WRA  Ob/Gyn per 10,000 WRA  FQHC  Medically Underserved Area  Number of Comm. Mental Health Centers	−0.18 (−17.0%) 4.5% 0.8% 0.5% 2.2% −26.6% 1.2% 0.4%	−0.17 (−16.6%) −10.0% −3.2% 1.3% −7.4% −0.3% 2.5% 0.6%	−0.00 (−0.1%) 1.7% 2.5% −1.2% 0.8% −3.6% −0.2% −0.2%	−0.04 (−2.8%) −1.1% 2.9% −1.0% 1.0% −3.8% −0.3% −0.5%
Year of Delivery	0.00 (0.0%)	0.00 (0.0%)	0.00 (0.1%)	0.00 (0.1%)
Hospital Fixed Effects	n/a	0.59 (56.0%)**	n/a	0.22 (16.1%)**
Unexplained gap	0.91 (86.4%)***	0.60 (57.1%)***	1.09 (80.7%)***	0.96 (71.0%)***

^*P<.05

^**P<.01

^***P<.001

^aAdjusted for all characteristics presented.

^bAdjusted for all characteristics presented and hospital fixed effects. Time invariant hospital characteristics of hospital ownership, teaching status, and perinatal level excluded from fixed effects models.

4. DISCUSSION

The rate of SMM in Georgia for 2016–2020 was higher for Black and White individuals in Medicaid (3.18 and 2.03 per 100 deliveries, respectively) compared to private insured (2.91 and 1.51 per 100 deliveries, respectively), but the Black-White SMM gap y was greater within Medicaid. For both insured groups, the Oaxaca-Blinder algebraic decomposition indicated that the percentage of the discriminatory gap accounted for increased markedly when hospital fixed-effects were included and that hospital fixed-effects represented the largest proportion of the explained Black-White SMM gap (47.9% for Medicaid, 14.1% for private). These findings indicate that unobserved, within-hospital characteristics or processes where Black individuals more often deliver may result in discriminatory effects on SMM and, further, that hospitals at which Medicaid-insured individuals in Georgia more often deliver may be more likely to have unobserved characteristics or processes that discriminate against non-Hispanic Black individuals.

Beyond hospital fixed-effects, the percent contribution of other factors to the discriminatory gap varied substantially for the two insured groups. For Medicaid, residential factors provided a large, significant protective effect whereas they were insignificant among private insured. For private insured, sociodemographic characteristics (especially marital status) contributed significantly to the discriminatory gap but not for Medicaid. Medical factors had a somewhat larger significant discriminatory effect within Medicaid (approximately 8.1%) compared to the private insured (approximately 5.1%) with gestational hypertension having a larger impact within Medicaid vs. private insured (approximately 4.5% vs. 2.1%).

Our results are consistent with national data demonstrating within-hospital racial differences in patient safety indicators and adverse events (Gangopadhyaya, 2021), and findings that between-hospital methods of analyzing disparities in adverse hospital outcomes underestimates them (Creanga et al., 2014). Our analysis provides significant new information on factors underlying racial disparities in SMM in a racially diverse southern state (Georgia) that has an overall SMM rate higher than most other states (Hirai et al., 2022) and 10^th highest among Medicaid insured (Admon et al., 2023).

Our results emphasize the importance of within-hospital processes among Medicaid insured, important to states like Georgia where 60% of Medicaid deliveries are to non-Hispanic Black individuals, and underscore the need for further research with hospital systems to investigate sources of racial differences in intra-hospital processes, which may be related to hospital-level teamwork and collaboration across disciplines, patient-centeredness of care, cultural competence of providers, emphasis upon patient safety, quality improvement, and professional development. A hospital culture lacking cultural competency is susceptible to implicit bias, which can contribute to racial disparities through differential treatment, quality of care, resource allocation, and referral patterns by race (Howell et al., 2017; Saluja & Bryan, 2021; Hailu et al., 2022). Across the nation, many hospitals are investing in ‘equity officers’ to monitor and document differences in care by race (Weissman et al., 2023).

Our findings also add emphasis to the importance of hospitals’ implementation of Alliance for Innovation (AIM) Peripartum Racial and Ethnic Disparities Patient Safety Bundles, which include development and implementation of dashboards to monitor disparities in process and outcome metrics, quality improvement projects around identified disparities, and staff-wide implicit bias training (ACOG, n.d; Howell et al., 2018). Combating implicit bias in healthcare is recognized as foundational for improving reproductive health outcomes and reducing racial disparities (Josiah et al., 2023).

Given the importance of delivery hospital in the occurrence of SMM, the use of maternity quality related measures as well as the current effort of CMS to identify ‘birthing friendly’ hospitals is needed. Unintentional and adverse effects of some policies on hospital revenue and quality of safety-net hospitals, which disproportionately serve minorities, should be monitored (Chatterjee et al., 2022; Gilman et al., 2015; Schmatlz et al., 2011; Werner et al., 2008). Relatedly, given the importance of payer in influencing access to care and hospital of delivery, understanding the Black-White SMM gap calls for better understanding of the factors affecting access to Medicaid physician participation and effects of potential ‘segmentation’ of Medicaid and private insured markets (Corbisiero et al., 2023; Ludomirksy et al., 2022). Nationally, additional requirements and metrics for marketplace plans are being promoted to assure that pregnant and postpartum patients do not face unnecessary barriers to care (Sabini et al, 2022). Georgia monitors the adequacy of its Medicaid risk-based managed care plans, called Care Management Organizations (CMOs), based on distance to providers and appointment wait times specific to obstetrical providers (Georgia Department of Community Health, n.d). Tabulating these plan details and Medicaid metrics by race may give further insights into the Black-White SMM gap.

Data Availability:

The data used in this analysis were made available to the researchers through the Georgia Department of Public Health (DPH) under a Data Use Agreement (DUA) with Emory University which does not allow further sharing of the data.